Cloudera Administration

Cloudera Administration
Cloudera Administration
Important Notice
© 2010-2017 Cloudera, Inc. All rights reserved.
Cloudera, the Cloudera logo, and any other product or service names or slogans contained
in this document are trademarks of Cloudera and its suppliers or licensors, and may not
be copied, imitated or used, in whole or in part, without the prior written permission
of Cloudera or the applicable trademark holder.
Hadoop and the Hadoop elephant logo are trademarks of the Apache Software
Foundation. All other trademarks, registered trademarks, product names and company
names or logos mentioned in this document are the property of their respective owners.
Reference to any products, services, processes or other information, by trade name,
trademark, manufacturer, supplier or otherwise does not constitute or imply
endorsement, sponsorship or recommendation thereof by us.
Complying with all applicable copyright laws is the responsibility of the user. Without
limiting the rights under copyright, no part of this document may be reproduced, stored
in or introduced into a retrieval system, or transmitted in any form or by any means
(electronic, mechanical, photocopying, recording, or otherwise), or for any purpose,
without the express written permission of Cloudera.
Cloudera may have patents, patent applications, trademarks, copyrights, or other
intellectual property rights covering subject matter in this document. Except as expressly
provided in any written license agreement from Cloudera, the furnishing of this document
does not give you any license to these patents, trademarks copyrights, or other
intellectual property. For information about patents covering Cloudera products, see
http://tiny.cloudera.com/patents.
The information in this document is subject to change without notice. Cloudera shall
not be liable for any damages resulting from technical errors or omissions which may
be present in this document, or from use of this document.
Cloudera, Inc.
1001 Page Mill Road, Bldg 3
Palo Alto, CA 94304
info@cloudera.com
US: 1-888-789-1488
Intl: 1-650-362-0488
www.cloudera.com
Release Information
Version: Cloudera Enterprise 5.5.x
Date: June 16, 2017
Table of Contents
About Cloudera Administration................................................................................7
Managing CDH and Managed Services.....................................................................8
Managing CDH and Managed Services Using Cloudera Manager........................................................................8
Configuration Overview.........................................................................................................................................................8
Managing Clusters...............................................................................................................................................................32
Managing Services...............................................................................................................................................................36
Managing Roles...................................................................................................................................................................44
Managing Hosts...................................................................................................................................................................49
Maintenance Mode..............................................................................................................................................................61
Managing CDH Using the Command Line..........................................................................................................64
Starting CDH Services Using the Command Line..................................................................................................................65
Stopping CDH Services Using the Command Line................................................................................................................70
Migrating Data between a CDH 4 and CDH 5 Cluster..........................................................................................................72
Managing Individual Services.............................................................................................................................78
Managing Flume..................................................................................................................................................................78
Managing HBase.................................................................................................................................................................80
Managing HDFS.................................................................................................................................................................134
Managing Hive...................................................................................................................................................................167
Managing Hue...................................................................................................................................................................176
Managing Impala..............................................................................................................................................................199
Managing Key-Value Store Indexer....................................................................................................................................210
Managing MapReduce and YARN......................................................................................................................................211
Managing Oozie.................................................................................................................................................................219
Managing Solr...................................................................................................................................................................227
Managing Spark.................................................................................................................................................................229
Managing the Sqoop 1 Client.............................................................................................................................................232
Managing Sqoop 2.............................................................................................................................................................233
Managing ZooKeeper.........................................................................................................................................................234
Configuring Services to Use the GPL Extras Parcel.............................................................................................................234
Resource Management........................................................................................236
Schedulers........................................................................................................................................................236
Cloudera Manager Resource Management.....................................................................................................236
Linux Control Groups (cgroups)........................................................................................................................238
Resource Management with Control Groups.....................................................................................................................240
Configuring Resource Parameters......................................................................................................................................241
Static Service Pools...........................................................................................................................................242
Dynamic Resource Pools..................................................................................................................................243
Managing Dynamic Resource Pools...................................................................................................................................244
YARN Pool Status and Configuration Options....................................................................................................................246
Assigning Applications and Queries to Resource Pools......................................................................................................247
Configuration Sets.............................................................................................................................................................249
Scheduling Rules................................................................................................................................................................250
Managing Impala Admission Control...............................................................................................................251
Managing the Impala Llama ApplicationMaster..............................................................................................253
Enabling Integrated Resource Management Using Cloudera Manager............................................................................254
Disabling Integrated Resource Management Using Cloudera Manager...........................................................................255
Configuring Llama Using Cloudera Manager.....................................................................................................................255
Impala Resource Management........................................................................................................................255
Admission Control and Query Queuing..............................................................................................................................255
Integrated Resource Management with YARN...................................................................................................................263
Performance Management...................................................................................265
Optimizing Performance in CDH.......................................................................................................................265
Choosing a Data Compression Format.............................................................................................................268
Tuning the Solr Server......................................................................................................................................269
Tuning to Complete During Setup......................................................................................................................................269
General Tuning...................................................................................................................................................................269
Other Resources.................................................................................................................................................................274
Tuning Spark Applications................................................................................................................................274
Tuning YARN.....................................................................................................................................................281
Overview............................................................................................................................................................................281
Cluster Configuration.........................................................................................................................................................285
YARN Configuration...........................................................................................................................................................286
MapReduce Configuration.................................................................................................................................................287
Step 7: MapReduce Configuration.....................................................................................................................................287
Step 7A: MapReduce Sanity Checking................................................................................................................................288
Configuring Your Cluster In Cloudera Manager.................................................................................................................288
High Availability...................................................................................................290
HDFS High Availability......................................................................................................................................290
Introduction to HDFS High Availability...............................................................................................................................290
Configuring Hardware for HDFS HA...................................................................................................................................292
Enabling HDFS HA..............................................................................................................................................................292
Disabling and Redeploying HDFS HA..................................................................................................................................305
Configuring Other CDH Components to Use HDFS HA.......................................................................................................309
Administering an HDFS High Availability Cluster...............................................................................................................311
Changing a Nameservice Name for Highly Available HDFS Using Cloudera Manager......................................................315
MapReduce (MRv1) and YARN (MRv2) High Availability..................................................................................315
YARN (MRv2) ResourceManager High Availability.............................................................................................................315
Work Preserving Recovery for YARN Components.............................................................................................................323
MapReduce (MRv1) JobTracker High Availability..............................................................................................................325
Cloudera Navigator Key Trustee Server High Availability.................................................................................337
Configuring Key Trustee Server High Availability Using Cloudera Manager......................................................................337
Configuring Key Trustee Server High Availability Using the Command Line......................................................................338
Recovering a Key Trustee Server........................................................................................................................................340
Key Trustee KMS High Availability....................................................................................................................340
High Availability for Other CDH Components...................................................................................................341
HBase High Availability......................................................................................................................................................341
Hive Metastore High Availability.......................................................................................................................................346
Hue High Availability .........................................................................................................................................................348
Llama High Availability......................................................................................................................................................351
Configuring Oozie for High Availability..............................................................................................................................352
Search High Availability.....................................................................................................................................................353
Configuring Cloudera Manager for High Availability With a Load Balancer.....................................................354
Introduction to Cloudera Manager Deployment Architecture...........................................................................................355
Prerequisites for Setting up Cloudera Manager High Availability......................................................................................356
High-Level Steps to Configure Cloudera Manager High Availability .................................................................................356
Database High Availability Configuration..........................................................................................................................383
TLS and Kerberos Configuration for Cloudera Manager High Availability.........................................................................384
Backup and Disaster Recovery..............................................................................386
Port Requirements for Backup and Disaster Recovery.....................................................................................386
Data Replication...............................................................................................................................................387
Designating a Replication Source......................................................................................................................................389
HDFS Replication................................................................................................................................................................390
Hive Replication.................................................................................................................................................................399
Impala Metadata Replication............................................................................................................................................406
Using Snapshots with Replication......................................................................................................................................407
Enabling Replication Between Clusters in Different Kerberos Realms................................................................................407
Replication of Encrypted Data...........................................................................................................................................409
HBase Replication..............................................................................................................................................................409
Snapshots.........................................................................................................................................................416
Cloudera Manager Snapshot Policies................................................................................................................................416
Managing HBase Snapshots..............................................................................................................................................419
Managing HDFS Snapshots................................................................................................................................................430
Cloudera Manager Administration........................................................................435
Starting, Stopping, and Restarting the Cloudera Manager Server...................................................................435
Configuring Cloudera Manager Server Ports....................................................................................................435
Moving the Cloudera Manager Server to a New Host.....................................................................................435
Managing the Cloudera Manager Server Log...................................................................................................436
Viewing the Log.................................................................................................................................................................436
Setting the Cloudera Manager Server Log Location..........................................................................................................437
Cloudera Manager Agents................................................................................................................................437
Starting, Stopping, and Restarting Cloudera Manager Agents..........................................................................................438
Configuring Cloudera Manager Agents.............................................................................................................................439
Managing Cloudera Manager Agent Logs.........................................................................................................................442
Changing Hostnames........................................................................................................................................443
Configuring Network Settings..........................................................................................................................445
Alerts................................................................................................................................................................445
Managing Alerts................................................................................................................................................................446
Managing Licenses...........................................................................................................................................453
Sending Usage and Diagnostic Data to Cloudera.............................................................................................458
Configuring a Proxy Server.................................................................................................................................................458
Managing Anonymous Usage Data Collection..................................................................................................................458
Managing Hue Analytics Data Collection..........................................................................................................................458
Diagnostic Data Collection.................................................................................................................................................459
Exporting and Importing Cloudera Manager Configuration.............................................................................461
Backing up Cloudera Manager.........................................................................................................................461
Backing up Databases........................................................................................................................................................463
Other Cloudera Manager Tasks and Settings...................................................................................................463
Settings..............................................................................................................................................................................463
Alerts..................................................................................................................................................................................464
Users..................................................................................................................................................................................464
Kerberos.............................................................................................................................................................................464
License...............................................................................................................................................................................464
User Interface Language....................................................................................................................................................464
Peers..................................................................................................................................................................................464
Cloudera Management Service........................................................................................................................464
Cloudera Navigator Data Management Component Administration......................470
Cloudera Navigator Audit Server......................................................................................................................470
Publishing Audit Events......................................................................................................................................................473
Cloudera Navigator Metadata Server...............................................................................................................474
Managing Metadata Extraction........................................................................................................................................480
Managing Metadata Policies.............................................................................................................................................481
About Cloudera Administration
About Cloudera Administration
This guide describes how to configure and administer a Cloudera deployment. Administrators manage resources,
availability, and backup and recovery configurations. In addition, this guide shows how to implement high availability,
and discusses integration.
Cloudera Administration | 7
Managing CDH and Managed Services
Managing CDH and Managed Services
If you use Cloudera Manager to manage your cluster, configuring and managing your cluster, as well as individual
services and hosts, uses a different paradigm than if you use CDH without Cloudera Manager. For this reason, many
of these configuration tasks offer two different subtasks, one each for clusters managed by Cloudera Manager and
one for clusters which do not use Cloudera Manager. Often, the tasks are not interchangeable. For instance, if you use
Cloudera Manager you cannot use standard Hadoop command-line utilities to start and stop services. Instead, you use
Cloudera Manager to perform these tasks.
Managing CDH and Managed Services Using Cloudera Manager
You manage CDH and managed services using the Cloudera Manager Admin Console and Cloudera Manager API.
The following sections focus on the Cloudera Manager Admin Console.
Configuration Overview
When Cloudera Manager configures a service, it allocates roles that are required for that service to the hosts in your
cluster. The role determines which service daemons run on a host.
For example, for an HDFS service instance, Cloudera Manager configures:
•
•
•
•
One host to run the NameNode role.
One host to run as the secondary NameNode role.
One host to run the Balancer role.
Remaining hosts as to run DataNode roles.
A role group is a set of configuration properties for a role type, as well as a list of role instances associated with that
group. Cloudera Manager automatically creates a default role group named Role Type Default Group for each role
type.
When you run the installation or upgrade wizard, Cloudera Manager configures the default role groups it adds, and
adds any other required role groups for a given role type. For example, a DataNode role on the same host as the
NameNode might require a different configuration than DataNode roles running on other hosts. Cloudera Manager
creates a separate role group for the DataNode role running on the NameNode host and uses the default configuration
for DataNode roles running on other hosts.
Cloudera Manager wizards autoconfigure role group properties based on the resources available on the hosts. For
properties that are not dependent on host resources, Cloudera Manager default values typically align with CDH default
values for that configuration. Cloudera Manager deviates when the CDH default is not a recommended configuration
or when the default values are illegal.
Cloudera Manager Configuration Layout
After running the Installation wizard, use Cloudera Manager to reconfigure the existing services and add and configure
additional hosts and services.
Cloudera Manager configuration screens offer two layout options: new (the default) and classic. You can switch between
layouts using the Switch to XXX layout link at the top right of the page. Keep the following in mind when you select a
layout:
• If you switch to the classic layout, Cloudera Manager preserves that setting when you upgrade to a new version.
• Selections made in one layout are not preserved when you switch.
• Certain features, including controls for configuring Navigator audit events and HDFS log redaction, are supported
only in the new layout.
8 | Cloudera Administration
Managing CDH and Managed Services
New layout pages contain controls that allow you filter configuration properties based on configuration status, category,
and group. For example, to display the JournalNode maximum log size property (JournalNode Max Log Size), click the
CATEGORY > JournalNode and GROUP > Logs filters:
When a configuration property has been set to a value different from the default, a reset to default value icon
displays.
Classic layout pages are organized by role group and categories within the role group. For example, to display the
JournalNode maximum log size property (JournalNode Max Log Size), select JournalNode Default Group > Logs.
Cloudera Administration | 9
Managing CDH and Managed Services
When a configuration property has been set to a value different from the default, a Reset to the default value link
displays.
There is no mechanism for resetting to an autoconfigured value. However, you can use the configuration history and
rollback feature to revert any configuration changes.
Modifying Configuration Properties Using Cloudera Manager
Note:
This topic discusses how to configure properties using the Cloudera Manager "new layout." The older
layout, called the "classic layout" is still available. For instructions on using the classic layout, see
Modifying Configuration Properties (Classic Layout) on page 15.
To switch between the layouts, click either the Switch to the new layout or Switch to the classic
layout links in the upper-right portion of all configuration pages.
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
When a service is added to Cloudera Manager, either through the installation or upgrade wizard or with the Add
Services workflow, Cloudera Manager automatically sets the configuration properties, based on the needs of the service
and characteristics of the cluster in which it will run. These configuration properties include both service-wide
configuration properties, as well as specific properties for each role type associated with the service, managed through
role groups. A role group is a set of configuration properties for a role type, as well as a list of role instances associated
with that group. Cloudera Manager automatically creates a default role group named Role Type Default Group for
each role type. See Role Groups on page 48.
Changing the Configuration of a Service or Role Instance
1. Go to the service status page. (Cluster > service name)
2. Click the Configuration tab.
3. Locate the property you want to edit. You can type all or part of the property name in the search box, or use the
filters on the left side of the screen:
• The Status section limits the displayed properties by their status. Possible statuses include:
10 | Cloudera Administration
Managing CDH and Managed Services
•
•
•
•
•
Error
Warning
Edited
Non-default
Has Overrides
• The Scope section of the left hand panel organizes the configuration properties by role types; first those that
are Service-Wide, followed by various role types within the service. When you select one of these roles, a
set of properties whose values are managed by the default role group for the role display. Any additional
role groups that apply to the property also appear in this panel and you can modify values for each role group
just as you can the default role group.
• The Category section of the left hand panel allows you to limit the displayed properties by category.
4. Edit the property value.
• To facilitate entering some types of values, you can specify not only the value, but also the units that apply
to the value. for example, to enter a setting that specifies bytes per second, you can choose to enter the
value in bytes (B), KiBs, MiBs, or GiBs—selected from a drop-down menu that appears when you edit the
value.
• If the property allows a list of values, click the icon to the right of the edit field to add an additional field.
An example of this is the HDFS DataNode Data Directory property, which can have a comma-delimited list of
directories as its value. To remove an item from such a list, click the icon to the right of the field you want
to remove.
Many configuration properties have different values that are configured by multiple role groups. (See Role Groups
on page 48).
To edit configuration values for multiple role groups:
1. Go to the property, For example, the configuration panel for the Heap Dump Directory property displays the
DataNode Default Group (a role group), and a link that says ... and 6 others.
2. Click the ... and 6 others link to display all of the role groups:
3. Click the Show fewer link to collapse the list of role groups.
If you edit the single value for this property, Cloudera Manager applies the value to all role groups. To edit
the values for one or more of these role groups individually, click Edit Individual Values. Individual fields
display where you can edit the values for each role group. For example:
Cloudera Administration | 11
Managing CDH and Managed Services
5. Click Save Changes to commit the changes. You can add a note that is included with the change in the Configuration
History. This changes the setting for the role group, and applies to all role instances associated with that role
group. Depending on the change you made, you may need to restart the service or roles associated with the
configuration you just changed. Or, you may need to redeploy your client configuration for the service. You should
see a message to that effect at the top of the Configuration page, and services will display an outdated configuration
(Restart Needed), (Refresh Needed), or outdated client configuration
the Stale Configurations on page 28 page.
indicator. Click the indicator to display
Searching for Properties
You can use the Search box to search for properties by name or label. The search also returns properties whose
description matches your search term.
Validation of Configuration Properties
Cloudera Manager validates the values you specify for configuration properties. If you specify a value that is outside
the recommended range of values or is invalid, Cloudera Manager displays a warning at the top of the Configuration
tab and in the text box after you click Save Changes. The warning is yellow if the value is outside the recommended
range of values and red if the value is invalid.
Overriding Configuration Properties
For role types that allow multiple instances, each role instance inherits its configuration properties from its associated
role group. While role groups provide a convenient way to provide alternate configuration properties for selected
groups of role instances, there may be situations where you want to make a one-off configuration change—for example
when a host has malfunctioned and you want to temporarily reconfigure it. In this case, you can override configuration
properties for a specific role instance:
1.
2.
3.
4.
5.
6.
Go to the Status page for the service whose role you want to change.
Click the Instances tab.
Click the role instance you want to change.
Click the Configuration tab.
Change the configuration values as appropriate.
Save your changes.
12 | Cloudera Administration
Managing CDH and Managed Services
You will most likely need to restart your service or role to have your configuration changes take effect. See Stale
Configuration Actions on page 29.
Viewing and Editing Overridden Configuration Properties
To see a list of all role instances that have an override value for a particular configuration setting, go to the Status page
for the service and select Status > Has overrides. A list of configuration properties where values have been overridden
displays. The panel for each configuration property displays the values for each role group or instance. You can edit
the value of this property for this instance, or, you can click the
icon next to an instance name to remove the overridden value.
Resetting Configuration Properties to the Default Value
To reset a property back to its default value, click the
icon. The default value is inserted and the icon turns into an Undo icon
(
Explicitly setting a configuration to the same value as its default (inherited value) has the same effect as using the
.)
icon.
There is no mechanism for resetting to an autoconfigured value. However, you can use the configuration history and
rollback feature to revert any configuration changes.
Viewing and Editing Host Overrides
You can override the properties of individual hosts in your cluster.
1.
2.
3.
4.
Click the Hosts tab.
Click the Configuration tab.
Use the Filters or Search box to locate the property that you want to override.
Click the Manage Host Overrides link.
The Manage Overrides dialog box displays.
Cloudera Administration | 13
Managing CDH and Managed Services
5. Select one or more hosts to override this property.
6. Click Update.
A new entry area displays where you can enter the override values. In the example below, servers
ed9-e.ent.cloudera.com and ed9-r.cloudera.com were selected for overrides. Note that the first set of
fields displays the value set for all hosts and the two sets of fields that follow allow you to edit the override values
for each specified host.
To remove the override, click the
icon next to the hostname.
To apply the same value to all hosts, click Edit Identical Values. Click Edit Individual Values to apply different
values to selected hosts.
7. If the property indicates Requires Agent Restart, restart the agent on the affected hosts.
Restarting Services and Instances after Configuration Changes
If you change the configuration properties after you start a service or instance, you may need to restart the service or
instance to have the configuration properties become active. If you change configuration properties at the service
level that affect a particular role only (such as all DataNodes but not the NameNodes), you can restart only that role;
you do not need to restart the entire service. If you changed the configuration for a particular role instance (such as
one of four DataNodes), you may need to restart only that instance.
1. Follow the instructions in Restarting a Service on page 41 or Starting, Stopping, and Restarting Role Instances on
page 46.
2. If you see a Finished status, the service or role instances have restarted.
3. Go to the Home > Status tab. The service should show a Status of Started for all instances and a health status of
Good.
For further information, see Stale Configurations on page 28.
Suppressing Configuration and Parameter Validation Warnings
You can suppress the warnings that Cloudera Manager issues when a configuration value is outside the recommended
range or is invalid. If a warning does not apply to your deployment, you might want to suppress it. Suppressed validation
warnings are still retained by Cloudera Manager, and you can unsuppress the warnings at any time. You can suppress
each warning when you view it, or you can configure suppression for a specific validation before warnings occur.
Suppressing a Configuration Validation in Cloudera Manager
1. Click the Suppress... link to suppress the warning.
14 | Cloudera Administration
Managing CDH and Managed Services
A dialog box opens where you can enter a comment about the suppression.
2. Click Confirm.
You can also suppress warnings from the All Configuration Issues screen:
1. Browse to the Home screen.
2. Click Configurations > Configuration Issues.
3. Locate the validation message in the list and click the Suppress... link.
A dialog box opens where you can enter a comment about the suppression.
4. Click Confirm.
The suppressed validation warning is now hidden.
Managing Suppressed Validations
On pages where you have suppressed validations, you see a link that says Show # Suppressed Warning(s). On this
screen, you can:
• Click the Show # Suppressed Warning(s) link to show the warnings.
Each suppressed warning displays an icon:
.
• Click the Unsuppress... link to unsuppress the configuration validation.
• Click the Hide Suppressed Warnings link to re-hide the suppressed warnings.
Suppressing Configuration Validations Before They Trigger Warnings
1. Go to the service or host with the configuration validation warnings you want to suppress.
2. Click Configuration.
3. In the filters on the left, select Category > Suppressions.
A list of suppression properties displays. The names of the properties begin with Suppress Parameter Validation
or Suppress Configuration Validator. You can also use the Search function to limit the number of properties that
display.
4. Select a suppression property to suppress the validation warning.
5. Click Save Changes to commit the changes.
Viewing a List of All Suppressed Validations
Do one of the following:
• From the Home page or the Status page of a cluster, select Configuration > Suppressed Health and Configuration
Issues.
• From the Status page of a service, select Configuration > Category > Suppressions and select Status > Non-default.
• From the Host tab, select Configuration > Category > Suppressions and select Status > Non-default.
Modifying Configuration Properties (Classic Layout)
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Note: As of Cloudera Manager version 5.2, a new layout of the pages where you configure Cloudera
Manager system properties was introduced. In Cloudera Manager version 5.4, this new layout displays
by default. This topic discusses how to configure properties using the older layout, called the "Classic
Layout". For instructions on using the new layout, see Modifying Configuration Properties Using
Cloudera Manager on page 10.
To switch between the layouts, click either the Switch to the new layout or Switch to the classic
layout links in the upper-right portion of all configuration pages.
Cloudera Administration | 15
Managing CDH and Managed Services
When a service is added to Cloudera Manager, either through the installation or upgrade wizard or with the Add
Services workflow, Cloudera Manager automatically sets the configuration properties, based on the needs of the service
and characteristics of the cluster in which it will run. These configuration properties include both service-wide
configuration properties, as well as specific properties for each role type associated with the service, managed through
role groups. A role group is a set of configuration properties for a role type, as well as a list of role instances associated
with that group. Cloudera Manager automatically creates a default role group named Role Type Default Group for
each role type. See Role Groups on page 48.
Changing the Configuration of a Service or Role Instance (Classic Layout)
1.
2.
3.
4.
Go to the service status page.
Click the Configuration tab.
Under the appropriate role group, select the category for the properties you want to change.
To search for a text string (such as "snippet"), in a property, value, or description, enter the text string in the
Search box at the top of the category list.
5. Moving the cursor over the value cell highlights the cell; click anywhere in the highlighted area to enable editing
of the value. Then type the new value in the field provided (or check or uncheck the box, as appropriate).
• To facilitate entering some types of values, you can specify not only the value, but also the units that apply
to the value. for example, to enter a setting that specifies bytes per second, you can choose to enter the
value in bytes (B), KiBs, MiBs, or GiBs—selected from a drop-down menu that appears when you edit the
value.
• If the property allows a list of values, click the icon to the right of the edit field to add an additional field.
An example of this is the HDFS DataNode Data Directory property, which can have a comma-delimited list of
directories as its value. To remove an item from such a list, click the icon to the right of the field you want
to remove.
6. Click Save Changes to commit the changes. You can add a note that will be included with the change in the
Configuration History. This will change the setting for the role group, and will apply to all role instances associated
with that role group. Depending on the change you made, you may need to restart the service or roles associated
with the configuration you just changed. Or, you may need to redeploy your client configuration for the service.
You should see a message to that effect at the top of the Configuration page, and services will display an outdated
configuration (Restart Needed), (Refresh Needed), or outdated client configuration
indicator to display the Stale Configurations on page 28 page.
indicator. Click the
Validation of Configuration Properties
Cloudera Manager validates the values you specify for configuration properties. If you specify a value that is outside
the recommended range of values or is invalid, Cloudera Manager displays a warning at the top of the Configuration
tab and in the text box after you click Save Changes. The warning is yellow if the value is outside the recommended
range of values and red if the value is invalid.
Overriding Configuration Properties
For role types that allow multiple instances, each role instance inherits its configuration properties from its associated
role group. While role groups provide a convenient way to provide alternate configuration properties for selected
groups of role instances, there may be situations where you want to make a one-off configuration change—for example
when a host has malfunctioned and you want to temporarily reconfigure it. In this case, you can override configuration
properties for a specific role instance:
1.
2.
3.
4.
5.
6.
Go to the Status page for the service whose role you want to change.
Click the Instances tab.
Click the role instance you want to change.
Click the Configuration tab.
Change the configuration values as appropriate.
Save your changes.
You will most likely need to restart your service or role to have your configuration changes take effect.
16 | Cloudera Administration
Managing CDH and Managed Services
Viewing and Editing Overridden Configuration Properties
To see a list of all role instances that have an override value for a particular configuration setting, go to the entry for
the configuration setting in the Status page, expand the Overridden by n instance(s) link in the value cell for the
overridden value.
To view the override values, and change them if appropriate, click the Edit Overrides link. This opens the Edit Overrides
page, and lists the role instances that have override properties for the selected configuration setting.
On the Edit Overrides page, you can do any of the following:
• View the list of role instances that have overridden the value specified in the role group. Use the selections on
the drop-down menu below the Value column header to view a list of instances that use the inherited value,
instances that use an override value, or all instances. This view is especially useful for finding inconsistent properties
in a cluster. You can also use the Host and Rack text boxes to filter the list.
• Change the override value for the role instances to the inherited value from the associated role group. To do so,
select the role instances you want to change, choose Inherited Value from the drop-down menu next to Change
value of selected instances to and click Apply.
• Change the override value for the role instances to a different value. To do so, select the role instances you want
to change, choose Other from the drop-down menu next to Change value of selected instances to. Enter the new
value in the text box and then click Apply.
Resetting Configuration Properties to the Default Value
To reset a property back to its default value, click the Reset to the default value link below the text box in the value
cell. The default value is inserted and both the text box and the Reset link disappear. Explicitly setting a configuration
to the same value as its default (inherited value) has the same effect as using the Reset to the default value link.
There is no mechanism for resetting to an autoconfigured value. However, you can use the configuration history and
rollback feature to revert any configuration changes.
Restarting Services and Instances after Configuration Changes
If you change the configuration properties after you start a service or instance, you may need to restart the service or
instance to have the configuration properties become active. If you change configuration properties at the service
level that affect a particular role only (such as all DataNodes but not the NameNodes), you can restart only that role;
you do not need to restart the entire service. If you changed the configuration for a particular role instance (such as
one of four DataNodes), you may need to restart only that instance.
Cloudera Administration | 17
Managing CDH and Managed Services
1. Follow the instructions in Restarting a Service on page 41 or Starting, Stopping, and Restarting Role Instances on
page 46.
2. If you see a Finished status, the service or role instances have restarted.
3. Go to the Home > Status tab. The service should show a Status of Started for all instances and a health status of
Good.
For further information, see Stale Configurations on page 28.
Autoconfiguration
Cloudera Manager provides several interactive wizards to automate common workflows:
•
•
•
•
•
•
Installation - used to bootstrap a Cloudera Manager deployment
Add Cluster - used when adding a new cluster
Add Service - used when adding a new service
Upgrade - used when upgrading to a new version of CDH
Static Service Pools - used when configuring static service pools
Import MapReduce - used when migrating from MapReduce to YARN
In some of these wizards, Cloudera Manager uses a set of rules to automatically configure certain settings to best suit
the characteristics of the deployment. For example, the number of hosts in the deployment drives the memory
requirements for certain monitoring daemons: the more hosts, the more memory is needed. Additionally, wizards that
are tasked with creating new roles will use a similar set of rules to determine an ideal host placement for those roles.
Scope
The following table shows, for each wizard, the scope of entities it affects during autoconfiguration and role-host
placement.
Wizard
Autoconfiguration Scope
Role-Host Placement Scope
Installation
New cluster, Cloudera Management
Service
New cluster, Cloudera Management
Service
Add Cluster
New cluster
New cluster
Add Service
New service
New service
Upgrade
Cloudera Management Service
Cloudera Management Service
Static Service Pools
Existing cluster
N/A
Import MapReduce
Existing YARN service
N/A
Certain autoconfiguration rules are unscoped, that is, they configure settings belonging to entities that aren't necessarily
the entities under the wizard's scope. These exceptions are explicitly listed.
Autoconfiguration
Cloudera Manager employs several different rules to drive automatic configuration, with some variation from wizard
to wizard. These rules range from the simple to the complex.
Configuration Scope
One of the points of complexity in autoconfiguration is configuration scope. The configuration hierarchy as it applies
to services is as follows: configurations may be modified at the service level (affecting every role in the service), role
group level (affecting every role instance in the group), or role level (affecting one role instance). A configuration found
in a lower level takes precedence over a configuration found in a higher level.
With the exception of the Static Service Pools, and the Import MapReduce wizard, all Cloudera Manager wizards follow
a basic pattern:
1. Every role in scope is moved into its own, new, role group.
18 | Cloudera Administration
Managing CDH and Managed Services
2. This role group is the receptacle for the role's "idealized" configuration. Much of this configuration is driven by
properties of the role's host, which can vary from role to role.
3. Once autoconfiguration is complete, new role groups with common configurations are merged.
4. The end result is a smaller set of role groups, each with an "idealized" configuration for some subset of the roles
in scope. A subset can have any number of roles; perhaps all of them, perhaps just one, and so on.
The Static Service Pools and Import MapReduce wizards configure role groups directly and do not perform any merging.
Static Service Pools
Certain rules are only invoked in the context of the Static Service Pools wizard. Additionally, the wizard autoconfigures
cgroup settings for certain kinds of roles:
•
•
•
•
•
•
•
•
•
HDFS DataNodes
HBase RegionServers
MapReduce TaskTrackers
YARN NodeManagers
Impala Daemons
Solr Servers
Spark Standalone Workers
Accumulo Tablet Servers
Add-on services
YARN
yarn.nodemanager.resource.cpu-vcores - For each NodeManager role group, set to number of cores,
including hyperthreads, on one NodeManager member's host * service percentage chosen in
wizard.
All Services
Cgroup cpu.shares - For each role group that supports cpu.shares, set to max(20, (service percentage
chosen in wizard) * 20).
Cgroup blkio.weight - For each role group that supports blkio.weight, set to max(100, (service percentage
chosen in wizard) * 10).
Data Directories
Several autoconfiguration rules work with data directories, and there's a common sub-rule used by all such rules to
determine, out of all the mountpoints present on a host, which are appropriate for data. The subrule works as follows:
• The initial set of mountpoints for a host includes all those that are disk-backed. Network-backed mountpoints are
excluded.
• Mountpoints beginning with /boot, /cdrom, /usr, /tmp, /home, or /dev are excluded.
• Mountpoints beginning with /media are excluded, unless the backing device's name contains /xvd somewhere
in it.
• Mountpoints beginning with /var are excluded, unless they are /var or /var/lib.
• The largest mount point (in terms of total space, not available space) is determined.
• Other mountpoints with less than 1% total space of the largest are excluded.
• Mountpoints beginning with /var or equal to / are excluded unless they’re the largest mount point.
• Remaining mountpoints are sorted lexicographically and retained for future use.
Memory
The rules used to autoconfigure memory reservations are perhaps the most complicated rules employed by Cloudera
Manager. When configuring memory, Cloudera Manager must take into consideration which roles are likely to enjoy
more memory, and must not over commit hosts if at all possible. To that end, it needs to consider each host as an
entire unit, partitioning its available RAM into segments, one segment for each role. To make matters worse, some
Cloudera Administration | 19
Managing CDH and Managed Services
roles have more than one memory segment. For example, a Solr server has two memory segments: a JVM heap used
for most memory allocation, and a JVM direct memory pool used for HDFS block caching. Here is the overall flow during
memory autoconfiguration:
1. The set of participants includes every host under scope as well as every {role, memory segment} pair on those
hosts. Some roles are under scope while others are not.
2. For each {role, segment} pair where the role is under scope, a rule is run to determine four different values for
that pair:
• Minimum memory configuration. Cloudera Manager must satisfy this minimum, possibly over-committing
the host if necessary.
• Minimum memory consumption. Like the above, but possibly scaled to account for inherent overhead. For
example, JVM memory values are multiplied by 1.3 to arrive at their consumption value.
• Ideal memory configuration. If RAM permits, Cloudera Manager will provide the pair with all of this memory.
• Ideal memory consumption. Like the above, but scaled if necessary.
3. For each {role, segment} pair where the role is not under scope, a rule is run to determine that pair's existing
memory consumption. Cloudera Manager will not configure this segment but will take it into consideration by
setting the pair's "minimum" and "ideal" to the memory consumption value.
4. For each host, the following steps are taken:
a. 20% of the host's available RAM is subtracted and reserved for the OS.
b. sum(minimum_consumption) and sum(ideal_consumption) are calculated.
c. An "availability ratio" is built by comparing the two sums against the host's available RAM.
a. If RAM < sum(minimum) ratio = 0
b. If RAM >= sum(ideal) ratio = 1
c. Otherwise, ratio is computed via: RAM - sum(minimum)) / (sum(ideal) - sum(minimum)
5. For each {role, segment} pair where the role is under scope, the segment is configured to be (minimum + ((ideal
- minimum) * (host availability ratio))). The value is rounded down to the nearest megabyte.
6. The {role, segment} pair is set with the value from the previous step. In the Static Service Pools wizard, the role
group is set just once (as opposed to each role).
7. Custom post-configuration rules are run.
Customization rules are applied in steps 2, 3 and 7. In step 2, there's a generic rule for most cases, as well as a series
of custom rules for certain {role, segment} pairs. Likewise, there's a generic rule to calculate memory consumption in
step 3 as well as some custom consumption functions for certain {role, segment} pairs.
Step 2 Generic Rule Excluding Static Service Pools Wizard
For every {role, segment} pair where the segment defines a default value, the pair's minimum is set to the segment's
minimum value (or 0 if undefined), and the ideal is set to the segment's default value.
Step 2 Custom Rules Excluding Static Service Pools Wizard
HDFS
For the NameNode and Secondary NameNode JVM heaps, the minimum is 50 MB and the ideal is max(1 GB,
sum_over_all(DataNode mountpoints’ available space) / 0.000008).
MapReduce
For the JobTracker JVM heap, the minimum is 50 MB and the ideal is max(1 GB, round((1 GB * 2.3717181092
* ln(number of TaskTrackers in MapReduce service)) - 2.6019933306)). If there are <=5 TaskTrackers,
the ideal is 1 GB.
For the mapper JVM heaps, the minimum is 1 and the ideal is (number of cores, including hyperthreads, on the
TaskTracker host). Note that memory consumption is scaled by mapred_child_java_opts_max_heap (the size of
a given task's heap).
20 | Cloudera Administration
Managing CDH and Managed Services
For the reducer JVM heaps, the minimum is 1 and the ideal is (number of cores, including hyperthreads, on the
TaskTracker host) / 2. Note that memory consumption is scaled by mapred_child_java_opts_max_heap (the size
of a given task's heap).
YARN
For the memory total allowed for containers, the minimum is 1 GB and the ideal is min(8 GB, (total RAM on
NodeManager host) * 0.8).
Hue
With the exception of the Beeswax Server (present only in CDH 4), Hue roles don’t have memory limits. Therefore,
Cloudera Manager treats them as roles that consume a fixed amount of memory by setting their minimum and ideal
consumption values, but not their configuration values. The two consumption values are set to 256 MB.
Impala
With the exception of the Impala Daemon, Impala roles don’t have memory limits. Therefore Cloudera Manager treats
them as roles that consume a fixed amount of memory by setting their minimum/ideal consumption values, but not
their configuration values. The two consumption values are set to 150 MB for the Catalog Server and 64 MB for the
StateStore.
For the Impala Daemon memory limit, the minimum is 256 MB and the ideal is (total RAM on daemon host) *
0.64.
Solr
For the Solr Server JVM heap, the minimum is 50 MB and the ideal is min(64 GB, (total RAM on Solr Server
host) * 0.64) / 2.6. For the Solr Server JVM direct memory segment, the minimum is 256 MB and the ideal is
min(64 GB, (total RAM on Solr Server host) * 0.64) / 2.
Cloudera Management Service
• Alert Publisher JVM heap - treated as if it consumed a fixed amount of memory by setting the minimum/ideal
consumption values, but not the configuration values. The two consumption values are set to 256 MB.
• Service and Host Monitor JVM heaps - the minimum is 50 MB and the ideal is either 256 MB (10 or fewer managed
hosts), 1 GB (100 or fewer managed hosts), or 2 GB (over 100 managed hosts).
• Event Server, Reports Manager, and Navigator Audit Server JVM heaps - the minimum is 50 MB and the ideal is
1 GB.
• Navigator Metadata Server JVM heap - the minimum is 512 MB and the ideal is 2 GB.
• Service and Host Monitor off-heap memory segments - the minimum is either 768 MB (10 or fewer managed
hosts), 2 GB (100 or fewer managed hosts), or 6 GB (over 100 managed hosts). The ideal is always twice the
minimum.
Step 2 Generic Rule for Static Service Pools Wizard
For every {role, segment} pair where the segment defines a default value and an autoconfiguration share, the pair's
minimum is set to the segment's default value, and the ideal is set to min((segment soft max (if exists) or
segment max (if exists) or 2^63-1), (total RAM on role's host * 0.8 / segment scale factor
* service percentage chosen in wizard * segment autoconfiguration share)).
Autoconfiguration shares are defined as follows:
•
•
•
•
•
•
HBase RegionServer JVM heap: 1
HDFS DataNode JVM heap: 1 in CDH 4, 0.2 in CDH 5
HDFS DataNode maximum locked memory: 0.8 (CDH 5 only)
Solr Server JVM heap: 0.5
Solr Server JVM direct memory: 0.5
Spark Standalone Worker JVM heap: 1
Cloudera Administration | 21
Managing CDH and Managed Services
• Accumulo Tablet Server JVM heap: 1
• Add-on services: any
Roles not mentioned here do not define autoconfiguration shares and thus aren't affected by this rule.
Additionally, there's a generic rule to handle cgroup.memory_limit_in_bytes, which is unused by Cloudera services
but is available for add-on services. Its behavior varies depending on whether the role in question has segments or
not.
With Segments
The minimum is the min(cgroup.memory_limit_in_bytes_min (if exists) or 0, sum_over_all(segment
minimum consumption)), and the ideal is the sum of all segment ideal consumptions.
Without Segments
The minimum is cgroup.memory_limit_in_bytes_min (if exists) or 0, and the ideal is (total RAM on
role's host * 0.8 * service percentage chosen in wizard).
Step 3 Custom Rules for Static Service Pools Wizard
YARN
For the memory total allowed for containers, the minimum is 1 GB and the ideal is min(8 GB, (total RAM on
NodeManager host) * 0.8 * service percentage chosen in wizard).
Impala
For the Impala Daemon memory limit, the minimum is 256 MB and the ideal is ((total RAM on Daemon host)
* 0.8 * service percentage chosen in wizard).
MapReduce
• Mapper JVM heaps - the minimum is 1 and the ideal is (number of cores, including hyperthreads, on the TaskTracker
host * service percentage chosen in wizard). Note that memory consumption is scaled by
mapred_child_java_opts_max_heap (the size of a given task's heap).
• Reducer JVM heaps - the minimum is 1 and the ideal is (number of cores, including hyperthreads on the TaskTracker
host * service percentage chosen in wizard) / 2. Note that memory consumption is scaled by
mapred_child_java_opts_max_heap (the size of a given task's heap).
Step 3 Generic Rule
For every {role, segment} pair, the segment's current value is converted into bytes, and then multiplied by the scale
factor (1.0 by default, 1.3 for JVM heaps, and freely defined for Custom Service Descriptor services).
Step 3 Custom Rules
Impala
For the Impala Daemon, the memory consumption is 0 if YARN Service for Resource Management is set. If the memory
limit is defined but not -1, its value is used verbatim. If it's defined but -1, the consumption is equal to the total RAM
on the Daemon host. If it is undefined, the consumption is (total RAM * 0.8).
MapReduce
See Step 3 Custom Rules for Static Service Pools Wizard on page 22.
22 | Cloudera Administration
Managing CDH and Managed Services
Solr
For the Solr Server JVM direct memory segment, the consumption is equal to the value verbatim provided
solr.hdfs.blockcache.enable and solr.hdfs.blockcache.direct.memory.allocation are both true.
Otherwise, the consumption is 0.
Step 7 Custom Rules
HDFS
• NameNode JVM heaps are equalized. For every pair of NameNodes in an HDFS service with different heap sizes,
the larger heap size is reset to the smaller one.
• JournalNode JVM heaps are equalized. For every pair of JournalNodes in an HDFS service with different heap sizes,
the larger heap size is reset to the smaller one.
• NameNode and Secondary NameNode JVM heaps are equalized. For every {NameNode, Secondary NameNode}
pair in an HDFS service with different heap sizes, the larger heap size is reset to the smaller one.
HBase
Master JVM heaps are equalized. For every pair of Masters in an HBase service with different heap sizes, the larger
heap size is reset to the smaller one.
Impala
If an Impala service has YARN Service for Resource Management set, every Impala Daemon memory limit is set to the
value of (yarn.nodemanager.resource.memory-mb * 1 GB) if there's a YARN NodeManager co-located with the
Impala Daemon.
MapReduce
JobTracker JVM heaps are equalized. For every pair of JobTrackers in an MapReduce service with different heap sizes,
the larger heap size is reset to the smaller one.
Oozie
Oozie Server JVM heaps are equalized. For every pair of Oozie Servers in an Oozie service with different heap sizes,
the larger heap size is reset to the smaller one.
YARN
ResourceManager JVM heaps are equalized. For every pair of ResourceManagers in a YARN service with different heap
sizes, the larger heap size is reset to the smaller one.
ZooKeeper
ZooKeeper Server JVM heaps are equalized. For every pair of servers in a ZooKeeper service with different heap sizes,
the larger heap size is reset to the smaller one.
General Rules
HBase
• hbase.replication - For each HBase service, set to true if there's a Key-Value Store Indexer service in the
cluster. This rule is unscoped; it can fire even if the HBase service is not under scope.
• replication.replicationsource.implementation - For each HBase service, set to
com.ngdata.sep.impl.SepReplicationSource if there's a Keystore Indexer service in the cluster. This rule
is unscoped; it can fire even if the HBase service is not under scope.
Cloudera Administration | 23
Managing CDH and Managed Services
HDFS
• dfs.datanode.du.reserved - For each DataNode, set to min((total space of DataNode host largest
mountpoint) / 10, 10 GB).
• dfs.namenode.name.dir - For each NameNode, set to the first two mountpoints on the NameNode host with
/dfs/nn appended.
• dfs.namenode.checkpoint.dir - For each Secondary NameNode, set to the first mountpoint on the Secondary
NameNode host with /dfs/snn appended.
• dfs.datanode.data.dir - For each DataNode, set to all the mountpoints on the host with /dfs/dn appended.
• dfs.journalnode.edits.dir - For each JournalNode, set to the first mountpoint on the JournalNode host
with /dfs/jn appended.
• dfs.datanode.failed.volumes.tolerated - For each DataNode, set to (number of mountpoints on DataNode
host) / 2.
• dfs.namenode.service.handler.count and dfs.namenode.handler.count - For each NameNode, set
to max(30, ln(number of DataNodes in this HDFS service) * 20).
• dfs.block.local-path-access.user - For each HDFS service, set to impala if there's an Impala service in
the cluster. This rule is unscoped; it can fire even if the HDFS service is not under scope.
• dfs.datanode.hdfs-blocks-metadata.enabled - For each HDFS service, set to true if there's an Impala
service in the cluster. This rule is unscoped; it can fire even if the HDFS service is not under scope.
• dfs.client.read.shortcircuit - For each HDFS service, set to true if there's an Impala service in the cluster.
This rule is unscoped; it can fire even if the HDFS service is not under scope.
• dfs.datanode.data.dir.perm - For each DataNode, set to 755 if there's an Impala service in the cluster and
the cluster isn’t Kerberized. This rule is unscoped; it can fire even if the HDFS service is not under scope.
• fs.trash.interval - For each HDFS service, set to 1.
Hue
• WebHDFS dependency - For each Hue service, set to either the first HttpFS role in the cluster, or, if there are
none, the first NameNode in the cluster.
• HBase Thrift Server dependency- For each Hue service in a CDH 4.4 or higher cluster, set to the first HBase Thrift
Server in the cluster.
Impala
For each Impala service, set Enable Audit Collection and Enable Lineage Collection to true if there's a Cloudera
Management Service with a Navigator Audit Server and Navigator Metadata Server roles. This rule is unscoped; it can
fire even if the Impala service is not under scope.
MapReduce
• mapred.local.dir - For each JobTracker, set to the first mountpoint on the JobTracker host with /mapred/jt
appended.
• mapred.local.dir - For each TaskTracker, set to all the mountpoints on the host with /mapred/local
appended.
• mapred.reduce.tasks - For each MapReduce service, set to max(1, sum_over_all(TaskTracker number
of reduce tasks (determined via mapred.tasktracker.reduce.tasks.maximum for that
TaskTracker, which is configured separately)) / 2).
• mapred.job.tracker.handler.count - For each JobTracker, set to max(10, ln(number of TaskTrackers
in this MapReduce service) * 20).
• mapred.submit.replication - If there's an HDFS service in the cluster, for each MapReduce service, set to
max(1, sqrt(number of DataNodes in the HDFS service)).
• mapred.tasktracker.instrumentation - If there's a management service, for each MapReduce service, set
to org.apache.hadoop.mapred.TaskTrackerCmonInst. This rule is unscoped; it can fire even if the MapReduce
service is not under scope.
24 | Cloudera Administration
Managing CDH and Managed Services
YARN
• yarn.nodemanager.local-dirs - For each NodeManager, set to all the mountpoints on the NodeManager
host with /yarn/nm appended.
• yarn.nodemanager.resource.cpu-vcores - For each NodeManager, set to the number of cores (including
hyperthreads) on the NodeManager host.
• mapred.reduce.tasks - For each YARN service, set to max(1,sum_over_all(NodeManager number of
cores, including hyperthreads) / 2).
• yarn.resourcemanager.nodemanagers.heartbeat-interval-ms - For each NodeManager, set to max(100,
10 * (number of NodeManagers in this YARN service)).
• yarn.scheduler.maximum-allocation-vcores - For each ResourceManager, set to
max_over_all(NodeManager number of vcores (determined via
yarn.nodemanager.resource.cpu-vcores for that NodeManager, which is configured
separately)).
• yarn.scheduler.maximum-allocation-mb - For each ResourceManager, set to max_over_all(NodeManager
amount of RAM (determined via yarn.nodemanager.resource.memory-mb for that NodeManager,
which is configured separately)).
• mapreduce.client.submit.file.replication - If there's an HDFS service in the cluster, for each YARN
service, set to max(1, sqrt(number of DataNodes in the HDFS service)).
All Services
If a service dependency is unset, and a service with the desired type exists in the cluster, set the service dependency
to the first such target service. Applies to all service dependencies except YARN Service for Resource Management.
Applies only to the Installation and Add Cluster wizards.
Role-Host Placement
Cloudera Manager employs the same role-host placement rule regardless of wizard. The set of hosts considered
depends on the scope. If the scope is a cluster, all hosts in the cluster are included. If a service, all hosts in the service's
cluster are included. If the Cloudera Management Service, all hosts in the deployment are included. The rules are as
follows:
1. The hosts are sorted from most to least physical RAM. Ties are broken by sorting on hostname (ascending) followed
by host identifier (ascending).
2. The overall number of hosts is used to determine which arrangement to use. These arrangements are hard-coded,
each dictating for a given "master" role type, what index (or indexes) into the sorted host list in step 1 to use.
3. Master role types are included based on several factors:
•
•
•
•
Is this role type part of the service (or services) under scope?
Does the service already have the right number of instances of this role type?
Does the cluster's CDH version support this role type?
Does the installed Cloudera Manager license allow for this role type to exist?
4. Master roles are placed on each host using the indexes and the sorted host list. If a host already has a given master
role, it is skipped.
5. An HDFS DataNode is placed on every host outside of the arrangement described in step 2, provided HDFS is one
of the services under scope.
6. Certain "worker" roles are placed on every host where an HDFS DataNode exists, either because it existed there
prior to the wizard, or because it was added in the previous step. The supported worker role types are:
•
•
•
•
•
MapReduce TaskTrackers
YARN NodeManagers
HBase RegionServers
Impala Daemons
Spark Workers
Cloudera Administration | 25
Managing CDH and Managed Services
7. Hive gateways are placed on every host, provided a Hive service is under scope and a gateway didn’t already exist
on a given host.
8. Spark on YARN gateways are placed on every host, provided a Spark on YARN service is under scope and a gateway
didn’t already exist on a given host.
This rule merely dictates the default placement of roles; you are free to modify it before it is applied by the wizard.
Custom Configuration
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Cloudera Manager exposes properties that allow you to insert custom configuration text into XML configuration,
property, and text files, or into an environment. The naming convention for these properties is: XXX Advanced
Configuration Snippet (Safety Valve) for YYY or XXX YYY Advanced Configuration Snippet (Safety Valve), where XXX
is a service or role and YYY is the target.
The values you enter into a configuration snippet must conform to the syntax of the target. For an XML configuration
file, the configuration snippet must contain valid XML property definitions. For a properties file, the configuration
snippet must contain valid property definitions. Some files simply require a list of host addresses.
The configuration snippet mechanism is intended for use in cases where there is configuration setting that is not
exposed as a configuration property in Cloudera Manager. Configuration snippets generally override normal configuration.
Contact Cloudera Support if you are required to use a configuration snippet that is not explicitly documented.
Service-wide configuration snippets apply to all roles in the service; a configuration snippet for a role group applies to
all instances of the role associated with that role group.
There are configuration snippets for servers and client configurations. In general after changing a server configuration
snippet you must restart the server, and after changing a client configuration snippet you must redeploy the client
configuration. Sometimes you can refresh instead of restart. In some cases however, you must restart a dependent
server after changing a client configuration. For example, changing a MapReduce client configuration marks the
dependent Hive server as stale, which must be restarted. The Admin Console displays an indicator when a server must
be restarted. In addition, the All Configuration Issues tab on the Home page lists the actions you must perform to
propagate the changes.
Configuration Snippet Types and Syntax
Type
Description
Syntax
Configuration
Set configuration properties in various
configuration files; the property name
indicates into which configuration file
the configuration will be placed.
Configuration files have the extension
.xml or .conf.
<property>
<name>property_name</name>
<value>property_value</value>
</property>
For example, to specify a MySQL connector library, put
this property definition in that configuration snippet:
<property>
<name>hive.aux.jars.path</name>
For example, there are several
configuration snippets for the Hive
<value>file:///usr/share/java/mysql-connector-java.jar</value>
service. One Hive configuration
</property>
snippet property is called the
HiveServer2 Advanced Configuration
Snippet for hive-site.xml;
configuration you enter here is
inserted verbatim into the
hive-site.xml file associated with
the HiveServer2 role group.
To see a list of configuration snippets
that apply to a specific configuration
file, enter the configuration file name
in the Search field in the top
26 | Cloudera Administration
Managing CDH and Managed Services
Type
Description
Syntax
navigation bar. For example, searching
for mapred-site.xml shows the
configuration snippets that have
mapred-site.xml in their name.
Environment
Specify key-value pairs for a service,
role, or client that are inserted into
the respective environment.
key=value
One example of using an environment
configuration snippet is to add a JAR
to a classpath. Place JARs in a custom
location such as /opt/myjars and
extend the classpath via the
appropriate service environment
configuration snippet. The value of a
JAR property must conform to the
syntax supported by its environment.
See Setting the class path.
AUX_CLASSPATH=/usr/share/java/mysql-connector-java.jar:\
/usr/share/java/oracle-connector-java.jar
For example, to add JDBC connectors to a Hive gateway
classpath, add
or
AUX_CLASSPATH=/usr/share/java/*
to Gateway Client Advanced Configuration Snippet for
hive-env.sh.
Do not place JARs inside locations such
as /opt/cloudera or
/usr/lib/{hadoop*,hbase*,hive*,etc.}
that are managed by Cloudera because
they are overwritten at upgrades.
Logging
Set log4j properties in a
log4j.properties file.
key1=value1
key2=value2
For example:
log4j.rootCategory=INFO, console
max.log.file.size=200MB
max.log.file.backup.index=10
Metrics
White and black
lists
Set properties to configure Hadoop
metrics in a
hadoop-metrics.properties or
hadoop-metrics2.properties file.
key1=value1
key2=value2
For example:
*.sink.foo.class=org.apache.hadoop.metrics2.sink.FileSink
namenode.sink.foo.filename=/tmp/namenode-metrics.out
secondarynamenode.sink.foo.filename=/tmp/secondarynamenode-metrics.out
Specify a list of host addresses that are host1.domain1
allowed or disallowed from accessing host2.domain2
a service.
Setting an Advanced Configuration Snippet
1.
2.
3.
4.
5.
6.
7.
Click a service.
Click the Configuration tab.
In the Search box, type Advanced Configuration Snippet.
Choose a property that contains the string Advanced Configuration Snippet (Safety Valve).
Specify the snippet properties.
Click Save Changes to commit the changes.
Restart the service or role or redeploy client configurations as indicated.
Setting Advanced Configuration Snippets for a Cluster or Clusters
1. Do one of the following
Cloudera Administration | 27
Managing CDH and Managed Services
• specific cluster
1. On the Home > Status tab, click a cluster name.
2. Select Configuration > Advanced Configuration Snippets.
• all clusters
1. Select Configuration > Advanced Configuration Snippets.
2. Specify the snippet properties.
3. Click Save Changes to commit the changes.
4. Restart the service or role or redeploy client configurations as indicated.
Stale Configurations
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
The Stale Configurations page provides differential views of changes made in a cluster. For any configuration change,
the page contains entries of all affected attributes. For example, the following File entry shows the change to the file
hdfs-site.xml when you update the property controlling how much disk space is reserved for non-HDFS use on
each DataNode:
To display the entities affected by a change, click the Show button at the right of the entry. The following dialog box
shows that three DataNodes were affected by the disk space change:
Viewing Stale Configurations
To view stale configurations, click the ,
Home Page or on a service status page.
Attribute Categories
The categories of attributes include:
28 | Cloudera Administration
, or
indicator next to a service on the Cloudera Manager Admin Console
Managing CDH and Managed Services
• Environment - represents environment variables set for the role. For example, the following entry shows the
change to the environment that occurs when you update the heap memory configuration of the
SecondaryNameNode.
• Files - represents configuration files used by the role.
• Process User & Group - represents the user and group for the role. Every role type has a configuration to specify
the user/group for the process. If you change a value for a user or group on any service's configuration page it
will appear in the Stale Configurations page.
• System Resources - represents system resources allocated for the role, including ports, directories, and cgroup
limits. For example, a change to the port of role instance will appear in the System Resources category.
• Client Configs Metadata - represents client configurations.
Filtering Stale Configurations
You filter the entries on the Stale Configurations page by selecting from one of the drop-down lists:
• Attribute - you can filter by an attribute category such as All Files or by a specific file such as topology.map or
yarn-site.xml.
• Service
• Role
After you make a selection, both the page and the drop-down show only entries that match that selection.
To reset the view, click Remove Filter or select All XXX, where XXX is Files, Services, or Roles, from the drop-down. For
example, to see all the files, select All Files.
Stale Configuration Actions
The Stale Configurations page displays action buttons. The action depends on what is required to bring the entire
cluster's configuration up to date. If you go to the page by clicking a (Refresh Needed) indicator, the action button
will say Restart Stale Services if one of the roles listed on the page need to be restarted.
•
•
•
•
Refresh Stale Services - Refreshes stale services.
Restart Stale Services - Restarts stale services.
Restart Cloudera Management Service - Runs the restart Cloudera Management Service action.
Deploy Client Configuration - Runs the cluster deploy client configurations action.
Client Configuration Files
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
To allow clients to use the HBase, HDFS, Hive, MapReduce, and YARN services, Cloudera Manager creates zip archives
of the configuration files containing the service properties. The zip archive is referred to as a client configuration file.
Each archive contains the set of configuration files needed to access the service: for example, the MapReduce client
configuration file contains copies of core-site.xml, hadoop-env.sh, hdfs-site.xml, log4j.properties, and
mapred-site.xml.
Client configuration files are generated automatically by Cloudera Manager based on the services and roles you have
installed and Cloudera Manager deploys these configurations automatically when you install your cluster, add a service
on a host, or add a gateway role on a host. Specifically, for each host that has a service role instance installed, and for
each host that is configured as a gateway role for that service, the deploy function downloads the configuration zip
file, unzips it into the appropriate configuration directory, and uses the Linux alternatives mechanism to set a given,
configurable priority level. If you are installing on a system that happens to have pre-existing alternatives, then it is
possible another alternative may have higher priority and will continue to be used. The alternatives priority of the
Cloudera Administration | 29
Managing CDH and Managed Services
Cloudera Manager client configuration is configurable under the Gateway scope of the Configuration tab for the
appropriate service.
You can also manually distribute client configuration files to the clients of a service.
The main circumstance that may require a redeployment of the client configuration files is when you have modified a
configuration. In this case you will typically see a message instructing you to redeploy your client configurations. The
affected service(s) will also display a
icon. Click the indicator to display the Stale Configurations on page 28 page.
How Client Configurations are Deployed
Client configuration files are deployed on any host that is a client for a service—that is, that has a role for the service
on that host. This includes roles such as DataNodes, TaskTrackers, RegionServers and so on as well as gateway roles
for the service.
If roles for multiple services are running on the same host (for example, a DataNode role and a TaskTracker role on
the same host) then the client configurations for both roles are deployed on that host, with the alternatives priority
determining which configuration takes precedence.
For example, suppose we have six hosts running roles as follows: host H1: HDFS-NameNode; host H2: MR-JobTracker;
host H3: HBase-Master; host H4: MR-TaskTracker, HDFS-DataNode, HBase-RegionServer; host H5: MR-Gateway; host
H6: HBase-Gateway. Client configuration files will be deployed on these hosts as follows: host H1: hdfs-clientconfig
(only); host H2: mapreduce-clientconfig, host H3: hbase-clientconfig; host H4: hdfs-clientconfig, mapreduce-clientconfig,
hbase-clientconfig; host H5: mapreduce-clientconfig; host H6: hbase-clientconfig
If the HDFS NameNode and MapReduce JobTracker were on the same host, then that host would have both
hdfs-clientconfig and mapreduce-clientconfig installed.
Downloading Client Configuration Files
1. Follow the appropriate procedure according to your starting point:
Page
Home
Procedure
1. On the Home > Status tab, click
to the right of the cluster name and select View Client Configuration URLs. A
pop-up window with links to the configuration files for the services you have
installed displays.
2. Click a link or save the link URL and download the file using wget or curl.
Service
1. Go to a service whose client configuration you want to download.
2. Select Actions > Download Client Configuration.
Manually Redeploying Client Configuration Files
Although Cloudera Manager will deploy client configuration files automatically in many cases, if you have modified
the configurations for a service, you may need to redeploy those configuration files.
If your client configurations were deployed automatically, the command described in this section will attempt to
redeploy them as appropriate.
Note: If you are deploying client configurations on a host that has multiple services installed, some
of the same configuration files, though with different configurations, will be installed in the conf
directories for each service. Cloudera Manager uses the priority parameter in the alternatives
--install command to ensure that the correct configuration directory is made active based on the
combination of services on that host. The priority order is YARN > MapReduce > HDFS. The priority
can be configured under the Gateway sections of the Configuration tab for the appropriate service.
30 | Cloudera Administration
Managing CDH and Managed Services
1. On the Home > Status tab, click
to the right of the cluster name and select Deploy Client Configuration.
2. Click Deploy Client Configuration.
Viewing and Reverting Configuration Changes
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Important: This feature is available only with a Cloudera Enterprise license; it is not available in
Cloudera Express. For information on Cloudera Enterprise licenses, see Managing Licenses on page
453.
Whenever you change and save a set of configuration settings for a service or role instance or a host, Cloudera Manager
saves a revision of the previous settings and the name of the user who made the changes. You can then view past
revisions of the configuration settings, and, if desired, roll back the settings to a previous state.
Viewing Configuration Changes
1. For a service, role, or host, click the Configuration tab.
2. Click the History and Rollback button. The most recent revision, currently in effect, is shown under Current
Revision. Prior revisions are shown under Past Revisions.
• By default, or if you click Show All, a list of all revisions is shown. If you are viewing a service or role instance,
all service/role group related revisions are shown. If you are viewing a host or all hosts, all host/all hosts
related revisions are shown.
• To list only the configuration revisions that were done in a particular time period, use the Time Range Selector
to select a time range. Then, click Show within the Selected Time Range.
3. Click the Details... link. The Revision Details dialog box displays.
Revision Details Dialog
For a service or role instance, shows the following:
•
•
•
•
•
A brief message describing the context of the changes
The date/time stamp of the change
The user who performed the change
The names of any role groups created
The names of any role groups deleted
For a host instance, shows just a message, date and time stamp, and the user.
The dialog box contains two tabs:
• Configuration Values - displays configuration value changes, where changes are organized under the role group
to which they were applied. (For example, if you changed a Service-Wide property, it will affect all role groups for
that service). For each modified property, the Value column shows the new value of the property and the previous
value.
• Group Membership - displays changes to the changed the group membership of a role instance (moved the
instance from one group to another). This tab is only shown for service and role configurations.
Reverting Configuration Changes
1.
2.
3.
4.
Select the current or past revision to which to roll back.
Click the Details... link. The Revision Details dialog box displays.
Click the Configuration Values tab.
Click the Revert Configuration Changes button. The revert action occurs immediately. You may need to restart
the service or the affected roles for the change to take effect.
Cloudera Administration | 31
Managing CDH and Managed Services
Important: This feature can only be used to revert changes to configuration values. You cannot use
this feature to:
• Revert NameNode high availability. You must perform this action by explicitly disabling high
availability.
• Disable Kerberos security.
• Revert role group actions (creating, deleting, or moving membership among groups). You must
perform these actions explicitly in the Role Groups on page 48 feature.
Exporting and Importing Cloudera Manager Configuration
You can use the Cloudera Manager API to programmatically export and import a definition of all the entities in your
Cloudera Manager-managed deployment—clusters, service, roles, hosts, users and so on. See the Cloudera Manager
API documentation on how to manage deployments using the /cm/deployment resource.
Managing Clusters
Cloudera Manager can manage multiple clusters, however each cluster can only be associated with a single Cloudera
Manager Server or Cloudera Manager HA pair. Once you have successfully installed your first cluster, you can add
additional clusters, running the same or a different version of CDH. You can then manage each cluster and its services
independently.
On the Home > Status tab you can access many cluster-wide actions by selecting
to the right of the cluster name: add a service, start, stop, restart, deploy client configurations, enable Kerberos, and
perform cluster refresh, rename, upgrade, and maintenance mode actions.
Note:
Cloudera Manager configuration screens offer two layout options: classic and new. The new layout
is the default; however, on each configuration page you can easily switch between layouts using the
Switch to XXX layout link at the top right of the page. For more information, see Configuration Overview
on page 8.
Adding and Deleting Clusters
Minimum Required Role: Full Administrator
Cloudera Manager can manage multiple clusters. Furthermore, the clusters do not need to run the same version of
CDH; you can manage both CDH 4 and CDH 5 clusters with Cloudera Manager.
Adding a Cluster
Action
New Hosts
Procedure
1. On the Home > Status tab, click
and select Add Cluster. This begins the Installation Wizard, just as if you were installing a
cluster for the first time. (See Cloudera Manager Deployment for detailed instructions.)
2. To find new hosts, not currently managed by Cloudera Manager, where you want to install
CDH, enter the hostnames or IP addresses, and click Search. Cloudera Manager lists the hosts
you can use to configure a new cluster. Managed hosts that already have services installed
will not be selectable.
32 | Cloudera Administration
Managing CDH and Managed Services
Action
Procedure
3. Click Continue to install the new cluster. At this point the installation continues through the
wizard the same as it did when you installed your first cluster. You will be asked to select the
version of CDH to install, which services you want and so on, just as previously.
4. Restart the Reports Manager role.
Managed Hosts
You may have hosts that are already "managed" but are not part of a cluster. You can have
managed hosts that are not part of a cluster when you have added hosts to Cloudera Manager
either through the Add Host wizard, or by manually installing the Cloudera Manager agent onto
hosts where you have not install any other services. This will also be the case if you remove all
services from a host so that it no longer is part of a cluster.
1. On the Home > Status tab, click
and select Add Cluster. This begins the Installation Wizard, just as if you were installing a
cluster for the first time. (See Cloudera Manager Deployment for detailed instructions.)
2. To see the list of the currently managed hosts, click the Currently Managed Hosts tab. This
tab does not appear if you have no currently managed hosts that are not part of a cluster.
3. To perform the installation, click Continue. Instead of searching for hosts, this will attempt
to install onto any hosts managed by Cloudera Manager that are not already part of a cluster.
It will proceed with the installation wizard as for a new cluster installation.
4. Restart the Reports Manager role.
Deleting a Cluster
1. Stop the cluster.
2. On the Home > Status tab, click
to the right of the cluster name and select Delete.
Starting, Stopping, Refreshing, and Restarting a Cluster
Minimum Required Role: Operator (also provided by Configurator, Cluster Administrator, Full Administrator)
Starting a Cluster
1. On the Home > Status tab, click
to the right of the cluster name and select Start.
2. Click Start that appears in the next screen to confirm. The Command Details window shows the progress of starting
services.
When All services successfully started appears, the task is complete and you can close the Command Details
window.
Note: The cluster-level Start action starts only CDH and other product services (Impala, Cloudera
Search). It does not start the Cloudera Management Service. You must start the Cloudera Management
Service separately if it is not already running.
Cloudera Administration | 33
Managing CDH and Managed Services
Stopping a Cluster
1. On the Home > Status tab, click
to the right of the cluster name and select Stop.
2. Click Stop in the confirmation screen. The Command Details window shows the progress of stopping services.
When All services successfully stopped appears, the task is complete and you can close the Command Details
window.
Note: The cluster-level Stop action does not stop the Cloudera Management Service. You must stop
the Cloudera Management Service separately.
Refreshing a Cluster
Runs a cluster refresh action to bring the configuration up to date without restarting all services. For example, certain
masters (for example NameNode and ResourceManager) have some configuration files (for example,
fair-scheduler.xml, mapred_hosts_allow.txt, topology.map) that can be refreshed. If anything changes in
those files then a refresh can be used to update them in the master. Here is a summary of the operations performed
in a refresh action:
To refresh a cluster, in the Home > Status tab, click
to the right of the cluster name and select Refresh Cluster.
Restarting a Cluster
1. On the Home > Status tab, click
to the right of the cluster name and select Restart.
2. Click Restart that appears in the next screen to confirm. The Command Details window shows the progress of
stopping services.
34 | Cloudera Administration
Managing CDH and Managed Services
When All services successfully started appears, the task is complete and you can close the Command Details
window.
Renaming a Cluster
Minimum Required Role: Full Administrator
1. On the Home > Status tab, click
to the right of the cluster name and select Rename Cluster.
2. Type the new cluster name and click Rename Cluster.
Cluster-Wide Configuration
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
To make configuration changes that apply to an entire cluster, do one of the following to open the configuration page:
• all clusters
1. Select Configuration and then select one of the following classes of properties:
•
•
•
•
•
•
•
•
•
•
•
Advanced Configuration Snippets
Databases
Disk Space Thresholds
Local Data Directories
Local Data Files
Log Directories
Navigator Settings
Non-default Values - properties whose value differs from the default value
Non-uniform Values - properties whose values are not uniform across the cluster or clusters
Port Configurations
Service Dependencies
You can also select Configuration Issues to view a list of configuration issues for all clusters.
• specific cluster
1. On the Home page, click a cluster name.
2. Select Configuration and then select one of the classes of properties listed above.
You can also apply the following filters to limit the displayed properties:
• Enter a search term in the Search box to search for properties by name or description.
• Expand the Status filter to select options that limit the displayed properties to those with errors or warnings,
properties that have been edited, properties with non-default values, or properties with overrides. Select All to
remove any filtering by Status.
• Expand the Scope filter to display a list of service types. Expand a service type heading to filter on Service-Wide
configurations for a specific service instance or select one of the default role groups listed under each service
type. Select All to remove any filtering by Scope.
• Expand the Category filter to filter using a sub-grouping of properties. Select All to remove any filtering by Category.
Moving a Host Between Clusters
Minimum Required Role: Full Administrator
Moving a host between clusters can be accomplished by:
1. Decommissioning the host (see Decommissioning Role Instances on page 46).
Cloudera Administration | 35
Managing CDH and Managed Services
2. Removing all roles from the host (except for the Cloudera Manager management roles). See Deleting Role Instances
on page 47.
3. Deleting the host from the cluster (see Deleting Hosts on page 61), specifically the section on removing a host
from a cluster but leaving it available to Cloudera Manager.
4. Adding the host to the new cluster (see Adding a Host to the Cluster on page 54).
5. Adding roles to the host (optionally using one of the host templates associated with the new cluster). See Adding
a Role Instance on page 45 and Host Templates on page 57.
Managing Services
Cloudera Manager service configuration features let you manage the deployment and configuration of CDH and
managed services. You can add new services and roles if needed, gracefully start, stop and restart services or roles,
and decommission and delete roles or services if necessary. Further, you can modify the configuration properties for
services or for individual role instances. If you have a Cloudera Enterprise license, you can view past configuration
changes and roll back to a previous revision. You can also generate client configuration files, enabling you to easily
distribute them to the users of a service.
The topics in this chapter describe how to configure and use the services on your cluster. Some services have unique
configuration requirements or provide unique features: those are covered in Managing Individual Services on page
78.
Note:
Cloudera Manager configuration screens offer two layout options: classic and new. The new layout
is the default; however, on each configuration page you can easily switch between layouts using the
Switch to XXX layout link at the top right of the page. For more information, see Configuration Overview
on page 8.
Adding a Service
Minimum Required Role: Full Administrator
After initial installation, you can use the Add a Service wizard to add and configure new service instances. For example,
you may want to add a service such as Oozie that you did not select in the wizard during the initial installation.
The binaries for the following services are not packaged in CDH 4 and CDH 5 and must be installed individually before
being adding the service:
Service
Installation Documentation
Accumulo
Apache Accumulo Documentation
Impala (not in CDH 4)
Installing Impala
Search (not in CDH 4)
Installing Search
Kafka
Installing Kafka
Key Trustee KMS
Installing Key Trustee KMS
If you do not add the binaries before adding the service, the service will fail to start.
To add a service:
1. On the Home > Status tab, click
to the right of the cluster name and select Add a Service. A list of service types display. You can add one type of
service at a time.
2. Click the radio button next to a service and click Continue. If you are missing required binaries, a pop-up displays
asking if you want to continue with adding the service.
36 | Cloudera Administration
Managing CDH and Managed Services
3. Select the radio button next to the services on which the new service should depend. All services must depend
on the same ZooKeeper service. Click Continue.
4. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of the
hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of hosts to
which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple hosts,
you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts dialog.
The following shortcuts for specifying hostname patterns are supported:
• Range of hostnames (without the domain portion)
Range Definition
Matching Hosts
10.1.1.[1-4]
10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4
host[1-3].company.com
host1.company.com, host2.company.com, host3.company.com
host[07-10].company.com
host07.company.com, host08.company.com, host09.company.com,
host10.company.com
• IP addresses
• Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
5. Review and modify configuration settings, such as data directory paths and heap sizes and click Continue. The
service is started.
6. Click Continue then click Finish. You are returned to the Home page.
7. Verify the new service is started properly by checking the health status for the new service. If the Health Status
is Good, then the service started properly.
Comparing Configurations for a Service Between Clusters
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
To compare the configuration settings for a particular service between two different clusters in a Cloudera Manager
deployment, perform the following steps:
1. On the Home > Status tab, click the name of the service you want to compare, or click the Clusters menu and
select the name of the service.
2. Click the Configuration tab.
3. Click the drop-down menu above the Filters pane, and select from one of the options that begins Diff with...:
• service on cluster - For example, HBASE-1 on Cluster 1. This is the default display setting. All properties are
displayed for the selected instance of the service.
• service on all clusters - For example, HBase on all clusters. All properties are displayed for all instances of the
service.
• Diff with service on cluster - For example, Diff with HBase on Cluster 2. Properties are displayed only if the
values for the instance of the service whose page you are on differ from the values for the instance selected
in the drop-down menu.
• Diff with service on all clusters - For example, Diff with HBase on all clusters. Properties are displayed if the
values for the instance of the service whose page you are on differ from the values for one or more other
instances in the Cloudera Manager deployment.
The service's properties will be displayed showing the values for each property for the selected clusters. The filters on
the left side can be used to limit the properties displayed.
Cloudera Administration | 37
Managing CDH and Managed Services
You can also view property configuration values that differ between clusters across a deployment by selecting
Non-uniform Values on the Configuration tab of the Cloudera Manager Home > Status tab. For more information, see
Cluster-Wide Configuration on page 35
Add-on Services
Minimum Required Role: Full Administrator
Cloudera Manager supports adding new types of services (referred to as an add-on service) to Cloudera Manager,
allowing such services to leverage Cloudera Manager distribution, configuration, monitoring, resource management,
and life-cycle management features. An add-on service can be provided by Cloudera or an independent software
vendor (ISV). If you have multiple clusters managed by Cloudera Manager, an add-on service can be deployed on any
of the clusters.
Note: If the add-on service is already installed and running on hosts that are not currently being
managed by Cloudera Manager, you must first add the hosts to a cluster that's under management.
See Adding a Host to the Cluster on page 54 for details.
Custom Service Descriptor Files
Integrating an add-on service requires a Custom Service Descriptor (CSD) file. A CSD file contains all the configuration
needed to describe and manage a new service. A CSD is provided in the form of a JAR file.
Depending on the service, the CSD and associated software may be provided by Cloudera or by an ISV. The integration
process assumes that the add-on service software (parcel or package) has been installed and is present on the cluster.
The recommended method is for the ISV to provide the software as a parcel, but the actual mechanism for installing
the software is up to the ISV. The instructions in Installing an Add-on Service on page 38 assume that you have obtained
the CSD file from the Cloudera repository or from an ISV. It also assumes you have obtained the service software,
ideally as a parcel, and have or will install it on your cluster either prior to installing the CSD or as part of the CSD
installation process.
Configuring the Location of Custom Service Descriptor Files
The default location for CSD files is /opt/cloudera/csd. You can change the location in the Cloudera Manager Admin
Console as follows:
1.
2.
3.
4.
Select Administration > Settings.
Click the Custom Service Descriptors category.
Edit the Local Descriptor Repository Path property.
Click Save Changes to commit the changes.
Installing an Add-on Service
An ISV may provide its software in the form or a parcel, or they may have a different way of installing their software
onto your cluster. If their installation process is not via a parcel, then you should install their software before adding
the CSD file. Follow the instructions from the ISV for installing the software, if you have not done so already. If the ISV
has provided their software as a parcel, they may also have included the location of their parcel repository in the CSD
they have provided. In that case, install the CSD first and then install the parcel.
Installing the Custom Service Descriptor File
1.
2.
3.
4.
Acquire the CSD file from Cloudera or an ISV.
Log on to the Cloudera Manager Server host, and place the CSD file under the location configured for CSD files.
Set the file ownership to cloudera-scm:cloudera-scm with permission 644.
Restart the Cloudera Manager Server:
service cloudera-scm-server restart
5. Log into the Cloudera Manager Admin Console and restart the Cloudera Management Service.
38 | Cloudera Administration
Managing CDH and Managed Services
a. Do one of the following:
•
1. Select Clusters > Cloudera Management Service > Cloudera Management Service.
2. Select Actions > Restart.
• On the Home > Status tab, click
to the right of Cloudera Management Service and select Restart.
b. Click Restart to confirm. The Command Details window shows the progress of stopping and then starting
the roles.
c. When Command completed with n/n successful subcommands appears, the task is complete. Click Close.
Installing the Parcel
If you have already installed the external software onto your cluster, you can skip these steps and proceed to Adding
an Add-on Service on page 39.
1. Click in the main navigation bar. If the vendor has included the location of the repository in the CSD, the parcel
should already be present and ready for downloading. If the parcel is available, skip to step 7.
2. Use one of the following methods to open the parcel settings page:
• Navigation bar
1. Click in the top navigation bar or click Hosts and click the Parcels tab.
2. Click the Edit Settings button.
• Menu
1. Select Administration > Settings.
2. Select Category > Parcels .
3. In the Remote Parcel Repository URLs list, click to open an additional row.
4. Enter the path to the repository.
5. Click Save Changes to commit the changes.
6. Click . The external parcel should appear in the set of parcels available for download.
7. Download, distribute, and activate the parcel. See Managing Parcels.
Adding an Add-on Service
Add the service following the procedure in Adding a Service on page 36.
Uninstalling an Add-on Service
1. Stop all instances of the service.
2. Delete the service from all clusters. If there are other services that depend on the service you are trying to delete,
you must delete those services first.
3. Log on to the Cloudera Manager Server host and remove the CSD file.
4. Restart the Cloudera Manager Server:
service cloudera-scm-server restart
5. After the server has restarted, log into the Cloudera Manager Admin Console and restart the Cloudera Management
Service.
6. Optionally remove the parcel.
Starting, Stopping, and Restarting Services
Minimum Required Role: Operator (also provided by Configurator, Cluster Administrator, Full Administrator)
Cloudera Administration | 39
Managing CDH and Managed Services
Starting and Stopping Services
It's important to start and stop services that have dependencies in the correct order. For example, because MapReduce
and YARN have a dependency on HDFS, you must start HDFS before starting MapReduce or YARN. The Cloudera
Management Service and Hue are the only two services on which no other services depend; although you can start
and stop them at anytime, their preferred order is shown in the following procedures.
The Cloudera Manager cluster actions start and stop services in the correct order. To start or stop all services in a
cluster, follow the instructions in Starting, Stopping, Refreshing, and Restarting a Cluster on page 33.
Starting a Service on All Hosts
1. On the Home > Status tab, click
to the right of the service name and select Start.
2. Click Start that appears in the next screen to confirm. When you see a Finished status, the service has started.
The order in which to start services is:
1. Cloudera Management Service
2. ZooKeeper
3. HDFS
4. Solr
5. Flume
6. HBase
7. Key-Value Store Indexer
8. MapReduce or YARN
9. Hive
10. Impala
11. Oozie
12. Sqoop
13. Hue
Note: If you are unable to start the HDFS service, it's possible that one of the roles instances, such
as a DataNode, was running on a host that is no longer connected to the Cloudera Manager Server
host, perhaps because of a hardware or network failure. If this is the case, the Cloudera Manager
Server will be unable to connect to the Cloudera Manager Agent on that disconnected host to start
the role instance, which will prevent the HDFS service from starting. To work around this, you can
stop all services, abort the pending command to start the role instance on the disconnected host, and
then restart all services again without that role instance. For information about aborting a pending
command, see Aborting a Pending Command on page 43.
Stopping a Service on All Hosts
1. On the Home > Status tab, click
to the right of the service name and select Stop.
2. Click Stop that appears in the next screen to confirm. When you see a Finished status, the service has stopped.
The order in which to stop services is:
1.
2.
3.
4.
Hue
Sqoop
Oozie
Impala
40 | Cloudera Administration
Managing CDH and Managed Services
5. Hive
6. MapReduce or YARN
7. Key-Value Store Indexer
8. HBase
9. Flume
10. Solr
11. HDFS
12. ZooKeeper
13. Cloudera Management Service
Restarting a Service
It is sometimes necessary to restart a service, which is essentially a combination of stopping a service and then starting
it again. For example, if you change the hostname or port where the Cloudera Manager is running, or you enable TLS
security, you must restart the Cloudera Management Service to update the URL to the Server.
1. On the Home > Status tab, click
to the right of the service name and select Restart.
2. Click Start on the next screen to confirm. When you see a Finished status, the service has restarted.
To restart all services, use the restart cluster action.
Rolling Restart
Minimum Required Role: Operator (also provided by Configurator, Cluster Administrator, Full Administrator)
Important: This feature is available only with a Cloudera Enterprise license; it is not available in
Cloudera Express. For information on Cloudera Enterprise licenses, see Managing Licenses on page
453.
Rolling restart allows you to conditionally restart the role instances of Flume, HBase, HDFS, Kafka, MapReduce, YARN,
and ZooKeeper services to update software or use a new configuration. If the service is not running, rolling restart is
not available for that service. You can do a rolling restart of each service individually.
If you have HDFS high availability enabled, you can also perform a cluster-level rolling restart. At the cluster level, the
rolling restart of worker hosts is performed on a host-by-host basis, rather than per service, to avoid all roles for a
service potentially being unavailable at the same time. During a cluster restart, in order to avoid having your NameNode
(and thus the cluster) being unavailable during the restart, Cloudera Manager will force a failover to the standby
NameNode.
MapReduce (MRv1) JobTracker High Availability on page 325 and YARN (MRv2) ResourceManager High Availability on
page 315 is not required for a cluster-level rolling restart. However, if you have JobTracker or ResourceManager high
availability enabled, Cloudera Manager will force a failover to the standby JobTracker or ResourceManager.
Performing a Service or Role Rolling Restart
You can initiate a rolling restart from either the Status page for one of the eligible services, or from the service's
Instances page, where you can select individual roles to be restarted.
1. Go to the service you want to restart.
2. Do one of the following:
• service - Select Actions > Rolling Restart.
• role 1. Click the Instances tab.
2. Select the roles to restart.
3. Select Actions for Selected > Rolling Restart.
Cloudera Administration | 41
Managing CDH and Managed Services
3. In the pop-up dialog box, select the options you want:
• Restart only roles whose configurations are stale
• Restart only roles that are running outdated software versions
• Which role types to restart
4. If you select an HDFS, HBase, MapReduce, or YARN service, you can have their worker roles restarted in batches.
You can configure:
• How many roles should be included in a batch - Cloudera Manager restarts the worker roles rack-by-rack in
alphabetical order, and within each rack, hosts are restarted in alphabetical order. If you are using the default
replication factor of 3, Hadoop tries to keep the replicas on at least 2 different racks. So if you have multiple
racks, you can use a higher batch size than the default 1. But you should be aware that using too high batch
size also means that fewer worker roles are active at any time during the upgrade, so it can cause temporary
performance degradation. If you are using a single rack only, you should only restart one worker node at a
time to ensure data availability during upgrade.
• How long should Cloudera Manager wait before starting the next batch.
• The number of batch failures that will cause the entire rolling restart to fail (this is an advanced feature). For
example if you have a very large cluster you can use this option to allow failures because if you know that
your cluster will be functional even if some worker roles are down.
Note:
• HDFS - If you do not have HDFS high availability configured, a warning appears reminding
you that the service will become unavailable during the restart while the NameNode is
restarted. Services that depend on that HDFS service will also be disrupted. It is recommended
that you restart the DataNodes one at a time—one host per batch, which is the default.
• HBase
– Administration operations such as any of the following should not be performed during
the rolling restart, to avoid leaving the cluster in an inconsistent state:
–
–
–
–
Split
Create, disable, enable, or drop table
Metadata changes
Create, clone, or restore a snapshot. Snapshots rely on the RegionServers being
up; otherwise the snapshot will fail.
– To increase the speed of a rolling restart of the HBase service, set the Region Mover
Threads property to a higher value. This increases the number of regions that can be
moved in parallel, but places additional strain on the HMaster. In most cases, Region
Mover Threads should be set to 5 or lower.
– Another option to increase the speed of a rolling restart of the HBase service is to set
the Skip Region Reload During Rolling Restart property to true. This setting can cause
regions to be moved around multiple times, which can degrade HBase client performance.
• MapReduce - If you restart the JobTracker, all current jobs will fail.
• YARN - If you restart ResourceManager and ResourceManager HA is enabled, current jobs
continue running: they do not restart or fail. ResourceManager HA is supported for CDH 5.2
and higher.
• ZooKeeper and Flume - For both ZooKeeper and Flume, the option to restart roles in batches
is not available. They are always restarted one by one.
5. Click Confirm to start the rolling restart.
42 | Cloudera Administration
Managing CDH and Managed Services
Performing a Cluster-Level Rolling Restart
You can perform a cluster-level rolling restart on demand from the Cloudera Manager Admin Console. A cluster-level
rolling restart is also performed as the last step in a rolling upgrade when the cluster is configured with HDFS high
availability enabled.
1. If you have not already done so, enable high availability. See HDFS High Availability on page 290 for instructions.
You do not need to enable automatic failover for rolling restart to work, though you can enable it if you wish.
Automatic failover does not affect the rolling restart operation.
2. For the cluster you want to restart select Actions > Rolling Restart.
3. In the pop-up dialog box, select the services you want to restart. Please review the caveats in the preceding section
for the services you elect to have restarted. The services that do not support rolling restart will simply be restarted,
and will be unavailable during their restart.
4. If you select an HDFS, HBase, or MapReduce service, you can have their worker roles restarted in batches. You
can configure:
• How many roles should be included in a batch - Cloudera Manager restarts the worker roles rack-by-rack in
alphabetical order, and within each rack, hosts are restarted in alphabetical order. If you are using the default
replication factor of 3, Hadoop tries to keep the replicas on at least 2 different racks. So if you have multiple
racks, you can use a higher batch size than the default 1. But you should be aware that using too high batch
size also means that fewer worker roles are active at any time during the upgrade, so it can cause temporary
performance degradation. If you are using a single rack only, you should only restart one worker node at a
time to ensure data availability during upgrade.
• How long should Cloudera Manager wait before starting the next batch.
• The number of batch failures that will cause the entire rolling restart to fail (this is an advanced feature). For
example if you have a very large cluster you can use this option to allow failures because if you know that
your cluster will be functional even if some worker roles are down.
5. Click Restart to start the rolling restart. While the restart is in progress, the Command Details page shows the
steps for stopping and restarting the services.
Aborting a Pending Command
Minimum Required Role: Operator (also provided by Configurator, Cluster Administrator, Full Administrator)
Commands will time out if they are unable to complete after a period of time.
If necessary, you can abort a pending command. For example, this may become necessary because of a hardware or
network failure where a host running a role instance becomes disconnected from the Cloudera Manager Server host.
In this case, the Cloudera Manager Server will be unable to connect to the Cloudera Manager Agent on that disconnected
host to start or stop the role instance which will prevent the corresponding service from starting or stopping. To work
around this, you can abort the command to start or stop the role instance on the disconnected host, and then you can
start or stop the service again.
To abort any pending command:
You can click the indicator ( ) with the blue badge, which shows the number of commands that are currently running
in your cluster (if any). This indicator is positioned just to the left of the Support link at the right hand side of the
navigation bar. Unlike the Commands tab for a role or service, this indicator includes all commands running for all
services or roles in the cluster. In the Running Commands window, click Abort to abort the pending command. For
more information, see Viewing Running and Recent Commands.
To abort a pending command for a service or role:
1. Go to the Service > Instances tab for the service where the role instance you want to stop is located. For example,
go to the HDFS Service > Instances tab if you want to abort a pending command for a DataNode.
2. In the list of instances, click the link for role instance where the command is running (for example, the instance
that is located on the disconnected host).
3. Go to the Commands tab.
4. Find the command in the list of Running Commands and click Abort Command to abort the running command.
Cloudera Administration | 43
Managing CDH and Managed Services
Deleting Services
Minimum Required Role: Full Administrator
1. Stop the service. For information on starting and stopping services, see Starting, Stopping, and Restarting Services
on page 39.
2. On the Home > Status tab, click
to the right of the service name and select Delete.
3. Click Delete to confirm the deletion. Deleting a service does not clean up the associated client configurations that
have been deployed in the cluster or the user data stored in the cluster. For a given "alternatives path" (for example
/etc/hadoop/conf) if there exist both "live" client configurations (ones that would be pushed out with deploy
client configurations for active services) and ones that have been "orphaned" client configurations (the service
they correspond to has been deleted), the orphaned ones will be removed from the alternatives database. In
other words, to trigger cleanup of client configurations associated with a deleted service you must create a service
to replace it. To remove user data, see Remove User Data.
Renaming a Service
Minimum Required Role: Full Administrator
A service is given a name upon installation, and that name is used as an identifier internally. However, Cloudera Manager
allows you to provide a display name for a service, and that name will appear in the Cloudera Manager Admin Console
instead of the original (internal) name.
1. On the Home > Status tab, click
to the right of the service name and select Rename.
2. Type the new name.
3. Click Rename.
The original service name will still be used internally, and may appear or be required in certain circumstances, such as
in log messages or in the API.
The rename action is recorded as an Audit event.
When looking at Audit or Event search results for the renamed service, it is possible that these search results might
contain either only the original (internal) name, or both the display name and the original name.
Configuring Maximum File Descriptors
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
You can setting the maximum file descriptor parameter for all daemon roles. When not specified, the role uses whatever
value it inherits from supervisor. When specified, configures soft and hard limits to the configured value.
1.
2.
3.
4.
5.
6.
Go to a service.
Click the Configuration tab.
In the Search box, type rlimit_fds.
Set the Maximum Process File Descriptors property for one or more roles.
Click Save Changes to commit the changes.
Restart the affected role instances.
Managing Roles
When Cloudera Manager configures a service, it configures hosts in your cluster with one or more functions (called
roles in Cloudera Manager) that are required for that service. The role determines which Hadoop daemons run on a
given host. For example, when Cloudera Manager configures an HDFS service instance it configures one host to run
the NameNode role, another host to run as the Secondary NameNode role, another host to run the Balancer role, and
some or all of the remaining hosts to run DataNode roles.
44 | Cloudera Administration
Managing CDH and Managed Services
Configuration settings are organized in role groups. A role group includes a set of configuration properties for a specific
group, as well as a list of role instances associated with that role group. Cloudera Manager automatically creates default
role groups.
For role types that allow multiple instances on multiple hosts, such as DataNodes, TaskTrackers, RegionServers (and
many others), you can create multiple role groups to allow one set of role instances to use different configuration
settings than another set of instances of the same role type. In fact, upon initial cluster setup, if you are installing on
identical hosts with limited memory, Cloudera Manager will (typically) automatically create two role groups for each
worker role — one group for the role instances on hosts with only other worker roles, and a separate group for the
instance running on the host that is also hosting master roles.
The HDFS service is an example of this: Cloudera Manager typically creates one role group (DataNode Default Group)
for the DataNode role instances running on the worker hosts, and another group (HDFS-1-DATANODE-1) for the
DataNode instance running on the host that is also running the master roles such as the NameNode, JobTracker, HBase
Master and so on. Typically the configurations for those two classes of hosts will differ in terms of settings such as
memory for JVMs.
Cloudera Manager configuration screens offer two layout options: classic and new. The new layout is the default;
however, on each configuration page you can easily switch between layouts using the Switch to XXX layout link at the
top right of the page. For more information, see Configuration Overview on page 8.
Gateway Roles
A gateway is a special type of role whose sole purpose is to designate a host that should receive a client configuration
for a specific service, when the host does not have any roles running on it. Gateway roles enable Cloudera Manager
to install and manage client configurations on that host. There is no process associated with a gateway role, and its
status will always be Stopped. You can configure gateway roles for HBase, HDFS, Hive, MapReduce, Solr, Spark, Sqoop
1 Client, and YARN.
Role Instances
Adding a Role Instance
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
After creating services, you can add role instances to the services. For example, after initial installation in which you
created the HDFS service, you can add a DataNode role instance to a host where one was not previously running. Upon
upgrading a cluster to a new version of CDH you might want to create a role instance for a role added in the new
version.
1. Go to the service for which you want to add a role instance. For example, to add a DataNode role instance, go to
the HDFS service.
2. Click the Instances tab.
3. Click the Add Role Instances button.
4. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of the
hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of hosts to
which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple hosts,
you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts dialog.
The following shortcuts for specifying hostname patterns are supported:
• Range of hostnames (without the domain portion)
Range Definition
Matching Hosts
10.1.1.[1-4]
10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4
host[1-3].company.com
host1.company.com, host2.company.com, host3.company.com
Cloudera Administration | 45
Managing CDH and Managed Services
Range Definition
Matching Hosts
host[07-10].company.com
host07.company.com, host08.company.com, host09.company.com,
host10.company.com
• IP addresses
• Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
5. Click Continue.
6. In the Review Changes page, review the configuration changes to be applied. Confirm the settings entered for file
system paths. The file paths required vary based on the services to be installed. For example, you might confirm
the NameNode Data Directory and the DataNode Data Directory for HDFS. Click Continue. The wizard finishes by
performing any actions necessary to prepare the cluster for the new role instances. For example, new DataNodes
are added to the NameNode dfs_hosts_allow.txt file. The new role instance is configured with the default
role group for its role type, even if there are multiple role groups for the role type. If you want to use a different
role group, follow the instructions in Managing Role Groups on page 48 for moving role instances to a different
role group. The new role instances are not started automatically.
Starting, Stopping, and Restarting Role Instances
Minimum Required Role: Operator (also provided by Configurator, Cluster Administrator, Full Administrator)
If the host for the role instance is currently decommissioned, you will not be able to start the role until the host has
been recommissioned.
1.
2.
3.
4.
Go to the service that contains the role instances to start, stop, or restart.
Click the Instances tab.
Check the checkboxes next to the role instances to start, stop, or restart (such as a DataNode instance).
Select Actions for Selected > Start, Stop, or Restart, and then click Start, Stop, or Restart again to start the process.
When you see a Finished status, the process has finished.
Also see Rolling Restart on page 41.
Decommissioning Role Instances
Minimum Required Role: Operator (also provided by Configurator, Cluster Administrator, Full Administrator)
You can remove a role instance such as a DataNode from a cluster while the cluster is running by decommissioning
the role instance. When you decommission a role instance, Cloudera Manager performs a procedure so that you can
safely retire a host without losing data. Role decommissioning applies to HDFS DataNode, MapReduce TaskTracker,
YARN NodeManager, and HBase RegionServer roles.
You cannot decommission a DataNode or a host with a DataNode if the number of DataNodes equals the replication
factor (which by default is three) of any file stored in HDFS. For example, if the replication factor of any file is three,
and you have three DataNodes, you cannot decommission a DataNode or a host with a DataNode. If you attempt to
decommission a DataNode or a host with a DataNode in such situations, the DataNode will be decommissioned, but
the decommission process will not complete. You will have to abort the decommission and recommission the DataNode.
A role will be decommissioned if its host is decommissioned. See Decommissioning and Recommissioning Hosts on
page 59 for more details.
To decommission role instances:
1. If you are decommissioning DataNodes, perform the steps in Tuning HDFS Prior to Decommissioning DataNodes
on page 59.
2. Click the service instance that contains the role instance you want to decommission.
3. Click the Instances tab.
4. Check the checkboxes next to the role instances to decommission.
46 | Cloudera Administration
Managing CDH and Managed Services
5. Select Actions for Selected > Decommission, and then click Decommission again to start the process. A
Decommission Command pop-up displays that shows each step or decommission command as it is run. In the
Details area, click to see the subcommands that are run. Depending on the role, the steps may include adding
the host to an "exclusions list" and refreshing the NameNode, JobTracker, or NodeManager; stopping the Balancer
(if it is running); and moving data blocks or regions. Roles that do not have specific decommission actions are
stopped.
You can abort the decommission process by clicking the Abort button, but you must recommission and restart
the role.
The Commission State facet in the Filters list displays
Decommissioning while decommissioning is in progress,
and
Decommissioned when the decommissioning process has finished. When the process is complete, a
added in front of Decommission Command.
is
Recommissioning Role Instances
Minimum Required Role: Operator (also provided by Configurator, Cluster Administrator, Full Administrator)
1.
2.
3.
4.
Click the service that contains the role instance you want to recommission.
Click the Instances tab.
Check the checkboxes next to the decommissioned role instances to recommission.
Select Actions for Selected > Recommission, and then click Recommission to start the process. A Recommission
Command pop-up displays that shows each step or recommission command as it is run. When the process is
complete, a is added in front of Recommission Command.
5. Restart the role instance.
Deleting Role Instances
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
1. Click the service instance that contains the role instance you want to delete. For example, if you want to delete
a DataNode role instance, click an HDFS service instance.
2. Click the Instances tab.
3. Check the checkboxes next to the role instances you want to delete.
4. If the role instance is running, select Actions for Selected > Stop and click Stop to confirm the action.
5. Select Actions for Selected > Delete. Click Delete to confirm the deletion.
Note: Deleting a role instance does not clean up the associated client configurations that have been
deployed in the cluster.
Configuring Roles to Use a Custom Garbage Collection Parameter
Every Java-based role in Cloudera Manager has a configuration setting called Java Configuration Options for role where
you can enter command line options. Commonly, garbage collection flags or extra debugging flags would be passed
here. To find the appropriate configuration setting, select the service you want to modify in the Cloudera Manager
Admin Console, then use the Search box to search for Java Configuration Options.
You can add configuration options for all instances of a given role by making this configuration change at the service
level. For example, to modify the setting for all DataNodes, select the HDFS service, then modify the Java Configuration
Options for DataNode setting.
To modify a configuration option for a given instance of a role, select the service, then select the particular role instance
(for example, a specific DataNode). The configuration settings you modify will apply to the selected role instance only.
For detailed instructions see Modifying Configuration Properties Using Cloudera Manager on page 10.
Cloudera Administration | 47
Managing CDH and Managed Services
Role Groups
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
A role group is a set of configuration properties for a role type, as well as a list of role instances associated with that
group. Cloudera Manager automatically creates a default role group named Role Type Default Group for each role
type.Each role instance can be associated with only a single role group.
Role groups provide two types of properties: those that affect the configuration of the service itself and those that
affect monitoring of the service, if applicable (the Monitoring subcategory). (Not all services have monitoring properties).
For more information about monitoring properties see Configuring Monitoring Settings.
When you run the installation or upgrade wizard, Cloudera Manager configures the default role groups it adds, and
adds any other required role groups for a given role type. For example, a DataNode role on the same host as the
NameNode might require a different configuration than DataNode roles running on other hosts. Cloudera Manager
creates a separate role group for the DataNode role running on the NameNode host and uses the default configuration
for DataNode roles running on other hosts.
You can modify the settings of the default role group, or you can create new role groups and associate role instances
to whichever role group is most appropriate. This simplifies the management of role configurations when one group
of role instances may require different settings than another group of instances of the same role type—for example,
due to differences in the hardware the roles run on. You modify the configuration for any of the service's role groups
through the Configuration tab for the service. You can also override the settings inherited from a role group for a role
instance.
If there are multiple role groups for a role type, you can move role instances from one group to another. When you
move a role instance to a different group, it inherits the configuration settings for its new group.
Creating a Role Group
1.
2.
3.
4.
5.
6.
Go to a service status page.
Click the Instances or Configuration tab.
Click Role Groups.
Click Create new group....
Provide a name for the group.
Select the role type for the group. You can select role types that allow multiple instances and that exist for the
service you have selected.
7. In the Copy From field, select the source of the basic configuration information for the role group:
• An existing role group of the appropriate type.
• None.... The role group is set up with generic default values that are not the same as the values Cloudera
Manager sets in the default role group, as Cloudera Manager specifically sets the appropriate configuration
properties for the services and roles it installs. After you create the group you must edit the configuration to
set missing properties (for example the TaskTracker Local Data Directory List property, which is not populated
if you select None) and clear other validation warnings and errors.
Managing Role Groups
You can rename or delete existing role groups, and move roles of the same role type from one group to another.
1.
2.
3.
4.
5.
Go to a service status page.
Click the Instances or Configuration tab.
Click Role Groups.
Click the group you want to manage. Role instances assigned to the role group are listed.
Perform the appropriate procedure for the action:
48 | Cloudera Administration
Managing CDH and Managed Services
Action
Rename
Procedure
1. Click the role group name, click
next to the name on the right and click Rename.
2. Specify the new name and click Rename.
Delete
You cannot delete any of the default groups. The group must first be empty; if you
want to delete a group you've created, you must move any role instances to a
different role group.
1. Click the role group name.
2. Click
next to the role group name on the right, select Delete, and confirm by clicking
Delete. Deleting a role group removes it from host templates.
Move
1. Select the role instance(s) to move.
2. Select Actions for Selected > Move To Different Role Group....
3. In the pop-up that appears, select the target role group and click Move.
Managing Hosts
Cloudera Manager provides a number of features that let you configure and manage the hosts in your clusters.
The Hosts screen has the following tabs:
The Status Tab
Viewing All Hosts
To display summary information about all the hosts managed by Cloudera Manager, click Hosts in the main navigation
bar. The All Hosts page displays with a list of all the hosts managed by Cloudera Manager.
The list of hosts shows the overall status of the Cloudera Manager-managed hosts in your cluster.
• The information provided varies depending on which columns are selected. To change the columns, click the
Columns: n Selected drop-down and select the checkboxes next to the columns to display.
• Clicking the to the left of the number of roles lists all the role instances running on that host. The balloon
annotation that appears when you move the cursor over a link indicates the service instance to which the role
belongs.
• Filter the hosts by typing a property value in the Search box or selecting a value from the facets at the left of the
page. If the Configuring Agent Heartbeat and Health Status Options on page 439 are configured as follows:
– Send Agent heartbeat every x
– Set health status to Concerning if the Agent heartbeats fail y
– Set health status to Bad if the Agent heartbeats fail z
The value v for a host's Last Heartbeat facet is computed as follows:
– v < x * y = Good
– v >= x * y and <= x * z = Concerning
– v >= x * z = Bad
Viewing the Hosts in a Cluster
Do one of the following:
Cloudera Administration | 49
Managing CDH and Managed Services
• Select Clusters > Cluster name > General > Hosts.
• In the Home screen, click
in a full form cluster table.
The All Hosts page displays with a list of the hosts filtered by the cluster name.
Viewing Individual Hosts
You can view detailed information about an individual host—resources (CPU/memory/storage) used and available,
which processes it is running, details about the host agent, and much more—by clicking a host link on the All Hosts
page. See Viewing Host Details on page 50.
The Configuration Tab
The Configuration tab lets you set properties related to parcels and to resource management, and also monitoring
properties for the hosts under management. The configuration settings you make here will affect all your managed
hosts. You can also configure properties for individual hosts from the Host Details page (see Viewing Host Details on
page 50) which will override the global properties set here).
To edit the Default configuration properties for hosts:
1. Click the Configuration tab.
For more information on making configuration changes, see Modifying Configuration Properties Using Cloudera Manager
on page 10.
The Roles and Disks Overview Tabs
Role Assignments
You can view the assignment of roles to hosts as follows:
1. Click the Roles tab.
2. Click a cluster name or All Clusters.
Disks Overview
Click the Disks Overview tab to display an overview of the status of all disks in the deployment. The statistics exposed
match or build on those in iostat, and are shown in a series of histograms that by default cover every physical disk
in the system.
Adjust the endpoints of the time line to see the statistics for different time periods. Specify a filter in the box to limit
the displayed data. For example, to see the disks for a single rack rack1, set the filter to: logicalPartition =
false and rackId = "rack1" and click Filter. Click a histogram to drill down and identify outliers. Mouse over
the graph and click to display additional information about the chart.
The Templates Tab
The Templates tab lets you create and manage host templates, which provide a way to specify a set of role configurations
that should be applied to a host. This greatly simplifies the process of adding new hosts, because it lets you specify the
configuration for multiple roles on a host in a single step, and then (optionally) start all those roles.
The Parcels Tab
In the Parcels tab you can download, distribute, and activate available parcels to your cluster. You can use parcels to
add new products to your cluster, or to upgrade products you already have installed.
Viewing Host Details
You can view detailed information about each host, including:
• Name, IP address, rack ID
• Health status of the host and last time the Cloudera Manager Agent sent a heartbeat to the Cloudera Manager
Server
50 | Cloudera Administration
Managing CDH and Managed Services
•
•
•
•
•
•
•
•
Number of cores
System load averages for the past 1, 5, and 15 minutes
Memory usage
File system disks, their mount points, and usage
Health test results for the host
Charts showing a variety of metrics and health test results over time.
Role instances running on the host and their health
CPU, memory, and disk resources used for each role instance
To view detailed host information:
1. Click the Hosts tab.
2. Click the name of one of the hosts. The Status page is displayed for the host you selected.
3. Click tabs to access specific categories of information. Each tab provides various categories of information about
the host, its services, components, and configuration.
From the status page you can view details about several categories of information.
Status
The Status page is displayed when a host is initially selected and provides summary information about the status of
the selected host. Use this page to gain a general understanding of work being done by the system, the configuration,
and health status.
If this host has been decommissioned or is in maintenance mode, you will see the following icon(s) (
bar of the page next to the status message.
,
) in the top
Details
This panel provides basic system configuration such as the host's IP address, rack, health status summary, and disk
and CPU resources. This information summarizes much of the detailed information provided in other panes on this
tab. To view details about the Host agent, click the Host Agent link in the Details section.
Health Tests
Cloudera Manager monitors a variety of metrics that are used to indicate whether a host is functioning as expected.
The Health Tests panel shows health test results in an expandable/collapsible list, typically with the specific metrics
that the test returned. (You can Expand All or Collapse All from the links at the upper right of the Health Tests panel).
• The color of the text (and the background color of the field) for a health test result indicates the status of the
results. The tests are sorted by their health status – Good, Concerning, Bad, or Disabled. The list of entries for
good and disabled health tests are collapsed by default; however, Bad or Concerning results are shown expanded.
• The text of a health test also acts as a link to further information about the test. Clicking the text will pop up a
window with further information, such as the meaning of the test and its possible results, suggestions for actions
you can take or how to make configuration changes related to the test. The help text for a health test also provides
a link to the relevant monitoring configuration section for the service. See Configuring Monitoring Settings for
more information.
Health History
The Health History provides a record of state transitions of the health tests for the host.
• Click the arrow symbol at the left to view the description of the health test state change.
• Click the View link to open a new page that shows the state of the host at the time of the transition. In this view
some of the status settings are greyed out, as they reflect a time in the past, not the current status.
File Systems
The File systems panel provides information about disks, their mount points and usage. Use this information to determine
if additional disk space is required.
Cloudera Administration | 51
Managing CDH and Managed Services
Roles
Use the Roles panel to see the role instances running on the selected host, as well as each instance's status and health.
Hosts are configured with one or more role instances, each of which corresponds to a service. The role indicates which
daemon runs on the host. Some examples of roles include the NameNode, Secondary NameNode, Balancer, JobTrackers,
DataNodes, RegionServers and so on. Typically a host will run multiple roles in support of the various services running
in the cluster.
Clicking the role name takes you to the role instance's status page.
You can delete a role from the host from the Instances tab of the Service page for the parent service of the role. You
can add a role to a host in the same way. See Role Instances on page 45.
Charts
Charts are shown for each host instance in your cluster.
See Viewing Charts for Cluster, Service, Role, and Host Instances for detailed information on the charts that are
presented, and the ability to search and display metrics of your choice.
Processes
The Processes page provides information about each of the processes that are currently running on this host. Use this
page to access management web UIs, check process status, and access log information.
Note: The Processes page may display exited startup processes. Such processes are cleaned up within
a day.
The Processes tab includes a variety of categories of information.
• Service - The name of the service. Clicking the service name takes you to the service status page. Using the triangle
to the right of the service name, you can directly access the tabs on the role page (such as the Instances, Commands,
Configuration, Audits, or Charts Library tabs).
• Instance - The role instance on this host that is associated with the service. Clicking the role name takes you to
the role instance's status page. Using the triangle to the right of the role name, you can directly access the tabs
on the role page (such as the Processes, Commands, Configuration, Audits, or Charts Library tabs) as well as the
status page for the parent service of the role.
• Name - The process name.
• Links - Links to management interfaces for this role instance on this system. These is not available in all cases.
• Status - The current status for the process. Statuses include stopped, starting, running, and paused.
• PID - The unique process identifier.
• Uptime - The length of time this process has been running.
• Full log file - A link to the full log (a file external to Cloudera Manager) for this host log entries for this host.
• Stderr - A link to the stderr log (a file external to Cloudera Manager) for this host.
• Stdout - A link to the stdout log (a file external to Cloudera Manager) for this host.
Resources
The Resources page provides information about the resources (CPU, memory, disk, and ports) used by every service
and role instance running on the selected host.
Each entry on this page lists:
•
•
•
•
The service name
The name of the particular instance of this service
A brief description of the resource
The amount of the resource being consumed or the settings for the resource
The resource information provided depends on the type of resource:
• CPU - An approximate percentage of the CPU resource consumed.
52 | Cloudera Administration
Managing CDH and Managed Services
• Memory - The number of bytes consumed.
• Disk - The disk location where this service stores information.
• Ports - The port number being used by the service to establish network connections.
Commands
The Commands page shows you running or recent commands for the host you are viewing. See Viewing Running and
Recent Commands for more information.
Configuration
Minimum Required Role: Full Administrator
The Configuration page for a host lets you set properties for the selected host. You can set properties in the following
categories:
• Advanced - Advanced configuration properties. These include the Java Home Directory, which explicitly sets the
value of JAVA_HOME for all processes. This overrides the auto-detection logic that is normally used.
• Monitoring - Monitoring properties for this host. The monitoring settings you make on this page will override the
global host monitoring settings you make on the Configuration tab of the Hosts page. You can configure monitoring
properties for:
–
–
–
–
health check thresholds
the amount of free space on the filesystem containing the Cloudera Manager Agent's log and process directories
a variety of conditions related to memory usage and other properties
alerts for health check events
For some monitoring properties, you can set thresholds as either a percentage or an absolute value (in bytes).
• Other - Other configuration properties.
• Parcels - Configuration properties related to parcels. Includes the Parcel Director property, the directory that
parcels will be installed into on this host. If the parcel_dir variable is set in the Agent's config.ini file, it will
override this value.
• Resource Management - Enables resource management using control groups (cgroups).
For more information, see the description for each or property or see Modifying Configuration Properties Using Cloudera
Manager on page 10.
Components
The Components page lists every component installed on this host. This may include components that have been
installed but have not been added as a service (such as YARN, Flume, or Impala).
This includes the following information:
• Component - The name of the component.
• Version - The version of CDH from which each component came.
• Component Version - The detailed version number for each component.
Audits
The Audits page lets you filter for audit events related to this host. See Lifecycle and Security Auditing for more
information.
Charts Library
The Charts Library page for a host instance provides charts for all metrics kept for that host instance, organized by
category. Each category is collapsible/expandable. See Viewing Charts for Cluster, Service, Role, and Host Instances
for more information.
Using the Host Inspector
Minimum Required Role: Full Administrator
Cloudera Administration | 53
Managing CDH and Managed Services
You can use the host inspector to gather information about hosts that Cloudera Manager is currently managing. You
can review this information to better understand system status and troubleshoot any existing issues. For example, you
might use this information to investigate potential DNS misconfiguration.
The inspector runs tests to gather information for functional areas including:
•
•
•
•
•
Networking
System time
User and group configuration
HDFS settings
Component versions
Common cases in which this information is useful include:
•
•
•
•
Installing components
Upgrading components
Adding hosts to a cluster
Removing hosts from a cluster
Running the Host Inspector
1. Click the Hosts tab.
2. Click Host Inspector. Cloudera Manager begins several tasks to inspect the managed hosts.
3. After the inspection completes, click Download Result Data or Show Inspector Results to review the results.
The results of the inspection displays a list of all the validations and their results, and a summary of all the components
installed on your managed hosts.
If the validation process finds problems, the Validations section will indicate the problem. In some cases the message
may indicate actions you can take to resolve the problem. If an issue exists on multiple hosts, you may be able to view
the list of occurrences by clicking a small triangle that appears at the end of the message.
The Version Summary section shows all the components that are available from Cloudera, their versions (if known)
and the CDH distribution to which they belong (CDH 4 or CDH 5).
If you are running multiple clusters with both CDH 4 and CDH 5, the lists will be organized by distribution (CDH 4 or
CDH 5). The hosts running that version are shown at the top of each list.
Viewing Past Host Inspector Results
You can view the results of a past host inspection by looking for the Host Inspector command using the Recent
Commands feature.
1. Click the Running Commands indicator ( ) just to the left of the Search box at the right hand side of the navigation
bar.
2. Click the Recent Commands button.
3. If the command is too far in the past, you can use the Time Range Selector to move the time range back to cover
the time period you want.
4. When you find the Host Inspector command, click its name to display its subcommands.
5. Click the Show Inspector Results button to view the report.
See Viewing Running and Recent Commands for more information about viewing past command activity.
Adding a Host to the Cluster
Minimum Required Role: Full Administrator
You can add one or more hosts to your cluster using the Add Hosts wizard, which installs the Oracle JDK, CDH, and
Cloudera Manager Agent software. After the software is installed and the Cloudera Manager Agent is started, the
Agent connects to the Cloudera Manager Server and you can use the Cloudera Manager Admin Console to manage
and monitor CDH on the new host.
54 | Cloudera Administration
Managing CDH and Managed Services
The Add Hosts wizard does not create roles on the new host; once you have successfully added the host(s) you can
either add roles, one service at a time, or apply a host template, which can define role configurations for multiple roles.
Important:
• All hosts in a single cluster must be running the same version of CDH.
• When you add a new host, you must install the same version of CDH to enable the new host to
work with the other hosts in the cluster. The installation wizard lets you select the version of CDH
to install, and you can choose a custom repository to ensure that the version you install matches
the version on the other hosts.
• If you are managing multiple clusters, select the version of CDH that matches the version in use
on the cluster where you plan to add the new host.
• When you add a new host, the following occurs:
– YARN topology.map is updated to include the new host
– Any service that includes topology.map in its configuration—Flume, Hive, Hue, Oozie, Solr,
Spark, Sqoop 2, YARN—is marked stale
At a convenient point after adding the host you should restart the stale services to pick up the
new configuration.
Use one of the following methods to add a new host:
Using the Add Hosts Wizard to Add Hosts
You can use the Add Hosts wizard to install CDH, Impala, and the Cloudera Manager Agent on a host.
Disable TLS Encryption or Authentication
If you have enabled TLS encryption or authentication for the Cloudera Manager Agents, you must disable both of them
before starting the Add Hosts wizard. Otherwise, skip to the next step.
Important: This step leaves the existing hosts in an unmanageable state; they are still configured to
use TLS, and so communicate with the Cloudera Manager Server.
1. From the Administration tab, select Settings.
2. Select the Security category.
3. Disable all levels of TLS that are currently enabled by deselecting the following options: Use TLS Encryption for
Agents, and Use TLS Authentication of Agents to Server.
4. Click Save Changes to save the settings.
5. Restart the Cloudera Management Server to have these changes take effect.
Using the Add Hosts Wizard
1. Click the Hosts tab.
2. Click the Add New Hosts button.
3. Follow the instructions in the wizard to install the Oracle JDK and Cloudera Manager Agent packages and start
the Agent.
4. In the Specify hosts for your CDH Cluster installation page, you can search for new hosts to add under the New
Hosts tab. However, if you have hosts that are already known to Cloudera Manager but have no roles assigned,
(for example, a host that was previously in your cluster but was then removed) these will appear under the
Currently Managed Hosts tab.
5. You will have an opportunity to add (and start) role instances to your newly-added hosts using a host template.
a. You can select an existing host template, or create a new one.
b. To create a new host template, click the + Create... button. This will open the Create New Host Template
pop-up. See Host Templates on page 57 for details on how you select the role groups that define the roles
Cloudera Administration | 55
Managing CDH and Managed Services
that should run on a host. When you have created the template, it will appear in the list of host templates
from which you can choose.
c. Select the host template you want to use.
d. By default Cloudera Manager will automatically start the roles specified in the host template on your newly
added hosts. To prevent this, uncheck the option to start the newly-created roles.
6. When the wizard is finished, you can verify the Agent is connecting properly with the Cloudera Manager Server
by clicking the Hosts tab and checking the health status for the new host. If the Health Status is Good and the
value for the Last Heartbeat is recent, then the Agent is connecting properly with the Cloudera Manager Server.
If you did not specify a host template during the Add Hosts wizard, then no roles will be present on your new hosts
until you add them. You can do this by adding individual roles under the Instances tab for a specific service, or by using
a host template. See Role Instances on page 45 for information about adding roles for a specific service. See Host
Templates on page 57 to create a host template that specifies a set of roles (from different services) that should run
on a host.
Enable TLS Encryption or Authentication
If you previously enabled TLS security on your cluster, you must re-enable the TLS options on the Administration page
and also configure TLS on each new host after using the Add Hosts wizard. Otherwise, you can ignore this step. For
instructions, see Configuring TLS Security for Cloudera Manager.
Enable TLS/SSL for CDH Components
If you have previously enabled TLS/SSL on your cluster, and you plan to start these roles on this new host, make sure
you install a new host certificate to be configured from the same path and naming convention as the rest of your hosts.
Since the new host and the roles configured on it are inheriting their configuration from the previous host, ensure that
the keystore or truststore passwords and locations are the same on the new host. For instructions on configuring
TLS/SSL, see Configuring TLS/SSL Encryption for CDH Services.
Enable Kerberos
If you have previously enabled Kerberos on your cluster:
• Install the packages required to kinit on the new host. For the list of packages required for each OS, see Kerberos
Prerequisites.
• If you have set up Cloudera Manager to manage krb5.conf, it will automatically deploy the file on the new host.
• If Cloudera Manager does not manage krb5.conf, you must manually update the file at /etc/krb5.conf.
Adding a Host by Installing the Packages Using Your Own Method
If you used a different mechanism to install the Oracle JDK, CDH, Cloudera Manager Agent packages, you can use that
same mechanism to install the Oracle JDK, CDH, Cloudera Manager Agent packages and then start the Cloudera Manager
Agent.
1. Install the Oracle JDK, CDH, and Cloudera Manager Agent packages using your own method. For instructions on
installing these packages, see Installation Path B - Manual Installation Using Cloudera Manager Packages.
2. After installation is complete, start the Cloudera Manager Agent. For instructions, see Starting, Stopping, and
Restarting Cloudera Manager Agents on page 438.
3. After the Agent is started, you can verify the Agent is connecting properly with the Cloudera Manager Server by
clicking the Hosts tab and checking the health status for the new host. If the Health Status is Good and the value
for the Last Heartbeat is recent, then the Agent is connecting properly with the Cloudera Manager Server.
4. If you have enabled TLS security on your cluster, you must enable and configure TLS on each new host. Otherwise,
ignore this step.
a. Enable and configure TLS on each new host by specifying 1 for the use_tls property in the
/etc/cloudera-scm-agent/config.ini configuration file.
b. Configure the same level(s) of TLS security on the new hosts by following the instructions in Configuring TLS
Security for Cloudera Manager.
56 | Cloudera Administration
Managing CDH and Managed Services
5. If you have previously enabled TLS/SSL on your cluster, and you plan to start these roles on this new host, make
sure you install a new host certificate to be configured from the same path and naming convention as the rest of
your hosts. Since the new host and the roles configured on it are inheriting their configuration from the previous
host, ensure that the keystore or truststore passwords and locations are the same on the new host. For instructions
on configuring TLS/SSL, see Configuring TLS/SSL Encryption for CDH Services.
6. If you have previously enabled Kerberos on your cluster:
• Install the packages required to kinit on the new host. For the list of packages required for each OS, see
Kerberos Prerequisites.
• If you have set up Cloudera Manager to manage krb5.conf, it will automatically deploy the file on the new
host.
• If Cloudera Manager does not manage krb5.conf, you must manually update the file at /etc/krb5.conf.
Specifying Racks for Hosts
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
To get maximum performance, it is important to configure CDH so that it knows the topology of your network. Network
locations such as hosts and racks are represented in a tree, which reflects the network “distance” between locations.
HDFS will use the network location to be able to place block replicas more intelligently to trade off performance and
resilience. When placing jobs on hosts, CDH will prefer within-rack transfers (where there is more bandwidth available)
to off-rack transfers; the MapReduce and YARN schedulers use network location to determine where the closest replica
is as input to a map task. These computations are performed with the assistance of rack awareness scripts.
Cloudera Manager includes internal rack awareness scripts, but you must specify the racks where the hosts in your
cluster are located. If your cluster contains more than 10 hosts, Cloudera recommends that you specify the rack for
each host. HDFS, MapReduce, and YARN will automatically use the racks you specify.
Cloudera Manager supports nested rack specifications. For example, you could specify the rack /rack3, or
/group5/rack3 to indicate the third rack in the fifth group. All hosts in a cluster must have the same number of path
components in their rack specifications.
To specify racks for hosts:
1.
2.
3.
4.
5.
Click the Hosts tab.
Check the checkboxes next to the host(s) for a particular rack, such as all hosts for /rack123.
Click Actions for Selected (n) > Assign Rack, where n is the number of selected hosts.
Enter a rack name or ID that starts with a slash /, such as /rack123 or /aisle1/rack123, and then click Confirm.
Optionally restart affected services. Rack assignments are not automatically updated for running services.
Host Templates
Minimum Required Role: Full Administrator
Host templates let you designate a set of role groups that can be applied in a single operation to a host or a set of
hosts. This significantly simplifies the process of configuring new hosts when you need to expand your cluster. Host
templates are supported for both CDH 4 and CDH 5 cluster hosts.
Important: A host template can only be applied on a host with a version of CDH that matches the
CDH version running on the cluster to which the host template belongs.
You can create and manage host templates under the Templates tab from the Hosts page.
1. Click the Hosts tab on the main Cloudera Manager navigation bar.
2. Click the Templates tab on the Hosts page.
Templates are not required; Cloudera Manager assigns roles and role groups to the hosts of your cluster when you
perform the initial cluster installation. However, if you want to add new hosts to your cluster, a host template can
make this much easier.
Cloudera Administration | 57
Managing CDH and Managed Services
If there are existing host templates, they are listed on the page, along with links to each role group included in the
template.
If you are managing multiple clusters, you must create separate host templates for each cluster, as the templates
specify role configurations specific to the roles in a single cluster. Existing host templates are listed under the cluster
to which they apply.
• You can click a role group name to be taken to the Edit configuration page for that role group, where you can
modify the role group settings.
• From the Actions menu associated with the template you can edit the template, clone it, or delete it.
Creating a Host Template
1. From the Templates tab, click Click here
2. In the Create New Host Template pop-up window that appears:
• Type a name for the template.
• For each role, select the appropriate role group. There may be multiple role groups for a given role type —
you want to select the one with the configuration that meets your needs.
3. Click Create to create the host template.
Editing a Host Template
1. From the Hosts tab, click the Templates tab.
2. Pull down the Actions menu for the template you want to modify, and click Edit. This put you into the Edit Host
Template pop-up window. This works exactly like the Create New Host Template window — you can modify they
template name or any of the role group selections.
3. Click OK when you have finished.
Applying a Host Template to a Host
You can use a host template to apply configurations for multiple roles in a single operation.
You can apply a template to a host that has no roles on it, or that has roles from the same services as those included
in the host template. New roles specified in the template that do not already exist on the host will be added. A role
on the host that is already a member of the role group specified in the template will be left unchanged. If a role on the
host matches a role in the template, but is a member of a different role group, it will be moved to the role group
specified by the template.
For example, suppose you have two role groups for a DataNode (DataNode Default Group and DataNode (1)). The host
has a DataNode role that belongs to DataNode Default Group. If you apply a host template that specifies the DataNode
(1) group, the role on the host will be moved from DataNode Default Group to DataNode (1).
However, if you have two instances of a service, such as MapReduce (for example, mr1 and mr2) and the host has a
TaskTracker role from service mr2, you cannot apply a TaskTracker role from service mr1.
A host may have no roles on it if you have just added the host to your cluster, or if you decommissioned a managed
host and removed its existing roles.
Also, the host must have the same version of CDH installed as is running on the cluster whose host templates you are
applying.
If a host belongs to a different cluster than the one for which you created the host template, you can apply the host
template if the "foreign" host either has no roles on it, or has only management roles on it. When you apply the host
template, the host will then become a member of the cluster whose host template you applied. The following instructions
assume you have already created the appropriate host template.
1.
2.
3.
4.
Go to the Hosts page, Status tab.
Select the host(s) to which you want to apply your host template.
From the Actions for Selected menu, select Apply Host Template.
In the pop-up window that appears, select the host template you want to apply.
58 | Cloudera Administration
Managing CDH and Managed Services
5. Optionally you can have Cloudera Manager start the roles created per the host template – check the box to enable
this.
6. Click Confirm to initiate the action.
Decommissioning and Recommissioning Hosts
Decommissioning a host decommissions and stops all roles on the host without requiring you to individually
decommission the roles on each service. Decommissioning applies to only to HDFS DataNode, MapReduce TaskTracker,
YARN NodeManager, and HBase RegionServer roles. If the host has other roles running on it, those roles are stopped.
After all roles on the host have been decommissioned and stopped, the host can be removed from service. You can
decommission multiple hosts in parallel.
Decommissioning Hosts
Minimum Required Role: Limited Operator (also provided by Operator, Configurator, Cluster Administrator, or Full
Administrator)
You cannot decommission a DataNode or a host with a DataNode if the number of DataNodes equals the replication
factor (which by default is three) of any file stored in HDFS. For example, if the replication factor of any file is three,
and you have three DataNodes, you cannot decommission a DataNode or a host with a DataNode. If you attempt to
decommission a DataNode or a host with a DataNode in such situations, the DataNode will be decommissioned, but
the decommission process will not complete. You will have to abort the decommission and recommission the DataNode.
To decommission hosts:
1.
2.
3.
4.
If the host has a DataNode, perform the steps in Tuning HDFS Prior to Decommissioning DataNodes on page 59.
Click the Hosts tab.
Select the checkboxes next to one or more hosts.
Select Actions for Selected > Hosts Decommission.
A confirmation pop-up informs you of the roles that will be decommissioned or stopped on the hosts you have
selected.
5. Click Confirm. A Decommission Command pop-up displays that shows each step or decommission command as
it is run, service by service. In the Details area, click to see the subcommands that are run for decommissioning
a given service. Depending on the service, the steps may include adding the host to an "exclusions list" and
refreshing the NameNode, JobTracker, or NodeManager; stopping the Balancer (if it is running); and moving data
blocks or regions. Roles that do not have specific decommission actions are stopped.
You can abort the decommission process by clicking the Abort button, but you must recommission and restart
each role that has been decommissioned.
The Commission State facet in the Filters lists displays
Decommissioning while decommissioning is in progress,
and
Decommissioned when the decommissioning process has finished. When the process is complete, a
added in front of Decommission Command.
is
You cannot start roles on a decommissioned host.
Note: When a DataNode is decommissioned, the data blocks are not removed from the storage
directories. You must delete the data manually.
Tuning HDFS Prior to Decommissioning DataNodes
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
When a DataNode is decommissioned, the NameNode ensures that every block from the DataNode will still be available
across the cluster as dictated by the replication factor. This procedure involves copying blocks from the DataNode in
small batches. If a DataNode has thousands of blocks, decommissioning can take several hours. Before decommissioning
hosts with DataNodes, you should first tune HDFS:
Cloudera Administration | 59
Managing CDH and Managed Services
1. Raise the heap size of the DataNodes. DataNodes should be configured with at least 4 GB heap size to allow for
the increase in iterations and max streams.
a.
b.
c.
d.
e.
Go to the HDFS service page.
Click the Configuration tab.
Select Scope > DataNode.
Select Category > Resource Management.
Set the Java Heap Size of DataNode in Bytes property as recommended.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
f. Click Save Changes to commit the changes.
2. Set the DataNode balancing bandwidth:
a.
b.
c.
d.
Select Scope > DataNode.
Expand the Category > Performance category.
Configure the DataNode Balancing Bandwidth property to the bandwidth you have on your disks and network.
Click Save Changes to commit the changes.
3. Increase the replication work multiplier per iteration to a larger number (the default is 2, however 10 is
recommended):
a. Select Scope > NameNode.
b. Expand the Category > Advanced category.
c. Configure the Replication Work Multiplier Per Iteration property to a value such as 10.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
d. Click Save Changes to commit the changes.
4. Increase the replication maximum threads and maximum replication thread hard limits:
a. Select Scope > NameNode.
b. Expand the Category > Advanced category.
c. Configure the Maximum number of replication threads on a DataNode and Hard limit on the number of
replication threads on a DataNode properties to 50 and 100 respectively.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
d. Click Save Changes to commit the changes.
5. Restart the HDFS service.
Tuning HBase Prior to Decommissioning DataNodes
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
To increase the speed of a rolling restart of the HBase service, set the Region Mover Threads property to a higher
value. This increases the number of regions that can be moved in parallel, but places additional strain on the HMaster.
In most cases, Region Mover Threads should be set to 5 or lower.
Recommissioning Hosts
Minimum Required Role: Operator (also provided by Configurator, Cluster Administrator, Full Administrator)
Only hosts that are decommissioned using Cloudera Manager can be recommissioned.
1. Click the Hosts tab.
2. Select one or more hosts to recommission.
60 | Cloudera Administration
Managing CDH and Managed Services
3. Select Actions for Selected > Recommission and Confirm. A Recommission Command pop-up displays that shows
each step or recommission command as it is run. When the process is complete, a is added in front of
Recommission Command. The host and roles are marked as commissioned, but the roles themselves are not
restarted.
Restarting All The Roles on a Host
Minimum Required Role: Operator (also provided by Configurator, Cluster Administrator, Full Administrator)
1. Click the Hosts tab.
2. Select one or more hosts on which to start all roles.
3. Select Actions for Selected > Start Roles on Hosts.
Deleting Hosts
Minimum Required Role: Full Administrator
You can remove a host from a cluster in two ways:
• Delete the host entirely from Cloudera Manager.
• Remove a host from a cluster, but leave it available to other clusters managed by Cloudera Manager.
Both methods decommission the hosts, delete roles, and remove managed service software, but preserve data
directories.
Deleting a Host from Cloudera Manager
1.
2.
3.
4.
In the Cloudera Manager Admin Console, click the Hosts tab.
Select the hosts to delete.
Select Actions for Selected > Decommission.
Stop the Agent on the host. For instructions, see Starting, Stopping, and Restarting Cloudera Manager Agents on
page 438.
5. In the Cloudera Manager Admin Console, click the Hosts tab.
6. Reselect the hosts you selected in step 2.
7. Select Actions for Selected > Delete.
Removing a Host From a Cluster
This procedure leaves the host managed by Cloudera Manager and preserves the Cloudera Management Service roles
(such as the Events Server, Activity Monitor, and so on).
1.
2.
3.
4.
In the Cloudera Manager Admin Console, click the Hosts tab.
Select the hosts to delete.
Select Actions for Selected > Remove From Cluster. The Remove Hosts From Cluster dialog box displays.
Leave the selections to decommission roles and skip removing the Cloudera Management Service roles. Click
Confirm to proceed with removing the selected hosts.
Maintenance Mode
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Maintenance mode allows you to suppress alerts for a host, service, role, or an entire cluster. This can be useful when
you need to take actions in your cluster (make configuration changes and restart various elements) and do not want
to see the alerts that will be generated due to those actions.
Putting an entity into maintenance mode does not prevent events from being logged; it only suppresses the alerts that
those events would otherwise generate. You can see a history of all the events that were recorded for entities during
the period that those entities were in maintenance mode.
Cloudera Administration | 61
Managing CDH and Managed Services
Explicit and Effective Maintenance Mode
When you enter maintenance mode on an entity (cluster, service, or host) that has subordinate entities (for example,
the roles for a service) the subordinate entities are also put into maintenance mode. These are considered to be in
effective maintenance mode, as they have inherited the setting from the higher-level entity.
For example:
• If you set the HBase service into maintenance mode, then its roles (HBase Master and all RegionServers) are put
into effective maintenance mode.
• If you set a host into maintenance mode, then any roles running on that host are put into effective maintenance
mode.
Entities that have been explicitly put into maintenance mode show the icon
. Entities that have entered effective
maintenance mode as a result of inheritance from a higher-level entity show the icon
.
When an entity (role, host or service) is in effective maintenance mode, it can only be removed from maintenance
mode when the higher-level entity exits maintenance mode. For example, if you put a service into maintenance mode,
the roles associated with that service are entered into effective maintenance mode, and remain in effective maintenance
mode until the service exits maintenance mode. You cannot remove them from maintenance mode individually.
Alternatively, an entity that is in effective maintenance mode can be put into explicit maintenance mode. In this case,
the entity remains in maintenance mode even when the higher-level entity exits maintenance mode. For example,
suppose you put a host into maintenance mode, (which puts all the roles on that host into effective maintenance
mode). You then select one of the roles on that host and put it explicitly into maintenance mode. When you have the
host exit maintenance mode, that one role remains in maintenance mode. You need to select it individually and
specifically have it exit maintenance mode.
Viewing Maintenance Mode Status
You can view the status of Maintenance Mode in your cluster by clicking
to the right of the cluster name and selecting View Maintenance Mode Status.
Entering Maintenance Mode
You can enable maintenance mode for a cluster, service, role, or host.
Putting a Cluster into Maintenance Mode
1.
Click
to the right of the cluster name and select Enter Maintenance Mode.
2. Confirm that you want to do this.
The cluster is put into explicit maintenance mode, as indicated by the
entered into effective maintenance mode, as indicated by the
icon. All services and roles in the cluster are
icon.
Putting a Service into Maintenance Mode
1.
Click
to the right of the service name and select Enter Maintenance Mode.
2. Confirm that you want to do this.
The service is put into explicit maintenance mode, as indicated by the
effective maintenance mode, as indicated by the
62 | Cloudera Administration
icon.
icon. All roles for the service are entered into
Managing CDH and Managed Services
Putting Roles into Maintenance Mode
1.
2.
3.
4.
5.
Go to the service page that includes the role.
Go to the Instances tab.
Select the role(s) you want to put into maintenance mode.
From the Actions for Selected menu, select Enter Maintenance Mode.
Confirm that you want to do this.
The roles will be put in explicit maintenance mode. If the roles were already in effective maintenance mode (because
its service or host was put into maintenance mode) the roles will now be in explicit maintenance mode. This means
that they will not exit maintenance mode automatically if their host or service exits maintenance mode; they must be
explicitly removed from maintenance mode.
Putting a Host into Maintenance Mode
1.
2.
3.
4.
Go to the Hosts page.
Select the host(s) you want to put into maintenance mode.
From the Actions for Selected menu, select Enter Maintenance Mode.
Confirm that you want to do this.
The confirmation pop-up lists the role instances that will be put into effective maintenance mode when the host goes
into maintenance mode.
Exiting Maintenance Mode
When you exit maintenance mode, the maintenance mode icons are removed and alert notification resumes.
Exiting a Cluster from Maintenance Mode
1.
Click
to the right of the cluster name and select Exit Maintenance Mode.
2. Confirm that you want to do this.
Exiting a Service from Maintenance Mode
1.
Click
to the right of the service name and select Exit Maintenance Mode.
2. Confirm that you want to do this.
Exiting Roles from Maintenance Mode
1.
2.
3.
4.
5.
Go to the services page that includes the role.
Go to the Instances tab.
Select the role(s) you want to exit from maintenance mode.
From the Actions for Selected menu, select Exit Maintenance Mode.
Confirm that you want to do this.
Exiting a Host from Maintenance Mode
1.
2.
3.
4.
Go to the Hosts page.
Select the host(s) you want to put into maintenance mode.
From the Actions for Selected menu, select Exit Maintenance Mode.
Confirm that you want to do this.
The confirmation pop-up lists the role instances that will be removed from effective maintenance mode when the host
exits maintenance mode.
Cloudera Administration | 63
Managing CDH and Managed Services
Managing CDH Using the Command Line
The following sections provide instructions and information on managing core Hadoop.
For installation and upgrade instructions, see the Cloudera Installation and Upgrade guide, which also contains initial
deployment and configuration instructions for core Hadoop and the CDH components, including:
• Cluster configuration and maintenance:
–
–
–
–
–
Ports Used by Components of CDH 5
Configuring Network Names
Deploying CDH 5 on a Cluster
Starting CDH Services Using the Command Line on page 65
Stopping CDH Services Using the Command Line on page 70
• Avro Usage
• Flume configuration
• HBase configuration:
– Configuration Settings for HBase
– HBase Replication on page 409
– HBase Snapshots
• HCatalog configuration
• Impala configuration
• Hive configuration:
– Configuring the Hive Metastore
– Configuring HiveServer2
– Configuring the Metastore to use HDFS High Availability
• HttpFS configuration
• Hue: Configuring CDH Components for Hue
• Oozie configuration:
– Configuring Oozie
• Parquet: Using the Parquet File Format with Impala, Hive, Pig, HBase, and MapReduce
• Snappy:
–
–
–
–
–
Using Snappy for MapReduce Compression
Using Snappy for Pig Compression
Using Snappy for Hive Compression
Using Snappy Compression in Sqoop 1 and Sqoop 2 Imports
Using Snappy Compression with HBase
• Spark configuration:
– Managing Spark Standalone Using the Command Line on page 230
– Running Spark Applications
– Building and Running a Crunch Application with Spark
• Sqoop configuration:
– Setting HADOOP_MAPRED_HOME for Sqoop
– Configuring Sqoop 2
• ZooKeeper: Maintaining a ZooKeeper Server
64 | Cloudera Administration
Managing CDH and Managed Services
Starting CDH Services Using the Command Line
You need to start and stop services in the right order to make sure everything starts or stops cleanly.
Note: The Oracle JDK is required for all Hadoop components.
START services in this order:
Order
Service
Comments
For instructions and more
information
1
ZooKeeper
Cloudera recommends
starting ZooKeeper before
starting HDFS; this is a
requirement in a
high-availability (HA)
deployment. In any case,
always start ZooKeeper
before HBase.
Installing the ZooKeeper
Server Package and Starting
ZooKeeper on a Single
Server; Installing ZooKeeper
in a Production
Environment; Deploying
HDFS High Availability on
page 302; Configuring High
Availability for the
JobTracker (MRv1)
2
HDFS
Start HDFS before all other Deploying HDFS on a
services except ZooKeeper. Cluster; HDFS High
If you are using HA, see the Availability on page 290
CDH 5 High Availability
Guide for instructions.
3
HttpFS
4a
MRv1
Start MapReduce before
Hive or Oozie. Do not start
MRv1 if YARN is running.
Deploying MapReduce v1
(MRv1) on a Cluster;
Configuring High Availability
for the JobTracker (MRv1)
4b
YARN
Start YARN before Hive or
Oozie. Do not start YARN if
MRv1 is running.
Deploying MapReduce v2
(YARN) on a Cluster
5
HBase
6
Hive
7
Oozie
Starting the Oozie Server
8
Flume 1.x
Running Flume
9
Sqoop
Sqoop Installation and
Sqoop 2 Installation
10
Hue
Hue Installation
HttpFS Installation
Starting and Stopping HBase
on page 87; Deploying
HBase in a Distributed
Cluster
Start the Hive metastore
Installing Hive
before starting HiveServer2
and the Hive console.
Cloudera Administration | 65
Managing CDH and Managed Services
Configuring init to Start Hadoop System Services
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
init(8) starts some daemons when the system is booted. Depending on the distribution, init executes scripts from
either the /etc/init.d directory or the /etc/rc2.d directory. The CDH packages link the files in init.d and rc2.d
so that modifying one set of files automatically updates the other.
To start system services at boot time and on restarts, enable their init scripts on the systems on which the services
will run, using the appropriate tool:
• chkconfig is included in the RHEL and CentOS distributions. Debian and Ubuntu users can install the chkconfig
package.
• update-rc.d is included in the Debian and Ubuntu distributions.
Configuring init to Start Core Hadoop System Services in an MRv1 Cluster
Important:
Cloudera does not support running MRv1 and YARN daemons on the same nodes at the same time;
it will degrade performance and may result in an unstable cluster deployment.
The chkconfig commands to use are:
$ sudo chkconfig hadoop-hdfs-namenode on
The update-rc.d commands to use on Ubuntu and Debian systems are:
66 | Cloudera Administration
Managing CDH and Managed Services
Where
Command
On the NameNode
On the JobTracker
On the Secondary NameNode (if used)
On each TaskTracker
On each DataNode
$ sudo update-rc.d hadoop-hdfs-namenode
defaults
$ sudo update-rc.d
hadoop-0.20-mapreduce-jobtracker
defaults
$ sudo update-rc.d
hadoop-hdfs-secondarynamenode defaults
$ sudo update-rc.d
hadoop-0.20-mapreduce-tasktracker
defaults
$ sudo update-rc.d hadoop-hdfs-datanode
defaults
Configuring init to Start Core Hadoop System Services in a YARN Cluster
Important:
Do not run MRv1 and YARN on the same set of nodes at the same time. This is not recommended; it
degrades your performance and may result in an unstable MapReduce cluster deployment.
The chkconfig commands to use are:
Where
On the NameNode
On the ResourceManager
On the Secondary NameNode (if used)
On each NodeManager
On each DataNode
On the MapReduce JobHistory node
Command
$ sudo chkconfig hadoop-hdfs-namenode
on
$ sudo chkconfig
hadoop-yarn-resourcemanager on
$ sudo chkconfig
hadoop-hdfs-secondarynamenode on
$ sudo chkconfig
hadoop-yarn-nodemanager on
$ sudo chkconfig hadoop-hdfs-datanode
on
$ sudo chkconfig
hadoop-mapreduce-historyserver on
The update-rc.d commands to use on Ubuntu and Debian systems are:
Cloudera Administration | 67
Managing CDH and Managed Services
Where
Command
On the NameNode
$ sudo update-rc.d hadoop-hdfs-namenode
defaults
On the ResourceManager
$ sudo update-rc.d
hadoop-yarn-resourcemanager defaults
On the Secondary NameNode (if used)
$ sudo update-rc.d
hadoop-hdfs-secondarynamenode defaults
On each NodeManager
$ sudo update-rc.d
hadoop-yarn-nodemanager defaults
On each DataNode
$ sudo update-rc.d hadoop-hdfs-datanode
defaults
On the MapReduce JobHistory node
$ sudo update-rc.d
hadoop-mapreduce-historyserver defaults
Configuring init to Start Non-core Hadoop System Services
Non-core Hadoop daemons can also be configured to start at init time using the chkconfig or update-rc.d
command.
The chkconfig commands are:
Component
Server
Hue
Hue server
Oozie
Oozie server
HBase
HBase master
On each HBase RegionServer
Hive Metastore
Hive Metastore server
HiveServer2
HiveServer2
68 | Cloudera Administration
Command
$ sudo chkconfig hue on
$ sudo chkconfig oozie
on
$ sudo chkconfig
hbase-master on
$ sudo chkconfig
hbase-regionserver on
$ sudo chkconfig
hive-metastore on
$ sudo chkconfig hive-server2
on
Managing CDH and Managed Services
Component
Server
Zookeeper
Zookeeper server
HttpFS
HttpFS server
Command
$ sudo chkconfig
zookeeper-server on
$ sudo chkconfig
hadoop-httpfs on
The update-rc.d commands to use on Ubuntu and Debian systems are:
Component
Server
Hue
Hue server
Oozie
Oozie server
HBase
HBase master
HBase RegionServer
Hive Metastore
Hive Metastore server
HiveServer2
HiveServer2
Zookeeper
Zookeeper server
HttpFS
HttpFS server
Command
$ sudo update-rc.d hue
defaults
$ sudo update-rc.d oozie
defaults
$ sudo update-rc.d
hbase-master defaults
$ sudo update-rc.d
hbase-regionserver
defaults
$ sudo update-rc.d
hive-metastore defaults
$ sudo update-rc.d
hive-server2 defaults
$ sudo update-rc.d
zookeeper-server
defaults
$ sudo update-rc.d
hadoop-httpfs defaults
Starting and Stopping HBase Using the Command Line
When starting and stopping CDH services, order is important. See Starting CDH Services Using the Command Line on
page 65 and Stopping CDH Services Using the Command Line on page 70 for details. If you use Cloudera Manager,
follow these instructions instead.
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
Starting HBase
When starting HBase, it is important to start the HMaster, followed by the RegionServers, then the Thrift server.
Cloudera Administration | 69
Managing CDH and Managed Services
1. To start a HBase cluster using the command line, start the HBase Master by using the sudo hbase-master
start command on RHEL or SuSE, or the sudo hadoop-hbase-regionserver start command on Ubuntu
or Debian. The HMaster starts the RegionServers automatically.
2. To start a RegionServer manually, use the sudo hbase-regionserver start command on RHEL or SuSE, or
the sudo hadoop-hbase-regionserver start command on Ubuntu or Debian.
3. To start the Thrift server, use the hbase-thrift start on RHEL or SuSE, or the hadoop-hbase-thrift
start on Ubuntu or Debian.
Stopping HBase
When stopping HBase, it is important to stop the Thrift server, followed by each RegionServer, followed by any backup
HMasters, and finally the main HMaster.
1. Shut down the Thrift server by using the hbase-thrift stop command on the Thrift server host. sudo service
hbase-thrift stop
2. Shut down each RegionServer by using the hadoop-hbase-regionserver stop command on the RegionServer
host.
sudo service hadoop-hbase-regionserver stop
3. Shut down backup HMasters, followed by the main HMaster, by using the hbase-master stop command.
sudo service hbase-master stop
Stopping CDH Services Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
To shut down all Hadoop Common system services (HDFS, YARN, MRv1), run the following on each host in the cluster:
$ for x in `cd /etc/init.d ; ls hadoop-*` ; do sudo service $x stop ; done
To verify that no Hadoop processes are running, run the following command on each host in the cluster:
# ps -aef | grep java
To stop system services individually, use the instructions in the table below.
Important: Stop services in the order listed in the table. (You can start services in the reverse order.)
70 | Cloudera Administration
Managing CDH and Managed Services
Order
Service
Comments
Instructions
1
Hue
sudo service hue stop
2
Impala
sudo service impala-server stop
sudo service impala-catalog stop
sudo service impala-state-store stop
3
Oozie
4
Hive
sudo service oozie stop
Exit the Hive console and
ensure no Hive scripts are
running. Stop the Hive
server, HCatalog, and
metastore daemon on each
client.
There is no Flume master.
sudo service hiveserver2 stop
sudo service hive-webhcat-server stop
sudo service hive-metastore stop
5
Flume 1.x
6
Sqoop 1
sudo service sqoop-metastore stop
6
Sqoop 2
sudo service sqoop2-server stop
7
Lily HBase Indexer
(Solr/HBase
Indexer)
10
Spark
sudo service flume-ng-agent stop
sudo service hbase-solr-indexer stop
sudo service spark-worker stop
sudo service spark-history-server stop
sudo service spark-master stop
8
Sentry
9
Solr Search
10
HBase
Only present on a secure
configuration.
sudo service sentry-store stop
sudo service solr-server stop
Stop the Thrift server and
clients, followed by
RegionServers and finally
the Master.
sudo service hbase-thrift stop
sudo service hbase-rest stop
sudo service hbase-regionserver stop
sudo service hbase-master stop
11
MapReduce v1
Stop the JobTracker service, For MRv1 HA setup:
then stop the TaskTracker
Cloudera Administration | 71
Managing CDH and Managed Services
Order
Service
Comments
service on all nodes where
it is running.
Instructions
sudo service
hadoop-0.20-mapreduce-jobtrackerha stop
For Non-HA setup:
sudo service
hadoop-0.20-mapreduce-jobtracker stop
For all types of MRv1 setups:
sudo service
hadoop-0.20-mapreduce-tasktracker stop
12
YARN
Stop the JobHistory server,
followed by the
ResourceManager and each
of the NodeManagers.
$ sudo service
hadoop-mapreduce-historyserver stop
$ sudo service
hadoop-yarn-resourcemanager stop
$ sudo service hadoop-yarn-nodemanager
stop
13
HDFS
Stop HttpFS and the NFS
Gateway (if present).
Stop the Secondary
NameNode, then the
primary NameNode,
followed by Journal nodes
(if present) and then each of
DataNodes.
sudo service hadoop-httpfs stop
sudo service hadoop-hdfs-nfs3 stop
sudo service
hadoop-hdfs-secondarynamenode stop
sudo service hadoop-hdfs-namenode stop
sudo service hadoop-hdfs-journalnode
stop
sudo service hadoop-hdfs-datanode stop
14
KMS (Key
Management
Server)
15
ZooKeeper
Only present if HDFS at rest
encryption is enabled
sudo service hadoop-kms-server stop
sudo service zookeeper-server stop
Migrating Data between a CDH 4 and CDH 5 Cluster
You can migrate the data from a CDH 4 (or any Apache Hadoop) cluster to a CDH 5 cluster by using a tool that copies
out data in parallel, such as the DistCp tool offered in CDH 5. This can be useful if you are not planning to upgrade your
CDH 4 cluster itself at this point. The following sections provide information and instructions:
• Requirements and Restrictions for Data Migration between CDH 4 and CDH 5 on page 73
• Copying Data Between Two Clusters Using Distcp on page 73
• Post-Migration Verification
72 | Cloudera Administration
Managing CDH and Managed Services
• Ports Used by DistCp
Requirements and Restrictions for Data Migration between CDH 4 and CDH 5
1. The CDH 5 cluster must have a MapReduce service running on it (MRv1 or YARN (MRv2)).
2. All the MapReduce nodes in the CDH 5 cluster should have full network access to all the nodes of the source
cluster. This allows you to perform the copy in a distributed manner.
3. To copy data between a secure and an insecure cluster, you must run the distcp command on the secure cluster.
4. To copy data from a CDH 4 to a CDH 5 cluster, you can do one of the following:
Note:
The term source in this case refers to the CDH 4 (or other Hadoop) cluster you want to migrate
or copy data from; and destination refers to the CDH 5 cluster.
• Running commands on the destination cluster, use the Hftp protocol for the source cluster, and HDFS for the
destination. (Hftp is read-only, so you must run DistCp on the destination cluster and pull the data from the
source cluster.) See Copying Data Between Two Clusters Using Distcp on page 73.
Note:
Do not use this method if one of the clusters is secure and the other is not.
• Running commands on the source cluster, use the HDFS or webHDFS protocol for the source cluster, and
webHDFS for the destination. See Copying Data between a Secure and an Insecure Cluster using DistCp and
WebHDFS on page 77.
• Running commands on the destination cluster, use webHDFS for the source cluster, and webHDFS for the
destination. See Copying Data between a Secure and an Insecure Cluster using DistCp and WebHDFS on page
77.
The following restrictions currently apply (see Apache Hadoop Known Issues):
• DistCp does not work between a secure cluster and an insecure cluster in some cases.
As of CDH 5.1.3, DistCp does work between a secure and an insecure cluster if you use the webHDFS protocol and
run the command from the secure cluster side after setting ipc.client.fallback-to-simple-auth-allowed
to true, as described under Copying Data between a Secure and an Insecure Cluster using DistCp and WebHDFS
on page 77.
• To use DistCp using Hftp from a secure cluster using SPNEGO, you must configure the dfs.https.port property
on the client to use the HTTP port (50070 by default).
Copying Data Between Two Clusters Using Distcp
The Distcp Command
The distributed copy command, distcp, is a general utility for copying large data sets between distributed filesystems
within and across clusters. The distcp command submits a regular MapReduce job that performs a file-by-file copy.
To see the distcp command options, run the built-in help:
$ hadoop distcp
Important:
• Do not run distcp as the hdfs user which is blacklisted for MapReduce jobs by default.
• Do not use Hadoop shell commands (such as cp, copyfromlocal, put, get) for large copying
jobs or you may experience I/O bottlenecks.
Cloudera Administration | 73
Managing CDH and Managed Services
Distcp Syntax and Examples
You can use distcp to copy files between compatible clusters in either direction, from or to the source or destination
clusters.
For example, when upgrading, say from CDH 4 to CDH 5, you should run distcp from the CDH 5 cluster in this manner:
$ hadoop distcp hftp://cdh4-namenode:50070/ hdfs://CDH5-nameservice/
$ hadoop distcp s3a://bucket/ hdfs://CDH5-nameservice/
You can also use a specific path, such as /hbase to move HBase data, for example:
$ hadoop distcp hftp://cdh4-namenode:50070/hbase hdfs://CDH5-nameservice/hbase
$ hadoop distcp s3a://bucket/file hdfs://CDH5-nameservice/bucket/file
HFTP Protocol
The HFTP protocol allows you to use FTP resources in an HTTP request. When copying with distcp across different
versions of CDH, use hftp:// for the source file system and hdfs:// for the destination file system, and run distcp
from the destination cluster. The default port for HFTP is 50070 and the default port for HDFS is 8020.
Example of a source URI: hftp://namenode-location:50070/basePath
• hftp:// is the source protocol.
• namenode-location is the CDH 4 (source) NameNode hostname as defined by its configured fs.default.name.
• 50070 is the NameNode's HTTP server port, as defined by the configured dfs.http.address.
Example of a destination URI: hdfs://nameservice-id/basePath or hdfs://namenode-location
• hdfs:// is the destination protocol
• nameservice-id or namenode-location is the CDH 5 (destination) NameNode hostname as defined by its
configured fs.defaultFS.
• basePath in both examples refers to the directory you want to copy, if one is specifically needed.
Important:
• HFTP is a read-only protocol and can only be used for the source cluster, not the destination.
• HFTP cannot be used when copying with distcp from an insecure cluster to a secure cluster.
S3 Protocol
Amazon S3 block and native filesystems are also supported with the s3a:// protocol.
Example of an Amazon S3 Block Filesystem URI: s3a://bucket_name/path/to/file
S3 credentials can be provided in a configuration file (for example, core-site.xml):
<property>
<name>fs.s3a.access.key</name>
<value>...</value>
</property>
<property>
<name>fs.s3a.secret.key</name>
<value>...</value>
</property>
or run on the command line as follows:
hadoop distcp -Dfs.s3a.access.key=... -Dfs.s3a.secret.key=... s3a://
74 | Cloudera Administration
Managing CDH and Managed Services
Enabling Fallback Configuration
To enable the fallback configuration, for copying between secure and insecure clusters, add the following to the HDFS
configuration file, core-default.xml, by using an advanced configuration snippet if you use Cloudera Manager, or
editing the file directly otherwise.
<property>
<name>ipc.client.fallback-to-simple-auth-allowed</name>
<value>true</value>
</property>
Protocol Support for Distcp
The following table lists the protocols supported with the distcp command on different versions of CDH. "Secure"
means that the cluster is configured to use Kerberos.
Note: Copying between a secure cluster and an insecure cluster is only supported with CDH 5.1.3
and higher (CDH 5.1.3+) in accordance with HDFS-6776.
Source
Destination
Where to
Issue distcp
Command
Source
Protocol
Source
Config
Destination
Protocol
Destination
Config
CDH 4
CDH 4
Destination
hftp
Secure
hdfs or
webhdfs
Secure
CDH 4
CDH 4
Source or
Destination
hdfs or
webhdfs
Secure
hdfs or
webhdfs
Secure
CDH 4
CDH 4
Source or
Destination
hdfs or
webhdfs
Insecure
hdfs or
webhdfs
Insecure
CDH 4
CDH 4
Destination
hftp
Insecure
hdfs or
webhdfs
Insecure
CDH 4
CDH 5
Destination
webhdfs or
hftp
Secure
webhdfs or
hdfs
Secure
CDH 4
CDH 5.1.3+
Destination
webhdfs
Insecure
webhdfs
Secure
CDH 4
CDH 5
Destination
webhdfs or
hftp
Insecure
webhdfs or
hdfs
Insecure
CDH 4
CDH 5
Source
hdfs or
webhdfs
Insecure
webhdfs
Insecure
CDH 5
CDH 4
Source or
Destination
webhdfs
Secure
webhdfs
Secure
CDH 5
CDH 4
Source
hdfs
Secure
webhdfs
Secure
CDH 5.1.3+
CDH 4
Source
hdfs or
webhdfs
Secure
webhdfs
Insecure
CDH 5
CDH 4
Source or
Destination
webhdfs
Insecure
webhdfs
Insecure
CDH 5
CDH 4
Destination
webhdfs
Insecure
hdfs
Insecure
CDH 5
CDH 4
Source
hdfs
Insecure
webhdfs
Insecure
CDH 5
CDH 4
Destination
hftp
Insecure
hdfs or
webhdfs
Insecure
Fallback
Config
Required
Yes
Yes
Cloudera Administration | 75
Managing CDH and Managed Services
Source
Destination
Where to
Issue distcp
Command
Source
Protocol
Source
Config
Destination
Protocol
Destination
Config
Fallback
Config
Required
CDH 5
CDH 5
Source or
Destination
hdfs or
webhdfs
Secure
hdfs or
webhdfs
Secure
CDH 5
CDH 5
Destination
hftp
Secure
hdfs or
webhdfs
Secure
CDH 5.1.3+
CDH 5
Source
hdfs or
webhdfs
Secure
hdfs or
webhdfs
Insecure
Yes
CDH 5
CDH 5.1.3+
Destination
hdfs or
webhdfs
Insecure
hdfs or
webhdfs
Secure
Yes
CDH 5
CDH 5
Source or
Destination
hdfs or
webhdfs
Insecure
hdfs or
webhdfs
Insecure
CDH 5
CDH 5
Destination
hftp
Insecure
hdfs or
webhdfs
Insecure
Distcp between Secure Clusters in Distinct Kerberos Realms
This section explains how to copy data between two secure clusters in distinct Kerberos realms.
Note: JDK version 1.7.x is required on both clusters when copying data between Kerberized clusters
that are in different realms. For information about supported JDK versions, see Supported JDK Versions.
Specify the Destination Parameters in krb5.conf
Edit the krb5.conf file on the client (where the distcp job will be submitted) to include the destination hostname
and realm.
[realms]
HADOOP.QA.domain.COM = { kdc = kdc.domain.com:88 admin_server = admin.test.com:749
default_domain = domain.com supported_enctypes = arcfour-hmac:normal des-cbc-crc:normal
des-cbc-md5:normal des:normal des:v4 des:norealm des:onlyrealm des:afs3 }
[domain_realm]
.domain.com = HADOOP.test.domain.COM
domain.com = HADOOP.test.domain.COM
test03.domain.com = HADOOP.QA.domain.COM
Configure HDFS RPC Protection and Acceptable Kerberos Principal Patterns
Set the hadoop.rpc.protection property to authentication in both clusters. You can modify this property either
in hdfs-site.xml, or using Cloudera Manager as follows:
1.
2.
3.
4.
5.
6.
7.
Open the Cloudera Manager Admin Console.
Go to the HDFS service.
Click the Configuration tab.
Select Scope > HDFS-1 (Service-Wide)
Select Category > Security.
Locate the Hadoop RPC Protection property and select authentication.
Click Save Changes to commit the changes.
The following steps are not required if the two realms are already set up to trust each other, or have the same principal
pattern. However, this isn't usually the case.
76 | Cloudera Administration
Managing CDH and Managed Services
Set the dfs.namenode.kerberos.principal.pattern property to * to allow distcp irrespective of the principal
patterns of the source and destination clusters. You can modify this property either in hdfs-site.xml on both
clusters, or using Cloudera Manager as follows:
1.
2.
3.
4.
5.
6.
Open the Cloudera Manager Admin Console.
Go to the HDFS service.
Click the Configuration tab.
Select Scope > Gateway
Select Category > Advanced.
Edit the HDFS Client Advanced Configuration Snippet (Safety Valve) for hdfs-site.xml property to add:
<property>
<name>dfs.namenode.kerberos.principal.pattern</name>
<value>*</value>
</property>
7. Click Save Changes to commit the changes.
(If TLS/SSL is enabled) Specify Truststore Properties
The following properties must be configured in the ssl-client.xml file on the client submitting the distcp job to
establish trust between the target and destination clusters.
<property>
<name>ssl.client.truststore.location</name>
<value>path_to_truststore</value>
</property>
<property>
<name>ssl.client.truststore.password</name>
<value>XXXXXX</value>
</property>
<property>
<name>ssl.client.truststore.type</name>
<value>jks</value>
</property>
Set HADOOP_CONF to the Destination Cluster
Set the HADOOP_CONF path to be the destination environment. If you are not using HFTP, set the HADOOP_CONF path
to the source environment instead.
Launch Distcp
Kinit on the client and launch the distcp job.
hadoop distcp hdfs://test01.domain.com:8020/user/alice
hdfs://test02.domain.com:8020/user/alice
If launching distcp fails, force Kerberos to use TCP instead of UDP by adding the following parameter to the krb5.conf
file on the client.
[libdefaults]
udp_preference_limit = 1
Copying Data between a Secure and an Insecure Cluster using DistCp and WebHDFS
You can use DistCp and WebHDFS to copy data between a secure cluster and an insecure cluster by doing the following:
1. Set ipc.client.fallback-to-simple-auth-allowed to true in core-site.xml on the secure cluster side:
<property>
<name>ipc.client.fallback-to-simple-auth-allowed</name>
Cloudera Administration | 77
Managing CDH and Managed Services
<value>true</value>
</property>
2. Use commands such as the following from the secure cluster side only:
distcp webhdfs://insecureCluster webhdfs://secureCluster
distcp webhdfs://secureCluster webhdfs://insecureCluster
Post-migration Verification
After migrating data between the two clusters, it is a good idea to use hadoop fs -ls /basePath to verify the
permissions, ownership and other aspects of your files, and correct any problems before using the files in your new
cluster.
Managing Individual Services
The following sections cover the configuration and management of individual CDH and other services that have specific
and unique requirements or options.
Managing Flume
The Flume packages are installed by the Installation wizard, but the service is not created. This page documents how
to add, configure, and start the Flume service.
Adding a Flume Service
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
1. On the Home > Status tab, click
to the right of the cluster name and select Add a Service. A list of service types display. You can add one type of
service at a time.
2. Select the Flume service and click Continue.
3. Select the radio button next to the services on which the new service should depend. All services must depend
on the same ZooKeeper service. Click Continue.
4. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of the
hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of hosts to
which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple hosts,
you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts dialog.
The following shortcuts for specifying hostname patterns are supported:
• Range of hostnames (without the domain portion)
Range Definition
Matching Hosts
10.1.1.[1-4]
10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4
host[1-3].company.com
host1.company.com, host2.company.com, host3.company.com
host[07-10].company.com
host07.company.com, host08.company.com, host09.company.com,
host10.company.com
• IP addresses
• Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
78 | Cloudera Administration
Managing CDH and Managed Services
Configuring the Flume Agents
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
After you create a Flume service, you must first configure the agents before you start them. For detailed information
about Flume agent configuration, see the Flume User Guide.
The default Flume agent configuration provided in the Configuration File property of the Agent default role group is
a configuration for a single agent in a single tier; you should replace this with your own configuration. When you add
new agent roles, they are placed (initially) in the Agent default role group.
Agents that share the same configuration should be members of the same agent role group. You can create new role
groups and can move agents between them. If your Flume configuration has multiple tiers, you must create an agent
role group for each tier, and move each agent to be a member of the appropriate role group for their tier.
A Flume agent role group Configuration File property can contain the configuration for multiple agents, since each
configuration property is prefixed by the agent name. You can set the agents' names using configuration overrides to
change the name of a specific agent without changing its role group membership. Different agents can have the same
name — agent names do not have to be unique.
1. Go to the Flume service.
2. Click the Configuration tab.
3. Select Scope > Agent . Settings you make to the default role group apply to all agent instances unless you associate
those instances with a different role group, or override them for specific agents. See Modifying Configuration
Properties Using Cloudera Manager on page 10.
4. Set the Agent Name property to the name of the agent (or one of the agents) defined in the flume.conf
configuration file. The agent name can be comprised of letters, numbers, and the underscore character. You can
specify only one agent name here — the name you specify will be used as the default for all Flume agent instances,
unless you override the name for specific agents. You can have multiple agents with the same name — they will
share the configuration specified in on the configuration file.
5. Copy the contents of the flume.conf file, in its entirety, into the Configuration File property. Unless overridden
for specific agent instances, this property applies to all agents in the group. You can provide multiple agent
configurations in this file and use agent name overrides to specify which configuration to use for each agent.
Important: The name-value property pairs in the Configuration File property must include an
equal sign (=). For example, tier1.channels.channel1.capacity = 10000.
Setting a Flume Agent Name
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
If you have specified multiple agent configurations in a Flume agent role group Configuration File property, you can
set the agent name for an agent that uses a different configuration. Overriding the agent name will point the agent to
the appropriate properties specified in the agent configuration.
1.
2.
3.
4.
5.
Go to the Flume service.
Click the Configuration tab.
Select Scope > Agent.
Locate the Agent Name property or search for it by typing its name in the Search box.
Enter a name for the agent.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
6. Click Save Changes to commit the changes.
Cloudera Administration | 79
Managing CDH and Managed Services
Using Flume with HDFS or HBase Sinks
If you want to use Flume with HDFS or HBase sinks, you can add a dependency to that service from the Flume
configuration page. This will automatically add the correct client configurations to the Flume agent's classpath.
Note: If you are using Flume with HBase, make sure that the /etc/zookeeper/conf/zoo.cfg file
either does not exist on the host of the Flume agent that is using an HBase sink, or that it contains
the correct ZooKeeper quorum.
Using Flume with Solr Sinks
Cloudera Manager provides a set of configuration settings under the Flume service to configure the Flume Morphline
Solr Sink. See Configuring the Flume Morphline Solr Sink for Use with the Solr Service on page 227 for detailed instructions.
Updating Flume Agent Configurations
Minimum Required Role: Full Administrator
If you modify the Configuration File property after you have started the Flume service, update the configuration across
Flume agents as follows:
1. Go to the Flume service.
2. Select Actions > Update Config.
Managing HBase
Managing HBase
Cloudera Manager requires certain additional steps to set up and configure the HBase service.
Creating the HBase Root Directory
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
When adding the HBase service, the Add Service wizard automatically creates a root directory for HBase in HDFS. If
you quit the Add Service wizard or it does not finish, you can create the root directory outside the wizard by doing
these steps:
1. Choose Create Root Directory from the Actions menu in the HBase > Status tab.
2. Click Create Root Directory again to confirm.
Graceful Shutdown
Minimum Required Role: Operator (also provided by Configurator, Cluster Administrator, Full Administrator)
A graceful shutdown of an HBase RegionServer allows the regions hosted by that RegionServer to be moved to other
RegionServers before stopping the RegionServer. Cloudera Manager provides the following configuration options to
perform a graceful shutdown of either an HBase RegionServer or the entire service.
To increase the speed of a rolling restart of the HBase service, set the Region Mover Threads property to a higher
value. This increases the number of regions that can be moved in parallel, but places additional strain on the HMaster.
In most cases, Region Mover Threads should be set to 5 or lower.
Gracefully Shutting Down an HBase RegionServer
1.
2.
3.
4.
5.
Go to the HBase service.
Click the Instances tab.
From the list of Role Instances, select the RegionServer you want to shut down gracefully.
Select Actions for Selected > Decommission (Graceful Stop).
Cloudera Manager attempts to gracefully shut down the RegionServer for the interval configured in the Graceful
Shutdown Timeout configuration option, which defaults to 3 minutes. If the graceful shutdown fails, Cloudera
80 | Cloudera Administration
Managing CDH and Managed Services
Manager forcibly stops the process by sending a SIGKILL (kill -9) signal. HBase will perform recovery actions
on regions that were on the forcibly stopped RegionServer.
6. If you cancel the graceful shutdown before the Graceful Shutdown Timeout expires, you can still manually stop
a RegionServer by selecting Actions for Selected > Stop, which sends a SIGTERM (kill -5) signal.
Gracefully Shutting Down the HBase Service
1. Go to the HBase service.
2. Select Actions > Stop. This tries to perform an HBase Master-driven graceful shutdown for the length of the
configured Graceful Shutdown Timeout (three minutes by default), after which it abruptly shuts down the whole
service.
Configuring the Graceful Shutdown Timeout Property
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
This timeout only affects a graceful shutdown of the entire HBase service, not individual RegionServers. Therefore, if
you have a large cluster with many RegionServers, you should strongly consider increasing the timeout from its default
of 180 seconds.
1.
2.
3.
4.
5.
Go to the HBase service.
Click the Configuration tab.
Select Scope > HBASE-1 (Service Wide)
Use the Search box to search for the Graceful Shutdown Timeout property and edit the value.
Click Save Changes to save this setting.
Configuring the HBase Thrift Server Role
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
The Thrift Server role is not added by default when you install HBase, but it is required before you can use certain other
features such as the Hue HBase browser. To add the Thrift Server role:
1.
2.
3.
4.
Go to the HBase service.
Click the Instances tab.
Click the Add Role Instances button.
Select the host(s) where you want to add the Thrift Server role (you only need one for Hue) and click Continue.
The Thrift Server role should appear in the instances list for the HBase server.
5. Select the Thrift Server role instance.
6. Select Actions for Selected > Start.
Enabling HBase Indexing
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
HBase indexing is dependent on the Key-Value Store Indexer service. The Key-Value Store Indexer service uses the Lily
HBase Indexer Service to index the stream of records being added to HBase tables. Indexing allows you to query data
stored in HBase with the Solr service.
1.
2.
3.
4.
5.
6.
Go to the HBase service.
Click the Configuration tab.
Select Scope > HBASE-1 (Service Wide)
Select Category > Backup.
Select the Enable Replication and Enable Indexing properties.
Click Save Changes.
Adding a Custom Coprocessor
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Cloudera Administration | 81
Managing CDH and Managed Services
The HBase coprocessor framework provides a way to extend HBase with custom functionality. To configure these
properties in Cloudera Manager:
1.
2.
3.
4.
5.
6.
Select the HBase service.
Click the Configuration tab.
Select Scope > All.
Select Category > All.
Type HBase Coprocessor in the Search box.
You can configure the values of the following properties:
• HBase Coprocessor Abort on Error (Service-Wide)
• HBase Coprocessor Master Classes (Master Default Group)
• HBase Coprocessor Region Classes (RegionServer Default Group)
7. Click Save Changes to commit the changes.
Enabling Hedged Reads on HBase
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
1.
2.
3.
4.
5.
Go to the HBase service.
Click the Configuration tab.
Select Scope > HBASE-1 (Service-Wide).
Select Category > Performance.
Configure the HDFS Hedged Read Threadpool Size and HDFS Hedged Read Delay Threshold properties. The
descriptions for each of these properties on the configuration pages provide more information.
6. Click Save Changes to commit the changes.
Advanced Configuration for Write-Heavy Workloads
HBase includes several advanced configuration parameters for adjusting the number of threads available to service
flushes and compactions in the presence of write-heavy workloads. Tuning these parameters incorrectly can severely
degrade performance and is not necessary for most HBase clusters. If you use Cloudera Manager, configure these
options using the HBase Service Advanced Configuration Snippet (Safety Valve) for hbase-site.xml.
hbase.hstore.flusher.count
The number of threads available to flush writes from memory to disk. Never increase
hbase.hstore.flusher.count to more of 50% of the number of disks available to HBase. For example, if you
have 8 solid-state drives (SSDs), hbase.hstore.flusher.count should never exceed 4. This allows scanners
and compactions to proceed even in the presence of very high writes.
hbase.regionserver.thread.compaction.large and hbase.regionserver.thread.compaction.small
The number of threads available to handle small and large compactions, respectively. Never increase either of these
options to more than 50% of the number of disks available to HBase.
Ideally, hbase.regionserver.thread.compaction.small should be greater than or equal to
hbase.regionserver.thread.compaction.large, since the large compaction threads do more intense work
and will be in use longer for a given operation.
In addition to the above, if you use compression on some column families, more CPU will be used when flushing these
column families to disk during flushes or compaction. The impact on CPU usage depends on the size of the flush or the
amount of data to be decompressed and compressed during compactions.
Managing HBase Security
This topic pulls together content also found elsewhere which relates to configuring and using HBase in a secure
environment. For the most part, securing an HBase cluster is a one-way operation, and moving from a secure to an
unsecure configuration should not be attempted without contacting Cloudera support for guidance.
82 | Cloudera Administration
Managing CDH and Managed Services
HBase Authentication
Warning: Disabling security on a production HBase system is difficult and could cause data loss.
Contact Cloudera Support if you need to disable security in HBase.
To configure HBase security, complete the following tasks:
1. Configure HBase Authentication: You must establish a mechanism for HBase servers and clients to securely identify
themselves with HDFS, ZooKeeper, and each other (called authentication). This ensures that, for example, a host
claiming to be an HBase RegionServer or a particular HBase client are in fact who they claim to be.
2. Configure HBase Authorization: You must establish rules for the resources that clients are allowed to access
(called authorization). For more information, see Configuring HBase Authorization.
For more background information, see this blog post.
The following sections describe how to use Apache HBase and CDH 5 with Kerberos security on your Hadoop cluster:
• Configuring Kerberos Authentication for HBase
• Configuring Secure HBase Replication
• Configuring the HBase Client TGT Renewal Period
Important:
To enable HBase to work with Kerberos security on your Hadoop cluster, make sure you perform the
installation and configuration steps in Configuring Hadoop Security in CDH 5 and ZooKeeper Security
Configuration.
Note:
These instructions have been tested with CDH and MIT Kerberos 5 only.
Important:
Although an HBase Thrift server can connect to a secured Hadoop cluster, access is not secured from
clients to the HBase Thrift server.
Configuring HBase Authorization
Warning: Disabling security on a production HBase system is difficult and could cause data loss.
Contact Cloudera Support if you need to disable security in HBase.
After you have configured HBase authentication as described in the previous section, you must establish authorization
rules for the resources that a client is allowed to access. HBase currently allows you to establish authorization rules at
the table, column and cell-level. for Cell-level authorization was added as an experimental feature in CDH 5.2 and is
still considered experimental.
Understanding HBase Access Levels
HBase access levels are granted independently of each other and allow for different types of operations at a given
scope.
•
•
•
•
Read (R) - can read data at the given scope
Write (W) - can write data at the given scope
Execute (X) - can execute coprocessor endpoints at the given scope
Create (C) - can create tables or drop tables (even those they did not create) at the given scope
Cloudera Administration | 83
Managing CDH and Managed Services
• Admin (A) - can perform cluster operations such as balancing the cluster or assigning regions at the given scope
The possible scopes are:
• Superuser - superusers can perform any operation available in HBase, to any resource. The user who runs HBase
on your cluster is a superuser, as are any principals assigned to the configuration property hbase.superuser in
hbase-site.xml on the HMaster.
• Global - permissions granted at global scope allow the admin to operate on all tables of the cluster.
• Namespace - permissions granted at namespace scope apply to all tables within a given namespace.
• Table - permissions granted at table scope apply to data or metadata within a given table.
• ColumnFamily - permissions granted at ColumnFamily scope apply to cells within that ColumnFamily.
• Cell - permissions granted at Cell scope apply to that exact cell coordinate. This allows for policy evolution along
with data. To change an ACL on a specific cell, write an updated cell with new ACL to the precise coordinates of
the original. If you have a multi-versioned schema and want to update ACLs on all visible versions, you'll need to
write new cells for all visible versions. The application has complete control over policy evolution. The exception
is append and increment processing. Appends and increments can carry an ACL in the operation. If one is
included in the operation, then it will be applied to the result of the append or increment. Otherwise, the ACL
of the existing cell being appended to or incremented is preserved.
The combination of access levels and scopes creates a matrix of possible access levels that can be granted to a user.
In a production environment, it is useful to think of access levels in terms of what is needed to do a specific job. The
following list describes appropriate access levels for some common types of HBase users. It is important not to grant
more access than is required for a given user to perform their required tasks.
• Superusers - In a production system, only the HBase user should have superuser access. In a development
environment, an administrator may need superuser access in order to quickly control and manage the cluster.
However, this type of administrator should usually be a Global Admin rather than a superuser.
• Global Admins - A global admin can perform tasks and access every table in HBase. In a typical production
environment, an admin should not have Read or Write permissions to data within tables.
– A global admin with Admin permissions can perform cluster-wide operations on the cluster, such as balancing,
assigning or unassigning regions, or calling an explicit major compaction. This is an operations role.
– A global admin with Create permissions can create or drop any table within HBase. This is more of a DBA-type
role.
In a production environment, it is likely that different users will have only one of Admin and Create permissions.
Warning:
In the current implementation, a Global Admin with Admin permission can grant himself Read
and Write permissions on a table and gain access to that table's data. For this reason, only grant
Global Admin permissions to trusted user who actually need them.
Also be aware that a Global Admin with Create permission can perform a Put operation on
the ACL table, simulating a grant or revoke and circumventing the authorization check for
Global Admin permissions. This issue (but not the first one) is fixed in CDH 5.3 and higher, as
well as CDH 5.2.1. It is not fixed in CDH 4.x or CDH 5.1.x.
Due to these issues, be cautious with granting Global Admin privileges.
• Table Admins - A table admin can perform administrative operations only on that table. A table admin with Create
permissions can create snapshots from that table or restore that table from a snapshot. A table admin with Admin
permissions can perform operations such as splits or major compactions on that table.
• Users - Users can read or write data, or both. Users can also execute coprocessor endpoints, if given Executable
permissions.
84 | Cloudera Administration
Managing CDH and Managed Services
Important:
If you are using Kerberos principal names when setting ACLs for users, note that Hadoop uses only
the first part (short) of the Kerberos principal when converting it to the user name. Hence, for the
principal ann/fully.qualified.domain.name@YOUR-REALM.COM, HBase ACLs should only be
set for user ann.
Table 1: Real-World Example of Access Levels
This table shows some typical job descriptions at a hypothetical company and the permissions they might require in
order to get their jobs done using HBase.
Job Title
Scope
Permissions
Description
Senior Administrator
Global
Admin, Create
Manages the cluster and
gives access to Junior
Administrators.
Junior Administrator
Global
Create
Creates tables and gives
access to Table
Administrators.
Table Administrator
Table
Admin
Maintains a table from an
operations point of view.
Data Analyst
Table
Read
Creates reports from HBase
data.
Web Application
Table
Read, Write
Puts data into HBase and
uses HBase data to perform
operations.
Further Reading
• Access Control Matrix
• Security - Apache HBase Reference Guide
Enable HBase Authorization
HBase authorization is built on top of the Coprocessors framework, specifically AccessController Coprocessor.
Note: Once the Access Controller coprocessor is enabled, any user who uses the HBase shell will be
subject to access control. Access control will also be in effect for native (Java API) client access to
HBase.
Enable HBase Authorization Using Cloudera Manager
1.
2.
3.
4.
Go to Clusters and select the HBase cluster.
Select Configuration.
Search for HBase Secure Authorization and select it.
Search for HBase Service Advanced Configuration Snippet (Safety Valve) for hbase-site.xml and enter the following
into it to enable hbase.security.exec.permission.checks. Without this option, all users will continue to
have access to execute endpoint coprocessors. This option is not enabled when you enable HBase Secure
Authorization for backward compatibility.
<property>
<name>hbase.security.exec.permission.checks</name>
Cloudera Administration | 85
Managing CDH and Managed Services
<value>true</value>
</property>
5. Optionally, search for and configure HBase Coprocessor Master Classes and HBase Coprocessor Region Classes.
Enable HBase Authorization Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
To enable HBase authorization, add the following properties to the hbase-site.xml file on every HBase server host
(Master or RegionServer):
<property>
<name>hbase.security.authorization</name>
<value>true</value>
</property>
<property>
<name>hbase.security.exec.permission.checks</name>
<value>true</value>
</property>
<property>
<name>hbase.coprocessor.master.classes</name>
<value>org.apache.hadoop.hbase.security.access.AccessController</value>
</property>
<property>
<name>hbase.coprocessor.region.classes</name>
<value>org.apache.hadoop.hbase.security.token.TokenProvider,org.apache.hadoop.hbase.security.access.AccessController</value>
</property>
Configure Access Control Lists for Authorization
Now that HBase has the security coprocessor enabled, you can set ACLs using the HBase shell. Start the HBase shell
as usual.
Important:
The host running the shell must be configured with a keytab file as described in Configuring Kerberos
Authentication for HBase.
The commands that control ACLs take the following form. Group names are prefixed with the @ symbol.
hbase> grant <user> <permissions> [ @<namespace> [ <table>[ <column family>[ <column
qualifier> ] ] ] ]
# grants permissions
hbase> revoke <user> <permissions> [ @<namespace> [ <table> [ <column family> [ <column
qualifier> ] ] ] # revokes permissions
hbase> user_permission <table>
# displays existing permissions
In the above commands, fields encased in <> are variables, and fields in [] are optional. The permissions variable
must consist of zero or more character from the set "RWCA".
• R denotes read permissions, which is required to perform Get, Scan, or Exists calls in a given scope.
• W denotes write permissions, which is required to perform Put, Delete, LockRow, UnlockRow,
IncrementColumnValue, CheckAndDelete, CheckAndPut, Flush, or Compact in a given scope.
• X denotes execute permissions, which is required to execute coprocessor endpoints.
86 | Cloudera Administration
Managing CDH and Managed Services
• C denotes create permissions, which is required to perform Create, Alter, or Drop in a given scope.
• A denotes admin permissions, which is required to perform Enable, Disable, Snapshot, Restore, Clone,
Split, MajorCompact, Grant, Revoke, and Shutdown in a given scope.
Access Control List Example Commands
grant 'user1', 'RWC'
grant 'user2', 'RW', 'tableA'
grant 'user3', 'C', '@my_namespace'
Be sure to review the information in Understanding HBase Access Levels to understand the implications of the different
access levels.
Configuring the HBase Thrift Server Role
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
The Thrift Server role is not added by default when you install HBase, but it is required before you can use certain other
features such as the Hue HBase browser. To add the Thrift Server role:
1.
2.
3.
4.
Go to the HBase service.
Click the Instances tab.
Click the Add Role Instances button.
Select the host(s) where you want to add the Thrift Server role (you only need one for Hue) and click Continue.
The Thrift Server role should appear in the instances list for the HBase server.
5. Select the Thrift Server role instance.
6. Select Actions for Selected > Start.
Other HBase Security Topics
• Using BulkLoad On A Secure Cluster on page 119
• Configuring Secure HBase Replication
Starting and Stopping HBase
Use these instructions to start, stop, restart, rolling restart, or decommission HBase clusters or individual hosts.
Starting or Restarting HBase
You can start HBase hosts individually or as an entire cluster.
Starting or Restarting HBase Using Cloudera Manager
1. Go to the HBase service.
2. Click the Actions button and select Start.
3. To restart a running cluster, click Actions and select Restart or Rolling Restart. A rolling restart, which restarts
each RegionServer, one at a time, after a grace period. To configure the grace period, see Configuring the Graceful
Shutdown Timeout Property on page 81.
4. The Thrift service has no dependencies and can be restarted at any time. To stop or restart the Thrift service:
•
•
•
•
Go to the HBase service.
Select Instances.
Select the HBase Thrift Server instance.
Select Actions for Selected and select either Stop or Restart.
Cloudera Administration | 87
Managing CDH and Managed Services
Starting or Restarting HBase Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
If you need the ability to perform a rolling restart, Cloudera recommends managing your cluster with Cloudera Manager.
1. To start a HBase cluster using the command line, start the HBase Master by using the sudo hbase-master
start command on RHEL or SuSE, or the sudo hadoop-hbase-regionserver start command on Ubuntu
or Debian. The HMaster starts the RegionServers automatically.
2. To start a RegionServer manually, use the sudo hbase-regionserver start command on RHEL or SuSE, or
the sudo hadoop-hbase-regionserver start command on Ubuntu or Debian. Running multiple RegionServer
processes on the same host is not supported.
3. The Thrift service has no dependencies and can be restarted at any time. To start the Thrift server, use the
hbase-thrift start on RHEL or SuSE, or the hadoop-hbase-thrift start on Ubuntu or Debian.
Stopping HBase
You can stop a single HBase host, all hosts of a given type, or all hosts in the cluster.
Stopping HBase Using Cloudera Manager
1. To stop or decommission a single RegionServer:
a.
b.
c.
d.
Go to the HBase service.
Click the Instances tab.
From the list of Role Instances, select the RegionServer or RegionServers you want to stop or decommission.
Select Actions for Selected and select either Decommission (Graceful Stop) or Stop.
• Graceful Stop causes the regions to be redistributed to other RegionServers, increasing availability during
the RegionServer outage. Cloudera Manager waits for an interval determined by the Graceful Shutdown
timeout interval, which defaults to three minutes. If the graceful stop does not succeed within this
interval, the RegionServer is stopped with a SIGKILL (kill -9) signal. Recovery will be initiated on
affected regions.
• Stop happens immediately and does not redistribute the regions. It issues a SIGTERM (kill -5) signal.
2. To stop or decommission a single HMaster, select the Master and go through the same steps as above.
3. To stop or decommission the entire cluster, select the Actions button at the top of the screen (not Actions for
selected) and select Decommission (Graceful Stop) or Stop.
Stopping HBase Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
1. Shut down the Thrift server by using the hbase-thrift stop command on the Thrift server host. sudo service
hbase-thrift stop
88 | Cloudera Administration
Managing CDH and Managed Services
2. Shut down each RegionServer by using the hadoop-hbase-regionserver stop command on the RegionServer
host.
sudo service hadoop-hbase-regionserver stop
3. Shut down backup HMasters, followed by the main HMaster, by using the hbase-master stop command.
sudo service hbase-master stop
Configuring the HBase Canary
The HBase canary is an optional service that periodically checks that a RegionServer is alive. This canary is different
from the Cloudera Service Monitoring canary and is provided by the HBase service. The HBase canary is disabled by
default. After enabling the canary, you can configure several different thresholds and intervals relating to it, as well
as exclude certain tables from the canary checks. The canary works on Kerberos-enabled clusters if you have the HBase
client configured to use Kerberos.
Configure the HBase Canary Using Cloudera Manager
Minimum Required Role: Full Administrator
1.
2.
3.
4.
5.
Go to the HBase service.
Click the Configuration tab.
Select Scope > HBase or HBase Service-Wide.
Select Category > Monitoring.
Locate the HBase Canary property or search for it by typing its name in the Search box. Several properties have
Canary in the property name.
6. Select the checkbox.
7. Review other HBase Canary properties to configure the specific behavior of the canary.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
8. Click Save Changes to commit the changes.
9. Restart the role.
10. Restart the service.
Configure the HBase Canary Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
The HBase canary is a Java class. To run it from the command line, in the foreground, issue a command similar to the
following, as the HBase user:
$ /usr/bin/hbase org.apache.hadoop.hbase.tool.Canary
To start the canary in the background, add the --daemon option. You can also use this option in your HBase startup
scripts.
$ /usr/bin/hbase org.apache.hadoop.hbase.tool.Canary --daemon
The canary has many options. To see usage instructions, add the --help parameter:
$ /usr/bin/hbase org.apache.hadoop.hbase.tool.Canary --help
Cloudera Administration | 89
Managing CDH and Managed Services
Checking and Repairing HBase Tables
HBaseFsck (hbck) is a command-line tool that checks for region consistency and table integrity problems and repairs
corruption. It works in two basic modes — a read-only inconsistency identifying mode and a multi-phase read-write
repair mode.
• Read-only inconsistency identification: In this mode, which is the default, a report is generated but no repairs
are attempted.
• Read-write repair mode: In this mode, if errors are found, hbck attempts to repair them.
You can run hbck manually or configure the hbck poller to run hbck periodically.
Always run HBase administrative commands such as the HBase Shell, hbck, or bulk-load commands as the HBase user
(typically hbase).
Running hbck Manually
The hbck command is located in the bin directory of the HBase install.
• With no arguments, hbck checks HBase for inconsistencies and prints OK if no inconsistencies are found, or the
number of inconsistencies otherwise.
• With the -details argument, hbck checks HBase for inconsistencies and prints a detailed report.
• To limit hbck to only checking specific tables, provide them as a space-separated list: hbck <table1> <table2>
Warning: The following hbck options modify HBase metadata and are dangerous. They are not
coordinated by the HMaster and can cause further corruption by conflicting with commands that are
currently in progress or coordinated by the HMaster. Even if the HMaster is down, it may try to recover
the latest operation when it restarts. These options should only be used as a last resort. The hbck
command can only fix actual HBase metadata corruption and is not a general-purpose maintenance
tool. Before running these commands, consider contacting Cloudera Support for guidance. In addition,
running any of these commands requires a HMaster restart.
• If region-level inconsistencies are found, use the -fix argument to direct hbck to try to fix them. The following
sequence of steps is followed:
1. The standard check for inconsistencies is run.
2. If needed, repairs are made to tables.
3. If needed, repairs are made to regions. Regions are closed during repair.
• You can also fix individual region-level inconsistencies separately, rather than fixing them automatically with the
-fix argument.
– -fixAssignments repairs unassigned, incorrectly assigned or multiply assigned regions.
– -fixMeta removes rows from hbase:meta when their corresponding regions are not present in HDFS and
adds new meta rows if regions are present in HDFS but not in hbase:meta.
– -repairHoles creates HFiles for new empty regions on the filesystem and ensures that the new regions
are consistent.
– -fixHdfsOrphans repairs a region directory that is missing a region metadata file (the .regioninfo file).
– -fixHdfsOverlaps fixes overlapping regions. You can further tune this argument using the following
options:
– -maxMerge <n> controls the maximum number of regions to merge.
– -sidelineBigOverlaps attempts to sideline the regions which overlap the largest number of other
regions.
– -maxOverlapsToSideline <n> limits the maximum number of regions to sideline.
• To try to repair all inconsistencies and corruption at once, use the -repair option, which includes all the region
and table consistency options.
For more details about the hbck command, see Appendix C of the HBase Reference Guide.
90 | Cloudera Administration
Managing CDH and Managed Services
Hedged Reads
Hadoop 2.4 introduced a new feature called hedged reads. If a read from a block is slow, the HDFS client starts up
another parallel, 'hedged' read against a different block replica. The result of whichever read returns first is used, and
the outstanding read is cancelled. This feature helps in situations where a read occasionally takes a long time rather
than when there is a systemic problem. Hedged reads can be enabled for HBase when the HFiles are stored in HDFS.
This feature is disabled by default.
Enabling Hedged Reads for HBase Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
1.
2.
3.
4.
5.
Go to the HBase service.
Click the Configuration tab.
Select Scope > HBASE-1 (Service-Wide).
Select Category > Performance.
Configure the HDFS Hedged Read Threadpool Size and HDFS Hedged Read Delay Threshold properties. The
descriptions for each of these properties on the configuration pages provide more information.
6. Click Save Changes to commit the changes.
Enabling Hedged Reads for HBase Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
To enable hedged reads for HBase, edit the hbase-site.xml file on each server. Set
dfs.client.hedged.read.threadpool.size to the number of threads to dedicate to running hedged threads,
and set the dfs.client.hedged.read.threshold.millis configuration property to the number of milliseconds
to wait before starting a second read against a different block replica. Set
dfs.client.hedged.read.threadpool.size to 0 or remove it from the configuration to disable the feature.
After changing these properties, restart your cluster.
The following is an example configuration for hedged reads for HBase.
<property>
<name>dfs.client.hedged.read.threadpool.size</name>
<value>20</value> <!-- 20 threads -->
</property>
<property>
<name>dfs.client.hedged.read.threshold.millis</name>
<value>10</value> <!-- 10 milliseconds -->
</property>
Monitoring the Performance of Hedged Reads
You can monitor the performance of hedged reads using the following metrics emitted by Hadoop when hedged reads
are enabled.
• hedgedReadOps - the number of hedged reads that have occurred
• hedgeReadOpsWin - the number of times the hedged read returned faster than the original read
Configuring the Blocksize for HBase
The blocksize is an important configuration option for HBase. HBase data is stored in one (after a major compaction)
or more (possibly before a major compaction) HFiles per column family per region. It determines both of the following:
• The blocksize for a given column family determines the smallest unit of data HBase can read from the column
family's HFiles.
• It is also the basic unit of measure cached by a RegionServer in the BlockCache.
Cloudera Administration | 91
Managing CDH and Managed Services
The default blocksize is 64 KB. The appropriate blocksize is dependent upon your data and usage patterns. Use the
following guidelines to tune the blocksize size, in combination with testing and benchmarking as appropriate.
Warning: The default blocksize is appropriate for a wide range of data usage patterns, and tuning
the blocksize is an advanced operation. The wrong configuration can negatively impact performance.
• Consider the average key/value size for the column family when tuning the blocksize. You can find the average
key/value size using the HFile utility:
$ hbase org.apache.hadoop.hbase.io.hfile.HFile -f /path/to/HFILE -m -v
...
Block index size as per heapsize: 296
reader=hdfs://srv1.example.com:9000/path/to/HFILE, \
compression=none, inMemory=false, \
firstKey=US6683275_20040127/mimetype:/1251853756871/Put, \
lastKey=US6684814_20040203/mimetype:/1251864683374/Put, \
avgKeyLen=37, avgValueLen=8, \
entries=1554, length=84447
...
• Consider the pattern of reads to the table or column family. For instance, if it is common to scan for 500 rows on
various parts of the table, performance might be increased if the blocksize is large enough to encompass 500-1000
rows, so that often, only one read operation on the HFile is required. If your typical scan size is only 3 rows,
returning 500-1000 rows would be overkill.
It is difficult to predict the size of a row before it is written, because the data will be compressed when it is written
to the HFile. Perform testing to determine the correct blocksize for your data.
Configuring the Blocksize for a Column Family
You can configure the blocksize of a column family at table creation or by disabling and altering an existing table. These
instructions are valid whether or not you use Cloudera Manager to manage your cluster.
hbase>
hbase>
hbase>
hbase>
create ‘test_table ,{NAME => ‘test_cf , BLOCKSIZE => '262144'}
disable 'test_table'
alter 'test_table', {NAME => 'test_cf', BLOCKSIZE => '524288'}
enable 'test_table'
After changing the blocksize, the HFiles will be rewritten during the next major compaction. To trigger a major
compaction, issue the following command in HBase Shell.
hbase> major_compact 'test_table'
Depending on the size of the table, the major compaction can take some time and have a performance impact while
it is running.
Monitoring Blocksize Metrics
Several metrics are exposed for monitoring the blocksize by monitoring the blockcache itself.
Configuring the HBase BlockCache
In the default configuration, HBase uses a single on-heap cache. If you configure the off-heap BucketCache, the
on-heap cache is used for Bloom filters and indexes,and the off-heap BucketCache is used to cache data blocks. This
is referred to as the Combined Blockcache configuration. The Combined BlockCache allows you to use a larger
in-memory cache while reducing the negative impact of garbage collection in the heap, because HBase manages the
BucketCache, rather than relying on the garbage collector.
Contents of the BlockCache
In order to size the BlockCache correctly, you need to understand what HBase places into it.
92 | Cloudera Administration
Managing CDH and Managed Services
• Your data: Each time a Get or Scan operation occurs, the result is added to the BlockCache if it was not already
cached there. If you use the BucketCache, data blocks are always cached in the BucketCache.
• Row keys: When a value is loaded into the cache, its row key is also cached. This is one reason to make your row
keys as small as possible. A larger row key takes up more space in the cache.
• hbase:meta: The hbase:meta catalog table keeps track of which RegionServer is serving which regions. It can
consume several megabytes of cache if you have a large number of regions, and has in-memory access priority,
which means HBase attempts to keep it in the cache as long as possible.
• Indexes of HFiles: HBase stores its data in HDFS in a format called HFile. These HFiles contain indexes which allow
HBase to seek for data within them without needing to open the entire HFile. The size of an index is a factor of
the block size, the size of your row keys, and the amount of data you are storing. For big data sets, the size can
exceed 1 GB per RegionServer, although the entire index is unlikely to be in the cache at the same time. If you
use the BucketCache, indexes are always cached on-heap.
• Bloom filters: If you use Bloom filters, they are stored in the BlockCache. If you use the BucketCache, Bloom filters
are always cached on-heap.
The sum of the sizes of these objects is highly dependent on your usage patterns and the characteristics of your data.
For this reason, the HBase Web UI and Cloudera Manager each expose several metrics to help you size and tune the
BlockCache.
Deciding Whether To Use the BucketCache
The HBase team has published the results of exhaustive BlockCache testing, which revealed the following guidelines.
• If the result of a Get or Scan typically fits completely in the heap, the default configuration, which uses the on-heap
LruBlockCache, is the best choice, as the L2 cache will not provide much benefit. If the eviction rate is low,
garbage collection can be 50% less than that of the BucketCache, and throughput can be at least 20% higher.
• Otherwise, if your cache is experiencing a consistently high eviction rate, use the BucketCache, which causes
30-50% of the garbage collection of LruBlockCache when the eviction rate is high.
• BucketCache using file mode on solid-state disks has a better garbage-collection profile but lower throughput
than BucketCache using off-heap memory.
Bypassing the BlockCache
If the data needed for a specific but atypical operation does not all fit in memory, using the BlockCache can be
counter-productive, because data that you are still using may be evicted, or even if other data is not evicted, excess
garbage collection can adversely effect performance. For this type of operation, you may decide to bypass the BlockCache.
To bypass the BlockCache for a given Scan or Get, use the setCacheBlocks(false) method.
In addition, you can prevent a specific column family's contents from being cached, by setting its BLOCKCACHE
configuration to false. Use the following syntax in HBase Shell:
hbase> alter 'myTable', CONFIGURATION => {NAME => 'myCF', BLOCKCACHE => 'false'}
Cache Eviction Priorities
Both the on-heap cache and the off-heap BucketCache use the same cache priority mechanism to decide which cache
objects to evict in order to make room for new objects. Three levels of block priority allow for scan-resistance and
in-memory column families. Objects evicted from the cache are subject to garbage collection.
• Single access priority: The first time a block is loaded from HDFS, that block is given single access priority, which
means that it will be part of the first group to be considered during evictions. Scanned blocks are more likely to
be evicted than blocks that are used more frequently.
• Multi access priority: If a block in the single access priority group is accessed again, that block is assigned multi
access priority, which moves it to the second group considered during evictions, and is therefore less likely to be
evicted.
• In-memory access priority: If the block belongs to a column family which is configured with the in-memory
configuration option, its priority is changed to in memory access priority, regardless of its access pattern. This
group is the last group considered during evictions, but is not guaranteed not to be evicted. Catalog tables are
configured with in-memory access priority.
Cloudera Administration | 93
Managing CDH and Managed Services
To configure a column family for in-memory access, use the following syntax in HBase Shell:
hbase> alter 'myTable', 'myCF', CONFIGURATION => {IN_MEMORY => 'true'}
To use the Java API to configure a column family for in-memory access, use the
HColumnDescriptor.setInMemory(true) method.
Sizing the BlockCache
When you use the LruBlockCache, the blocks needed to satisfy each read are cached, evicting older cached objects
if the LruBlockCache is full. The size cached objects for a given read may be significantly larger than the actual result
of the read. For instance, if HBase needs to scan through 20 HFile blocks to return a 100 byte result, and the HFile
blocksize is 100 KB, the read will add 20 * 100 KB to the LruBlockCache.
Because the LruBlockCache resides entirely within the Java heap, the amount of which is available to HBase and
what percentage of the heap is available to the LruBlockCache strongly impact performance. By default, the amount
of HBase heap reserved for LruBlockCache (hfile.block.cache.size) is .40, or 40%. To determine the amount
of heap available for the LruBlockCache, use the following formula. The 0.99 factor allows 1% of heap to be available
as a "working area" for evicting items from the cache. If you use the BucketCache, the on-heap LruBlockCache only
stores indexes and Bloom filters, and data blocks are cached in the off-heap BucketCache.
number of RegionServers * heap size * hfile.block.cache.size * 0.99
To tune the size of the LruBlockCache, you can add RegionServers or increase the total Java heap on a given
RegionServer to increase it, or you can tune hfile.block.cache.size to reduce it. Reducing it will cause cache
evictions to happen more often, but will reduce the time it takes to perform a cycle of garbage collection. Increasing
the heap will cause garbage collection to take longer but happen less frequently.
About the off-heap BucketCache
If the BucketCache is enabled, it stores data blocks, leaving the on-heap cache free for storing indexes and Bloom
filters. The physical location of the BucketCache storage can be either in memory (off-heap) or in a file stored in a
fast disk.
• Off-heap: This is the default configuration.
• File-based: You can use the file-based storage mode to store the BucketCache on an SSD or FusionIO device,
Starting in CDH 5.4 (HBase 1.0), you can configure a column family to keep its data blocks in the L1 cache instead of
the BucketCache, using the HColumnDescriptor.cacheDataInL1(true) method or by using the following syntax
in HBase Shell:
hbase> alter 'myTable', CONFIGURATION => {CACHE_DATA_IN_L1 => 'true'}}
Configuring the off-heap BucketCache
This table summaries the important configuration properties for the BucketCache. To configure the BucketCache,
see Configuring the off-heap BucketCache Using Cloudera Manager on page 98 or Configuring the off-heap BucketCache
Using the Command Line on page 99. The table is followed by three diagrams that show the impacts of different
blockcache settings.
Table 2: BucketCache Configuration Properties
Property
Default
hbase.bucketcache.combinedcache.enabled true
Description
When BucketCache is enabled, use
it as a L2 cache for LruBlockCache.
If set to true, indexes and Bloom filters
are kept in the LruBlockCache and
the data blocks are kept in the
BucketCache
94 | Cloudera Administration
Managing CDH and Managed Services
Property
Default
Description
hbase.bucketcache.ioengine
none (BucketCache is disabled by
default)
Where to store the contents of the
BucketCache. Either onheap or
file:/path/to/file.
hfile.block.cache.size
0.4
A float between 0.0 and 1.0. This
factor multiplied by the Java heap size
is the size of the L1 cache. In other
words, the percentage of the Java
heap to use for the L1 cache.
hbase.bucketcache.size
not set
When using BucketCache, this is a
float that represents one of two
different values, depending on
whether it is a floating-point decimal
less than 1.0 or an integer greater than
1.0.
• If less than 1.0, it represents a
percentage of total heap memory
size to give to the cache.
• If greater than 1.0, it represents
the capacity of the cache in
megabytes
hbase.bucketcache.bucket.sizes 4, 8, 16, 32, 40, 48, 56, 64, A comma-separated list of sizes for
96, 128, 192, 256, 384, 512 buckets for the BucketCache if you
-XX:MaxDirectMemorySize
KB
prefer to use multiple sizes. The sizes
should be multiples of the default
blocksize, ordered from smallest to
largest. The sizes you use will depend
on your data patterns. This parameter
is experimental.
not set
A JVM option to configure the
maximum amount of direct memory
available to the JVM. If you use the
offheap block cache, this value should
be larger than the amount of memory
assigned to the BucketCache, plus
some extra memory to accommodate
buffers used for HDFS short-circuit
reads.
Cloudera Administration | 95
Managing CDH and Managed Services
96 | Cloudera Administration
Managing CDH and Managed Services
Cloudera Administration | 97
Managing CDH and Managed Services
Configuring the off-heap BucketCache Using Cloudera Manager
1. Go to the HBase service.
2. Click the Configuration tab.
3. Edit the parameter HBASE_OFFHEAPSIZE in the HBase Service Advanced Configuration Snippet for hbase-env.shand
set it to a value (such as 5G) which will accommodate your desired L2 cache size, in addition to space reserved for
cache management.
4. Edit the parameter HBASE_OPTS in the HBase Service Advanced Configuration Snippet for hbase-env.sh and add
the JVM option -XX:MaxDirectMemorySize=<size>G, replacing <size> with a value large enough to contain
your heap and off-heap BucketCache, expressed as a number of gigabytes.
5. Add the following settings to the HBase Service Advanced Configuration Snippet for hbase-site.xml, using values
appropriate to your situation. See Table 2: BucketCache Configuration Properties on page 94.
<property>
<name>hbase.bucketcache.ioengine</name>
<value>offheap</value>
</property>
<property>
<name>hfile.block.cache.size</name>
<value>0.2</value>
</property>
<property>
<name>hbase.bucketcache.size</name>
<value>4096</value>
</property>
6. Click Save Changes to commit the changes.
7. Restart or rolling restart your RegionServers for the changes to take effect.
98 | Cloudera Administration
Managing CDH and Managed Services
Configuring the off-heap BucketCache Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
1. First, verify the RegionServer's off-heap size, and if necessary, tune it by editing the hbase-env.sh file and adding
a line like the following:
HBASE_OFFHEAPSIZE=5G
Set it to a value which will accommodate your desired L2 cache size, in addition to space reserved for cache
management.
2. Edit the parameter HBASE_OPTS in the hbase-env.sh file and add the JVM option
-XX:MaxDirectMemorySize=<size>G, replacing <size> with a value large enough to contain your heap and
off-heap BucketCache, expressed as a number of gigabytes.
3. Next, configure the properties in Table 2: BucketCache Configuration Properties on page 94 as appropriate, using
the example below as a model.
<property>
<name>hbase.bucketcache.ioengine</name>
<value>offheap</value>
</property>
<property>
<name>hfile.block.cache.size</name>
<value>0.2</value>
</property>
<property>
<name>hbase.bucketcache.size</name>
<value>4194304</value>
</property>
4. Restart each RegionServer for the changes to take effect.
Monitoring the BlockCache
Cloudera Manager provides metrics to monitor the performance of the BlockCache, to assist you in tuning your
configuration.
You can view further detail and graphs using the RegionServer UI. To access the RegionServer UI in Cloudera Manager,
go to the Cloudera Manager page for the host, click the RegionServer process, and click HBase RegionServer Web UI.
If you do not use Cloudera Manager, access the BlockCache reports at
http://regionServer_host:22102/rs-status#memoryStats, replacing regionServer_host with the
hostname or IP address of your RegionServer.
Configuring the HBase Scanner Heartbeat
A scanner heartbeat check enforces a time limit on the execution of scan RPC requests. This helps prevent scans from
taking too long and causing a timeout at the client.
When the server receives a scan RPC request, a time limit is calculated to be half of the smaller of two values:
hbase.client.scanner.timeout.period and hbase.rpc.timeout (which both default to 60000 milliseconds,
or one minute). When the time limit is reached, the server returns the results it has accumulated up to that point. This
result set may be empty. If your usage pattern includes that scans will take longer than a minute, you can increase
these values.
Cloudera Administration | 99
Managing CDH and Managed Services
To make sure the timeout period is not too short, you can configure hbase.cells.scanned.per.heartbeat.check
to a minimum number of cells that must be scanned before a timeout check occurs. The default value is 10000. A
smaller value causes timeout checks to occur more often.
Configure the Scanner Heartbeat Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
1.
2.
3.
4.
5.
6.
7.
Go to the HBase service.
Click the Configuration tab.
Select HBase or HBase Service-Wide.
Select Category > Main.
Locate the RPC Timeout property or search for it by typing its name in the Search box.
Edit the property.
To modify the default values for hbase.client.scanner.timeout.period or
hbase.cells.scanned.per.heartbeat.check, search for HBase Service Advanced Configuration Snippet
(Safety Valve) for hbase-site.xml. Paste one or both of the following properties into the field and modify the
values as needed.
<property>
<name>hbase.client.scanner.timeout.period</name>
<value>60000</value>
</property>
<property>
<name>hbase.cells.scanned.per.heartbeat.check</name>
<value>10000</value>
</property>
8. Click Save Changes to commit the changes.
9. Restart the role.
10. Restart the service.
Configure the Scanner Heartbeat Using the Command Line
1. Edit hbase-site.xml and add the following properties, modifying the values as needed.
<property>
<name>hbase.rpc.timeout</name>
<value>60000</value>
</property>
<property>
<name>hbase.client.scanner.timeout.period</name>
<value>60000</value>
</property>
<property>
<name>hbase.cells.scanned.per.heartbeat.check</name>
<value>10000</value>
</property>
2. Distribute the modified hbase-site.xml to all your cluster hosts and restart the HBase master and RegionServer
processes for the change to take effect.
Limiting the Speed of Compactions
You can limit the speed at which HBase compactions run, by configuring
hbase.regionserver.throughput.controller and its related settings. The default controller is
org.apache.hadoop.hbase.regionserver.compactions.PressureAwareCompactionThroughputController,
which uses the following algorithm:
1. If compaction pressure is greater than 1.0, there is no speed limitation.
2. In off-peak hours, use a fixed throughput limitation, configured using
hbase.hstore.compaction.throughput.offpeak, hbase.offpeak.start.hour, and
hbase.offpeak.end.hour.
100 | Cloudera Administration
Managing CDH and Managed Services
3. In normal hours, the max throughput is tuned between
hbase.hstore.compaction.throughput.higher.bound and
hbase.hstore.compaction.throughput.lower.bound (which default to 20 MB/sec and 10 MB/sec
respectively), using the following formula, where compactionPressure is between 0.0 and 1.0. The
compactionPressure refers to the number of store files that require compaction.
lower + (higer - lower) * compactionPressure
To disable compaction speed limits, set hbase.regionserver.throughput.controller to
org.apache.hadoop.hbase.regionserver.compactions.NoLimitCompactionThroughputController.
Configure the Compaction Speed Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
1.
2.
3.
4.
Go to the HBase service.
Click the Configuration tab.
Select HBase or HBase Service-Wide.
Search for HBase Service Advanced Configuration Snippet (Safety Valve) for hbase-site.xml. Paste the relevant
properties into the field and modify the values as needed. See Configure the Compaction Speed Using the Command
Line on page 101 for an explanation of the properties.
5. Click Save Changes to commit the changes.
6. Restart the role.
7. Restart the service.
Configure the Compaction Speed Using the Command Line
1. Edit hbase-site.xml and add the relevant properties, modifying the values as needed. Default values are shown.
hbase.offpeak.start.hour and hbase.offpeak.end.hour have no default values; this configuration sets
the off-peak hours from 20:00 (8 PM) to 6:00 (6 AM).
<property>
<name>hbase.regionserver.throughput.controller</name>
<value>org.apache.hadoop.hbase.regionserver.compactions.PressureAwareCompactionThroughputController</value>
</property>
<property>
<name>hbase.hstore.compaction.throughput.higher.bound</name>
<value>20971520</value>
<description>The default is 20 MB/sec</description>
</property>
<property>
<name>hbase.hstore.compaction.throughput.lower.bound</name>
<value>10485760</value>
<description>The default is 10 MB/sec</description>
</property>
<property>
<name>hbase.hstore.compaction.throughput.offpeak</name>
<value>9223372036854775807</value>
<description>The default is Long.MAX_VALUE, which effectively means no
limitation</description>
</property>
<property>
<name>hbase.offpeak.start.hour</name>
<value>20</value>
<value>When to begin using off-peak compaction settings, expressed as an integer
between 0 and 23.</value>
</property>
<property>
<name>hbase.offpeak.start.hour</name>
<value>6</value>
<value>When to stop using off-peak compaction settings, expressed as an integer between
0 and 23.</value>
</property>
<property>
<name>hbase.hstore.compaction.throughput.tune.period</name>
Cloudera Administration | 101
Managing CDH and Managed Services
<value>60000</value>
<description>
</property>
2. Distribute the modified hbase-site.xml to all your cluster hosts and restart the HBase master and RegionServer
processes for the change to take effect.
Reading Data from HBase
Get and Scan are the two ways to read data from HBase, aside from manually parsing HFiles. A Get is simply a Scan
limited by the API to one row. A Scan fetches zero or more rows of a table. By default, a Scan reads the entire table
from start to end. You can limit your Scan results in several different ways, which affect the Scan's load in terms of
IO, network, or both, as well as processing load on the client side. This topic is provided as a quick reference. Refer to
the API documentation for Scan for more in-depth information. You can also perform Gets and Scan using the HBase
Shell.
• Specify a startrow or stoprow or both. Neither startrow nor stoprow need to exist. Because HBase sorts
rows lexicographically, it will return the first row after startrow would have occurred, and will stop returning
rows after stoprow would have occurred.The goal is to reduce IO and network.
– The startrow is inclusive and the stoprow is exclusive. Given a table with rows a, b, c, d, e, f, and startrow
of c and stoprow of f, rows c-e are returned.
– If you omit startrow, the first row of the table is the startrow.
– If you omit the stoprow, all results after startrow (including startrow) are returned.
– If startrow is lexicographically after stoprow, and you set Scan setReversed(boolean reversed) to
true, the results are returned in reverse order. Given the same table above, with rows a-f, if you specify c
as the stoprow and f as the startrow, rows f, e, and d are returned.
Scan()
Scan(byte[] startRow)
Scan(byte[] startRow, byte[] stopRow)
• Specify a scanner cache that will be filled before the Scan result is returned, setting setCaching to the number
of rows to cache before returning the result. By default, the caching setting on the table is used. The goal is to
balance IO and network load.
public Scan setCaching(int caching)
• To limit the number of columns if your table has very wide rows (rows with a large number of columns), use
setBatch(int batch) and set it to the number of columns you want to return in one batch. A large number of columns
is not a recommended design pattern.
public Scan setBatch(int batch)
• To specify a maximum result size, use setMaxResultSize(long), with the number of bytes. The goal is to
reduce IO and network.
public Scan setMaxResultSize(long maxResultSize)
• When you use setCaching and setMaxResultSize together, single server requests are limited by either number
of rows or maximum result size, whichever limit comes first.
• You can limit the scan to specific column families or columns by using addFamily or addColumn. The goal is to
reduce IO and network. IO is reduced because each column family is represented by a Store on each RegionServer,
and only the Stores representing the specific column families in question need to be accessed.
public Scan addColumn(byte[] family,
byte[] qualifier)
public Scan addFamily(byte[] family)
102 | Cloudera Administration
Managing CDH and Managed Services
• You can specify a range of timestamps or a single timestamp by specifying setTimeRange or setTimestamp.
public Scan setTimeRange(long minStamp,
long maxStamp)
throws IOException
public Scan setTimeStamp(long timestamp)
throws IOException
• You can retrieve a maximum number of versions by using setMaxVersions.
public Scan setMaxVersions(int maxVersions)
• You can use a filter by using setFilter. Filters are discussed in detail in HBase Filtering on page 104 and the Filter
API.
public Scan setFilter(Filter filter)
• You can disable the server-side block cache for a specific scan using the API setCacheBlocks(boolean). This
is an expert setting and should only be used if you know what you are doing.
Perform Scans Using HBase Shell
You can perform scans using HBase Shell, for testing or quick queries. Use the following guidelines or issue the scan
command in HBase Shell with no parameters for more usage information. This represents only a subset of possibilities.
# Display usage information
hbase> scan
# Scan all rows of table 't1'
hbase> scan 't1'
# Specify a startrow, limit the result to 10 rows, and only return selected columns
hbase> scan 't1', {COLUMNS => ['c1', 'c2'], LIMIT => 10, STARTROW => 'xyz'}
# Specify a timerange
hbase> scan 't1', {TIMERANGE => [1303668804, 1303668904]}
# Specify a custom filter
hbase> scan 't1', {FILTER => org.apache.hadoop.hbase.filter.ColumnPaginationFilter.new(1,
0)}
# Disable the block cache for a specific scan (experts only)
hbase> scan 't1', {COLUMNS => ['c1', 'c2'], CACHE_BLOCKS => false}
Hedged Reads
Hadoop 2.4 introduced a new feature called hedged reads. If a read from a block is slow, the HDFS client starts up
another parallel, 'hedged' read against a different block replica. The result of whichever read returns first is used, and
the outstanding read is cancelled. This feature helps in situations where a read occasionally takes a long time rather
than when there is a systemic problem. Hedged reads can be enabled for HBase when the HFiles are stored in HDFS.
This feature is disabled by default.
Enabling Hedged Reads for HBase Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
To enable hedged reads for HBase, edit the hbase-site.xml file on each server. Set
dfs.client.hedged.read.threadpool.size to the number of threads to dedicate to running hedged threads,
Cloudera Administration | 103
Managing CDH and Managed Services
and set the dfs.client.hedged.read.threshold.millis configuration property to the number of milliseconds
to wait before starting a second read against a different block replica. Set
dfs.client.hedged.read.threadpool.size to 0 or remove it from the configuration to disable the feature.
After changing these properties, restart your cluster.
The following is an example configuration for hedged reads for HBase.
<property>
<name>dfs.client.hedged.read.threadpool.size</name>
<value>20</value> <!-- 20 threads -->
</property>
<property>
<name>dfs.client.hedged.read.threshold.millis</name>
<value>10</value> <!-- 10 milliseconds -->
</property>
HBase Filtering
When reading data from HBase using Get or Scan operations, you can use custom filters to return a subset of results
to the client. While this does not reduce server-side IO, it does reduce network bandwidth and reduces the amount
of data the client needs to process. Filters are generally used using the Java API, but can be used from HBase Shell for
testing and debugging purposes.
For more information on Gets and Scans in HBase, see Reading Data from HBase on page 102.
Dynamically Loading a Custom Filter
CDH 5.5 and higher adds (and enables by default) the ability to dynamically load a custom filter by adding a JAR with
your filter to the directory specified by the hbase.dynamic.jars.dir property (which defaults to the lib/ directory
under the HBase root directory).
To disable automatic loading of dynamic JARs, set hbase.use.dynamic.jars to false in the advanced configuration
snippet for hbase-site.xml if you use Cloudera Manager, or to hbase-site.xml otherwise.
Filter Syntax Guidelines
HBase filters take zero or more arguments, in parentheses. Where the argument is a string, it is surrounded by single
quotes ('string').
Logical Operators, Comparison Operators and Comparators
Filters can be combined together with logical operators. Some filters take a combination of comparison operators and
comparators. Following is the list of each.
Logical Operators
•
•
•
•
•
AND - the key-value must pass both the filters to be included in the results.
OR - the key-value must pass at least one of the filters to be included in the results.
SKIP - for a particular row, if any of the key-values do not pass the filter condition, the entire row is skipped.
WHILE - For a particular row, it continues to emit key-values until a key-value is reached that fails the filter condition.
Compound Filters - Using these operators, a hierarchy of filters can be created. For example:
(Filter1 AND Filter2)OR(Filter3 AND Filter4)
Comparison Operators
•
•
•
•
•
LESS (<)
LESS_OR_EQUAL (<=)
EQUAL (=)
NOT_EQUAL (!=)
GREATER_OR_EQUAL (>=)
104 | Cloudera Administration
Managing CDH and Managed Services
• GREATER (>)
• NO_OP (no operation)
Comparators
• BinaryComparator - lexicographically compares against the specified byte array using the
Bytes.compareTo(byte[], byte[]) method.
• BinaryPrefixComparator - lexicographically compares against a specified byte array. It only compares up to the
length of this byte array.
• RegexStringComparator - compares against the specified byte array using the given regular expression. Only EQUAL
and NOT_EQUAL comparisons are valid with this comparator.
• SubStringComparator - tests whether or not the given substring appears in a specified byte array. The comparison
is case insensitive. Only EQUAL and NOT_EQUAL comparisons are valid with this comparator.
Examples
Example1: >, 'binary:abc' will match everything that is lexicographically greater than
"abc"
Example2: =, 'binaryprefix:abc' will match everything whose first 3 characters are
lexicographically equal to "abc"
Example3: !=, 'regexstring:ab*yz' will match everything that doesn't begin with "ab"
and ends with "yz"
Example4: =, 'substring:abc123' will match everything that begins with the substring
"abc123"
Compound Operators
Within an expression, parentheses can be used to group clauses together, and parentheses have the highest order of
precedence.
SKIP and WHILE operators are next, and have the same precedence.
The AND operator is next.
The OR operator is next.
Examples
A filter string of the form: “Filter1 AND Filter2 OR Filter3” will be evaluated as:
“(Filter1 AND Filter2) OR Filter3”
A filter string of the form: “Filter1 AND SKIP Filter2 OR Filter3” will be evaluated
as: “(Filter1 AND (SKIP Filter2)) OR Filter3”
Filter Types
HBase includes several filter types, as well as the ability to group filters together and create your own custom filters.
• KeyOnlyFilter - takes no arguments. Returns the key portion of each key-value pair.
Syntax: KeyOnlyFilter ()
• FirstKeyOnlyFilter - takes no arguments. Returns the key portion of the first key-value pair.
Syntax: FirstKeyOnlyFilter ()
• PrefixFilter - takes a single argument, a prefix of a row key. It returns only those key-values present in a row that
start with the specified row prefix
Syntax:
PrefixFilter (‘<row_prefix>’)
Example: PrefixFilter (‘Row’)
Cloudera Administration | 105
Managing CDH and Managed Services
• ColumnPrefixFilter - takes a single argument, a column prefix. It returns only those key-values present in a column
that starts with the specified column prefix.
Syntax:
ColumnPrefixFilter (‘<column_prefix>’)
Example: ColumnPrefixFilter (‘Col’)
• MultipleColumnPrefixFilter - takes a list of column prefixes. It returns key-values that are present in a column
that starts with any of the specified column prefixes.
Syntax: MultipleColumnPrefixFilter (‘<column_prefix>’, ‘<column_prefix>’, …,
‘<column_prefix>’)
Example: MultipleColumnPrefixFilter (‘Col1’, ‘Col2’)
• ColumnCountGetFilter - takes one argument, a limit. It returns the first limit number of columns in the table.
Syntax:
ColumnCountGetFilter (‘<limit>’)
Example: ColumnCountGetFilter (4)
• PageFilter - takes one argument, a page size. It returns page size number of rows from the table.
Syntax:
PageFilter (‘<page_size>’)
Example: PageFilter (2)
• ColumnPaginationFilter - takes two arguments, a limit and offset. It returns limit number of columns after offset
number of columns. It does this for all the rows.
Syntax:
ColumnPaginationFilter (‘<limit>’, ‘<offset>’)
Example: ColumnPaginationFilter (3, 5)
• InclusiveStopFilter - takes one argument, a row key on which to stop scanning. It returns all key-values present
in rows up to and including the specified row.
Syntax:
InclusiveStopFilter (‘<stop_row_key>’)
Example: InclusiveStopFilter (‘Row2’)
• TimeStampsFilter - takes a list of timestamps. It returns those key-values whose timestamps matches any of the
specified timestamps.
Syntax:
TimeStampsFilter (<timestamp>, <timestamp>, ... ,<timestamp>)
Example: TimeStampsFilter (5985489, 48895495, 58489845945)
• RowFilter - takes a compare operator and a comparator. It compares each row key with the comparator using
the compare operator and if the comparison returns true, it returns all the key-values in that row.
Syntax:
RowFilter (<compareOp>, ‘<row_comparator>’)
Example: RowFilter (<=, ‘binary:xyz)
• FamilyFilter - takes a compare operator and a comparator. It compares each family name with the comparator
using the compare operator and if the comparison returns true, it returns all the key-values in that family.
Syntax:
FamilyFilter (<compareOp>, ‘<family_comparator>’)
Example: FamilyFilter (>=, ‘binaryprefix:FamilyB’)
106 | Cloudera Administration
Managing CDH and Managed Services
• QualifierFilter - takes a compare operator and a comparator. It compares each qualifier name with the comparator
using the compare operator and if the comparison returns true, it returns all the key-values in that column.
Syntax:
QualifierFilter (<compareOp>, ‘<qualifier_comparator>’)
Example: QualifierFilter (=, ‘substring:Column1’)
• ValueFilter - takes a compare operator and a comparator. It compares each value with the comparator using the
compare operator and if the comparison returns true, it returns that key-value.
Syntax:
ValueFilter (<compareOp>, ‘<value_comparator>’)
Example: ValueFilter (!=, ‘binary:Value’)
• DependentColumnFilter - takes two arguments required arguments, a family and a qualifier. It tries to locate this
column in each row and returns all key-values in that row that have the same timestamp. If the row does not
contain the specified column, none of the key-values in that row will be returned.
The filter can also take an optional boolean argument, dropDependentColumn. If set to true, the column used
for the filter does not get returned.
The filter can also take two more additional optional arguments, a compare operator and a value comparator,
which are further checks in addition to the family and qualifier. If the dependent column is found, its value should
also pass the value check. If it does pass the value check, only then is its timestamp taken into consideration.
Syntax: DependentColumnFilter (‘<family>’, ‘<qualifier>’, <boolean>, <compare operator>,
‘<value comparator’)
DependentColumnFilter (‘<family>’, ‘<qualifier>’, <boolean>)
DependentColumnFilter (‘<family>’, ‘<qualifier>’)
Example: DependentColumnFilter (‘conf’, ‘blacklist’, false, >=, ‘zebra’)
DependentColumnFilter (‘conf’, ‘blacklist’, true)
DependentColumnFilter (‘conf’, ‘blacklist’)
• SingleColumnValueFilter - takes a column family, a qualifier, a compare operator and a comparator. If the specified
column is not found, all the columns of that row will be emitted. If the column is found and the comparison with
the comparator returns true, all the columns of the row will be emitted. If the condition fails, the row will not
be emitted.
This filter also takes two additional optional boolean arguments, filterIfColumnMissing and
setLatestVersionOnly.
If the filterIfColumnMissing flag is set to true, the columns of the row will not be emitted if the specified
column to check is not found in the row. The default value is false.
If the setLatestVersionOnly flag is set to false, it will test previous versions (timestamps) in addition to the
most recent. The default value is true.
These flags are optional and dependent on each other. You must set neither or both of them together.
Syntax: SingleColumnValueFilter (‘<family>’, ‘<qualifier>’, <compare operator>,
‘<comparator>’, <filterIfColumnMissing_boolean>, <latest_version_boolean>)
Syntax: SingleColumnValueFilter (‘<family>’, ‘<qualifier>’, <compare operator>,
‘<comparator>’)
Example: SingleColumnValueFilter (‘FamilyA’, ‘Column1’, <=, ‘abc’, true, false)
Example: SingleColumnValueFilter ('FamilyA’, ‘Column1’, <=, ‘abc’)
Cloudera Administration | 107
Managing CDH and Managed Services
• SingleColumnValueExcludeFilter - takes the same arguments and behaves same as SingleColumnValueFilter.
However, if the column is found and the condition passes, all the columns of the row will be emitted except for
the tested column value.
Syntax: SingleColumnValueExcludeFilter (<family>, <qualifier>, <compare operators>,
<comparator>, <latest_version_boolean>, <filterIfColumnMissing_boolean>)
Syntax: SingleColumnValueExcludeFilter (<family>, <qualifier>, <compare operator>
<comparator>)
Example: SingleColumnValueExcludeFilter (‘FamilyA’, ‘Column1’, ‘<=’, ‘abc’, ‘false’,
‘true’)
Example: SingleColumnValueExcludeFilter (‘FamilyA’, ‘Column1’, ‘<=’, ‘abc’)
• ColumnRangeFilter - takes either minColumn, maxColumn, or both. Returns only those keys with columns that
are between minColumn and maxColumn. It also takes two boolean variables to indicate whether to include the
minColumn and maxColumn or not. If you don’t want to set the minColumn or the maxColumn, you can pass in
an empty argument.
Syntax: ColumnRangeFilter (‘<minColumn >’, <minColumnInclusive_bool>, ‘<maxColumn>’,
<maxColumnInclusive_bool>)
Example: ColumnRangeFilter (‘abc’, true, ‘xyz’, false)
• Custom Filter - You can create a custom filter by implementing the Filter class. The JAR must be available on all
RegionServers.
HBase Shell Example
This example scans the 'users' table for rows where the contents of the cf:name column equals the string 'abc'.
hbase> scan 'users', { FILTER => SingleColumnValueFilter.new(Bytes.toBytes('cf'),
Bytes.toBytes('name'), CompareFilter::CompareOp.valueOf('EQUAL'),
BinaryComparator.new(Bytes.toBytes('abc')))}
Java API Example
This example, taken from the HBase unit test found in
hbase-server/src/test/java/org/apache/hadoop/hbase/filter/TestSingleColumnValueFilter.java
, shows how to use the Java API to implement several different filters..
/**
*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
*
http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.hadoop.hbase.filter;
import static org.junit.Assert.assertFalse;
import static org.junit.Assert.assertTrue;
import java.util.regex.Pattern;
108 | Cloudera Administration
Managing CDH and Managed Services
import
import
import
import
import
import
import
org.apache.hadoop.hbase.KeyValue;
org.apache.hadoop.hbase.SmallTests;
org.apache.hadoop.hbase.filter.CompareFilter.CompareOp;
org.apache.hadoop.hbase.util.Bytes;
org.junit.Before;
org.junit.Test;
org.junit.experimental.categories.Category;
/**
* Tests the value filter
*/
@Category(SmallTests.class)
public class TestSingleColumnValueFilter {
private static final byte[] ROW = Bytes.toBytes("test");
private static final byte[] COLUMN_FAMILY = Bytes.toBytes("test");
private static final byte [] COLUMN_QUALIFIER = Bytes.toBytes("foo");
private static final byte[] VAL_1 = Bytes.toBytes("a");
private static final byte[] VAL_2 = Bytes.toBytes("ab");
private static final byte[] VAL_3 = Bytes.toBytes("abc");
private static final byte[] VAL_4 = Bytes.toBytes("abcd");
private static final byte[] FULLSTRING_1 =
Bytes.toBytes("The quick brown fox jumps over the lazy dog.");
private static final byte[] FULLSTRING_2 =
Bytes.toBytes("The slow grey fox trips over the lazy dog.");
private static final String QUICK_SUBSTR = "quick";
private static final String QUICK_REGEX = ".+quick.+";
private static final Pattern QUICK_PATTERN = Pattern.compile("QuIcK",
Pattern.CASE_INSENSITIVE | Pattern.DOTALL);
Filter
Filter
Filter
Filter
Filter
basicFilter;
nullFilter;
substrFilter;
regexFilter;
regexPatternFilter;
@Before
public void setUp() throws Exception {
basicFilter = basicFilterNew();
nullFilter = nullFilterNew();
substrFilter = substrFilterNew();
regexFilter = regexFilterNew();
regexPatternFilter = regexFilterNew(QUICK_PATTERN);
}
private Filter basicFilterNew() {
return new SingleColumnValueFilter(COLUMN_FAMILY, COLUMN_QUALIFIER,
CompareOp.GREATER_OR_EQUAL, VAL_2);
}
private Filter nullFilterNew() {
return new SingleColumnValueFilter(COLUMN_FAMILY, COLUMN_QUALIFIER,
CompareOp.NOT_EQUAL,
new NullComparator());
}
private Filter substrFilterNew() {
return new SingleColumnValueFilter(COLUMN_FAMILY, COLUMN_QUALIFIER,
CompareOp.EQUAL,
new SubstringComparator(QUICK_SUBSTR));
}
private Filter regexFilterNew() {
return new SingleColumnValueFilter(COLUMN_FAMILY, COLUMN_QUALIFIER,
CompareOp.EQUAL,
new RegexStringComparator(QUICK_REGEX));
}
private Filter regexFilterNew(Pattern pattern) {
return new SingleColumnValueFilter(COLUMN_FAMILY, COLUMN_QUALIFIER,
CompareOp.EQUAL,
new RegexStringComparator(pattern.pattern(), pattern.flags()));
}
Cloudera Administration | 109
Managing CDH and Managed Services
private void basicFilterTests(SingleColumnValueFilter filter)
throws Exception {
KeyValue kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_2);
assertTrue("basicFilter1", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_3);
assertTrue("basicFilter2", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_4);
assertTrue("basicFilter3", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertFalse("basicFilterNotNull", filter.filterRow());
filter.reset();
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_1);
assertTrue("basicFilter4", filter.filterKeyValue(kv) == Filter.ReturnCode.NEXT_ROW);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_2);
assertTrue("basicFilter4", filter.filterKeyValue(kv) == Filter.ReturnCode.NEXT_ROW);
assertFalse("basicFilterAllRemaining", filter.filterAllRemaining());
assertTrue("basicFilterNotNull", filter.filterRow());
filter.reset();
filter.setLatestVersionOnly(false);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_1);
assertTrue("basicFilter5", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_2);
assertTrue("basicFilter5", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertFalse("basicFilterNotNull", filter.filterRow());
}
private void nullFilterTests(Filter filter) throws Exception {
((SingleColumnValueFilter) filter).setFilterIfMissing(true);
KeyValue kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, FULLSTRING_1);
assertTrue("null1", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertFalse("null1FilterRow", filter.filterRow());
filter.reset();
kv = new KeyValue(ROW, COLUMN_FAMILY, Bytes.toBytes("qual2"), FULLSTRING_2);
assertTrue("null2", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertTrue("null2FilterRow", filter.filterRow());
}
private void substrFilterTests(Filter filter)
throws Exception {
KeyValue kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER,
FULLSTRING_1);
assertTrue("substrTrue",
filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER,
FULLSTRING_2);
assertTrue("substrFalse", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertFalse("substrFilterAllRemaining", filter.filterAllRemaining());
assertFalse("substrFilterNotNull", filter.filterRow());
}
private void regexFilterTests(Filter filter)
throws Exception {
KeyValue kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER,
FULLSTRING_1);
assertTrue("regexTrue",
filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER,
FULLSTRING_2);
assertTrue("regexFalse", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertFalse("regexFilterAllRemaining", filter.filterAllRemaining());
assertFalse("regexFilterNotNull", filter.filterRow());
}
private void regexPatternFilterTests(Filter filter)
throws Exception {
KeyValue kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER,
FULLSTRING_1);
110 | Cloudera Administration
Managing CDH and Managed Services
assertTrue("regexTrue",
filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertFalse("regexFilterAllRemaining", filter.filterAllRemaining());
assertFalse("regexFilterNotNull", filter.filterRow());
}
private Filter serializationTest(Filter filter)
throws Exception {
// Decompose filter to bytes.
byte[] buffer = filter.toByteArray();
// Recompose filter.
Filter newFilter = SingleColumnValueFilter.parseFrom(buffer);
return newFilter;
}
/**
* Tests identification of the stop row
* @throws Exception
*/
@Test
public void testStop() throws Exception {
basicFilterTests((SingleColumnValueFilter) basicFilter);
nullFilterTests(nullFilter);
substrFilterTests(substrFilter);
regexFilterTests(regexFilter);
regexPatternFilterTests(regexPatternFilter);
}
/**
* Tests serialization
* @throws Exception
*/
@Test
public void testSerialization() throws Exception {
Filter newFilter = serializationTest(basicFilter);
basicFilterTests((SingleColumnValueFilter)newFilter);
newFilter = serializationTest(nullFilter);
nullFilterTests(newFilter);
newFilter = serializationTest(substrFilter);
substrFilterTests(newFilter);
newFilter = serializationTest(regexFilter);
regexFilterTests(newFilter);
newFilter = serializationTest(regexPatternFilter);
regexPatternFilterTests(newFilter);
}
}
Writing Data to HBase
To write data to HBase, you use methods of the HTableInterface class. You can use the Java API directly, or use
HBase Shell, Thrift API, REST API, or another client which uses the Java API indirectly. When you issue a Put, the
coordinates of the data are the row, the column, and the timestamp. The timestamp is unique per version of the cell,
and can be generated automatically or specified programmatically by your application, and must be a long integer.
Variations on Put
There are several different ways to write data into HBase. Some of them are listed below.
• A Put operation writes data into HBase.
• A Delete operation deletes data from HBase. What actually happens during a Delete depends upon several
factors.
• A CheckAndPut operation performs a Scan before attempting the Put, and only does the Put if a value matches
what is expected, and provides row-level atomicity.
• A CheckAndDelete operation performs a Scan before attempting the Delete, and only does the Delete if a
value matches what is expected.
Cloudera Administration | 111
Managing CDH and Managed Services
• An Increment operation increments values of one or more columns within a single row, and provides row-level
atomicity.
Refer to the API documentation for a full list of methods provided for writing data to HBase.Different methods require
different access levels and have other differences.
Versions
When you put data into HBase, a timestamp is required. The timestamp can be generated automatically by the
RegionServer or can be supplied by you. The timestamp must be unique per version of a given cell, because the
timestamp identifies the version. To modify a previous version of a cell, for instance, you would issue a Put with a
different value for the data itself, but the same timestamp.
HBase's behavior regarding versions is highly configurable. The maximum number of versions defaults to 1 in CDH 5,
and 3 in previous versions. You can change the default value for HBase by configuring hbase.column.max.version
in hbase-site.xml, either using an advanced configuration snippet if you use Cloudera Manager, or by editing the
file directly otherwise.
You can also configure the maximum and minimum number of versions to keep for a given column, or specify a default
time-to-live (TTL), which is the number of seconds before a version is deleted. The following examples all use alter
statements in HBase Shell to create new column families with the given characteristics, but you can use the same
syntax when creating a new table or to alter an existing column family. This is only a fraction of the options you can
specify for a given column family.
hbase> alter ‘t1 , NAME => ‘f1 , VERSIONS => 5
hbase> alter ‘t1 , NAME => ‘f1 , MIN_VERSIONS => 2
hbase> alter ‘t1 , NAME => ‘f1 , TTL => 15
HBase sorts the versions of a cell from newest to oldest, by sorting the timestamps lexicographically. When a version
needs to be deleted because a threshold has been reached, HBase always chooses the "oldest" version, even if it is in
fact the most recent version to be inserted. Keep this in mind when designing your timestamps. Consider using the
default generated timestamps and storing other version-specific data elsewhere in the row, such as in the row key. If
MIN_VERSIONS and TTL conflict, MIN_VERSIONS takes precedence.
Deletion
When you request for HBase to delete data, either explicitly using a Delete method or implicitly using a threshold such
as the maximum number of versions or the TTL, HBase does not delete the data immediately. Instead, it writes a
deletion marker, called a tombstone, to the HFile, which is the physical file where a given RegionServer stores its region
of a column family. The tombstone markers are processed during major compaction operations, when HFiles are
rewritten without the deleted data included.
Even after major compactions, "deleted" data may not actually be deleted. You can specify the KEEP_DELETED_CELLS
option for a given column family, and the tombstones will be preserved in the HFile even after major compaction. One
scenario where this approach might be useful is for data retention policies.
Another reason deleted data may not actually be deleted is if the data would be required to restore a table from a
snapshot which has not been deleted. In this case, the data is moved to an archive during a major compaction, and
only deleted when the snapshot is deleted. This is a good reason to monitor the number of snapshots saved in HBase.
Examples
This abbreviated example writes data to an HBase table using HBase Shell and then scans the table to show the result.
hbase> put 'test', 'row1', 'cf:a', 'value1'
0 row(s) in 0.1770 seconds
hbase> put 'test', 'row2', 'cf:b', 'value2'
0 row(s) in 0.0160 seconds
hbase> put 'test', 'row3', 'cf:c', 'value3'
0 row(s) in 0.0260 seconds
112 | Cloudera Administration
Managing CDH and Managed Services
hbase> scan 'test'
ROW
COLUMN+CELL
row1
column=cf:a, timestamp=1403759475114, value=value1
row2
column=cf:b, timestamp=1403759492807, value=value2
row3
column=cf:c, timestamp=1403759503155, value=value3
3 row(s) in 0.0440 seconds
This abbreviated example uses the HBase API to write data to an HBase table, using the automatic timestamp created
by the Region Server.
publicstaticfinalbyte[] CF = "cf".getBytes();
publicstaticfinalbyte[] ATTR = "attr".getBytes();
...
Put put = new Put(Bytes.toBytes(row));
put.add(CF, ATTR, Bytes.toBytes( data));
htable.put(put);
This example uses the HBase API to write data to an HBase table, specifying the timestamp.
publicstaticfinalbyte[] CF = "cf".getBytes();
publicstaticfinalbyte[] ATTR = "attr".getBytes();
...
Put put = new Put( Bytes.toBytes(row));
long explicitTimeInMs = 555; // just an example
put.add(CF, ATTR, explicitTimeInMs, Bytes.toBytes(data));
htable.put(put);
Further Reading
• Refer to the HTableInterface and HColumnDescriptor API documentation for more details about configuring tables
and columns, as well as reading and writing to HBase.
• Refer to the Apache HBase Reference Guide for more in-depth information about HBase, including details about
versions and deletions not covered here.
Importing Data Into HBase
The method you use for importing data into HBase depends on several factors:
•
•
•
•
The location, size, and format of your existing data
Whether you need to import data once or periodically over time
Whether you want to import the data in bulk or stream it into HBase regularly
How fresh the HBase data needs to be
This topic helps you choose the correct method or composite of methods and provides example workflows for each
method.
Always run HBase administrative commands as the HBase user (typically hbase).
Choosing the Right Import Method
If the data is already in an HBase table:
• To move the data from one HBase cluster to another, use snapshot and either the clone_snapshot or
ExportSnapshot utility; or, use the CopyTable utility.
• To move the data from one HBase cluster to another without downtime on either cluster, use replication.
• To migrate data between HBase version that are not wire compatible, such as from CDH 4 to CDH 5, see Importing
HBase Data From CDH 4 to CDH 5 on page 114.
If the data currently exists outside HBase:
• If possible, write the data to HFile format, and use a BulkLoad to import it into HBase. The data is immediately
available to HBase and you can bypass the normal write path, increasing efficiency.
Cloudera Administration | 113
Managing CDH and Managed Services
• If you prefer not to use bulk loads, and you are using a tool such as Pig, you can use it to import your data.
If you need to stream live data to HBase instead of import in bulk:
• Write a Java client using the Java API, or use the Apache Thrift Proxy API to write a client in a language supported
by Thrift.
• Stream data directly into HBase using the REST Proxy API in conjunction with an HTTP client such as wget or curl.
• Use Flume or Spark.
Most likely, at least one of these methods works in your situation. If not, you can use MapReduce directly. Test the
most feasible methods with a subset of your data to determine which one is optimal.
Using CopyTable
CopyTable uses HBase read and write paths to copy part or all of a table to a new table in either the same cluster or
a different cluster. CopyTable causes read load when reading from the source, and write load when writing to the
destination. Region splits occur on the destination table in real time as needed. To avoid these issues, use snapshot
and export commands instead of CopyTable. Alternatively, you can pre-split the destination table to avoid excessive
splits. The destination table can be partitioned differently from the source table. See this section of the Apache HBase
documentation for more information.
Edits to the source table after the CopyTable starts are not copied, so you may need to do an additional CopyTable
operation to copy new data into the destination table. Run CopyTable as follows, using --help to see details about
possible parameters.
$ ./bin/hbase org.apache.hadoop.hbase.mapreduce.CopyTable --help
Usage: CopyTable [general options] [--starttime=X] [--endtime=Y] [--new.name=NEW]
[--peer.adr=ADR] <tablename>
The starttime/endtime and startrow/endrow pairs function in a similar way: if you leave out the first of the pair,
the first timestamp or row in the table is the starting point. Similarly, if you leave out the second of the pair, the
operation continues until the end of the table. To copy the table to a new table in the same cluster, you must specify
--new.name, unless you want to write the copy back to the same table, which would add a new version of each cell
(with the same data), or just overwrite the cell with the same value if the maximum number of versions is set to 1 (the
default in CDH 5). To copy the table to a new table in a different cluster, specify --peer.adr and optionally, specify
a new table name.
The following example creates a new table using HBase Shell in non-interactive mode, and then copies data in two
ColumnFamilies in rows starting with timestamp 1265875194289 and including the last row before the CopyTable
started, to the new table.
$ echo create 'NewTestTable', 'cf1', 'cf2', 'cf3' | bin/hbase shell --non-interactive
$ bin/hbase org.apache.hadoop.hbase.mapreduce.CopyTable --starttime=1265875194289
--families=cf1,cf2,cf3 --new.name=NewTestTable TestTable
In CDH 5, snapshots are recommended instead of CopyTable for most situations.
Importing HBase Data From CDH 4 to CDH 5
CDH 4 and CDH 5 are not wire-compatible, so import methods such as CopyTable will not work. Instead, you can use
separate export and import operations using distcp, or you can copy the table's HFiles using HDFS utilities and upgrade
the HFiles in place. The first option is preferred unless the size of the table is too large to be practical and the export
or import will take too long. The import/export mechanism gives you flexibility and allows you to run exports as often
as you need, for an ongoing period of time. This would allow you to test CDH 5 with your production data before
finalizing your upgrade, for instance.
Import and Export Data Using DistCP
1. Both Import and Export applications have several command-line options which you can use to control their
behavior, such as limiting the import or export to certain column families or modifying the output directory. Run
114 | Cloudera Administration
Managing CDH and Managed Services
the commands without arguments to view the usage instructions. The output below is an example, and may be
different for different HBase versions.
$ bin/hbase org.apache.hadoop.hbase.mapreduce.Import
Usage: Import [options] <tablename> <inputdir>
By default Import will load data directly into HBase. To instead generate
HFiles of data to prepare for a bulk data load, pass the option:
-Dimport.bulk.output=/path/for/output
To apply a generic org.apache.hadoop.hbase.filter.Filter to the input, use
-Dimport.filter.class=<name of filter class>
-Dimport.filter.args=<comma separated list of args for filter
NOTE: The filter will be applied BEFORE doing key renames using the
HBASE_IMPORTER_RENAME_CFS property. Futher, filters will only use the
Filter#filterRowKey(byte[] buffer, int offset, int length) method to identify
whether the current row needs to be ignored completely
for processing and Filter#filterKeyValue(KeyValue) method to determine if the
KeyValue should be added; Filter.ReturnCode#INCLUDE
and #INCLUDE_AND_NEXT_COL will be considered as including the KeyValue.
To import data exported from HBase 0.94, use
-Dhbase.import.version=0.94
For performance consider the following options:
-Dmapreduce.map.speculative=false
-Dmapreduce.reduce.speculative=false
-Dimport.wal.durability=<Used while writing data to hbase. Allowed values
are the supported durability values like SKIP_WAL/ASYNC_WAL/SYNC_WAL/...>
$ /usr/bin/hbase org.apache.hadoop.hbase.mapreduce.Export
ERROR: Wrong number of arguments: 0
Usage: Export [-D <property=value>]* <tablename> <outputdir> [<versions> [<starttime>
[<endtime>]] [^[regex pattern] or [Prefix] to filter]]
Note: -D properties will be applied to the conf used.
For example:
-D mapreduce.output.fileoutputformat.compress=true
-D
mapreduce.output.fileoutputformat.compress.codec=org.apache.hadoop.io.compress.GzipCodec
-D mapreduce.output.fileoutputformat.compress.type=BLOCK
Additionally, the following SCAN properties can be specified
to control/limit what is exported..
-D hbase.mapreduce.scan.column.family=<familyName>
-D hbase.mapreduce.include.deleted.rows=true
-D hbase.mapreduce.scan.row.start=<ROWSTART>
-D hbase.mapreduce.scan.row.stop=<ROWSTOP>
For performance consider the following properties:
-Dhbase.client.scanner.caching=100
-Dmapreduce.map.speculative=false
-Dmapreduce.reduce.speculative=false
For tables with very wide rows consider setting the batch size as below:
-Dhbase.export.scanner.batch=10
2. On the CDH 4 cluster, export the contents of the table to sequence files in a given directory using a command like
the following.
$ sudo -u hdfs hbase org.apache.hadoop.hbase.mapreduce.Export <tablename>
/export_directory
The sequence files are located in the /export_directory directory.
3. Copy the contents of /export_directory to the CDH 5 cluster using distcp or through a filesystem accessible
from hosts on both clusters. If you use distcp, the following is an example command.
$ sudo -u hdfs hadoop distcp -p -update -skipcrccheck
hftp://cdh4-namenode:port/export_directory hdfs://cdh5-namenode/import_directory
4. Create the table on the CDH 5 cluster using HBase Shell. Column families must be identical to the table on the
CDH 4 cluster.
Cloudera Administration | 115
Managing CDH and Managed Services
5. Import the sequence file into the newly-created table.
$ sudo -u hdfs hbase -Dhbase.import.version=0.94 org.apache.hadoop.hbase.mapreduce.Import
t1 /import_directory
Copy and Upgrade the HFiles
If exporting and importing the data is not feasible because of the size of the data or other reasons, or you know that
the import will be a one-time occurrence, you can copy the HFiles directly from the CDH 4 cluster's HDFS filesystem
to the CDH 5 cluster's HDFS filesystem, and upgrade the HFiles in place.
Warning: Only use this procedure if the destination cluster is a brand new HBase cluster with empty
tables, and is not currently hosting any data. If this is not the case, or if you are unsure, contact Cloudera
Support before following this procedure.
1. Use the distcp command on the CDH 5 cluster to copy the HFiles from the CDH 4 cluster.
$ sudo -u hdfs hadoop distcp -p -update -skipcrccheck
webhdfs://cdh4-namenode:http-port/hbase hdfs://cdh5-namenode:rpc-port/hbase
2. In the destination cluster, upgrade the HBase tables. In Cloudera Manager, go to Cluster > HBase and choose
Upgrade HBase from the Action menu. This checks that the HBase tables can be upgraded, and then upgrades
them.
3. Start HBase on the CDH 5 cluster. The upgraded tables are available. Verify the data and confirm that no errors
are logged.
Using Snapshots
As of CDH 4.7, Cloudera recommends snapshots instead of CopyTable where possible. A snapshot captures the state
of a table at the time the snapshot was taken. Because no data is copied when a snapshot is taken, the process is very
quick. As long as the snapshot exists, cells in the snapshot are never deleted from HBase, even if they are explicitly
deleted by the API. Instead, they are archived so that the snapshot can restore the table to its state at the time of the
snapshot.
After taking a snapshot, use the clone_snapshot command to copy the data to a new (immediately enabled) table
in the same cluster, or the Export utility to create a new table based on the snapshot, in the same cluster or a new
cluster. This is a copy-on-write operation. The new table shares HFiles with the original table until writes occur in the
new table but not the old table, or until a compaction or split occurs in either of the tables. This can improve performance
in the short term compared to CopyTable.
To export the snapshot to a new cluster, use the ExportSnapshot utility, which uses MapReduce to copy the snapshot
to the new cluster. Run the ExportSnapshot utility on the source cluster, as a user with HBase and HDFS write
permission on the destination cluster, and HDFS read permission on the source cluster. This creates the expected
amount of IO load on the destination cluster. Optionally, you can limit bandwidth consumption, which affects IO on
the destination cluster. After the ExportSnapshot operation completes, you can see the snapshot in the new cluster
using the list_snapshot command, and you can use the clone_snapshot command to create the table in the
new cluster from the snapshot.
For full instructions for the snapshot and clone_snapshot HBase Shell commands, run the HBase Shell and type
help snapshot. The following example takes a snapshot of a table, uses it to clone the table to a new table in the
same cluster, and then uses the ExportSnapshot utility to copy the table to a different cluster, with 16 mappers and
limited to 200 Mb/sec bandwidth.
$ bin/hbase shell
hbase(main):005:0> snapshot 'TestTable', 'TestTableSnapshot'
0 row(s) in 2.3290 seconds
hbase(main):006:0> clone_snapshot 'TestTableSnapshot', 'NewTestTable'
0 row(s) in 1.3270 seconds
116 | Cloudera Administration
Managing CDH and Managed Services
hbase(main):007:0> describe 'NewTestTable'
DESCRIPTION
ENABLED
'NewTestTable', {NAME => 'cf1', DATA_BLOCK_ENCODING true
=> 'NONE', BLOOMFILTER => 'ROW', REPLICATION_SCOPE
=> '0', VERSIONS => '1', COMPRESSION => 'NONE', MI
N_VERSIONS => '0', TTL => 'FOREVER', KEEP_DELETED_C
ELLS => 'false', BLOCKSIZE => '65536', IN_MEMORY =>
'false', BLOCKCACHE => 'true'}, {NAME => 'cf2', DA
TA_BLOCK_ENCODING => 'NONE', BLOOMFILTER => 'ROW',
REPLICATION_SCOPE => '0', VERSIONS => '1', COMPRESS
ION => 'NONE', MIN_VERSIONS => '0', TTL => 'FOREVER
', KEEP_DELETED_CELLS => 'false', BLOCKSIZE => '655
36', IN_MEMORY => 'false', BLOCKCACHE => 'true'}
1 row(s) in 0.1280 seconds
hbase(main):008:0> quit
$ hbase org.apache.hadoop.hbase.snapshot.ExportSnapshot -snapshot TestTableSnapshot
-copy-to file:///tmp/hbase -mappers 16 -bandwidth 200
14/10/28 21:48:16 INFO snapshot.ExportSnapshot: Copy Snapshot Manifest
14/10/28 21:48:17 INFO client.RMProxy: Connecting to ResourceManager at
a1221.halxg.cloudera.com/10.20.188.121:8032
14/10/28 21:48:19 INFO snapshot.ExportSnapshot: Loading Snapshot 'TestTableSnapshot'
hfile list
14/10/28 21:48:19 INFO Configuration.deprecation: hadoop.native.lib is deprecated.
Instead, use io.native.lib.available
14/10/28 21:48:19 INFO util.FSVisitor: No logs under
directory:hdfs://a1221.halxg.cloudera.com:8020/hbase/.hbase-snapshot/TestTableSnapshot/WALs
14/10/28 21:48:20 INFO mapreduce.JobSubmitter: number of splits:0
14/10/28 21:48:20 INFO mapreduce.JobSubmitter: Submitting tokens for job:
job_1414556809048_0001
14/10/28 21:48:20 INFO impl.YarnClientImpl: Submitted application
application_1414556809048_0001
14/10/28 21:48:20 INFO mapreduce.Job: The url to track the job:
http://a1221.halxg.cloudera.com:8088/proxy/application_1414556809048_0001/
14/10/28 21:48:20 INFO mapreduce.Job: Running job: job_1414556809048_0001
14/10/28 21:48:36 INFO mapreduce.Job: Job job_1414556809048_0001 running in uber mode
: false
14/10/28 21:48:36 INFO mapreduce.Job: map 0% reduce 0%
14/10/28 21:48:37 INFO mapreduce.Job: Job job_1414556809048_0001 completed successfully
14/10/28 21:48:37 INFO mapreduce.Job: Counters: 2
Job Counters
Total time spent by all maps in occupied slots (ms)=0
Total time spent by all reduces in occupied slots (ms)=0
14/10/28 21:48:37 INFO snapshot.ExportSnapshot: Finalize the Snapshot Export
14/10/28 21:48:37 INFO snapshot.ExportSnapshot: Verify snapshot integrity
14/10/28 21:48:37 INFO Configuration.deprecation: fs.default.name is deprecated. Instead,
use fs.defaultFS
14/10/28 21:48:37 INFO snapshot.ExportSnapshot: Export Completed: TestTableSnapshot
The bold italic line contains the URL from which you can track the ExportSnapshot job. When it finishes, a new set
of HFiles, comprising all of the HFiles that were part of the table when the snapshot was taken, is created at the HDFS
location you specified.
You can use the SnapshotInfo command-line utility included with HBase to verify or debug snapshots.
Using BulkLoad
HBase uses the well-known HFile format to store its data on disk. In many situations, writing HFiles programmatically
with your data, and bulk-loading that data into HBase on the RegionServer, has advantages over other data ingest
mechanisms. BulkLoad operations bypass the write path completely, providing the following benefits:
• The data is available to HBase immediately but does cause additional load or latency on the cluster when it appears.
• BulkLoad operations do not use the write-ahead log (WAL) and do not cause flushes or split storms.
• BulkLoad operations do not cause excessive garbage collection.
Note: Because they bypass the WAL, BulkLoad operations are not propagated between clusters
using replication. If you need the data on all replicated clusters, you must perform the BulkLoad
on each cluster.
Cloudera Administration | 117
Managing CDH and Managed Services
If you use BulkLoads with HBase, your workflow is similar to the following:
1. Extract your data from its existing source. For instance, if your data is in a MySQL database, you might run the
mysqldump command. The process you use depends on your data. If your data is already in TSV or CSV format,
skip this step and use the included ImportTsv utility to process your data into HFiles. See the ImportTsv
documentation for details.
2. Process your data into HFile format. See http://hbase.apache.org/book.html#_hfile_format_2 for details about
HFile format. Usually you use a MapReduce job for the conversion, and you often need to write the Mapper
yourself because your data is unique. The job must to emit the row key as the Key, and either a KeyValue, a Put,
or a Delete as the Value. The Reducer is handled by HBase; configure it using
HFileOutputFormat.configureIncrementalLoad() and it does the following:
•
•
•
•
•
Inspects the table to configure a total order partitioner
Uploads the partitions file to the cluster and adds it to the DistributedCache
Sets the number of reduce tasks to match the current number of regions
Sets the output key/value class to match HFileOutputFormat requirements
Sets the Reducer to perform the appropriate sorting (either KeyValueSortReducer or PutSortReducer)
3. One HFile is created per region in the output folder. Input data is almost completely re-written, so you need
available disk space at least twice the size of the original data set. For example, for a 100 GB output from
mysqldump, you should have at least 200 GB of available disk space in HDFS. You can delete the original input
file at the end of the process.
4. Load the files into HBase. Use the LoadIncrementalHFiles command (more commonly known as the
completebulkload tool), passing it a URL that locates the files in HDFS. Each file is loaded into the relevant region
on the RegionServer for the region. You can limit the number of versions that are loaded by passing the
--versions= N option, where N is the maximum number of versions to include, from newest to oldest
(largest timestamp to smallest timestamp).
If a region was split after the files were created, the tool automatically splits the HFile according to the new
boundaries. This process is inefficient, so if your table is being written to by other processes, you should load as
soon as the transform step is done.
The following illustration shows the full BulkLoad process.
Extra Steps for BulkLoad With Encryption Zones
When using BulkLoad to import data into HBase in the a cluster using encryption zones, the following information is
important.
• Both the staging directory and the directory into which you place your generated HFiles need to be within HBase's
encryption zone (generally under the /hbase directory). Before you can do this, you need to change the permissions
of /hbase to be world-executable but not world-readable (rwx--x--x, or numeric mode 711).
118 | Cloudera Administration
Managing CDH and Managed Services
• You also need to configure the HMaster to set the permissions of the HBase root directory correctly. If you use
Cloudera Manager, edit the Master Advanced Configuration Snippet (Safety Valve) for hbase-site.xml. Otherwise,
edit hbase-site.xml on the HMaster. Add the following:
<property>
<name>hbase.rootdir.perms</name>
<value>711</value>
</property>
If you skip this step, a previously-working BulkLoad setup will start to fail with permission errors when you restart
the HMaster.
Use Cases for BulkLoad
• Loading your original dataset into HBase for the first time - Your initial dataset might be quite large, and bypassing
the HBase write path can speed up the process considerably.
• Incremental Load - To load new data periodically, use BulkLoad to import it in batches at your preferred intervals.
This alleviates latency problems and helps you to achieve service-level agreements (SLAs). However, one trigger
for compaction is the number of HFiles on a RegionServer. Therefore, importing a large number of HFiles at
frequent intervals can cause major compactions to happen more often than they otherwise would, negatively
impacting performance. You can mitigate this by tuning the compaction settings such that the maximum number
of HFiles that can be present without triggering a compaction is very high, and relying on other factors, such as
the size of the Memstore, to trigger compactions.
• Data needs to originate elsewhere - If an existing system is capturing the data you want to have in HBase and
needs to remain active for business reasons, you can periodically BulkLoad data from the system into HBase so
that you can perform operations on it without impacting the system.
Using BulkLoad On A Secure Cluster
If you use security, HBase allows you to securely BulkLoad data into HBase. For a full explanation of how secure BulkLoad
works, see HBase Transparent Encryption at Rest.
First, configure a hbase.bulkload.staging.dir which will be managed by HBase and whose subdirectories will
be writable (but not readable) by HBase users. Next, add the
org.apache.hadoop.hbase.security.access.SecureBulkLoadEndpoint coprocessor to your configuration,
so that users besides the hbase user can BulkLoad files into HBase. This functionality is available in CDH 5.5 and higher.
<property>
<name>hbase.bulkload.staging.dir</name>
<value>/tmp/hbase-staging</value>
</property>
<property>
<name>hbase.coprocessor.region.classes</name>
<value>org.apache.hadoop.hbase.security.access.SecureBulkLoadEndpoint</value>
</property>
More Information about BulkLoad
For more information and examples, as well as an explanation of the ImportTsv utility, which can be used to import
data in text-delimited formats such as CSV, see this post on the Cloudera Blog.
Using Cluster Replication
If your data is already in an HBase cluster, replication is useful for getting the data into additional HBase clusters. In
HBase, cluster replication refers to keeping one cluster state synchronized with that of another cluster, using the
write-ahead log (WAL) of the source cluster to propagate the changes. Replication is enabled at column family granularity.
Before enabling replication for a column family, create the table and all column families to be replicated, on the
destination cluster.
Cloudera Administration | 119
Managing CDH and Managed Services
Cluster replication uses an active-push methodology. An HBase cluster can be a source (also called active, meaning
that it writes new data), a destination (also called passive, meaning that it receives data using replication), or can fulfill
both roles at once. Replication is asynchronous, and the goal of replication is consistency.
When data is replicated from one cluster to another, the original source of the data is tracked with a cluster ID, which
is part of the metadata. In CDH 5, all clusters that have already consumed the data are also tracked. This prevents
replication loops.
Common Replication Topologies
• A central source cluster might propagate changes to multiple destination clusters, for failover or due to geographic
distribution.
• A source cluster might push changes to a destination cluster, which might also push its own changes back to the
original cluster.
• Many different low-latency clusters might push changes to one centralized cluster for backup or resource-intensive
data-analytics jobs. The processed data might then be replicated back to the low-latency clusters.
• Multiple levels of replication can be chained together to suit your needs. The following diagram shows a hypothetical
scenario. Use the arrows to follow the data paths.
At the top of the diagram, the San Jose and Tokyo clusters, shown in red, replicate changes to each other, and each
also replicates changes to a User Data and a Payment Data cluster.
Each cluster in the second row, shown in blue, replicates its changes to the All Data Backup 1 cluster, shown in
grey. The All Data Backup 1 cluster replicates changes to the All Data Backup 2 cluster (also shown in grey),
as well as the Data Analysis cluster (shown in green). All Data Backup 2 also propagates any of its own changes
back to All Data Backup 1.
The Data Analysis cluster runs MapReduce jobs on its data, and then pushes the processed data back to the San
Jose and Tokyo clusters.
Configuring Clusters for Replication
To configure your clusters for replication, see HBase Replication on page 409 and Configuring Secure HBase Replication.
The following is a high-level overview of the steps to enable replication.
120 | Cloudera Administration
Managing CDH and Managed Services
Important: To run replication-related HBase comands, your user must have HBase administrator
permissions. If ZooKeeper uses Kerberos, configure HBase Shell to authenticate to ZooKeeper using
Kerberos before attempting to run replication-related commands. There are currently no
replication-related ACLs.
1. Configure and start the source and destination clusters.
2. Create tables with the same names and column families on both the source and destination clusters, so that the
destination cluster knows where to store data it receives. All hosts in the source and destination clusters should
be reachable to each other. See Creating the Empty Table On the Destination Cluster on page 413.
3. On the source cluster, enable replication in Cloudera Manager, or by setting hbase.replication to true in
hbase-site.xml.
4. On the source cluster, in HBase Shell, add the destination cluster as a peer, using the add_peer command. The
syntax is as follows:
add_peer 'ID', 'CLUSTER_KEY'
The ID must be a short integer. To compose the CLUSTER_KEY, use the following template:
hbase.zookeeper.quorum:hbase.zookeeper.property.clientPort:zookeeper.znode.parent
If both clusters use the same ZooKeeper cluster, you must use a different zookeeper.znode.parent, because they
cannot write in the same folder.
5. On the source cluster, configure each column family to be replicated by setting its REPLICATION_SCOPE to 1,
using commands such as the following in HBase Shell.
hbase> disable 'example_table'
hbase> alter 'example_table', {NAME => 'example_family', REPLICATION_SCOPE => '1'}
hbase> enable 'example_table'
6. Verify that replication is occurring by examining the logs on the source cluster for messages such as the following.
Considering 1 rs, with ratio 0.1
Getting 1 rs from peer cluster # 0
Choosing peer 10.10.1.49:62020
7. To verify the validity of replicated data, use the included VerifyReplication MapReduce job on the source
cluster, providing it with the ID of the replication peer and table name to verify. Other options are available, such
as a time range or specific families to verify.
The command has the following form:
hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication
[--starttime=timestamp1] [--stoptime=timestamp] [--families=comma separated list of
families] <peerId> <tablename>
The VerifyReplication command prints GOODROWS and BADROWS counters to indicate rows that did and did
not replicate correctly.
Note:
Some changes are not replicated and must be propagated by other means, such as Snapshots or
CopyTable. See Initiating Replication When Data Already Exists on page 413 for more details.
• Data that existed in the master before replication was enabled.
• Operations that bypass the WAL, such as when using BulkLoad or API calls such as
writeToWal(false).
• Table schema modifications.
Cloudera Administration | 121
Managing CDH and Managed Services
Using Pig and HCatalog
Apache Pig is a platform for analyzing large data sets using a high-level language. Apache HCatalog is a sub-project of
Apache Hive, which enables reading and writing of data from one Hadoop utility to another. You can use a combination
of Pig and HCatalog to import data into HBase. The initial format of your data and other details about your infrastructure
determine the steps you follow to accomplish this task. The following simple example assumes that you can get your
data into a TSV (text-separated value) format, such as a tab-delimited or comma-delimited text file.
1. Format the data as a TSV file. You can work with other file formats; see the Pig and HCatalog project documentation
for more details.
The following example shows a subset of data from Google's NGram Dataset, which shows the frequency of specific
phrases or letter-groupings found in publications indexed by Google. Here, the first column has been added to
this dataset as the row ID. The first column is formulated by combining the n-gram itself (in this case, Zones) with
the line number of the file in which it occurs (z_LINE_NUM). This creates a format such as "Zones_z_6230867."
The second column is the n-gram itself, the third column is the year of occurrence, the fourth column is the
frequency of occurrence of that Ngram in that year, and the fifth column is the number of distinct publications.
This extract is from the z file of the 1-gram dataset from version 20120701. The data is truncated at the ... mark,
for the sake of readability of this document. In most real-world scenarios, you will not work with tables that have
five columns. Most HBase tables have one or two columns.
Zones_z_6230867
Zones_z_6230868
Zones_z_6230869
Zones_z_6230870
...
Zones_z_6231150
Zones_z_6231151
Zones_z_6231152
Zones_z_6231153
Zones_z_6231154
Zones_z_6231155
Zones_z_6231156
Zones_z_6231157
Zones_z_6231158
Zones_z_6231159
Zones_z_6231160
Zones_z_6231161
Zones_z_6231162
Zones
Zones
Zones
Zones
1507
1638
1656
1681
1
1
2
8
1
1
1
2
Zones
Zones
Zones
Zones
Zones
Zones
Zones
Zones
Zones
Zones
Zones
Zones
Zones
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
17868
21296
20365
20288
22996
20469
21338
29724
23334
24300
22362
22101
21037
4356
4675
4972
5021
5714
5470
5946
6446
6524
6580
6707
6798
6328
2. Using the hadoop fs command, put the data into HDFS. This example places the file into an /imported_data/
directory.
$ hadoop fs -put zones_frequency.tsv /imported_data/
3. Create and register a new HBase table in HCatalog, using the hcat command, passing it a DDL file to represent
your table. You could also register an existing HBase table, using the same command. The DDL file format is
specified as part of the Hive REST API. The following example illustrates the basic mechanism.
CREATE TABLE
zones_frequency_table (id STRING, ngram STRING, year STRING, freq STRING, sources STRING)
STORED BY 'org.apache.hcatalog.hbase.HBaseHCatStorageHandler'
TBLPROPERTIES (
'hbase.table.name' = 'zones_frequency_table',
'hbase.columns.mapping' = 'd:ngram,d:year,d:freq,d:sources',
'hcat.hbase.output.bulkMode' = 'true'
);
$ hcat -f zones_frequency_table.ddl
4. Create a Pig file to process the TSV file created in step 1, using the DDL file created in step 3. Modify the file names
and other parameters in this command to match your values if you use data different from this working example.
122 | Cloudera Administration
Managing CDH and Managed Services
USING PigStorage('\t') indicates that the input file is tab-delimited. For more details about Pig syntax, see
the Pig Latin reference documentation.
A = LOAD 'hdfs:///imported_data/zones_frequency.tsv' USING PigStorage('\t') AS
(id:chararray, ngram:chararray, year:chararray, freq:chararray, sources:chararray);
-- DUMP A;
STORE A INTO 'zones_frequency_table' USING org.apache.hcatalog.pig.HCatStorer();
Save the file as zones.bulkload.pig.
5. Use the pig command to bulk-load the data into HBase.
$ pig -useHCatalog zones.bulkload.pig
The data is now in HBase and is available to use.
Using the Java API
The Java API is the most common mechanism for getting data into HBase, through Put operations. The Thrift and REST
APIs, as well as the HBase Shell, use the Java API. The following simple example ouses the Java API to put data into an
HBase table. The Java API traverses the entire write path and can cause compactions and region splits, which can
adversely affect performance.
...
HTable table = null;
try {
table = myCode.createTable(tableName, fam);
int i = 1;
List<Put> puts = new ArrayList<Put>();
for (String labelExp : labelExps) {
Put put = new Put(Bytes.toBytes("row" + i));
put.add(fam, qual, HConstants.LATEST_TIMESTAMP, value);
puts.add(put);
i++;
}
table.put(puts);
} finally {
if (table != null) {
table.flushCommits();
}
}
...
Using the Apache Thrift Proxy API
The Apache Thrift library provides cross-language client-server remote procedure calls (RPCs), using Thrift bindings. A
Thrift binding is client code generated by the Apache Thrift Compiler for a target language (such as Python) that allows
communication between the Thrift server and clients using that client code. HBase includes an Apache Thrift Proxy
API, which allows you to write HBase applications in Python, C, C++, or another language that Thrift supports. The Thrift
Proxy API is slower than the Java API and may have fewer features. T use the Thrift Proxy API, you need to configure
and run the HBase Thrift server on your cluster. See Installing and Starting the HBase Thrift Server. You also need to
install the Apache Thrift compiler on your development system.
After the Thrift server is configured and running, generate Thrift bindings for the language of your choice, using an IDL
file. A HBase IDL file named HBase.thrift is included as part of HBase. After generating the bindings, copy the Thrift
libraries for your language into the same directory as the generated bindings. In the following Python example, these
libraries provide the thrift.transport and thrift.protocol libraries. These commands show how you might
generate the Thrift bindings for Python and copy the libraries on a Linux system.
$
$
$
$
$
mkdir HBaseThrift
cd HBaseThrift/
thrift -gen py /path/to/Hbase.thrift
mv gen-py/* .
rm -rf gen-py/
Cloudera Administration | 123
Managing CDH and Managed Services
$ mkdir thrift
$ cp -rp ~/Downloads/thrift-0.9.0/lib/py/src/* ./thrift/
The following iexample shows a simple Python application using the Thrift Proxy API.
from
from
from
from
thrift.transport import TSocket
thrift.protocol import TBinaryProtocol
thrift.transport import TTransport
hbase import Hbase
# Connect to HBase Thrift server
transport = TTransport.TBufferedTransport(TSocket.TSocket(host, port))
protocol = TBinaryProtocol.TBinaryProtocolAccelerated(transport)
# Create and open the client connection
client = Hbase.Client(protocol)
transport.open()
# Modify a single row
mutations = [Hbase.Mutation(
column='columnfamily:columndescriptor', value='columnvalue')]
client.mutateRow('tablename', 'rowkey', mutations)
# Modify a batch of rows
# Create a list of mutations per work of Shakespeare
mutationsbatch = []
for line in myDataFile:
rowkey = username + "-" + filename + "-" + str(linenumber).zfill(6)
mutations = [
Hbase.Mutation(column=messagecolumncf, value=line.strip()),
Hbase.Mutation(column=linenumbercolumncf, value=encode(linenumber)),
Hbase.Mutation(column=usernamecolumncf, value=username)
]
mutationsbatch.append(Hbase.BatchMutation(row=rowkey,mutations=mutations))
# Run the mutations for all the lines in myDataFile
client.mutateRows(tablename, mutationsbatch)
transport.close()
The Thrift Proxy API does not support writing to HBase clusters that are secured using Kerberos.
This example was modified from the following two blog posts on http://www.cloudera.com. See them for more details.
• Using the HBase Thrift Interface, Part 1
• Using the HBase Thrift Interface, Part 2
Using the REST Proxy API
After configuring and starting the HBase REST Server on your cluster, you can use the HBase REST Proxy API to stream
data into HBase, from within another application or shell script, or by using an HTTP client such as wget or curl. The
REST Proxy API is slower than the Java API and may have fewer features. This approach is simple and does not require
advanced development experience to implement. However, like the Java and Thrift Proxy APIs, it uses the full write
path and can cause compactions and region splits.
Specified addresses without existing data create new values. Specified addresses with existing data create new versions,
overwriting an existing version if the row, column:qualifier, and timestamp all match that of the existing value.
$ curl -H "Content-Type: text/xml" http://localhost:8000/test/testrow/test:testcolumn
The REST Proxy API does not support writing to HBase clusters that are secured using Kerberos.
For full documentation and more examples, see the REST Proxy API documentation.
124 | Cloudera Administration
Managing CDH and Managed Services
Using Flume
Apache Flume is a fault-tolerant system designed for ingesting data into HDFS, for use with Hadoop. You can configure
Flume to write data directly into HBase. Flume includes two different sinks designed to work with HBase: HBaseSink
(org.apache.flume.sink.hbase.HBaseSink) and AsyncHBaseSink (org.apache.flume.sink.hbase.AsyncHBaseSink). HBaseSink
supports HBase IPC calls introduced in HBase 0.96, and allows you to write data to an HBase cluster that is secured by
Kerberos, whereas AsyncHBaseSink does not. However, AsyncHBaseSink uses an asynchronous model and guarantees
atomicity at the row level.
You configure HBaseSink and AsyncHBaseSink nearly identically. Following is an example configuration for each. Bold
lines highlight differences in the configurations. For full documentation about configuring HBaseSink and AsyncHBaseSink,
see the Flume documentation. The table, columnFamily, and column parameters correlate to the HBase table,
column family, and column where the data is to be imported. The serializer is the class that converts the data at the
source into something HBase can use. Configure your sinks in the Flume configuration file.
In practice, you usually need to write your own serializer, which implements either AsyncHBaseEventSerializer or
HBaseEventSerializer. The HBaseEventSerializer converts Flume Events into one or more HBase Puts, sends them to
the HBase cluster, and is closed when the HBaseSink stops. AsyncHBaseEventSerializer starts and listens for Events.
When it receives an Event, it calls the setEvent method and then calls the getActions and getIncrements methods.
When the AsyncHBaseSink is stopped, the serializer cleanUp method is called. These methods return PutRequest and
AtomicIncrementRequest, which are part of the asynchbase API.
AsyncHBaseSink:
#Use the AsyncHBaseSink
host1.sinks.sink1.type = org.apache.flume.sink.hbase.AsyncHBaseSink
host1.sinks.sink1.channel = ch1
host1.sinks.sink1.table = transactions
host1.sinks.sink1.columnFamily = clients
host1.sinks.sink1.column = charges
host1.sinks.sink1.batchSize = 5000
#Use the SimpleAsyncHbaseEventSerializer that comes with Flume
host1.sinks.sink1.serializer = org.apache.flume.sink.hbase.SimpleAsyncHbaseEventSerializer
host1.sinks.sink1.serializer.incrementColumn = icol
host1.channels.ch1.type=memory
HBaseSink:
#Use the HBaseSink
host1.sinks.sink1.type = org.apache.flume.sink.hbase.HBaseSink
host1.sinks.sink1.channel = ch1
host1.sinks.sink1.table = transactions
host1.sinks.sink1.columnFamily = clients
host1.sinks.sink1.column = charges
host1.sinks.sink1.batchSize = 5000
#Use the SimpleHbaseEventSerializer that comes with Flume
host1.sinks.sink1.serializer = org.apache.flume.sink.hbase.SimpleHbaseEventSerializer
host1.sinks.sink1.serializer.incrementColumn = icol
host1.channels.ch1.type=memory
The following serializer, taken from an Apache Flume blog post by Dan Sandler, splits the event body based on a
delimiter and inserts each split into a different column. The row is defined in the event header. When each event is
received, a counter is incremented to track the number of events received.
/**
* A serializer for the AsyncHBaseSink, which splits the event body into
* multiple columns and inserts them into a row whose key is available in
* the headers
*/
public class SplittingSerializer implements AsyncHbaseEventSerializer {
private byte[] table;
private byte[] colFam;
private Event currentEvent;
private byte[][] columnNames;
private final List<PutRequest> puts = new ArrayList<PutRequest>();
private final List<AtomicIncrementRequest> incs = new
Cloudera Administration | 125
Managing CDH and Managed Services
ArrayList<AtomicIncrementRequest>();
private byte[] currentRowKey;
private final byte[] eventCountCol = "eventCount".getBytes();
@Override
public void initialize(byte[] table, byte[] cf) {
this.table = table;
this.colFam = cf;
}
@Override
public void setEvent(Event event) {
// Set the event and verify that the rowKey is not present
this.currentEvent = event;
String rowKeyStr = currentEvent.getHeaders().get("rowKey");
if (rowKeyStr == null) {
throw new FlumeException("No row key found in headers!");
}
currentRowKey = rowKeyStr.getBytes();
}
@Override
public List<PutRequest> getActions() {
// Split the event body and get the values for the columns
String eventStr = new String(currentEvent.getBody());
String[] cols = eventStr.split(",");
puts.clear();
for (int i = 0; i < cols.length; i++) {
//Generate a PutRequest for each column.
PutRequest req = new PutRequest(table, currentRowKey, colFam,
columnNames[i], cols[i].getBytes());
puts.add(req);
}
return puts;
}
@Override
public List<AtomicIncrementRequest> getIncrements() {
incs.clear();
//Increment the number of events received
incs.add(new AtomicIncrementRequest(table, "totalEvents".getBytes(), colFam,
eventCountCol));
return incs;
}
@Override
public void cleanUp() {
table = null;
colFam = null;
currentEvent = null;
columnNames = null;
currentRowKey = null;
}
@Override
public void configure(Context context) {
//Get the column names from the configuration
String cols = new String(context.getString("columns"));
String[] names = cols.split(",");
byte[][] columnNames = new byte[names.length][];
int i = 0;
for(String name : names) {
columnNames[i++] = name.getBytes();
}
}
@Override
public void configure(ComponentConfiguration conf) {
}
}
Using Spark
You can write data to HBase from Apache Spark by using def saveAsHadoopDataset(conf: JobConf): Unit.
This example is adapted from a post on the spark-users mailing list.
// Note: mapred package is used, instead of the
// mapreduce package which contains new hadoop APIs.
import org.apache.hadoop.hbase.mapred.TableOutputFormat
import org.apache.hadoop.hbase.client
// ... some other settings
val conf = HBaseConfiguration.create()
// general hbase settings
126 | Cloudera Administration
Managing CDH and Managed Services
conf.set("hbase.rootdir",
"hdfs://" + nameNodeURL + ":" + hdfsPort + "/hbase")
conf.setBoolean("hbase.cluster.distributed", true)
conf.set("hbase.zookeeper.quorum", hostname)
conf.setInt("hbase.client.scanner.caching", 10000)
// ... some other settings
val jobConfig: JobConf = new JobConf(conf, this.getClass)
// Note: TableOutputFormat is used as deprecated code
// because JobConf is an old hadoop API
jobConfig.setOutputFormat(classOf[TableOutputFormat])
jobConfig.set(TableOutputFormat.OUTPUT_TABLE, outputTable)
Next, provide the mapping between how the data looks in Spark and how it should look in HBase. The following example
assumes that your HBase table has two column families, col_1 and col_2, and that your data is formatted in sets of
three in Spark, like (row_key, col_1, col_2).
def convert(triple: (Int, Int, Int)) = {
val p = new Put(Bytes.toBytes(triple._1))
p.add(Bytes.toBytes("cf"),
Bytes.toBytes("col_1"),
Bytes.toBytes(triple._2))
p.add(Bytes.toBytes("cf"),
Bytes.toBytes("col_2"),
Bytes.toBytes(triple._3))
(new ImmutableBytesWritable, p)
}
To write the data from Spark to HBase, you might use:
new PairRDDFunctions(localData.map(convert)).saveAsHadoopDataset(jobConfig)
Using Spark and Kafka
This example, written in Scala, uses Apache Spark in conjunction with the Apache Kafka message bus to stream data
from Spark to HBase. The example was provided in SPARK-944. It produces some random words and then stores them
in an HBase table, creating the table if necessary.
package org.apache.spark.streaming.examples
import java.util.Properties
import kafka.producer._
import
}
import
import
import
import
import
import
import
import
import
import
import
org.apache.hadoop.hbase.{ HBaseConfiguration, HColumnDescriptor, HTableDescriptor
org.apache.hadoop.hbase.client.{ HBaseAdmin, Put }
org.apache.hadoop.hbase.io.ImmutableBytesWritable
org.apache.hadoop.hbase.mapred.TableOutputFormat
org.apache.hadoop.hbase.mapreduce.TableInputFormat
org.apache.hadoop.hbase.util.Bytes
org.apache.hadoop.mapred.JobConf
org.apache.spark.SparkContext
org.apache.spark.rdd.{ PairRDDFunctions, RDD }
org.apache.spark.streaming._
org.apache.spark.streaming.StreamingContext._
org.apache.spark.streaming.kafka._
object MetricAggregatorHBase {
def main(args : Array[String]) {
if (args.length < 6) {
System.err.println("Usage: MetricAggregatorTest <master> <zkQuorum> <group> <topics>
<destHBaseTableName> <numThreads>")
System.exit(1)
}
val Array(master, zkQuorum, group, topics, hbaseTableName, numThreads) = args
Cloudera Administration | 127
Managing CDH and Managed Services
val conf = HBaseConfiguration.create()
conf.set("hbase.zookeeper.quorum", zkQuorum)
// Initialize hBase table if necessary
val admin = new HBaseAdmin(conf)
if (!admin.isTableAvailable(hbaseTableName)) {
val tableDesc = new HTableDescriptor(hbaseTableName)
tableDesc.addFamily(new HColumnDescriptor("metric"))
admin.createTable(tableDesc)
}
// setup streaming context
val ssc = new StreamingContext(master, "MetricAggregatorTest", Seconds(2),
System.getenv("SPARK_HOME"), StreamingContext.jarOfClass(this.getClass))
ssc.checkpoint("checkpoint")
val topicpMap = topics.split(",").map((_, numThreads.toInt)).toMap
val lines = KafkaUtils.createStream(ssc, zkQuorum, group, topicpMap)
.map { case (key, value) => ((key, Math.floor(System.currentTimeMillis() /
60000).toLong * 60), value.toInt) }
val aggr = lines.reduceByKeyAndWindow(add _, Minutes(1), Minutes(1), 2)
aggr.foreach(line => saveToHBase(line, zkQuorum, hbaseTableName))
ssc.start
ssc.awaitTermination
}
def add(a : Int, b : Int) = { (a + b) }
def saveToHBase(rdd : RDD[((String, Long), Int)], zkQuorum : String, tableName :
String) = {
val conf = HBaseConfiguration.create()
conf.set("hbase.zookeeper.quorum", zkQuorum)
val jobConfig = new JobConf(conf)
jobConfig.set(TableOutputFormat.OUTPUT_TABLE, tableName)
jobConfig.setOutputFormat(classOf[TableOutputFormat])
new PairRDDFunctions(rdd.map { case ((metricId, timestamp), value) =>
createHBaseRow(metricId, timestamp, value) }).saveAsHadoopDataset(jobConfig)
}
def createHBaseRow(metricId : String, timestamp : Long, value : Int) = {
val record = new Put(Bytes.toBytes(metricId + "~" + timestamp))
record.add(Bytes.toBytes("metric"), Bytes.toBytes("col"),
Bytes.toBytes(value.toString))
(new ImmutableBytesWritable, record)
}
}
// Produces some random words between 1 and 100.
object MetricDataProducer {
def main(args : Array[String]) {
if (args.length < 2) {
System.err.println("Usage: MetricDataProducer <metadataBrokerList> <topic>
<messagesPerSec>")
System.exit(1)
}
val Array(brokers, topic, messagesPerSec) = args
// ZooKeeper connection properties
val props = new Properties()
props.put("metadata.broker.list", brokers)
props.put("serializer.class", "kafka.serializer.StringEncoder")
128 | Cloudera Administration
Managing CDH and Managed Services
val config = new ProducerConfig(props)
val producer = new Producer[String, String](config)
// Send some messages
while (true) {
val messages = (1 to messagesPerSec.toInt).map { messageNum =>
{
val metricId = scala.util.Random.nextInt(10)
val value = scala.util.Random.nextInt(1000)
new KeyedMessage[String, String](topic, metricId.toString, value.toString)
}
}.toArray
producer.send(messages : _*)
Thread.sleep(100)
}
}
}
Using a Custom MapReduce Job
Many of the methods to import data into HBase use MapReduce implicitly. If none of those approaches fit your needs,
you can use MapReduce directly to convert data to a series of HFiles or API calls for import into HBase. In this way,
you can import data from Avro, Parquet, or another format into HBase, or export data from HBase into another format,
using API calls such as TableOutputFormat, HFileOutputFormat, and TableInputFormat.
Configuring HBase MultiWAL Support
CDH supports multiple write-ahead logs (MultiWAL) for HBase. (For more information, see HBASE-5699.)
Without MultiWAL support, each region on a RegionServer writes to the same WAL. A busy RegionServer might host
several regions, and each write to the WAL is serial because HDFS only supports sequentially written files. This causes
the WAL to negatively impact performance.
MultiWAL allows a RegionServer to write multiple WAL streams in parallel by using multiple pipelines in the underlying
HDFS instance, which increases total throughput during writes.
Note: In the current implementation of MultiWAL, incoming edits are partitioned by Region. Therefore,
throughput to a single Region is not increased.
To configure MultiWAL for a RegionServer, set the value of the property hbase.wal.provider to multiwal and
restart the RegionServer. To disable MultiWAL for a RegionServer, unset the property and restart the RegionServer.
RegionServers using the original WAL implementation and those using the MultiWAL implementation can each handle
recovery of either set of WALs, so a zero-downtime configuration update is possible through a rolling restart.
Configuring MultiWAL Support Using Cloudera Manager
1.
2.
3.
4.
5.
6.
7.
8.
Go to the HBase service.
Click the Configuration tab.
Select Scope > RegionServer.
Select Category > Main.
Set WAL Provider to MultiWAL.
Set the Per-RegionServer Number of WAL Pipelines to a value greater than 1.
Click Save Changes to commit the changes.
Restart the RegionServer roles.
Cloudera Administration | 129
Managing CDH and Managed Services
Configuring MultiWAL Support Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
1. Edit hbase-site.xml on each RegionServer where you want to enable MultiWAL. Add the following property
by pasting the XML.
<property>
<name>hbase.wal.provider</name>
<value>multiwal</value>
</property>
2. Stop and restart the RegionServer.
Storing Medium Objects (MOBs) in HBase
Data comes in many sizes, and saving all of your data in HBase, including binary data such as images and documents,
is convenient. HBase can technically handle binary objects with cells that are up to 10 MB in size. However, HBase
normal read and write paths are optimized for values smaller than 100 KB in size. When HBase handles large numbers
of values up to 10 MB (medium objects, or MOBs), performance is degraded because of write amplification caused by
splits and compactions.
One way to solve this problem is by storing objects larger than 100KB directly in HDFS, and storing references to their
locations in HBase. CDH 5.4 and higher includes optimizations for storing MOBs directly in HBase) based on HBASE-11339.
To use MOB, you must use HFile version 3. Optionally, you can configure the MOB file reader's cache settings
Service-Wide and for each RegionServer, and then configure specific columns to hold MOB data. No change to client
code is required for HBase MOB support.
Enabling HFile Version 3 Using Cloudera Manager
Minimum Required Role: Full Administrator
To enable HFile version 3 using Cloudera Manager, edit the HBase Service Advanced Configuration Snippet for HBase
Service-Wide.
1.
2.
3.
4.
Go to the HBase service.
Click the Configuration tab.
Search for the property HBase Service Advanced Configuration Snippet (Safety Valve) for hbase-site.xml.
Paste the following XML into the Value field and save your changes.
<property>
<name>hfile.format.version</name>
<value>3</value>
</property>
Changes will take effect after the next major compaction.
Enabling HFile Version 3 Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
130 | Cloudera Administration
Managing CDH and Managed Services
Paste the following XML into hbase-site.xml.
<property>
<name>hfile.format.version</name>
<value>3</value>
</property>
Restart HBase. Changes will take effect for a given region during its next major compaction.
Configuring Columns to Store MOBs
Use the following options to configure a column to store MOBs:
• IS_MOB is a Boolean option, which specifies whether or not the column can store MOBs.
• MOB_THRESHOLD configures the number of bytes at which an object is considered to be a MOB. If you do not
specify a value for MOB_THRESHOLD, the default is 100 KB. If you write a value larger than this threshold, it is
treated as a MOB.
You can configure a column to store MOBs using the HBase Shell or the Java API.
Using HBase Shell:
hbase> create 't1', {NAME => 'f1', IS_MOB => true, MOB_THRESHOLD => 102400}
hbase> alter 't1', {NAME => 'f1', IS_MOB => true, MOB_THRESHOLD =>
102400}
Using the Java API:
HColumnDescriptor hcd = new HColumnDescriptor(“f”);
hcd.setMobEnabled(true);
hcd.setMobThreshold(102400L);
HBase MOB Cache Properties
Because there can be a large number of MOB files at any time, as compared to the number of HFiles, MOB files are
not always kept open. The MOB file reader cache is a LRU cache which keeps the most recently used MOB files open.
The following properties are available for tuning the HBase MOB cache.
Table 3: HBase MOB Cache Properties
Property
Default
Description
hbase.mob.file.cache.size
1000
The of opened file handlers to cache.
A larger value will benefit reads by
providing more file handlers per MOB
file cache and would reduce frequent
file opening and closing of files.
However, if the value is too high,
errors such as "Too many opened file
handlers" may be logged.
hbase.mob.cache.evict.period 3600
The amount of time in seconds after
a file is opened before the MOB cache
evicts cached files. The default value
is 3600 seconds.
hbase.mob.cache.evict.remain.ratio 0.5f
The ratio, expressed as a float
between 0.0 and 1.0, that controls
how manyfiles remain cached after an
eviction is triggered due to the number
of cached files exceeding the
Cloudera Administration | 131
Managing CDH and Managed Services
Property
Default
Description
hbase.mob.file.cache.size. The
default value is 0.5f.
Configuring the MOB Cache Using Cloudera Manager
To configure the MOB cache within Cloudera Manager, edit the HBase Service advanced aonfiguration Snippet for the
cluster. Cloudera recommends testing your configuration with the default settings first.
1.
2.
3.
4.
Go to the HBase service.
Click the Configuration tab.
Search for the property HBase Service Advanced Configuration Snippet (Safety Valve) for hbase-site.xml.
Paste your configuration into the Value field and save your changes. The following example sets the
hbase.mob.cache.evict.period property to 5000 seconds. See Table 3: HBase MOB Cache Properties on
page 131 for a full list of configurable properties for HBase MOB.
<property>
<name>hbase.mob.cache.evict.period</name>
<value>5000</value>
</property>
5. Restart your cluster for the changes to take effect.
Configuring the MOB Cache Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
Because there can be a large number of MOB files at any time, as compared to the number of HFiles, MOB files are
not always kept open. The MOB file reader cache is a LRU cache which keeps the most recently used MOB files open.
To customize the configuration of the MOB file reader's cache on each RegionServer, configure the MOB cache properties
in the RegionServer's hbase-site.xml. Customize the configuration to suit your environment, and restart or rolling
restart the RegionServer. Cloudera recommends testing your configuration with the default settings first. The following
example sets the hbase.mob.cache.evict.period property to 5000 seconds. See Table 3: HBase MOB Cache
Properties on page 131 for a full list of configurable properties for HBase MOB.
<property>
<name>hbase.mob.cache.evict.period</name>
<value>5000</value>
</property>
Testing MOB Storage and Retrieval Performance
HBase provides the Java utility org.apache.hadoop.hbase.IntegrationTestIngestMOB to assist with testing
the MOB feature and deciding on appropriate configuration values for your situation. The utility is run as follows:
$ sudo -u hbase hbase org.apache.hadoop.hbase.IntegrationTestIngestMOB \
-threshold 102400 \
-minMobDataSize 512 \
-maxMobDataSize 5120
• threshold is the threshold at which cells are considered to be MOBs. The default is 1 kB, expressed in bytes.
• minMobDataSize is the minimum value for the size of MOB data. The default is 512 B, expressed in bytes.
• maxMobDataSize is the maximum value for the size of MOB data. The default is 5 kB, expressed in bytes.
132 | Cloudera Administration
Managing CDH and Managed Services
Compacting MOB Files Manually
You can trigger manual compaction of MOB files manually, rather than waiting for them to be triggered by your
configuration, using the HBase Shell commands compact_mob and major_compact_mob. Each of these commands
requires the first parameter to be the table name, and takes an optional column family name as the second argument.
If the column family is provided, only that column family's files are compacted. Otherwise, all MOB-enabled column
families' files are compacted.
hbase>
hbase>
hbase>
hbase>
compact_mob 't1'
compact_mob 't1', 'f1'
major_compact_mob 't1'
major_compact_mob 't1', 'f1'
This functionality is also available using the API, using the Admin.compact and Admin.majorCompact methods.
Exposing HBase Metrics to a Ganglia Server
Ganglia is a popular open-source monitoring framework. You can expose HBase metrics to a Ganglia instance so that
Ganglia can detect potential problems with your HBase cluster.
Expose HBase Metrics to Ganglia Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
1. Go to the HBase service.
2. Click the Configuration tab.
3. Select the HBase Master or RegionServer role. To monitor both, configure each role as described in the rest of
the procedure.
4. Select Category > Metrics.
5. Locate the Hadoop Metrics2 Advanced Configuration Snippet (Safety Valve) property or search for it by typing
its name in the Search box.
6. Edit the property. Add the following, substituting the server information with your own.
hbase.sink.ganglia.class=org.apache.hadoop.metrics2.sink.ganglia.GangliaSink31
hbase.sink.ganglia.servers=<Ganglia server>:<port>
hbase.sink.ganglia.period=10
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
7. Click Save Changes to commit the changes.
8. Restart the role.
9. Restart the service.
Expose HBase Metrics to Ganglia Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
1. Edit /etc/hbase/conf/hadoop-metrics2-hbase.properties on the master or RegionServers you want to
monitor, and add the following properties, substituting the server information with your own:
hbase.sink.ganglia.class=org.apache.hadoop.metrics2.sink.ganglia.GangliaSink31
hbase.sink.ganglia.servers=<Ganglia server>:<port>
hbase.sink.ganglia.period=10
2. Restart the master or RegionServer.
Cloudera Administration | 133
Managing CDH and Managed Services
Managing HDFS
The section contains configuration tasks for the HDFS service. For information on configuring HDFS for high availability,
see HDFS High Availability on page 290.
Managing Federated Nameservices
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
Cloudera Manager supports the configuration of multiple nameservices managing separate HDFS namespaces, all of
which share the storage available on the set of DataNodes. These nameservices are federated, meaning each nameservice
is independent and does not require coordination with other nameservices. See HDFS Federation for more information.
It is simplest to add a second nameservice if high availability is already enabled. The process of enabling high availability
creates a nameservice as part of the enable high availability workflow.
Important: Configuring a new nameservice shut downs the services that depend upon HDFS. Once
the new nameservice has been started, the services that depend upon HDFS must be restarted, and
the client configurations must be redeployed. (This can be done as part of the Add Nameservice
workflow, as an option.)
Configuring the First Nameservice Using Cloudera Manager
Follow the instructions below to define the first nameservice.
1. Go to the HDFS service.
2. Click the Configuration tab.
3. Type nameservice in the Search field. The nameservice properties for the NameNode and SecondaryNameNode
display.
4. In the NameNode Nameservice field, type a name for the nameservice. The name must be unique and can contain
only alphanumeric characters.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
5. In the Mountpoints field, change the mount point from "/" to a list of mount points that are in the namespace
that this nameservice will manage. (You can enter this as a comma-separated list — for example, /hbase, /tmp,
/user or by clicking the
icon to add each mount point in its own field.) You can determine the list of mount
points by running the command hadoop fs -ls / from the CLI on the NameNode host.
6. In the SecondaryNameNode Nameservice field, type the nameservice name that you provided for the NameNode
Nameservice property.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
7. Click Save Changes to commit the changes.
8. Click the Instances tab. The Federation and High Availability section displays with the nameservice listed.
Editing the List of Mountpoints for a Nameservice Using Cloudera Manager
1. Go to the HDFS service.
2. Click the Instances tab. The Federation and High Availability section displays with the nameservices listed.
3. Select Actions > Edit. In the Mount Points field, change the mount point to a list of mount points in the namespace
that the nameservice will manage.
4. Click OK.
Adding a Nameservice Using Cloudera Manager
The instructions below for adding a nameservice assume that one nameservice is already set up. The first nameservice
cam be set up either by configuring the first nameservice or by enabling HDFS high availability.
1. Go to the HDFS service.
134 | Cloudera Administration
Managing CDH and Managed Services
2. Click the Instances tab. At the top of this page you should see the Federation and High Availability section. If this
section does not appear, it means you do not have any nameservices configured. You must have one nameservice
already configured in order to add a nameservice.
3. Click the Add Nameservice button.
a. In the Nameservice Name field, enter a name for the new nameservice. The name must be unique and can
contain only alphanumeric characters.
b. In the Mount Points field, enter at least one mount point for the nameservice. This defines the portion of
HDFS that will be managed under the new nameservice. (Click the to the right of the field to add a new
mount point). You cannot use "/" as a mount point; you must specify HDFS directories by name.
• The mount points must be unique for this nameservice; you cannot specify any of the same mount points
you have used for other nameservices.
• You can specify mount points that do not yet exist, and create the corresponding directories in a later
step in this procedure.
• If you want to use a mount point previously associated with another nameservice you must first remove
that mount point from that service. You can do this using the Edit command from the Actions menu for
that nameservice, and later add the mount point to the new nameservice.
• After you have brought up the new nameservice, you must create the directories that correspond with
the mount points you specified in the new namespace.
• If a mount point corresponds to a directory that formerly was under a different nameservice, you must
also move any contents of that directory, if appropriate as described in step 8.
• If an HBase service is set to depend on the federated HDFS service, edit the mount points of the existing
nameservice to reference:
– HBase root directory (default /hbase)
– MapReduce system directory (default /tmp/mapred/system)
– MapReduce JobTracker staging root directory (default value /user).
c. If you want to configure high availability for the nameservice, leave the Highly Available checkbox checked.
d. Click Continue.
4. Select the hosts on which the new NameNode and Secondary NameNodes will be created. (These must be hosts
that are not already running other NameNode or SecondaryNameNode instances, and their /dfs/nn and /dfs/snn
directories should be empty if they exist. Click Continue.
5. Enter or confirm the directory property values (these will differ depending on whether you are enabling high
availability for this nameservice, or not).
6. Select the Start Dependent Services checkbox if you need to create directories or move data onto the new
nameservice. Leave this checked if you want the workflow to restart services and redeploy the client configurations
as the last steps in the workflow.
7. Click Continue. If the process finished successfully, click Finish. The new nameservice displays in the Federation
and High Availability section in the Instances tab of the HDFS service.
8. Create the directories you want under the new nameservice using the CLI:
a. To create a directory in the new namespace, use the command hadoop fs -mkdir
/nameservices/nameservice/directory where nameservice is the new nameservice you just created
and directory is the directory that corresponds to a mount point you specified.
b. To move data from one nameservice to another, use distcp or manual export/import. dfs -cp and dfs
-mv will not work.
c. Verify that the directories and data are where you expect them to be.
9. Restart the dependent services.
Note: The monitoring configurations at the HDFS level apply to all nameservices. If you have two
nameservices, it is not possible to disable a check on one but not the other. Likewise, it's not possible
to have different event thresholds for the two nameservices.
Cloudera Administration | 135
Managing CDH and Managed Services
Also see Changing a Nameservice Name for Highly Available HDFS Using Cloudera Manager on page 315.
Nameservice and Quorum-based Storage
With Quorum-based Storage, JournalNodes are shared across nameservices. So, if JournalNodes are present in an
HDFS service, all nameservices will have Quorum-based Storage enabled. To override this:
• The dfs.namenode.shared.edits.dir configuration of the two NameNodes of a high availability nameservice
should be configured to include the value of the dfs.namenode.name.dirs setting, or
• The dfs.namenode.edits.dir configuration of the one NameNode of a non-high availability nameservice
should be configured to include the value of the dfs.namenode.name.dirs setting.
NameNodes
NameNodes maintain the namespace tree for HDFS and a mapping of file blocks to DataNodes where the data is stored.
A simple HDFS cluster can have only one primary NameNode, supported by a secondary NameNode that periodically
compresses the NameNode edits log file that contains a list of HDFS metadata modifications. This reduces the amount
of disk space consumed by the log file on the NameNode, which also reduces the restart time for the primary NameNode.
A high availability cluster contains two NameNodes: active and standby.
Formatting the NameNode and Creating the /tmp Directory
Formatting the NameNode and Creating the /tmp Directory Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
When you add an HDFS service, the wizard automatically formats the NameNode and creates the /tmp directory on
HDFS. If you quit the wizard or it does not finish, you can format the NameNode and create the /tmp directory outside
the wizard by doing these steps:
1.
2.
3.
4.
5.
6.
Stop the HDFS service if it is running. See Starting, Stopping, and Restarting Services on page 39.
Click the Instances tab.
Click the NameNode role instance.
Select Actions > Format.
Start the HDFS service.
Select Actions > Create /tmp Directory.
Formatting the NameNode and Creating the /tmp Directory Using the Command Line
See Formatting the NameNode.
Backing Up and Restoring HDFS Metadata
Backing Up HDFS Metadata Using Cloudera Manager
HDFS metadata backups can be used to restore a NameNode when both NameNode roles have failed. In addition,
Cloudera recommends backing up HDFS metadata before a major upgrade.
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
This backup method requires you to shut down the cluster.
1. Note the active NameNode.
2. Stop the cluster. It is particularly important that the NameNode role process is not running so that you can make
a consistent backup.
3. Go to the HDFS service.
4. Click the Configuration tab.
5. In the Search field, search for "NameNode Data Directories" and note the value.
6. On the active NameNode host, back up the directory listed in the NameNode Data Directories property. If more
than one is listed, make a backup of one directory, since each directory is a complete copy. For example, if the
NameNode data directory is /data/dfs/nn, do the following as root:
# cd /data/dfs/nn
# tar -cvf /root/nn_backup_data.tar .
136 | Cloudera Administration
Managing CDH and Managed Services
You should see output like this:
./
./current/
./current/fsimage
./current/fstime
./current/VERSION
./current/edits
./image/
./image/fsimage
If there is a file with the extension lock in the NameNode data directory, the NameNode most likely is still running.
Repeat the steps, starting by shutting down the NameNode role.
Restoring HDFS Metadata From a Backup Using Cloudera Manager
The following process assumes a scenario where both NameNode hosts have failed and you must restore from a
backup.
1.
2.
3.
4.
5.
6.
7.
Remove the NameNode, JournalNode, and Failover Controller roles from the HDFS service.
Add the host on which the NameNode role will run.
Create the NameNode data directory, ensuring that the permissions, ownership, and group are set correctly.
Copy the backed up files to the NameNode data directory.
Add the NameNode role to the host.
Add the Secondary NameNode role to another host.
Enable high availability. If not all roles are started after the wizard completes, restart the HDFS service. Upon
startup, the NameNode reads the fsimage file and loads it into memory. If the JournalNodes are up and running
and there are edit files present, any edits newer than the fsimage are applied.
Moving NameNode Roles
This section describes two procedures for moving NameNode roles. Both procedures require cluster downtime. If
highly availability is enabled for the NameNode, you can use a Cloudera Manager wizard to automate the migration
process. Otherwise you must manually delete and add the NameNode role to a new host.
After moving a NameNode, if you have a Hive or Impala service, perform the steps in NameNode Post-Migration Steps
on page 139.
Moving Highly Available NameNode, Failover Controller, and JournalNode Roles Using the Migrate Roles Wizard
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
The Migrate Roles wizard allows you to move roles of a highly available HDFS service from one host to another. You
can use it to move NameNode, JournalNode, and Failover Controller roles.
Requirements and Limitations
• Nameservice federation (multiple namespaces) is not supported.
• This procedure requires cluster downtime, not a shutdown. The services discussed in this list must be running for
the migration to complete.
• The configuration of HDFS and services that depend on it must be valid.
• The destination host must be commissioned and healthy.
• The NameNode must be highly available using quorum-based storage.
• HDFS automatic failover must be enabled, and the cluster must have a running ZooKeeper service.
• If a Hue service is present in the cluster, its HDFS Web Interface Role property must refer to an HttpFS role, not
to a NameNode role.
• A majority of configured JournalNode roles must be running.
• The Failover Controller role that is not located on the source host must be running.
Before You Begin
Do the following before you run the wizard:
Cloudera Administration | 137
Managing CDH and Managed Services
•
•
•
•
On hosts running active and standby NameNodes, back up the data directories.
On hosts running JournalNodes, back up the JournalNode edits directory.
If the source host is not functioning properly, or is not reliably reachable, decommission the host.
If CDH and HDFS metadata was recently upgraded, and the metadata upgrade was not finalized, finalize the
metadata upgrade.
Running the Migrate Roles Wizard
1. If the host to which you want to move the NameNode is not in the cluster, follow the instructions in Adding a Host
to the Cluster on page 54 to add the host.
2. Go to the HDFS service.
3. Click the Instances tab.
4. Click the Migrate Roles button.
5. Click the Source Host text field and specify the host running the roles to migrate. In the Search field optionally
enter hostnames to filter the list of hosts and click Search.
The following shortcuts for specifying hostname patterns are supported:
• Range of hostnames (without the domain portion)
Range Definition
Matching Hosts
10.1.1.[1-4]
10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4
host[1-3].company.com
host1.company.com, host2.company.com, host3.company.com
host[07-10].company.com
host07.company.com, host08.company.com, host09.company.com,
host10.company.com
• IP addresses
• Rack name
6.
7.
8.
9.
Select the checkboxes next to the desired host. The list of available roles to migrate displays. Deselect any roles
you do not want to migrate. When migrating a NameNode, the co-located Failover Controller must be migrated
as well.
Click the Destination Host text field and specify the host to which the roles will be migrated. On destination hosts,
indicate whether to delete data in the NameNode data directories and JournalNode edits directory. If you choose
not to delete data and such role data exists, the Migrate Roles command will not complete successfully.
Acknowledge that the migration process incurs service unavailability by selecting the Yes, I am ready to restart
the cluster now checkbox.
Click Continue. The Command Progress screen displays listing each step in the migration process.
When the migration completes, click Finish.
Moving a NameNode to a Different Host Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
Note: This procedure requires cluster downtime.
1. If the host to which you want to move the NameNode is not in the cluster, follow the instructions in Adding a Host
to the Cluster on page 54 to add the host.
2. Stop all cluster services.
3. Make a backup of the dfs.name.dir directories on the existing NameNode host. Make sure you back up the
fsimage and edits files. They should be the same across all of the directories specified by the dfs.name.dir
property.
4. Copy the files you backed up from dfs.name.dir directories on the old NameNode host to the host where you
want to run the NameNode.
138 | Cloudera Administration
Managing CDH and Managed Services
5. Go to the HDFS service.
6. Click the Instances tab.
7. Select the checkbox next to the NameNode role instance and then click the Delete button. Click Delete again to
confirm.
8. In the Review configuration changes page that appears, click Skip.
9. Click Add Role Instances to add a NameNode role instance.
10. Select the host where you want to run the NameNode and then click Continue.
11. Specify the location of the dfs.name.dir directories where you copied the data on the new host, and then click
Accept Changes.
12. Start cluster services. After the HDFS service has started, Cloudera Manager distributes the new configuration
files to the DataNodes, which will be configured with the IP address of the new NameNode host.
NameNode Post-Migration Steps
After moving a NameNode, if you have a Hive or Impala service, perform the following steps:
1.
2.
3.
4.
Go to the Hive service.
Stop the Hive service.
Select Actions > Update Hive Metastore NameNodes.
If you have an Impala service, restart the Impala service or run an INVALIDATE METADATA query.
DataNodes
DataNodes store data in a Hadoop cluster and is the name of the daemon that manages the data. File data is replicated
on multiple DataNodes for reliability and so that localized computation can be executed near the data.
How NameNode Manages Blocks on a Failed DataNode
After a period without any heartbeats (which by default is 10.5 minutes), a DataNode is assumed to be failed. The
following describes how the NameNode manages block replication in such cases.
1. NameNode determines which blocks were on the failed DataNode.
2. NameNode locates other DataNodes with copies of these blocks.
3. The DataNodes with block copies are instructed to copy those blocks to other DataNodes to maintain the configured
replication factor.
4. Follow the procedure in Replacing a Disk on a DataNode Host on page 139 or Performing Disk Hot Swap for
DataNodes on page 141 to bring a repaired DataNode back online.
Replacing a Disk on a DataNode Host
Minimum Required Role: Operator (also provided by Configurator, Cluster Administrator, Full Administrator)
For CDH 5.3 and higher, see Performing Disk Hot Swap for DataNodes on page 141.
If one of your DataNode hosts experiences a disk failure, follow this process to replace the disk:
1.
2.
3.
4.
5.
Stop managed services.
Decommission the DataNode role instance.
Replace the failed disk.
Recommission the DataNode role instance.
Run the HDFS fsck utility to validate the health of HDFS. The utility normally reports over-replicated blocks
immediately after a DataNode is reintroduced to the cluster, which is automatically corrected over time.
6. Start managed services.
Adding and Removing Storage Directories for DataNodes
Adding and Removing Storage Directories Using Cloudera Manager
Adding Storage Directories
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Cloudera Administration | 139
Managing CDH and Managed Services
1.
2.
3.
4.
Go to the HDFS service.
Click the Configuration tab.
Select Scope > DataNode.
Add the new storage directory to the DataNode Data Directory property.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
5. Click Save Changes to commit the changes.
Removing Storage Directories
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
1.
2.
3.
4.
5.
Stop the cluster.
Go to the HDFS service.
Click the Configuration tab.
Select Scope > DataNode.
Remove the current directories and add new ones to the DataNode Data Directory property.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
6. Click Save Changes to commit the changes.
7. Copy the contents under the old directory to the new directory.
8. Start the cluster.
Configuring Storage-Balancing for DataNodes
You can configure HDFS to distribute writes on each DataNode in a manner that balances out available storage among
that DataNode's disk volumes.
By default a DataNode writes new block replicas to disk volumes solely on a round-robin basis. You can configure a
volume-choosing policy that causes the DataNode to take into account how much space is available on each volume
when deciding where to place a new replica.
You can configure
• how much DataNode volumes are allowed to differ in terms of bytes of free disk space before they are considered
imbalanced, and
• what percentage of new block allocations will be sent to volumes with more available disk space than others.
Configuring Storage-Balancing for DataNodes Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
1.
2.
3.
4.
5.
Go to the HDFS service.
Click the Configuration tab.
Select Scope > DataNode.
Select Category > Advanced.
Configure the following properties (you can use the Search box to locate the properties):
Property
Value
dfs.datanode.
fsdataset.
volume.choosing.
policy
org.apache.hadoop.
Enables storage balancing among the DataNode's
hdfs.server.datanode.
volumes.
fsdataset.
AvailableSpaceVolumeChoosingPolicy
dfs.datanode.
available-spacevolume-choosingpolicy.balancedspace-threshold
10737418240 (default)
140 | Cloudera Administration
Description
The amount by which volumes are allowed to differ
from each other in terms of bytes of free disk space
before they are considered imbalanced. The default
is 10737418240 (10 GB).
Managing CDH and Managed Services
Property
Value
Description
If the free space on each volume is within this range
of the other volumes, the volumes will be
considered balanced and block assignments will be
done on a pure round-robin basis.
dfs.datanode.
available-spacevolume-choosingpolicy.balancedspace-preferencefraction
0.75 (default)
What proportion of new block allocations will be
sent to volumes with more available disk space than
others. The allowable range is 0.0-1.0, but set it in
the range 0.5 - 1.0 (that is, 50-100%), since there
should be no reason to prefer that volumes with
less available disk space receive more block
allocations.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
6. Click Save Changes to commit the changes.
7. Restart the role.
Configuring Storage-Balancing for DataNodes Using the Command Line
See Configuring Storage-Balancing for DataNodes.
Performing Disk Hot Swap for DataNodes
This section describes how to replace HDFS disks without shutting down a DataNode. This is referred to as hot swap.
Warning: Requirements and Limitations
•
•
•
•
Hot swap is supported for CDH 5.4 and higher.
Hot swap can only add disks with empty data directories.
Removing a disk does not move the data off the disk, which could potentially result in data loss.
Do not perform hot swap on multiple hosts at the same time.
Performing Disk Hot Swap for DataNodes Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
1. Configure data directories to remove the disk you are swapping out:
a.
b.
c.
d.
e.
f.
g.
Go to the HDFS service.
Click the Instances tab.
Click the affected DataNode.
Click the Configuration tab.
Select Scope > DataNode.
Select Category > Main.
Change the value of the DataNode Data Directory property to remove the directories that are mount points
for the disk you are removing.
Warning: Change the value of this property only for the specific DataNode instance where
you are planning to hot swap the disk. Do not edit the role group value for this property.
Doing so will cause data loss.
2. Click Save Changes to commit the changes.
3. Refresh the affected DataNode. Select Actions > Refresh Data Directories.
Cloudera Administration | 141
Managing CDH and Managed Services
4. Remove the old disk and add the replacement disk.
5. Change the value of the DataNode Data Directory property to add back the directories that are mount points for
the disk you added.
6. Click Save Changes to commit the changes.
7. Refresh the affected DataNode. Select Actions > Refresh Data Directories.
8. Run the HDFS fsck utility to validate the health of HDFS.
Performing Disk Hot Swap for DataNodes Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
Use these instructions to perform hot swap of disks in a cluster that is not managed by Cloudera Manager
To add and remove disks:
1. If you are adding disks, format and mount them.
2. Change the value of dfs.datanode.data.dir in hdfs-site.xml on the DataNode to reflect the directories
that will be used from now on (add new points and remove obsolete ones). For more information, see the
instructions for DataNodes under Configuring Local Storage Directories.
3. Start the reconfiguration process:
• If Kerberos is enabled:
$ kinit -kt /path/to/hdfs.keytab hdfs/<fully.qualified.domain.name@YOUR-REALM.COM> &&
dfsadmin -reconfig datanode HOST:PORT start
• If Kerberos is not enabled:
$ sudo -u hdfs hdfs dfsadmin -reconfig datanode HOST:PORT start
where HOST:PORT is the DataNode's dfs.datanode.ipc.address (or its hostname and the port specified in
dfs.datanode.ipc.address; for example dnhost1.example.com:5678)
To check on the progress of the reconfiguration, you can use the status option of the command; for example,
if Kerberos is not enabled:
$ sudo -u hdfs hdfs dfsadmin -reconfig datanode HOST:PORT status
4. Once the reconfiguration is complete, unmount any disks you have removed from the configuration.
5. Run the HDFS fsck utility to validate the health of HDFS.
To perform maintenance on a disk:
1. Change the value of dfs.datanode.data.dir in hdfs-site.xml on the DataNode to exclude the mount point
directories that reside on the affected disk and reflect only the directories that will be used during the maintenance
window. For more information, see the instructions for DataNodes under Configuring Local Storage Directories.
2. Start the reconfiguration process:
• If Kerberos is enabled:
$ kinit -kt /path/to/hdfs.keytab hdfs/<fully.qualified.domain.name@YOUR-REALM.COM> &&
dfsadmin -reconfig datanode HOST:PORT start
142 | Cloudera Administration
Managing CDH and Managed Services
• If Kerberos is not enabled:
$ sudo -u hdfs hdfs dfsadmin -reconfig datanode HOST:PORT start
where HOST:PORT is the DataNode's dfs.datanode.ipc.address, or its hostname and the port specified in
dfs.datanode.ipc.address.
To check on the progress of the reconfiguration, you can use the status option of the command; for example,
if Kerberos is not enabled:
$ sudo -u hdfs hdfs dfsadmin -reconfig datanode HOST:PORT status
3.
4.
5.
6.
7.
8.
Once the reconfiguration is complete, unmount the disk.
Perform maintenance on the disk.
Remount the disk.
Change the value of dfs.datanode.data.dir again to reflect the original set of mount points.
Repeat step 2.
Run the HDFS fsck utility to validate the health of HDFS.
JournalNodes
High-availabilty clusters use JournalNodes to synchronize active and standby NameNodes. The active NameNode writes
to each JournalNode with changes, or "edits," to HDFS namespace metadata. During failover, the standby NameNode
applies all edits from the JournalNodes before promoting itself to the active state.
Moving the JournalNode Edits Directory
Moving the JournalNode Edits Directory for an Role Instance Using Cloudera Manager
To change the location of the edits directory for one JournalNode instance:
1. Reconfigure the JournalNode Edits Directory.
a.
b.
c.
d.
e.
f.
Go to the HDFS service in Cloudera Manager.
Click JournalNode under Status Summary.
Click the JournalNode link for the instance you are changing.
Click the Configuration tab.
Set dfs.journalnode.edits.dir to the path of the new jn directory.
Click Save Changes.
2. Move the location of the JournalNode (jn) directory at the command line:
a. Connect to host of the JournalNode.
b. Copy the JournalNode (jn) directory to its new location with the -a option to preserve permissions:
cp -a /<old_path_to_jn_dir>/jn /<new_path_to_jn_dir>/jn
c. Rename the old jn directory to avoid confusion:
mv /<old_path_to_jn_dir>/jn /<old_path_to_jn_dir>/jn_to_delete
3. Redeploy the HDFS client configuration:
a. Go to the HDFS service.
b. Select Actions > Deploy Client Configuration.
4. Perform a Rolling Restart on page 41 for HDFS by selecting Actions > Rolling Restart. Use the default settings.
5. From the command line, delete the old jn_to_delete directory.
Cloudera Administration | 143
Managing CDH and Managed Services
Moving the JournalNode Edits Directory for a Role Group Using Cloudera Manager
To change the location of the edits directory for each JournalNode in the JournalNode Default Group:
1. Stop all services on the cluster in Cloudera Manager:
a. Go to the Cluster.
b. Select Actions > Stop.
2. Find the list of JournalNode hosts:
a. Go to the HDFS service.
b. Click JournalNode under Status Summary.
3. Move the location of each JournalNode (jn) directory at the command line:
a. Connect to each host with a JournalNode.
b. Per host, copy the JournalNode (jn) directory to its new location with the -a option to preserve permissions:
cp -a /<old_path_to_jn_dir>/jn /<new_path_to_jn_dir>/jn
c. Per host, rename the old jn directory to avoid confusion:
mv /<old_path_to_jn_dir>/jn /<old_path_to_jn_dir>/jn_to_delete
4. Reconfigure the JournalNode Default Group:
a.
b.
c.
d.
e.
Go to the HDFS service.
Click the Configuration tab.
Click JournalNode under Scope.
Set dfs.journalnode.edits.dir to the path of the new jn directory for all JournalNodes in the group.
Click Save Changes.
5. Redeploy the client configuration for the cluster:
a. Go to the Cluster.
b. Select Actions > Deploy Client Configuration.
6. Start all services on the cluster by selecting Actions > Start.
7. Delete the old jn_to_delete directories from the command line.
Moving JournalNodes Across Hosts
To move JournalNodes to a new host, see Moving Highly Available NameNode, Failover Controller, and JournalNode
Roles Using the Migrate Roles Wizard on page 137.
Configuring Short-Circuit Reads
So-called "short-circuit" reads bypass the DataNode, allowing a client to read the file directly, as long as the client is
co-located with the data. Short-circuit reads provide a substantial performance boost to many applications and help
improve HBase random read profile and Impala performance.
Short-circuit reads require libhadoop.so (the Hadoop Native Library) to be accessible to both the server and the
client. libhadoop.so is not available if you have installed from a tarball. You must install from an .rpm, .deb, or
parcel in order to use short-circuit local reads.
Configuring Short-Circuit Reads Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Note: Short-circuit reads are enabled by default in Cloudera Manager.
144 | Cloudera Administration
Managing CDH and Managed Services
1.
2.
3.
4.
5.
Go to the HDFS service.
Click the Configuration tab.
Select Scope > Gateway or HDFS (Service-Wide).
Select Category > Performance.
Locate the Enable HDFS Short Circuit Read property or search for it by typing its name in the Search box. Check
the box to enable it.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
6. Click Save Changes to commit the changes.
Configuring Short-Circuit Reads Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
Configure the following properties in hdfs-site.xml to enable short-circuit reads in a cluster that is not managed
by Cloudera Manager:
<property>
<name>dfs.client.read.shortcircuit</name>
<value>true</value>
</property>
<property>
<name>dfs.client.read.shortcircuit.streams.cache.size</name>
<value>1000</value>
</property>
<property>
<name>dfs.client.read.shortcircuit.streams.cache.expiry.ms</name>
<value>10000</value>
</property>
<property>
<name>dfs.domain.socket.path</name>
<value>/var/run/hadoop-hdfs/dn._PORT</value>
</property>
Note: The text _PORT appears just as shown; you do not need to substitute a number.
If /var/run/hadoop-hdfs/ is group-writable, make sure its group is root.
Configuring HDFS Trash
The Hadoop trash feature helps prevent accidental deletion of files and directories. If trash is enabled and a file or
directory is deleted using the Hadoop shell, the file is moved to the .Trash directory in the user's home directory
instead of being deleted. Deleted files are initially moved to the Current sub-directory of the .Trash directory, and
their original path is preserved. If trash checkpointing is enabled, the Current directory is periodically renamed using
a timestamp. Files in .Trash are permanently removed after a user-configurable time delay. Files and directories in
the trash can be restored simply by moving them to a location outside the .Trash directory.
Cloudera Administration | 145
Managing CDH and Managed Services
Important:
• The trash feature is disabled by default. Cloudera recommends that you enable it on all production
clusters.
• The trash feature works by default only for files and directories deleted using the Hadoop shell.
Files or directories deleted programmatically using other interfaces (WebHDFS or the Java APIs,
for example) are not moved to trash, even if trash is enabled, unless the program has implemented
a call to the trash functionality. (Hue, for example, implements trash as of CDH 4.4.)
Users can bypass trash when deleting files using the shell by specifying the -skipTrash option
to the hadoop fs -rm -r command. This can be useful when it is necessary to delete files that
are too large for the user's quota.
Configuring HDFS Trash Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Enabling and Disabling Trash
1.
2.
3.
4.
Go to the HDFS service.
Click the Configuration tab.
Select Scope > Gateway.
Select or deselect the Use Trash checkbox.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
5. Click Save Changes to commit the changes.
6. Restart the cluster and deploy the cluster client configuration.
Setting the Trash Interval
1.
2.
3.
4.
Go to the HDFS service.
Click the Configuration tab.
Select Scope > NameNode.
Specify the Filesystem Trash Interval property, which controls the number of minutes after which a trash checkpoint
directory is deleted and the number of minutes between trash checkpoints. For example, to enable trash so that
deleted files are deleted after 24 hours, set the value of the Filesystem Trash Interval property to 1440.
Note: The trash interval is measured from the point at which the files are moved to trash, not
from the last time the files were modified.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
5. Click Save Changes to commit the changes.
6. Restart all NameNodes.
Configuring HDFS Trash Using the Command Line
See Enabling Trash.
HDFS Balancers
HDFS data might not always be distributed uniformly across DataNodes. One common reason is addition of new
DataNodes to an existing cluster. HDFS provides a balancer utility that analyzes block placement and balances data
across the DataNodes. The balancer moves blocks until the cluster is deemed to be balanced, which means that the
146 | Cloudera Administration
Managing CDH and Managed Services
utilization of every DataNode (ratio of used space on the node to total capacity of the node) differs from the utilization
of the cluster (ratio of used space on the cluster to total capacity of the cluster) by no more than a given threshold
percentage. The balancer does not balance between individual volumes on a single DataNode.
Configuring and Running the HDFS Balancer Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
In Cloudera Manager, the HDFS balancer utility is implemented by the Balancer role. The Balancer role usually shows
a health of None on the HDFS Instances tab because it does not run continuously.
The Balancer role is normally added (by default) when the HDFS service is installed. If it has not been added, you must
add a Balancer role in order to rebalance HDFS and to see the Rebalance action.
Configuring the Balancer Threshold
The Balancer has a default threshold of 10%, which ensures that disk usage on each DataNode differs from the overall
usage in the cluster by no more than 10%. For example, if overall usage across all the DataNodes in the cluster is 40%
of the cluster's total disk-storage capacity, the script ensures that DataNode disk usage is between 30% and 50% of
the DataNode disk-storage capacity. To change the threshold:
1.
2.
3.
4.
5.
Go to the HDFS service.
Click the Configuration tab.
Select Scope > Balancer.
Select Category > Main.
Set the Rebalancing Threshold property.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
6. Click Save Changes to commit the changes.
Running the Balancer
1.
2.
3.
4.
Go to the HDFS service.
Ensure the service has a Balancer role.
Select Actions > Rebalance.
Click Rebalance to confirm. If you see a Finished status, the Balancer ran successfully.
Configuring and Running the HDFS Balancer Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
The HDFS balancer re-balances data across the DataNodes, moving blocks from overutilized to underutilized nodes.
As the system administrator, you can run the balancer from the command-line as necessary -- for example, after adding
new DataNodes to the cluster.
Points to note:
• The balancer requires the capabilities of an HDFS superuser (for example, the hdfs user) to run.
• The balancer does not balance between individual volumes on a single DataNode.
• You can run the balancer without parameters, as follows:
sudo -u hdfs hdfs balancer
Cloudera Administration | 147
Managing CDH and Managed Services
Note: If Kerberos is enabled, do not use commands in the form sudo -u <user> hadoop
<command>; they will fail with a security error. Instead, use the following commands: $ kinit
<user> (if you are using a password) or $ kinit -kt <keytab> <principal> (if you are
using a keytab) and then, for each command executed by this user, $ <command>
This runs the balancer with a default threshold of 10%, meaning that the script will ensure that disk usage on each
DataNode differs from the overall usage in the cluster by no more than 10%. For example, if overall usage across
all the DataNodes in the cluster is 40% of the cluster's total disk-storage capacity, the script ensures that each
DataNode's disk usage is between 30% and 50% of that DataNode's disk-storage capacity.
• You can run the script with a different threshold; for example:
sudo -u hdfs hdfs balancer -threshold 5
This specifies that each DataNode's disk usage must be (or will be adjusted to be) within 5% of the cluster's overall
usage.
• You can adjust the network bandwidth used by the balancer, by running the dfsadmin -setBalancerBandwidth
command before you run the balancer; for example:
dfsadmin -setBalancerBandwidth newbandwidth
where newbandwidth is the maximum amount of network bandwidth, in bytes per second, that each DataNode
can use during the balancing operation. For more information about the bandwidth command, see
BalancerBandwidthCommand.
• The balancer can take a long time to run, especially if you are running it for the first time, or do not run it regularly.
Enabling WebHDFS
Enabling WebHDFS Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
To enable WebHDFS, proceed as follows:
1.
2.
3.
4.
5.
6.
Select the HDFS service.
Click the Configuration tab.
Select Scope > HDFS-1 (Service Wide)
Select the Enable WebHDFS property.
Click the Save Changes button.
Restart the HDFS service.
You can find a full explanation of the WebHDFS API in the WebHDFS API documentation.
Enabling WebHDFS Using the Command Line
See Enabling WebHDFS.
Adding HttpFS
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
Apache Hadoop HttpFS is a service that provides HTTP access to HDFS.
HttpFS has a REST HTTP API supporting all HDFS filesystem operations (both read and write).
Common HttpFS use cases are:
• Read and write data in HDFS using HTTP utilities (such as curl or wget) and HTTP libraries from languages other
than Java (such as Perl).
• Transfer data between HDFS clusters running different versions of Hadoop (overcoming RPC versioning issues),
for example using Hadoop DistCp.
148 | Cloudera Administration
Managing CDH and Managed Services
• Read and write data in HDFS in a cluster behind a firewall. (The HttpFS server acts as a gateway and is the only
system that is allowed to send and receive data through the firewall).
HttpFS supports Hadoop pseudo-authentication, HTTP SPNEGO Kerberos, and additional authentication mechanisms
using a plugin API. HttpFS also supports Hadoop proxy user functionality.
The webhdfs client file system implementation can access HttpFS using the Hadoop filesystem command (hadoop
fs), by using Hadoop DistCp, and from Java applications using the Hadoop file system Java API.
The HttpFS HTTP REST API is interoperable with the WebHDFS REST HTTP API.
For more information about HttpFS, see Hadoop HDFS over HTTP.
The HttpFS role is required for Hue when you enable HDFS high availability.
Adding the HttpFS Role
1.
2.
3.
4.
5.
6.
7.
Go to the HDFS service.
Click the Instances tab.
Click Add Role Instances.
Click the text box below the HttpFS field. The Select Hosts dialog box displays.
Select the host on which to run the role and click OK.
Click Continue.
Check the checkbox next to the HttpFS role and select Actions for Selected > Start.
Using Load Balancer with HttpFS
To configure a load balancer, select Clusters > HDFS > Configuration > Category > HttpFS and enter the hostname and
port number of the load balancer in the HTTPFS Load Balancer property in the format hostname:port number.
Note:
When you set this property, Cloudera Manager regenerates the keytabs for HttpFS roles. The principal
in these keytabs contains the load balancer hostname.
If there is a Hue service that depends on this HDFS service, the Hue service has the option to use the
load balancer as its HDFS Web Interface Role.
Adding and Configuring an NFS Gateway
The NFSv3 gateway allows a client to mount HDFS as part of the client's local file system. The gateway machine can
be any host in the cluster, including the NameNode, a DataNode, or any HDFS client. The client can be any
NFSv3-client-compatible machine.
Important:
HDFS does not currently provide ACL support for an NFS gateway.
After mounting HDFS to his or her local filesystem, a user can:
• Browse the HDFS file system as though it were part of the local file system
• Upload and download files from the HDFS file system to and from the local file system.
• Stream data directly to HDFS through the mount point.
File append is supported, but random write is not.
Adding and Configuring an NFS Gateway Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
The NFS Gateway role implements an NFSv3 gateway. It is an optional role for a CDH 5 HDFS service.
Cloudera Administration | 149
Managing CDH and Managed Services
Requirements and Limitations
• The NFS gateway works only with the following operating systems and Cloudera Manager and CDH versions:
– With Cloudera Manager 5.0.1 or higher and CDH 5.0.1 or higher, the NFS gateway works on all operating
systems supported by Cloudera Manager.
– With Cloudera Manager 5.0.0 or CDH 5.0.0, the NFS gateway only works on RHEL and similar systems.
– The NFS gateway is not supported on versions lower than Cloudera Manager 5.0.0 and CDH 5.0.0.
• If any NFS server is already running on the NFS Gateway host, it must be stopped before the NFS Gateway role is
started.
• There are two configuration options related to NFS Gateway role: Temporary Dump Directory and Allowed Hosts
and Privileges. The Temporary Dump Directory is automatically created by the NFS Gateway role and should be
configured before starting the role.
• The Access Time Precision property in the HDFS service must be enabled.
Adding and Configuring the NFS Gateway Role
1. Go to the HDFS service.
2. Click the Instances tab.
3. Click Add Role Instances.
4. Click the text box below the NFS Gateway field. The Select Hosts dialog box displays.
5. Select the host on which to run the role and click OK.
6. Click Continue.
7. Click the NFS Gateway role.
8. Click the Configuration tab.
9. Select Scope > NFS Gateway.
10. Select Category > Main.
11. Ensure that the requirements on the directory set in the Temporary Dump Directory property are met.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
12. Optionally edit Allowed Hosts and Privileges.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
13. Click Save Changes to commit the changes.
14. Click the Instances tab.
15. Check the checkbox next to the NFS Gateway role and select Actions for Selected > Start.
Configuring an NFSv3 Gateway Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
The subsections that follow provide information on installing and configuring the gateway.
150 | Cloudera Administration
Managing CDH and Managed Services
Note: Install Cloudera Repository
Before using the instructions on this page to install or upgrade, install the Cloudera yum, zypper/YaST
or apt repository, and install or upgrade CDH 5 and make sure it is functioning correctly. For
instructions, see Installing the Latest CDH 5 Release and Upgrading Unmanaged CDH Using the
Command Line.
Upgrading from a CDH 5 Beta Release
If you are upgrading from a CDH 5 Beta release, you must first remove the hadoop-hdfs-portmap package. Proceed
as follows.
1. Unmount existing HDFS gateway mounts. For example, on each client, assuming the file system is mounted on
/hdfs_nfs_mount:
$ umount /hdfs_nfs_mount
2. Stop the services:
$ sudo service hadoop-hdfs-nfs3 stop
$ sudo hadoop-hdfs-portmap stop
3. Remove the hadoop-hdfs-portmap package.
• On a RHEL-compatible system:
$ sudo yum remove hadoop-hdfs-portmap
• On a SLES system:
$ sudo zypper remove hadoop-hdfs-portmap
• On an Ubuntu or Debian system:
$ sudo apt-get remove hadoop-hdfs-portmap
4. Install the new version
• On a RHEL-compatible system:
$ sudo yum install hadoop-hdfs-nfs3
• On a SLES system:
$ sudo zypper install hadoop-hdfs-nfs3
• On an Ubuntu or Debian system:
$ sudo apt-get install hadoop-hdfs-nfs3
5. Start the system default portmapper service:
$ sudo service portmap start
6. Now proceed with Starting the NFSv3 Gateway on page 153, and then remount the HDFS gateway mounts.
Installing the Packages for the First Time
On RHEL and similar systems:
Cloudera Administration | 151
Managing CDH and Managed Services
Install the following packages on the cluster host you choose for NFSv3 Gateway machine (we'll refer to it as the NFS
server from here on).
• nfs-utils
• nfs-utils-lib
• hadoop-hdfs-nfs3
The first two items are standard NFS utilities; the last is a CDH package.
Use the following command:
$ sudo yum install nfs-utils nfs-utils-lib hadoop-hdfs-nfs3
On SLES:
Install nfs-utils on the cluster host you choose for NFSv3 Gateway machine (referred to as the NFS server from
here on):
$ sudo zypper install nfs-utils
On an Ubuntu or Debian system:
Install nfs-common on the cluster host you choose for NFSv3 Gateway machine (referred to as the NFS server from
here on):
$ sudo apt-get install nfs-common
Configuring the NFSv3 Gateway
Proceed as follows to configure the gateway.
1. Add the following property to hdfs-site.xml on the NameNode:
<property>
<name>dfs.namenode.accesstime.precision</name>
<value>3600000</value>
<description>The access time for an HDFS file is precise up to this value. The
default value is 1 hour.
Setting a value of 0 disables access times for HDFS.</description>
</property>
2. Add the following property to hdfs-site.xml on the NFS server:
<property>
<name>dfs.nfs3.dump.dir</name>
<value>/tmp/.hdfs-nfs</value>
</property>
Note:
You should change the location of the file dump directory, which temporarily saves out-of-order
writes before writing them to HDFS. This directory is needed because the NFS client often reorders
writes, and so sequential writes can arrive at the NFS gateway in random order and need to be
saved until they can be ordered correctly. After these out-of-order writes have exceeded 1MB in
memory for any given file, they are dumped to the dfs.nfs3.dump.dir (the memory threshold
is not currently configurable).
Make sure the directory you choose has enough space. For example, if an application uploads 10
files of 100MB each, dfs.nfs3.dump.dir should have roughly 1GB of free space to allow for
a worst-case reordering of writes to every file.
152 | Cloudera Administration
Managing CDH and Managed Services
3. Configure the user running the gateway (normally the hdfs user as in this example) to be a proxy for other users.
To allow the hdfs user to be a proxy for all other users, add the following entries to core-site.xml on the
NameNode:
<property>
<name>hadoop.proxyuser.hdfs.groups</name>
<value>*</value>
<description>
Set this to '*' to allow the gateway user to proxy any group.
</description>
</property>
<property>
<name>hadoop.proxyuser.hdfs.hosts</name>
<value>*</value>
<description>
Set this to '*' to allow requests from any hosts to be proxied.
</description>
</property>
4. Restart the NameNode.
Starting the NFSv3 Gateway
Do the following on the NFS server.
1. First, stop the default NFS services, if they are running:
$ sudo service nfs stop
2. Start the HDFS-specific services:
$ sudo service hadoop-hdfs-nfs3 start
Verifying that the NFSv3 Gateway is Working
To verify that the NFS services are running properly, you can use the rpcinfo command on any host on the local
network:
$ rpcinfo -p <nfs_server_ip_address>
You should see output such as the following:
program
vers
proto
port
100005
100005
100005
100000
100000
100005
100005
100003
100005
1
2
2
2
2
3
1
3
3
tcp
udp
tcp
tcp
udp
udp
udp
tcp
tcp
4242
4242
4242
111
111
4242
4242
2049
4242
mountd
mountd
mountd
portmapper
portmapper
mountd
mountd
nfs
mountd
To verify that the HDFS namespace is exported and can be mounted, use the showmount command.
$ showmount -e <nfs_server_ip_address>
You should see output similar to the following:
Exports list on <nfs_server_ip_address>:
/ (everyone)
Cloudera Administration | 153
Managing CDH and Managed Services
Mounting HDFS on an NFS Client
To import the HDFS file system on an NFS client, use a mount command such as the following on the client:
$ mount -t
nfs
-o vers=3,proto=tcp,nolock <nfs_server_hostname>:/ /hdfs_nfs_mount
Note:
When you create a file or directory as user hdfs on the client (that is, in the HDFS file system imported
using the NFS mount), the ownership may differ from what it would be if you had created it in HDFS
directly. For example, ownership of a file created on the client might be hdfs:hdfs when the same
operation done natively in HDFS resulted in hdfs:supergroup. This is because in native HDFS, BSD
semantics determine the group ownership of a newly-created file: it is set to the same group as the
parent directory where the file is created. When the operation is done over NFS, the typical Linux
semantics create the file with the group of the effective GID (group ID) of the process creating the
file, and this characteristic is explicitly passed to the NFS gateway and HDFS.
Setting HDFS Quotas
You can set quotas in HDFS for:
• The number of file and directory names used
• The amount of space used by given directories
Points to note:
•
•
•
•
The quotas for names and the quotas for space are independent of each other.
File and directory creation fails if the creation would cause the quota to be exceeded.
The Reports Manager must index a file or directory before you can set a quota for it.
Allocation fails if the quota would prevent a full block from being written; keep this in mind if you are using a large
block size.
• If you are using replication, remember that each replica of a block counts against the quota.
About file count limits
• The file count quota is a limit on the number of file and directory names in the directory configured.
• A directory counts against its own quota, so a quota of 1 forces the directory to remain empty.
• File counts are based on the intended replication factor for the files; changing the replication factor for a file will
credit or debit quotas.
About disk space limits
• The space quota is a hard limit on the number of bytes used by files in the tree rooted at the directory being
configured.
• Each replica of a block counts against the quota.
• The disk space quota calculation takes replication into account, so it uses the replicated size of each file, not the
user-facing size.
• The disk space quota calculation includes open files (files presently being written), as well as files already written.
• Block allocations for files being written will fail if the quota would not allow a full block to be written.
Setting HDFS Quotas Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
1. From the HDFS service page, select the File Browser tab.
2. Browse the file system to find the directory for which you want to set quotas.
3. Click the directory name so that it appears in the gray panel above the listing of its contents and in the detail
section to the right of the File Browser table.
154 | Cloudera Administration
Managing CDH and Managed Services
4. Click the Edit Quota button for the directory. A Manage Quota pop-up displays, where you can set file count or
disk space limits for the directory you have selected.
5. When you have set the limits you want, click OK.
Setting HDFS Quotas Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
To set space quotas on a directory:
dfsadmin -setSpaceQuota n directory
where n is a number of bytes and directory is the directory the quota applies to. You can specify multiple directories
in a single command; napplies to each.
To remove space quotas from a directory:
dfsadmin -clrSpaceQuota directory
You can specify multiple directories in a single command.
To set name quotas on a directory:
dfsadmin -setQuota n directory
where n is the number of file and directory names in directory. You can specify multiple directories in a single command;
napplies to each.
To remove name quotas from a directory:
dfsadmin -clrQuota directory
You can specify multiple directories in a single command.
For More Information
For more information, see the HDFS Quotas Guide.
Configuring Mountable HDFS
CDH 5 includes a FUSE (Filesystem in Userspace) interface into HDFS. The hadoop-hdfs-fuse package enables you
to use your HDFS cluster as if it were a traditional filesystem on Linux. Proceed as follows.
Note: FUSE does not currently support file append operations.
Before you start: You must have a working HDFS cluster and know the hostname and port that your NameNode exposes.
To install hadoop-hdfs-fuses On Red Hat-compatible systems:
$ sudo yum install hadoop-hdfs-fuse
To install hadoop-hdfs-fuse on Ubuntu systems:
$ sudo apt-get install hadoop-hdfs-fuse
Cloudera Administration | 155
Managing CDH and Managed Services
To install hadoop-hdfs-fuse on SLES systems:
$ sudo zypper install hadoop-hdfs-fuse
You now have everything you need to begin mounting HDFS on Linux.
To set up and test your mount point in a non-HA installation:
$ mkdir -p <mount_point>
$ hadoop-fuse-dfs dfs://<name_node_hostname>:<namenode_port> <mount_point>
where namenode_port is the NameNode's RPC port, dfs.namenode.servicerpc-address.
To set up and test your mount point in an HA installation:
$ mkdir -p <mount_point>
$ hadoop-fuse-dfs dfs://<nameservice_id> <mount_point>
where nameservice_id is the value of fs.defaultFS. In this case the port defined for
dfs.namenode.rpc-address.[nameservice ID].[name node ID] is used automatically. See Enabling HDFS
HA on page 292 for more information about these properties.
You can now run operations as if they are on your mount point. Press Ctrl+C to end the fuse-dfs program, and
umount the partition if it is still mounted.
Note:
To find its configuration directory, hadoop-fuse-dfs uses the HADOOP_CONF_DIR configured at
the time the mount command is invoked.
To clean up your test:
$ umount <mount_point>
You can now add a permanent HDFS mount which persists through reboots.
To add a system mount:
1. Open /etc/fstab and add lines to the bottom similar to these:
hadoop-fuse-dfs#dfs://<name_node_hostname>:<namenode_port> <mount_point> fuse
allow_other,usetrash,rw 2 0
For example:
hadoop-fuse-dfs#dfs://localhost:8020 /mnt/hdfs fuse allow_other,usetrash,rw 2 0
Note:
In an HA deployment, use the HDFS nameservice instead of the NameNode URI; that is, use the
value of dfs.nameservices in hdfs-site.xml.
2. Test to make sure everything is working properly:
$ mount <mount_point>
Your system is now configured to allow you to use the ls command and use that mount point as if it were a normal
system disk.
156 | Cloudera Administration
Managing CDH and Managed Services
For more information, see the help for hadoop-fuse-dfs:
$ hadoop-fuse-dfs --help
Optimizing Mountable HDFS
• Cloudera recommends that you use the -obig_writes option on kernels later than 2.6.26. This option allows
for better performance of writes.
• By default, the CDH 5 package installation creates the /etc/default/hadoop-fuse file with a maximum heap
size of 128 MB. You might need to change the JVM minimum and maximum heap size for better performance.
For example:
export LIBHDFS_OPTS="-Xms64m -Xmx256m"
Be careful not to set the minimum to a higher value than the maximum.
Configuring Centralized Cache Management in HDFS
Centralized cache management in HDFS is an explicit caching mechanism that allows users to specify paths to be cached
by HDFS. The NameNode will communicate with DataNodes that have the desired blocks on disk, and instruct them
to cache the blocks in off-heap caches.
This has several advantages:
• Explicit pinning prevents frequently used data from being evicted from memory. This is particularly important
when the size of the working set exceeds the size of main memory, which is common for many HDFS workloads.
• Since DataNode caches are managed by the NameNode, applications can query the set of cached block locations
when making task placement decisions. Co-locating a task with a cached block replica improves read performance.
• When block has been cached by a DataNode, clients can use a new, more-efficient, zero-copy read API. Since
checksum verification of cached data is done once by the DataNode, clients can incur essentially zero overhead
when using this new API.
• Centralized caching can improve overall cluster memory utilization. When relying on the OS buffer cache at each
DataNode, repeated reads of a block will result in all n replicas of the block being pulled into buffer cache. With
centralized cache management, you can explicitly pin only m of the n replicas, saving n-m memory.
Use Cases
Centralized cache management is best used for files that are accessed repeatedly. For example, a fact table in Hive
that is often used in JOIN clauses is a good candidate for caching. Caching the input of an annual reporting query is
probably less useful, as the historical data might be read only once.
Centralized cache management is also useful for mixed workloads with performance SLAs. Caching the working set of
a high-priority workload insures that it does not contend for disk I/O with a low-priority workload.
Cloudera Administration | 157
Managing CDH and Managed Services
Architecture
In this architecture, the NameNode is responsible for coordinating all the DataNode off-heap caches in the cluster. The
NameNode periodically receives a "cache report" from each DataNode which describes all the blocks cached on a given
DataNode. The NameNode manages DataNode caches by piggybacking cache and uncache commands on the DataNode
heartbeat.
The NameNode queries its set of cache directives to determine which paths should be cached. Cache directives are
persistently stored in the fsimage and edit log, and can be added, removed, and modified using Java and command-line
APIs. The NameNode also stores a set of cache pools, which are administrative entities used to group cache directives
together for resource management and enforcing permissions.
The NameNode periodically rescans the namespace and active cache directories to determine which blocks need to
be cached or uncached and assign caching to DataNodes. Rescans can also be triggered by user actions like adding or
removing a cache directive or removing a cache pool.
We do not currently cache blocks which are under construction, corrupt, or otherwise incomplete. If a cache directive
covers a symlink, the symlink target is not cached. Caching is currently done on a per-file basis, although we would like
to add block-level granularity in the future.
Concepts
Cache Directive
A cache directive defines a path that should be cached. Paths can be either directories or files. Directories are cached
non-recursively, meaning only files in the first-level listing of the directory will be cached. Directives also specify
additional parameters, such as the cache replication factor and expiration time. The replication factor specifies the
number of block replicas to cache. If multiple cache directives refer to the same file, the maximum cache replication
factor is applied.
The expiration time is specified on the command line as a time-to-live (TTL), a relative expiration time in the future.
After a cache directive expires, it is no longer considered by the NameNode when making caching decisions.
Cache Pool
A cache pool is an administrative entity used to manage groups of cache directives. Cache pools have UNIX-like
permissions that restrict which users and groups have access to the pool. Write permissions allow users to add and
remove cache directives to the pool. Read permissions allow users to list the cache directives in a pool, as well as
additional metadata. Execute permissions are unused.
158 | Cloudera Administration
Managing CDH and Managed Services
Cache pools are also used for resource management. Pools can enforce a maximum limit, which restricts the number
of bytes that can be cached in aggregate by directives in the pool. Normally, the sum of the pool limits will approximately
equal the amount of aggregate memory reserved for HDFS caching on the cluster. Cache pools also track a number of
statistics to help cluster users determine what is and should be cached.
Pools also enforce a maximum time-to-live. This restricts the maximum expiration time of directives being added to
the pool.
cacheadmin Command-Line Interface
On the command-line, administrators and users can interact with cache pools and directives using the hdfs cacheadmin
subcommand. Cache directives are identified by a unique, non-repeating 64-bit integer ID. IDs are not reused even if
a cache directive is later removed. Cache pools are identified by a unique string name.
Cache Directive Commands
addDirective
Description: Add a new cache directive.
Usage: hdfs cacheadmin -addDirective -path <path> -pool <pool-name> [-force] [-replication
<replication>] [-ttl <time-to-live>]
Where, path: A path to cache. The path can be a directory or a file.
pool-name: The pool to which the directive will be added. You must have write permission on the cache pool in order
to add new directives.
force: Skips checking of cache pool resource limits.
replication: The cache replication factor to use. Defaults to 1.
time-to-live: Time period for which the directive is valid. Can be specified in seconds, minutes, hours, and days,
for example: 30m, 4h, 2d. The value never indicates a directive that never expires. If unspecified, the directive never
expires.
removeDirective
Description: Remove a cache directive.
Usage: hdfs cacheadmin -removeDirective <id>
Where, id: The id of the cache directive to remove. You must have write permission on the pool of the directive in
order to remove it. To see a list of PathBasedCache directive IDs, use the -listDirectives command.
removeDirectives
Description: Remove every cache directive with the specified path.
Usage: hdfs cacheadmin -removeDirectives <path>
Where, path: The path of the cache directives to remove. You must have write permission on the pool of the directive
in order to remove it.
listDirectives
Description: List PathBasedCache directives.
Usage: hdfs cacheadmin -listDirectives [-stats] [-path <path>] [-pool <pool>]
Where, path: List only PathBasedCache directives with this path. Note that if there is a PathBasedCache directive for
path in a cache pool that we do not have read access for, it will not be listed.
pool: List only path cache directives in that pool.
stats: List path-based cache directive statistics.
Cloudera Administration | 159
Managing CDH and Managed Services
Cache Pool Commands
addPool
Description: Add a new cache pool.
Usage: hdfs cacheadmin -addPool <name> [-owner <owner>] [-group <group>] [-mode <mode>]
[-limit <limit>] [-maxTtl <maxTtl>]
Where, name: Name of the new pool.
owner: Username of the owner of the pool. Defaults to the current user.
group: Group of the pool. Defaults to the primary group name of the current user.
mode: UNIX-style permissions for the pool. Permissions are specified in octal, for example: 0755. By default, this is set
to 0755.
limit: The maximum number of bytes that can be cached by directives in this pool, in aggregate. By default, no limit
is set.
maxTtl: The maximum allowed time-to-live for directives being added to the pool. This can be specified in seconds,
minutes, hours, and days, for example: 120s, 30m, 4h, 2d. By default, no maximum is set. A value of never specifies
that there is no limit.
modifyPool
Description: Modify the metadata of an existing cache pool.
Usage: hdfs cacheadmin -modifyPool <name> [-owner <owner>] [-group <group>] [-mode <mode>]
[-limit <limit>] [-maxTtl <maxTtl>]
Where, name: Name of the pool to modify.
owner: Username of the owner of the pool.
group: Groupname of the group of the pool.
mode: Unix-style permissions of the pool in octal.
limit: Maximum number of bytes that can be cached by this pool.
maxTtl: The maximum allowed time-to-live for directives being added to the pool.
removePool
Description: Remove a cache pool. This also uncaches paths associated with the pool.
Usage: hdfs cacheadmin -removePool <name>
Where, name: Name of the cache pool to remove.
listPools
Description: Display information about one or more cache pools, for example: name, owner, group, permissions, and
so on.
Usage: hdfs cacheadmin -listPools [-stats] [<name>]
Where, name: If specified, list only the named cache pool.
stats: Display additional cache pool statistics.
help
Description: Get detailed help about a command.
160 | Cloudera Administration
Managing CDH and Managed Services
Usage: hdfs cacheadmin -help <command-name>
Where, command-name: The command for which to get detailed help. If no command is specified, print detailed help
for all commands.
Configuration
Native Libraries
In order to lock block files into memory, the DataNode relies on native JNI code found in libhadoop.so. Be sure to
enable JNI if you are using HDFS centralized cache management.
Configuration Properties
Required
Be sure to configure the following in /etc/default/hadoop/conf/hdfs-default.xml:
• dfs.datanode.max.locked.memory: The maximum amount of memory a DataNode will use for caching (in
bytes). The "locked-in-memory size" ulimit (ulimit -l) of the DataNode user also needs to be increased to
match this parameter (see OS Limits). When setting this value, remember that you will need space in memory for
other things as well, such as the DataNode and application JVM heaps and the operating system page cache.
Optional
The following properties are not required, but may be specified for tuning:
• dfs.namenode.path.based.cache.refresh.interval.ms: The NameNode uses this as the amount of
milliseconds between subsequent path cache rescans. This calculates the blocks to cache and each DataNode
containing a replica of the block that should cache it. By default, this parameter is set to 300000, which is five
minutes.
• dfs.datanode.fsdatasetcache.max.threads.per.volume: The DataNode uses this as the maximum
number of threads per volume to use for caching new data. By default, this parameter is set to 4.
• dfs.cachereport.intervalMsec: The DataNode uses this as the amount of milliseconds between sending a
full report of its cache state to the NameNode. By default, this parameter is set to 10000, which is 10 seconds.
• dfs.namenode.path.based.cache.block.map.allocation.percent: The percentage of the Java heap
which we will allocate to the cached blocks map. The cached blocks map is a hash map which uses chained hashing.
Smaller maps may be accessed more slowly if the number of cached blocks is large; larger maps will consume
more memory. By default, this parameter is set to 0.25 percent.
OS Limits
If you get the error Cannot start datanode because the configured max locked memory size... is
more than the datanode's available RLIMIT_MEMLOCK ulimit, that means that the operating system is
imposing a lower limit on the amount of memory that you can lock than what you have configured. To fix this, you
must adjust the ulimit -l value that the DataNode runs with. Usually, this value is configured in
/etc/security/limits.conf. However, it will vary depending on what operating system and distribution you are
using.
You will know that you have correctly configured this value when you can run ulimit -l from the shell and get back
either a higher value than what you have configured with dfs.datanode.max.locked.memory, or the string
unlimited, indicating that there is no limit. Note that it's typical for ulimit -l to output the memory lock limit in
KB, but dfs.datanode.max.locked.memory must be specified in bytes.
Using CDH with Isilon Storage
EMC Isilon is a storage service with a distributed file system that can used in place of HDFS to provide storage for CDH
services.
Cloudera Administration | 161
Managing CDH and Managed Services
Note: This documentation covers only the Cloudera Manager portion of using EMC Isilon storage
with Cloudera Manager. For information about tasks performed on Isilon OneFS, see the information
hub for Cloudera on the EMC Community Network: https://community.emc.com/docs/DOC-39529.
Supported Versions
The following versions of Cloudera and Isilon products are supported:
CDH Version
Isilon OneFS Version
5.2
7.2.x releases starting with 7.2.0.1 and higher
Cloudera recommends 7.2.1.1
5.3
7.2.x releases starting with 7.2.0.2 and higher
Cloudera recommends 7.2.1.1
5.4
7.2.x releases starting with 7.2.0.3 and higher
Cloudera recommends 7.2.1.1
5.5
7.2.x releases starting with 7.2.0.3 and higher
Cloudera recommends 7.2.1.1
Note: Cloudera Navigator is not supported in this release.
Differences between Isilon HDFS and CDH HDFS
The following features of HDFS are not implemented with Isilon OneFS:
• HDFS caching
• HDFS encryption
• HDFS ACLs
Preliminary Steps on the Isilon Service
Before installing a Cloudera Manager cluster to use Isilon storage, perform the following steps on the Isilon OneFS
system. For detailed information on setting up Isilon OneFS for Cloudera Manager, see the Isilon documentation at
https://community.emc.com/docs/DOC-39529.
1. Create an Isilon access zone with HDFS support.
Example:
/ifs/your-access-zone/hdfs
Note: The above is simply an example; the HDFS root directory does not have to begin with ifs
or end with hdfs.
2. Create two directories that will be used by all CDH services:
a. Create a tmp directory in the access zone.
• Create supergroup group and hdfs user.
162 | Cloudera Administration
Managing CDH and Managed Services
• Create a tmp directory and set ownership to hdfs:supergroup, and permissions to 1777.
Example:
cd hdfs_root_directory
isi_run -z zone_id mkdir tmp
isi_run -z zone_id chown hdfs:supergroup tmp
isi_run -z zone_id chmod 1777 tmp
b. Create a user directory in the access zone and set ownership to hdfs:supergroup, and permissions to
755
Example:
cd hdfs_root_directory
isi_run -z zone_id mkdir user
isi_run -z zone_id chown hdfs:supergroup user
isi_run -z zone_id chmod 755 user
3. Create the service-specific users, groups, or directories for each CDH service you plan to use. Create the directories
under the access zone you have created.
Note: Many of the values provided in the examples below are default values in Cloudera Manager
and must match the Cloudera Manager configuration settings. The format for the examples is:
dir user:group permission . Create the directories below under the access zone you have
created, for example, /ifs/ your-access-zone /hdfs/
• ZooKeeper: nothing required.
• HBase
– Create hbase group with hbase user.
– Create the root directory for HBase:
Example:
hdfs_root_directory/hbase hbase:hbase 755
• YARN (MR2)
– Create mapred group with mapred user.
– Create history directory for YARN:
Example:
hdfs_root_directory/user/history mapred:hadoop 777
– Create the remote application log directory for YARN:
Example:
hdfs_root_directory/tmp/logs mapred:hadoop 775
• Oozie
– Create oozie group with oozie user.
– Create the user directory for Oozie:
Example:
hdfs_root_directory/user/oozie oozie:oozie 775
• Flume
Cloudera Administration | 163
Managing CDH and Managed Services
– Create flume group with flume user.
– Create the user directory for Flume:
Example:
hdfs_root_directory/user/flume flume:flume 775
• Hive
– Create hive group with hive user.
– Create the user directory for Hive:
Example:
hdfs_root_directory/user/hive hive:hive 775
– Create the warehouse directory for Hive:
Example:
hdfs_root_directory/user/hive/warehouse hive:hive 1777
– Create a temporary directory for Hive:
Example:
hdfs_root_directory/tmp/hive hive:supergroup 777
• Solr
– Create solr group with solr user.
– Create the data directory for Solr:
Example:
hdfs_root_directory/solr solr:solr 775
• Sqoop
– Create sqoop group with sqoop2 user.
– Create the user directory for Sqoop:
Example:
hdfs_root_directory/user/sqoop2 sqoop2:sqoop 775
• Hue
– Create hue group with hue user.
– Create sample group with sample user.
• Spark
– Create spark group with spark user.
– Create the user directory for Spark:
Example:
hdfs_root_directory/user/spark spark:spark 751
– Create application history directory for Spark:
Example:
hdfs_root_directory/user/spark/applicationHistory spark:spark 1777
164 | Cloudera Administration
Managing CDH and Managed Services
Once the users, groups, and directories are created in Isilon OneFS, you are ready to install Cloudera Manager.
Installing Cloudera Manager with Isilon
To install Cloudera Manager follow the instructions provided in Installation.
• The simplest installation procedure, suitable for development or proof of concept, is Installation Path A, which
uses embedded databases that are installed as part of the Cloudera Manager installation process.
• For production environments, Installation Path B - Manual Installation Using Cloudera Manager Packages describes
configuring external databases for Cloudera Manager and CDH storage needs.
If you choose parcel installation on the Cluster Installation screen, the installation wizard will point to the latest parcels
of CDH available.
On the installation wizard's Cluster Setup page, choose Custom Services, and choose the services you want installed
in the cluster. Be sure to choose Isilon among the selected services, do not select the HDFS service, and do not check
Include Cloudera Navigator at the bottom of the Cluster Setup page. Also, on the Role Assignments page, be sure to
specify the hosts that will serve as gateway roles for the Isilon service. You can add gateway roles to one, some, or all
nodes in the cluster.
Installing a Secure Cluster with Isilon
To set up a secure cluster with Isilon using Kerberos, perform the following steps:
1. Create an unsecure Cloudera Manager cluster as described above in Installing Cloudera Manager with Isilon on
page 165.
2. Follow the Isilon documentation to enable Kerberos for your access zone:
https://community.emc.com/docs/DOC-39529. This includes adding a Kerberos authentication provider to your
Isilon access zone.
3. Add the following proxy users in Isilon if your Cloudera Manager cluster includes the corresponding CDH services.
The procedure for configuring proxy users is described in the Isilon documentation,
https://community.emc.com/docs/DOC-39529.
•
•
•
•
•
•
•
proxy user hdfs for hdfs user.
proxy user mapred for mapred user.
proxy user hive for hive user.
proxy user impala for impala user.
proxy user oozie for oozie user
proxy user flume for flume user
proxy user hue for hue user
4. Follow the Cloudera Manager documentation for information on configuring a secure cluster with Kerberos:
Configuring Authentication in Cloudera Manager.
Upgrading a Cluster with Isilon
To upgrade CDH and Cloudera Manager in a cluster that uses Isilon:
1. If required, upgrade OneFS to a version compatible with the version of CDH to which you are upgrading. For
compatibility information, see Product Compatibility Matrix for EMC Isilon. For OneFS upgrade instructions, see
the EMC Isilon documentation.
2. (Optional) Upgrade Cloudera Manager. See Upgrading Cloudera Manager.
The Cloudera Manager minor version must always be equal to or greater than the CDH minor version because
older versions of Cloudera Manager may not support features in newer versions of CDH. For example, if you want
to upgrade to CDH 5.4.8 you must first upgrade to Cloudera Manager 5.4 or higher.
3. Upgrade CDH. See Upgrading CDH and Managed Services Using Cloudera Manager.
Cloudera Administration | 165
Managing CDH and Managed Services
Using Impala with Isilon Storage
You can use Impala to query data files that reside on EMC Isilon storage devices, rather than in HDFS. This capability
allows convenient query access to a storage system where you might already be managing large volumes of data. The
combination of the Impala query engine and Isilon storage is certified on CDH 5.4.4 or higher.
Because the EMC Isilon storage devices use a global value for the block size rather than a configurable value for each
file, the PARQUET_FILE_SIZE query option has no effect when Impala inserts data into a table or partition residing
on Isilon storage. Use the isi command to set the default block size globally on the Isilon device. For example, to set
the Isilon default block size to 256 MB, the recommended size for Parquet data files for Impala, issue the following
command:
isi hdfs settings modify --default-block-size=256MB
The typical use case for Impala and Isilon together is to use Isilon for the default filesystem, replacing HDFS entirely.
In this configuration, when you create a database, table, or partition, the data always resides on Isilon storage and you
do not need to specify any special LOCATION attribute. If you do specify a LOCATION attribute, its value refers to a
path within the Isilon filesystem. For example:
-- If the default filesystem is Isilon, all Impala data resides there
-- and all Impala databases and tables are located there.
CREATE TABLE t1 (x INT, s STRING);
-- You can specify LOCATION for database, table, or partition,
-- using values from the Isilon filesystem.
CREATE DATABASE d1 LOCATION '/some/path/on/isilon/server/d1.db';
CREATE TABLE d1.t2 (a TINYINT, b BOOLEAN);
Impala can write to, delete, and rename data files and database, table, and partition directories on Isilon storage.
Therefore, Impala statements such as CREATE TABLE, DROP TABLE, CREATE DATABASE, DROP DATABASE, ALTER
TABLE, and INSERT work the same with Isilon storage as with HDFS.
When the Impala spill-to-disk feature is activated by a query that approaches the memory limit, Impala writes all the
temporary data to a local (not Isilon) storage device. Because the I/O bandwidth for the temporary data depends on
the number of local disks, and clusters using Isilon storage might not have as many local disks attached, pay special
attention on Isilon-enabled clusters to any queries that use the spill-to-disk feature. Where practical, tune the queries
or allocate extra memory for Impala to avoid spilling. Although you can specify an Isilon storage device as the destination
for the temporary data for the spill-to-disk feature, that configuration is not recommended due to the need to transfer
the data both ways using remote I/O.
When tuning Impala queries on HDFS, you typically try to avoid any remote reads. When the data resides on Isilon
storage, all the I/O consists of remote reads. Do not be alarmed when you see non-zero numbers for remote read
measurements in query profile output. The benefit of the Impala and Isilon integration is primarily convenience of not
having to move or copy large volumes of data to HDFS, rather than raw query performance. You can increase the
performance of Impala I/O for Isilon systems by increasing the value for the num_remote_hdfs_io_threads
configuration parameter, in the Cloudera Manager user interface for clusters using Cloudera Manager, or through the
--num_remote_hdfs_io_threads startup option for the impalad daemon on clusters not using Cloudera Manager.
For information about managing Isilon storage devices through Cloudera Manager, see Using CDH with Isilon Storage
on page 161.
Required Configurations
Specify the following configurations in Cloudera Manager on the Clusters > Isilon Service > Configuration tab:
• In HDFS Client Advanced Configuration Snippet (Safety Valve) for hdfs-site.xml hdfs-site.xml and the
Cluster-wide Advanced Configuration Snippet (Safety Valve) for core-site.xml properties for the Isilon service,
set the value of the dfs.client.file-block-storage-locations.timeout.millis property to 10000.
• In the Isilon Cluster-wide Advanced Configuration Snippet (Safety Valve) for core-site.xml property for the Isilon
service, set the value of the hadoop.security.token.service.use_ip property to FALSE.
• If you see errors that reference the .Trash directory, make sure that the Use Trash property is selected.
166 | Cloudera Administration
Managing CDH and Managed Services
Managing Hive
Use the following procedures to manage HiveServer2 and the Hive metastore. To configure high availability for the
Hive metastore, see Hive Metastore High Availability on page 346.
Heap Size and Garbage Collection for Hive Components
Hive Component Memory Recommendations
HiveServer2 and the Hive metastore require sufficient memory in order to run correctly. The default heap size of 256
MB for each component is inadequate for production workloads. Consider the following guidelines for sizing the heap
for each component, based upon your cluster size.
Number of Concurrent Connections
HiveServer2 Heap Size Minimum
Recommendation
Hive Metastore Heap Size Minimum
Recommendation
Up to 40 concurrent connections
12 GB
(Cloudera recommends splitting
HiveServer2 into multiple instances
and load balancing once you start
allocating >12 GB to HiveServer2. The
objective is to size to reduce impact
of Java garbage collection on active
processing by the service.
12 GB
Up to 20 concurrent connections
6 GB
10 GB
Up to 10 concurrent connections
4 GB
8 GB
Single connection
2 GB
4 GB
Important: These numbers are general guidance only, and may be affected by factors such as number
of columns, partitions, complex joins, and client activity among other things. It is important to review
and refine through testing based on your anticipated deployment to arrive at best values for your
environment.
In addition, the Beeline CLI should use a heap size of at least 2 GB.
The permGenSize should be set to 512M for all.
Configuring Heap Size and Garbage Collection for Hive Components
To configure the heap size for HiveServer2 and Hive metastore, set the -Xmx parameter in the HADOOP_OPTS variable
to the desired maximum heap size in the hive-env.sh advanced configuration snippet if you use Cloudera Manager
or otherwise edit /etc/hive/hive-env.sh.
To configure the heap size for the Beeline CLI, set the HADOOP_HEAPSIZE environment variable in the hive-env.sh
advanced configuration snippet if you use Cloudera Manager or otherwise edit /etc/hive/hive-env.sh before
starting the Beeline CLI.
The following example shows a configuration with the following settings:
• HiveServer2 uses 12 GB heap
• Hive metastore uses 12 GB heap
• Hive clients use 2 GB heap
Cloudera Administration | 167
Managing CDH and Managed Services
The settings to change are in bold. All of these lines are commented out (prefixed with a # character) by default.
Uncomment the lines by removing the # character.
if [ "$SERVICE" = "cli" ]; then
if [ -z "$DEBUG" ]; then
export HADOOP_OPTS="$HADOOP_OPTS -XX:NewRatio=12 -Xmx12288m -Xms10m
-XX:MaxHeapFreeRatio=40 -XX:MinHeapFreeRatio=15 -XX:+useParNewGC -XX:-useGCOverheadLimit"
else
export HADOOP_OPTS="$HADOOP_OPTS -XX:NewRatio=12 -Xmx12288m -Xms10m
-XX:MaxHeapFreeRatio=40 -XX:MinHeapFreeRatio=15 -XX:-useGCOverheadLimit"
fi
fi
export HADOOP_HEAPSIZE=2048
You can choose whether to use the Concurrent Collector or the New Parallel Collector for garbage collection, by passing
-XX:+useParNewGC or -XX:+useConcMarkSweepGC in the HADOOP_OPTS lines above, and you can tune the garbage
collection overhead limit by setting -XX:-useGCOverheadLimit. To enable the garbage collection overhead limit,
remove the setting or change it to -XX:+useGCOverheadLimit.
Configuration for WebHCat
If you want to use WebHCat, you need to set the PYTHON_CMD variable in /etc/default/hive-webhcat-server
after installing Hive; for example:
export PYTHON_CMD=/usr/bin/python
Transaction (ACID) Support in Hive
The CDH distribution of Hive does not support transactions (HIVE-5317). Currently, transaction support in Hive is an
experimental feature that only works with the ORC file format. Cloudera recommends using the Parquet file format,
which works across many tools. Merge updates in Hive tables using existing functionality, including statements such
as INSERT, INSERT OVERWRITE, and CREATE TABLE AS SELECT.
Managing Hive Using Cloudera Manager
There are two Hive service roles:
• Hive Metastore Server - manages the metastore process when Hive is configured with a remote metastore. You
are strongly encouraged to read Configuring the Hive Metastore (CDH 4) or Configuring the Hive Metastore (CDH
5).
• HiveServer2 - supports a Thrift API tailored for JDBC and ODBC clients, Kerberos authentication, and multi-client
concurrency. There is also a CLI for HiveServer2 named Beeline. Cloudera recommends that you deploy HiveServer2
whenever possible. You can use the original HiveServer, and run it concurrently with HiveServer2. However,
Cloudera Manager does not manage HiveServer, so you must configure and manage it outside Cloudera Manager.
See HiveServer2 documentation (CDH 4) or HiveServer2 documentation (CDH 5) for more information.
How Hive Configurations are Propagated to Hive Clients
Because the Hive service does not have worker roles, another mechanism is needed to enable the propagation of client
configurations to the other hosts in your cluster. In Cloudera Manager gateway roles fulfill this function. Whether you
add a Hive service at installation time or at a later time, ensure that you assign the gateway roles to hosts in the cluster.
If you do not have gateway roles, client configurations are not deployed.
The Hive Metastore Server
Cloudera recommends using a remote Hive metastore, especially for CDH 4.2 and later. Since the remote metastore
is recommended, Cloudera Manager treats the Hive Metastore Server as a required role for all Hive services. Here are
a couple key reasons why the remote metastore setup is advantageous, especially in production settings:
168 | Cloudera Administration
Managing CDH and Managed Services
• The Hive metastore database password and JDBC drivers don’t need to be shared with every Hive client; only the
Hive Metastore Server does. Sharing passwords with many hosts is a security concern.
• You can control activity on the Hive metastore database. To stop all activity on the database, just stop the Hive
Metastore Server. This makes it easy to perform tasks such as backup and upgrade, which require all Hive activity
to stop.
Information about the initial configuration of a remote Hive metastore database with Cloudera Manager can be found
at Cloudera Manager and Managed Service Datastores.
Considerations When Upgrading CDH
Hive has undergone major version changes from CDH 4.0 to 4.1 and between CDH 4.1 and 4.2. (CDH 4.0 had Hive 0.8.0,
CDH 4.1 used Hive 0.9.0, and CDH 4.2 or later has 0.10.0). This requires that you manually back up and upgrade the
Hive metastore database when upgrading between major Hive versions.
You should follow the steps in the appropriate in the Cloudera Manager procedure for upgrading CDH to upgrade the
metastore before you restart the Hive service. This applies whether you are upgrading to packages or parcels. The
procedure for upgrading CDH using packages is at Upgrading CDH 4 Using Packages. The procedure for upgrading with
parcels is at Upgrading CDH 4 Using Parcels.
Considerations When Upgrading Cloudera Manager
Cloudera Manager 4.5 added support for Hive, which includes the Hive Metastore Server role type. This role manages
the metastore process when Hive is configured with a remote metastore.
When upgrading from Cloudera Manager versions before 4.5, Cloudera Manager automatically creates new Hive
services to capture the previous implicit Hive dependency from Hue and Impala. Your previous services continue to
function without impact. If Hue was using a Hive metastore backed by a Derby database, the newly created Hive
Metastore Server also uses Derby. Because Derby does not allow concurrent connections, Hue continues to work, but
the new Hive Metastore Server does not run. The failure is harmless (because nothing uses this new Hive Metastore
Server at this point) and intentional, to preserve the set of cluster functionality as it was before upgrade. Cloudera
discourages the use of a Derby-backed Hive metastore due to its limitations and recommends switching to a different
supported database.
Cloudera Manager provides a Hive configuration option to bypass the Hive Metastore Server. When this configuration
is enabled, Hive clients, Hue, and Impala connect directly to the Hive metastore database. Prior to Cloudera Manager
4.5, Hue and Impala connected directly to the Hive metastore database, so the bypass mode is enabled by default
when upgrading to Cloudera Manager 4.5 or later. This ensures that the upgrade does not disrupt your existing setup.
You should plan to disable the bypass mode, especially when using CDH 4.2 or later. Using the Hive Metastore Server
is the recommended configuration, and the WebHCat Server role requires the Hive Metastore Server to not be bypassed.
To disable bypass mode, see Disabling Bypass Mode on page 169.
Cloudera Manager 4.5 or later also supports HiveServer2 with CDH 4.2. In CDH 4, HiveServer2 is not added by default,
but can be added as a new role under the Hive service (see Role Instances on page 45). In CDH 5, HiveServer2 is a
mandatory role.
Disabling Bypass Mode
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
In bypass mode Hive clients directly access the metastore database instead of using the Hive Metastore Server for
metastore information.
1.
2.
3.
4.
5.
6.
7.
Go to the Hive service.
Click the Configuration tab.
Select Scope > HIVE service_name (Service-Wide)
Select Category > Advanced.
Deselect the Bypass Hive Metastore Server property.
Click Save Changes to commit the changes.
Re-deploy Hive client configurations.
Cloudera Administration | 169
Managing CDH and Managed Services
8. Restart Hive and any Hue or Impala services configured to use that Hive service.
Using Hive Gateways
Because the Hive service does not have worker roles, another mechanism is needed to enable the automatic propagation
of client configurations to the other hosts in your cluster. Gateway roles fulfill this function. Gateways in fact aren't
really roles and do not have state, but they act as indicators for where client configurations should be placed. Hive
gateways are created by default when the Hive service is added.
Hive Table Statistics
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
If your cluster has Impala then you can use the Impala implementation to compute statistics. The Impala implementation
to compute table statistics is available in CDH 5.0.0 or higher and in Impala version 1.2.2 or higher. The Impala
implementation of COMPUTE STATS requires no setup steps and is preferred over the Hive implementation. See
Overview of Table Statistics. If you are running an older version of Impala, you can collect statistics on a Hive table by
running the following command from a Beeline client connected to HiveServer2:
analyze table <table name> compute statistics;
analyze table <table name> compute statistics for columns <all columns of a table>;
Managing User-Defined Functions (UDFs) with HiveServer2
Hive's query language (HiveQL) can be extended with Java-based user-defined functions (UDFs). See the Apache Hive
Language Manual UDF page for information about Hive built-in UDFs. To create customized UDFs, see the Apache Hive
wiki. After creating a new Java class to extend the com.example.hive.udf package, you must compile your code
into a Java archive file (JAR), and add it to the Hive classpath with the ADD JAR command. The ADD JAR command
does not work with HiveServer2 and the Beeline client when Beeline runs on a different host. As an alternative to ADD
JAR, Hive's auxiliary paths functionality should be used.
Perform one of the following procedures depending on whether you want to create permanent or temporary functions.
Blacklist for Built-in UDFs
HiveServer2 maintains a blacklist for built-in UDFs to secure itself against attacks in a multiuser scenario where the
hive user's credentials can be used to execute any Java code.
hive.server2.builtin.udf.blacklist A comma separated list of built-in UDFs that are not allowed to be executed.
A UDF that is included in the list will return an error if invoked from a query.
Default value: Empty
User-Defined Functions (UDFs) with HiveServer2 Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Creating Permanent Functions
1. Copy the JAR file to HDFS and make sure the hive user can access this JAR file.
2. Copy the JAR file to the host on which HiveServer2 is running. Save the JARs to any directory you choose, give the
hive user read, write, and execute access to this directory, and make a note of the path (for example,
/opt/local/hive/lib/).
Note: If the Hive Metastore is running on a different host, create the same directory there that
you created on the HiveServer2 host. You do not need to copy the JAR file onto the Hive Metastore
host, but the same directory must be there. For example, if you copied the JAR file to
/opt/local/hive/lib/ on the HiveServer2 host, you must create the same directory on the
Hive Metastore host. If the same directory is not present on the Hive Metastore host, Hive
Metastore service will not start.
170 | Cloudera Administration
Managing CDH and Managed Services
3.
4.
5.
6.
7.
In the Cloudera Manager Admin Console, go to the Hive service.
Click the Configuration tab.
Expand the Hive (Service-Wide) scope.
Click the Advanced category.
Configure the Hive Auxiliary JARs Directory property with the HiveServer2 host path and the Hive Metastore host
path from Step 2, for example /opt/local/hive/lib/. Setting this property overwrites hive.aux.jars.path,
even if this variable has been previously set in the HiveServer2 advanced configuration snippet.
8. Click Save Changes. The JARs are added to HIVE_AUX_JARS_PATH environment variable.
9. Redeploy the Hive client configuration.
a. In the Cloudera Manager Admin Console, go to the Hive service.
b. From the Actions menu at the top right of the service page, select Deploy Client Configuration.
c. Click Deploy Client Configuration.
10. Restart the Hive service.
11. With Sentry enabled - Grant privileges on the JAR files to the roles that require access. Log in to Beeline as user
hive and use the Hive SQL GRANT statement to do so. For example:
GRANT ALL ON URI 'file:///opt/local/hive/lib/my.jar' TO ROLE EXAMPLE_ROLE
You must also grant privilege to the JAR on HDFS:
GRANT ALL ON URI 'hdfs:///path/to/jar' TO ROLE EXAMPLE_ROLE
12. Run the CREATE FUNCTION command to create the UDF from the JAR file and point to the JAR file location in
HDFS. For example:
CREATE FUNCTION addfunc AS 'com.example.hiveserver2.udf.add' USING JAR
'hdfs:///path/to/jar'
Creating Temporary Functions
1. Copy the JAR file to the host on which HiveServer2 is running. Save the JARs to any directory you choose, give the
hive user read, write, and execute access to this directory, and make a note of the path (for example,
/opt/local/hive/lib/).
Note: If the Hive Metastore is running on a different host, create the same directory there that
you created on the HiveServer2 host. You do not need to copy the JAR file onto the Hive Metastore
host, but the same directory must be there. For example, if you copied the JAR file to
/opt/local/hive/lib/ on the HiveServer2 host, you must create the same directory on the
Hive Metastore host. If the same directory is not present on the Hive Metastore host, Hive
Metastore service will not start.
2.
3.
4.
5.
6.
In the Cloudera Manager Admin Console, go to the Hive service.
Click the Configuration tab.
Expand the Hive (Service-Wide) scope.
Click the Advanced category.
Configure the Hive Auxiliary JARs Directory property with the HiveServer2 host path and the Hive Metastore host
path from Step 1, for example /opt/local/hive/lib/. Setting this property overwrites hive.aux.jars.path,
even if this variable has been previously set in the HiveServer2 advanced configuration snippet.
7. Click Save Changes. The JARs are added to HIVE_AUX_JARS_PATH environment variable.
8. Redeploy the Hive client configuration.
a. In the Cloudera Manager Admin Console, go to the Hive service.
b. From the Actions menu at the top right of the service page, select Deploy Client Configuration.
c. Click Deploy Client Configuration.
Cloudera Administration | 171
Managing CDH and Managed Services
9. Restart the Hive service.
10. With Sentry enabled - Grant privileges on the JAR files to the roles that require access. Log in to Beeline as user
hive and use the Hive SQL GRANT statement to do so. For example:
GRANT ALL ON URI 'file:///opt/local/hive/lib/my.jar' TO ROLE EXAMPLE_ROLE
You must also grant privilege to the JAR on HDFS:
GRANT ALL ON URI 'hdfs:///path/to/jar' TO ROLE EXAMPLE_ROLE
11. Run the CREATE TEMPORARY FUNCTION command. For example:
CREATE TEMPORARY FUNCTION addfunc AS 'com.example.hiveserver2.udf.add'
User-Defined Functions (UDFs) with HiveServer2 Using the Command Line
The following sections describe how to create permanent and temporary functions using the command line.
Creating Permanent Functions
1. Copy the JAR file to HDFS and make sure the hive user can access this jar file.
2. On the Beeline client machine, in /etc/hive/conf/hive-site.xml, set the hive.aux.jars.path property
to a comma-separated list of the fully-qualified paths to the JAR file and any dependent libraries.
hive.aux.jars.path=file:///opt/local/hive/lib/my.jar
3. Copy the JAR file (and its dependent libraries) to the host running HiveServer2/Impala. Make sure the hive user
has read, write, and execute access to these files on the HiveServer2/Impala host.
4. On the HiveServer2/Impala host, open /etc/default/hive-server2 and set the AUX_CLASSPATH variable
to a comma-separated list of the fully-qualified paths to the JAR file and any dependent libraries.
AUX_CLASSPATH=/opt/local/hive/lib/my.jar
5. Restart HiveServer2.
6. If Sentry is enabled - Grant privileges on the JAR files to the roles that require access. Login to Beeline as user
hive and use the Hive SQL GRANT statement to do so. For example:
GRANT ALL ON URI 'file:///opt/local/hive/lib/my.jar' TO ROLE EXAMPLE_ROLE
You must also grant privilege to the JAR on HDFS:
GRANT ALL ON URI 'hdfs:///path/to/jar' TO ROLE EXAMPLE_ROLE
If you are using Sentry policy files, you can grant the URI privilege as follows:
udf_r = server=server1->uri=file:///opt/local/hive/lib
udf_r = server=server1->uri=hdfs:///path/to/jar
7. Run the CREATE FUNCTION command and point to the JAR from Hive:
CREATE FUNCTION addfunc AS 'com.example.hiveserver2.udf.add' USING JAR
'hdfs:///path/to/jar'
172 | Cloudera Administration
Managing CDH and Managed Services
Creating Temporary Functions
1. On the Beeline client machine, in /etc/hive/conf/hive-site.xml, set the hive.aux.jars.path property
to a comma-separated list of the fully-qualified paths to the JAR file and any dependent libraries.
hive.aux.jars.path=file:///opt/local/hive/lib/my.jar
2. Copy the JAR file (and its dependent libraries) to the host running HiveServer2/Impala. Make sure the hive user
has read, write, and execute access to these files on the HiveServer2/Impala host.
3. On the HiveServer2/Impala host, open /etc/default/hive-server2 and set the AUX_CLASSPATH variable
to a comma-separated list of the fully-qualified paths to the JAR file and any dependent libraries.
AUX_CLASSPATH=/opt/local/hive/lib/my.jar
4. If Sentry is enabled - Grant privileges on the local JAR files to the roles that require access. Login to Beeline as
user hive and use the Hive SQL GRANT statement to do so. For example:
GRANT ALL ON URI 'file:///opt/local/hive/lib/my.jar' TO ROLE EXAMPLE_ROLE
If you are using Sentry policy files, you can grant the URI privilege as follows:
udf_r = server=server1->uri=file:///opt/local/hive/lib
5. Restart HiveServer2.
6. Run the CREATE TEMPORARY FUNCTION command and point to the JAR from Hive:
CREATE TEMPORARY FUNCTION addfunc AS 'com.example.hiveserver2.udf.add'
Running Hive on Spark
This section explains how to set up Hive on Spark. It assumes that your cluster is managed by Cloudera Manager.
Important: Hive on Spark is included in CDH 5.5 but is not currently supported nor recommended
for production use. To try this feature in CDH 5.5, use it in a test environment.
Configuring Hive on Spark
Important: Hive on Spark is included in CDH 5.5 but is not currently supported nor recommended
for production use. To try this feature in CDH 5.5, use it in a test environment.
This topic explains the configuration properties you set up to run Hive on Spark.
Note: We recommend that you use HiveServer2 with Beeline. The following content, except for
Configuring Hive on Spark for Hive CLI on page 175, is based on this assumption.
Installation Considerations
For Hive to work on Spark, you must deploy Spark gateway roles on the same machine that hosts HiveServer2. Otherwise,
Hive on Spark cannot read from Spark configurations and cannot submit Spark jobs. For more information about
gateway roles, see Managing Roles on page 44.
After installation, run the following command in Hive so that Hive will use Spark as the back-end engine for all subsequent
queries.
set hive.execution.engine=spark;
Cloudera Administration | 173
Managing CDH and Managed Services
Enabling Hive on Spark
By default, Hive on Spark is not enabled. To enable Hive on Spark, perform the following steps in Cloudera Manager.
1.
2.
3.
4.
5.
6.
Go to the Hive service.
Click the Configuration tab.
Enter Enable Hive on Spark in the Search field.
Check the box for Enable Hive on Spark (Unsupported).
Locate the Spark On YARN Service and click SPARK_ON_YARN.
Click Save Changes to commit the changes.
Configuration Properties
Property
Description
hive.stats.collect.rawdatasize
Hive on Spark uses statistics to determine the threshold
for converting common join to map join. There are two
types of statistics about the size of data:
• totalSize: The approximate size of data on the disk
• rawDataSize: The approximate size of data in
memory
When both metrics are available, Hive chooses
rawDataSize.
Default: True
hive.auto.convert.join.noconditionaltask.size The threshold for the sum of all the small table size (by
default, rawDataSize), for map join conversion. You can
increase the value if you want better performance by
converting more common joins to map joins. However, if
you set this value too high, tasks may fail because too
much memory is being used by data from small tables.
Default: 20MB
Configuring Hive
For improved performance, Cloudera recommends that you configure the following additional properties for Hive. In
Cloudera Manager, set these properties in the advanced configuration snippet for HiveServer2.
• hive.stats.fetch.column.stats=true
• hive.optimize.index.filter=true
Configuring Executor Memory Size
For general Spark configuration recommendations, see Configuring Spark on YARN Applications.
Executor memory size can have a number of effects on Hive. Increasing executor memory increases the number of
queries for which Hive can enable mapjoin optimization. However, if there's too much executor memory, it takes longer
to perform garbage collection. Also, some experiments shows that HDFS doesn’t handle concurrent writers well, so it
may face a race condition if there are too many executor cores.
Cloudera recommends that you set the value for spark.executor.cores to 5, 6, or 7, depending on what the host
is divisible by. For example, if yarn.nodemanager.resource.cpu-vcores is 19, then you would set the value to
6. Executors must have the same number of cores. If you set the value to 5, three executors with 5 cores each can be
launched, leaving four cores unused. If you set the value to 7, only two executors are used, and five cores are unused.
If the number of cores is 20, set the value to 5 so that each executor gets four cores, and no cores are unused.
Cloudera also recommends the following:
174 | Cloudera Administration
Managing CDH and Managed Services
• Compute a memory size equal to yarn.nodemanager.resource.memory-mb * (spark.executor.cores
/ yarn.nodemanager.resource.cpu-vcores) and then split that between spark.executor.memory and
spark.yarn.executor.memoryOverhead.
• spark.yarn.executor.memoryOverhead is 15-20% of the total memory size.
Troubleshooting Hive on Spark
Important: Hive on Spark is included in CDH 5.5 but is not currently supported nor recommended
for production use. To try this feature in CDH 5.5, use it in a test environment.
Problem: Delayed result from the first query after starting a new Hive on Spark session
The first query after starting a new Hive on Spark session might be delayed due to the start-up time for the Spark
on YARN cluster. The query waits for YARN containers to initialize. Subsequent queries will be faster.
Problem: Exception Error: org.apache.thrift.transport.TTransportException (state=08S01,code=0)
and HiveServer2 is down
HiveServer2 memory is set too small. For more information, see STDOUT for HiveServer2. To fix this issue:
1. In Cloudera Manager, go to HIVE.
2. Click Configuration.
3. Search for Java Heap Size of HiveServer2 in Bytes, and change it to be a larger value. Cloudera
recommends a minimum value of 256 MB.
4. Restart HiveServer2.
Problem: Out-of-memory error
You might get an out-of-memory error similar to the following:
15/03/19 03:43:17 WARN channel.DefaultChannelPipeline: An exception was thrown by a user
handler while handling an exception event ([id: 0x9e79a9b1, /10.20.118.103:45603 =>
/10.20.120.116:39110] EXCEPTION: java.lang.OutOfMemoryError: Java heap space)
java.lang.OutOfMemoryError: Java heap space
This error indicates that the Spark driver does not have enough off-heap memory. Increase the off-heap memory
by setting spark.yarn.driver.memoryOverhead or spark.driver.memory.
Problem: Hive on Spark does not work with HBase
Hive on Spark with HBase is not supported. If you use HBase, use Hive on MapReduce instead of Hive on Spark.
Problem: Spark applications stay alive forever and occupy cluster resources
This can occur if there are multiple concurrent Hive sessions. To manually terminate the Spark applications:
1. Find the YARN application IDs for the applications by going to Cloudera Manager and clicking Yarn >
ResourceManager > ResourceManager Web UI.
2. Log in to the YARN ResourceManager host.
3. Open a terminal and run:
yarn application -kill <applicationID>
applicationID is each YARN application ID you found in step 1.
Configuring Hive on Spark for Hive CLI
You no longer need to configure Hive on Spark for the Hive command line interface (CLI). This feature is now configured
automatically.
Important: Hive on Spark is included in CDH 5.5 but is not currently supported nor recommended
for production use. To try this feature in CDH 5.5, use it in a test environment.
Cloudera Administration | 175
Managing CDH and Managed Services
Managing Hue
Hue is a set of web UIs that enable you to interact with a CDH cluster. This section describes tasks for managing Hue.
Adding a Hue Service and Role Instance
Adding the Hue Service
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
After initial installation, you can use the Add a Service wizard to add and configure a new Hue service instance.
1. On the Home > Status tab, click
to the right of the cluster name and select Add a Service. A list of service types display.
2. Select Hue.
3. Click Continue.
A page displays where you can specify the dependencies for the Hue service.
4. Select the row with the Hue dependencies required for your cluster. For more information, see Hue Dependencies.
5. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of the
hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of hosts to
which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple hosts,
you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts dialog.
The following shortcuts for specifying hostname patterns are supported:
• Range of hostnames (without the domain portion)
Range Definition
Matching Hosts
10.1.1.[1-4]
10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4
host[1-3].company.com
host1.company.com, host2.company.com, host3.company.com
host[07-10].company.com
host07.company.com, host08.company.com, host09.company.com,
host10.company.com
• IP addresses
• Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
6. Click Continue.
Cloudera Manager starts the Hue service.
7. Click Continue.
8. Click Finish.
9. If your cluster uses Kerberos, Cloudera Manager will automatically add a Hue Kerberos Ticket Renewer role to
each host where you assigned the Hue Server role instance. Also see, Enable Hue to Work with Hadoop Security
using Cloudera Manager.
Adding a Hue Role Instance
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
1. In Cloudera Manager Administration Console, go to the Hue service.
2. Click the Instances tab.
3. Click the Add Role Instances button.
176 | Cloudera Administration
Managing CDH and Managed Services
4. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of the
hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of hosts to
which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple hosts,
you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts dialog.
The following shortcuts for specifying hostname patterns are supported:
• Range of hostnames (without the domain portion)
Range Definition
Matching Hosts
10.1.1.[1-4]
10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4
host[1-3].company.com
host1.company.com, host2.company.com, host3.company.com
host[07-10].company.com
host07.company.com, host08.company.com, host09.company.com,
host10.company.com
• IP addresses
• Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
5. If your cluster uses Kerberos, you must add the Hue Kerberos Ticket Renewer role to each host where you assigned
the Hue Server role instance. Cloudera Manager will throw a validation error if the new Hue Server role does not
have a colocated KT Renewer role. Also see, Enable Hue to Work with Hadoop Security using Cloudera Manager.
6. Click Continue.
Hue and High Availability
If your cluster has HDFS high availability enabled, you must configure the Hue HDFS Web Interface Role property to
use HttpFS. See Configuring Hue to Work with HDFS HA on page 310 for detailed instructions.
To configure the Hue service itself for high availability, see Hue High Availability on page 348.
Managing Hue Analytics Data Collection
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Hue tracks anonymized pages and application versions to collect information used to compare each application's usage
levels. The data collected does not include hostnames or IDs; For example, the data has the format /2.3.0/pig,
/2.5.0/beeswax/execute. You can restrict data collection as follows:
1.
2.
3.
4.
5.
Go to the Hue service.
Click the Configuration tab.
Select Scope > Hue.
Locate the Enable Usage Data Collection property or search for it by typing its name in the Search box.
Deselect the Enable Usage Data Collection checkbox.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
6. Click Save Changes to commit the changes.
7. Restart the Hue service.
Enabling Hue Applications Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Most Hue applications are configured by default, based on the services you have installed. Cloudera Manager selects
the service instance that Hue depends on. If you have more than one service, you may want to verify or change the
Cloudera Administration | 177
Managing CDH and Managed Services
service dependency for Hue. Also, if you add a service such as Sqoop 2 or Oozie after you have set up Hue, you need
to set the dependency because it is not done automatically. To add a dependency:
1.
2.
3.
4.
5.
6.
7.
Go to the Hue service.
Click the Configuration tab.
Select Scope > Hue (Service-Wide).
Select Category > Main.
Select each service name Service property to set the dependency. Select none to remove the dependency.
Click Save Changes to commit the changes.
Restart the Hue service.
Enabling the Sqoop 2 Application
If you are upgrading Cloudera Manager from a release 4.6 or lower, you need to set the Hue dependency to enable
the Sqoop 2 application.
Enabling the HBase Browser Application with doAs Impersonation
Minimum Required Role: Full Administrator
The Hue HBase application communicates through a proxy server called the HBase Thrift Server, which then forwards
commands to HBase. Because Hue stands between the Thrift server and the actual user, all HBase operations appear
to come from the hue user and not the actual user. To secure these interactions, you must do the following:
• Ensure that users logged into Hue perform operations with their own privileges, and not those of the impersonating
hue user.
• Once Hue can impersonate other users, ensure that only the Hue server can send commands to the HBase Thrift
server. To ensure this, use Kerberos to authenticate the hue user to the HBase Thrift server.
To enable the HBase browser application:
1. Add the HBase Thrift Server role.
2. If you have a Kerberos-enabled cluster, enable impersonation by configuring the following HBase properties:
a.
b.
c.
d.
e.
Select the HBase service.
Click the Configuration tab.
Select Scope > Service-Wide.
Select Category > Security.
For the HBase Thrift Authentication property, make sure it is set to one of the following values:
• auth-conf: authentication, integrity and confidentiality checking
• auth-int: authentication and integrity checking
• auth: authentication only
f. Select Category > Main.
g. Check the Enable HBase Thrift Http Server and Enable HBase Thrift Proxy Users properties checkboxes.
h. Click Save Changes to commit the changes.
3. Enable TLS/SSL for the HBase Thrift Server.
4. Configure Hue to point to the Thrift Server and to a valid HBase configuration directory:
a.
b.
c.
d.
e.
Select the Hue service.
Click the Configuration tab.
Select Scope > All.
Select Category > Main.
For the HBase Service property, make sure it is set to the HBase service for which you enabled the Thrift
Server role (if you have more than one HBase service instance).
f. In the HBase Thrift Server property, click the edit field and select the Thrift Server role for Hue to use.
178 | Cloudera Administration
Managing CDH and Managed Services
g. Select Category > Advanced.
h. Locate the Hue Service Advanced Configuration Snippet (Safety Valve) for hue_safety_valve.ini property
and add the following property:
[hbase]
hbase_conf_dir=/etc/hbase/conf
i. Click Save Changes to commit the changes.
Enabling the Solr Search Application
To use the Solr Search application with Hue, you must update the URL for the Solr Server in the Hue Server advanced
configuration snippet. In addition, if you are using parcels with CDH 4.3, you must register the "hue-search" application
manually, or access will fail. See Deploying Solr with Hue on page 227 for detailed instructions.
Using an External Database for Hue
Cloudera strongly recommends an external database for clusters with multiple Hue users, especially clusters in a
production environment. The default database, SQLite, works best with a single user and a small dataset. For supported
databases, see:
• CDH 4 supported databases
• CDH 5 supported databases
Using an External Database for Hue Using Cloudera Manager
Minimum Required Role: Full Administrator
The Hue server requires an SQL database to store small amounts of data such as user account information, job
submissions, and Hive queries. Hue supports a lightweight embedded (SQLite) database and several types of external
databases. This page explains how to configure Hue with a selection of Supported Databases.
Important: Cloudera strongly recommends an external database for clusters with multiple Hue
servers (or "hue" users). With the default embedded database (one per server), in a multi-server
environment, the data on server "A" appears lost when working on server "B" and vice versa. Use an
external database, and configure each server to point to it to ensure that no matter which server is
being used by Hue, your data is always accessible.
To configure Hue with any of the supported external databases, the high-level steps are:
1.
2.
3.
4.
5.
Stop Hue service.
Backup default SQLite database (if applicable).
Install database software and dependencies.
Create and configure database and load data.
Start Hue service.
See the tasks on this page for details. If you do not need to migrate a SQLite database, you can skip the steps on
dumping the database and editing the JSON objects.
Configuring the Hue Server to Store Data in MariaDB
For information about installing and configuring a MariaDB database , see MariaDB Database.
1.
2.
3.
4.
In the Cloudera Manager Admin Console, go to the Hue service status page.
Select Actions > Stop. Confirm you want to stop the service by clicking Stop.
Select Actions > Dump Database. Confirm you want to dump the database by clicking Dump Database.
Note the host to which the dump was written under Step in the Dump Database Command window. You can also
find it by selecting Commands > Recent Commands > Dump Database.
5. Open a terminal window for the host and go to the dump file in /tmp/hue_database_dump.json.
Cloudera Administration | 179
Managing CDH and Managed Services
6. Remove all JSON objects with useradmin.userprofile in the model field, for example:
{
"pk": 14,
"model": "useradmin.userprofile",
"fields":
{ "creation_method": "EXTERNAL", "user": 14, "home_directory": "/user/tuser2" }
},
7. Set strict mode in /etc/my.cnf and restart MySQL:
[mysqld]
sql_mode=STRICT_ALL_TABLES
8. Create a new database and grant privileges to a Hue user to manage this database. For example:
mysql> create database hue;
Query OK, 1 row affected (0.01 sec)
mysql> grant all on hue.* to 'hue'@'localhost' identified by 'secretpassword';
Query OK, 0 rows affected (0.00 sec)
9. In the Cloudera Manager Admin Console, click the Hue service.
10. Click the Configuration tab.
11. Select Scope > All.
12. Select Category > Database.
13. Specify the settings for Hue Database Type, Hue Database Hostname, Hue Database Port, Hue Database Username,
Hue Database Password, and Hue Database Name. For example, for a MySQL database on the local host, you
might use the following values:
•
•
•
•
•
•
Hue Database Type = mysql
Hue Database Hostname = host
Hue Database Port = 3306
Hue Database Username = hue
Hue Database Password = secretpassword
Hue Database Name = hue
14. Optionally restore the Hue data to the new database:
a. Select Actions > Synchronize Database.
b. Determine the foreign key ID.
$ mysql -uhue -psecretpassword
mysql > SHOW CREATE TABLE auth_permission;
c. (InnoDB only) Drop the foreign key that you retrieved in the previous step.
mysql > ALTER TABLE auth_permission DROP FOREIGN KEY content_type_id_refs_id_XXXXXX;
d. Delete the rows in the django_content_type table.
mysql > DELETE FROM hue.django_content_type;
e. In Hue service instance page, click Actions > Load Database. Confirm you want to load the database by clicking
Load Database.
f. (InnoDB only) Add back the foreign key.
mysql > ALTER TABLE auth_permission ADD FOREIGN KEY (content_type_id) REFERENCES
django_content_type (id);
180 | Cloudera Administration
Managing CDH and Managed Services
15. Start the Hue service.
Configuring the Hue Server to Store Data in MySQL
Note: Cloudera recommends InnoDB over MyISAM as the Hue MySQL engine. On CDH 5, Hue requires
InnoDB.
For information about installing and configuring a MySQL database , see MySQL Database.
1.
2.
3.
4.
In the Cloudera Manager Admin Console, go to the Hue service status page.
Select Actions > Stop. Confirm you want to stop the service by clicking Stop.
Select Actions > Dump Database. Confirm you want to dump the database by clicking Dump Database.
Note the host to which the dump was written under Step in the Dump Database Command window. You can also
find it by selecting Commands > Recent Commands > Dump Database.
5. Open a terminal window for the host and go to the dump file in /tmp/hue_database_dump.json.
6. Remove all JSON objects with useradmin.userprofile in the model field, for example:
{
"pk": 14,
"model": "useradmin.userprofile",
"fields":
{ "creation_method": "EXTERNAL", "user": 14, "home_directory": "/user/tuser2" }
},
7. Set strict mode in /etc/my.cnf and restart MySQL:
[mysqld]
sql_mode=STRICT_ALL_TABLES
8. Create a new database and grant privileges to a Hue user to manage this database. For example:
mysql> create database hue;
Query OK, 1 row affected (0.01 sec)
mysql> grant all on hue.* to 'hue'@'localhost' identified by 'secretpassword';
Query OK, 0 rows affected (0.00 sec)
9. In the Cloudera Manager Admin Console, click the Hue service.
10. Click the Configuration tab.
11. Select Scope > All.
12. Select Category > Database.
13. Specify the settings for Hue Database Type, Hue Database Hostname, Hue Database Port, Hue Database Username,
Hue Database Password, and Hue Database Name. For example, for a MySQL database on the local host, you
might use the following values:
•
•
•
•
•
•
Hue Database Type = mysql
Hue Database Hostname = host
Hue Database Port = 3306
Hue Database Username = hue
Hue Database Password = secretpassword
Hue Database Name = hue
14. Optionally restore the Hue data to the new database:
a. Select Actions > Synchronize Database.
b. Determine the foreign key ID.
$ mysql -uhue -psecretpassword
mysql > SHOW CREATE TABLE auth_permission;
Cloudera Administration | 181
Managing CDH and Managed Services
c. (InnoDB only) Drop the foreign key that you retrieved in the previous step.
mysql > ALTER TABLE auth_permission DROP FOREIGN KEY content_type_id_refs_id_XXXXXX;
d. Delete the rows in the django_content_type table.
mysql > DELETE FROM hue.django_content_type;
e. In Hue service instance page, click Actions > Load Database. Confirm you want to load the database by clicking
Load Database.
f. (InnoDB only) Add back the foreign key.
mysql > ALTER TABLE auth_permission ADD FOREIGN KEY (content_type_id) REFERENCES
django_content_type (id);
15. Start the Hue service.
Configuring the Hue Server to Store Data in PostgreSQL
For information about installing and configuring an external PostgreSQL database , see External PostgreSQL Database.
1.
2.
3.
4.
In the Cloudera Manager Admin Console, go to the Hue service status page.
Select Actions > Stop. Confirm you want to stop the service by clicking Stop.
Select Actions > Dump Database. Confirm you want to dump the database by clicking Dump Database.
Note the host to which the dump was written under Step in the Dump Database Command window. You can also
find it by selecting Commands > Recent Commands > Dump Database.
5. Open a terminal window for the host and go to the dump file in /tmp/hue_database_dump.json.
6. Remove all JSON objects with useradmin.userprofile in the model field, for example:
{
"pk": 14,
"model": "useradmin.userprofile",
"fields":
{ "creation_method": "EXTERNAL", "user": 14, "home_directory": "/user/tuser2" }
},
7. Install the PostgreSQL server.
RHEL
$ sudo yum install postgresql-server
SLES
$ sudo zypper install postgresql-server
Ubuntu or Debian
$ sudo apt-get install postgresql
8. Initialize the data directories.
$ service postgresql initdb
9. Configure client authentication.
a. Edit /var/lib/pgsql/data/pg_hba.conf.
182 | Cloudera Administration
Managing CDH and Managed Services
b. Set the authentication methods for local to trust and for host to password and add the following line at
the end.
host hue hue 0.0.0.0/0 md5
10. Start the PostgreSQL server.
$ su - postgres
# /usr/bin/postgres -D /var/lib/pgsql/data > logfile 2>&1 &
11. Configure PostgreSQL to listen on all network interfaces.
a. Edit /var/lib/pgsql/data/postgresql.conf and set list_addresses.
listen_addresses = ‘0.0.0.0’
# Listen on all addresses
12. Create the hue database and grant privileges to a hue user to manage the database.
# psql -U postgres
postgres=# create database hue;
postgres=# \c hue;
You are now connected to database 'hue'.
postgres=# create user hue with password 'secretpassword';
postgres=# grant all privileges on database hue to hue;
postgres=# \q
13. Restart the PostgreSQL server.
$ sudo service postgresql restart
14. Verify connectivity.
psql –h localhost –U hue –d hue
Password for user hue: secretpassword
15. Configure the PostgreSQL server to start at boot.
RHEL
$ sudo /sbin/chkconfig postgresql on
$ sudo /sbin/chkconfig --list postgresql
postgresql
0:off
1:off
2:on
3:on
4:on
5:on
6:off
SLES
$ sudo chkconfig --add postgresql
Ubuntu or Debian
$ sudo chkconfig postgresql on
16. Configure the Hue database:
a.
b.
c.
d.
In the Cloudera Manager Admin Console, click the HUE service.
Click the Configuration tab.
Select Scope > Hue Server.
Select Category > Advanced.
Cloudera Administration | 183
Managing CDH and Managed Services
e. Set Hue Server Advanced Configuration Snippet (Safety Valve) for hue_safety_valve_server.ini with the
following:
[desktop]
[[database]]
engine=postgresql_psycopg2
name=hue
host=localhost
port=5432
user=hue
password=secretpassword
Note: If you set Hue Database Hostname, Hue Database Port, Hue Database Username,
and Hue Database Password at the service-level, under Service-Wide > Database, you can
omit those properties from the server-lever configuration above and avoid storing the Hue
password as plain text. In either case, set engine and name in the server-level safety-valve.
f. Click Save Changes.
17. Optionally restore the Hue data to the new database:
a. Select Actions > Synchronize Database.
b. Determine the foreign key ID.
bash# su – postgres
$ psql –h localhost –U hue –d hue
postgres=# \d auth_permission;
c. Drop the foreign key that you retrieved in the previous step.
postgres=# ALTER TABLE auth_permission DROP CONSTRAINT content_type_id_refs_id_XXXXXX;
d. Delete the rows in the django_content_type table.
postgres=# TRUNCATE django_content_type CASCADE;
e. In Hue service instance page, Actions > Load Database. Confirm you want to load the database by clicking
Load Database.
f. Add back the foreign key you dropped.
bash# su – postgres
$ psql –h localhost –U hue –d hue
postgres=# ALTER TABLE auth_permission ADD CONSTRAINT content_type_id_refs_id_XXXXXX
FOREIGN KEY (content_type_id) REFERENCES django_content_type(id) DEFERRABLE INITIALLY
DEFERRED;
18. Start the Hue service.
Configuring the Hue Server to Store Data in Oracle (Parcel Installation)
Use the following instructions to configure the Hue Server with an Oracle database if you are working on a parcel-based
deployment. If you are using packages, see Configuring the Hue Server to Store Data in Oracle (Package Installation)
on page 186.
For information about installing and configuring an Oracle database , see Oracle Database.
Important: Configure the database for character set AL32UTF8 and national character set UTF8.
1. Install the required packages.
184 | Cloudera Administration
Managing CDH and Managed Services
RHEL
$ sudo yum install gcc python-devel python-pip python-setuptools libaio
SLES:
Python devel packages are not included in SLES. Add the SLES Software Development Kit (SDK) as a repository and
then install:
$ zypper install gcc libaio python-pip python-setuptools python-devel
Ubuntu or Debian
$ sudo apt-get install gcc python-dev python-pip python-setuptools libaio1
2. Download and add the Oracle Client parcel to the Cloudera Manager remote parcel repository URL list and
download, distribute, and activate the parcel.
3. For CDH versions lower than 5.3, install the Python Oracle library:
Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ HUE_HOME/build/env/bin/pip install cx_Oracle
4. For CDH versions lower than 5.3, upgrade django south:
$ HUE_HOME/build/env/bin/pip install south --upgrade
5. In the Cloudera Manager Admin Console, go to the Hue service status page.
6. Select Actions > Stop. Confirm you want to stop the service by clicking Stop.
7. Select Actions > Dump Database. Confirm you want to dump the database by clicking Dump Database.
8. Click the Configuration tab.
9. Select Scope > All.
10. Select Category > Advanced.
11. Set the Hue Service Advanced Configuration Snippet (Safety Valve) for hue_safety_valve.ini property.
Note: If you set Hue Database Hostname, Hue Database Port, Hue Database Username, and
Hue Database Password at the service-level, under Service-Wide > Database, you can omit those
properties from the server-lever configuration above and avoid storing the Hue password as plain
text. In either case, set engine and name in the server-level safety-valve.
Add the following options (and modify accordingly for your setup):
[desktop]
[[database]]
host=localhost
port=1521
engine=oracle
user=hue
password=secretpassword
name=<SID of the Oracle database, for example, 'XE'>
Cloudera Administration | 185
Managing CDH and Managed Services
For CDH 5.1 and higher you can use an Oracle service name. To use the Oracle service name instead of the SID,
use the following configuration instead:
port=0
engine=oracle
user=hue
password=secretpassword
name=oracle.example.com:1521/orcl.example.com
The directive port=0 allows Hue to use a service name. The name string is the connect string, including hostname,
port, and service name.
To add support for a multithreaded environment, set the threaded option to true under the
[desktop]>[[database]] section.
options={"threaded":true}
12. Grant required permissions to the hue user in Oracle:
GRANT
GRANT
GRANT
GRANT
GRANT
GRANT
GRANT
GRANT
CREATE <sequence> TO <user>;
CREATE <session> TO <user>;
CREATE <table> TO <user>;
CREATE <view> TO <user>;
CREATE <procedure> TO <user>;
CREATE <trigger> TO <user>;
EXECUTE ON sys.dbms_crypto TO <user>;
EXECUTE ON SYS.DBMS_LOB TO <user>;
13. Go to the Hue Server instance in Cloudera Manager and select Actions > Synchronize Database.
14. Ensure you are connected to Oracle as the hue user, then run the following command to delete all data from
Oracle tables:
> set pagesize 100;
> SELECT 'DELETE FROM ' || table_name || ';' FROM user_tables;
15. Run the statements generated in the preceding step.
16. Commit your changes.
commit;
17. Load the data that you dumped. Go to the Hue Server instance and select Actions > Load Database. This step is
not necessary if you have a fresh Hue install with no data or if you don’t want to save the Hue data.
18. Start the Hue service.
Configuring the Hue Server to Store Data in Oracle (Package Installation)
If you have a parcel-based environment, see Configuring the Hue Server to Store Data in Oracle (Parcel Installation)
on page 184.
Important: Configure the database for character set AL32UTF8 and national character set UTF8.
1. Download the Oracle libraries at Instant Client for Linux x86-64 Version 11.1.0.7.0, Basic and SDK (with headers)
zip files to the same directory.
2. Unzip the Oracle client zip files.
3. Set environment variables to reference the libraries.
$ export ORACLE_HOME=oracle_download_directory
$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$ORACLE_HOME
186 | Cloudera Administration
Managing CDH and Managed Services
4. Create a symbolic link for the shared object:
$ cd $ORACLE_HOME
$ ln -sf libclntsh.so.11.1 libclntsh.so
5. Install the required packages.
RHEL
$ sudo yum install gcc python-devel python-pip python-setuptools libaio
SLES:
Python devel packages are not included in SLES. Add the SLES Software Development Kit (SDK) as a repository and
then install:
$ zypper install gcc libaio python-pip python-setuptools python-devel
Ubuntu or Debian
$ sudo apt-get install gcc python-dev python-pip python-setuptools libaio1
6. For CDH versions lower than 5.3, install the Python Oracle library:
Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ HUE_HOME/build/env/bin/pip install cx_Oracle
7. For CDH versions lower than 5.3, upgrade django south:
$ HUE_HOME/build/env/bin/pip install south --upgrade
8. In the Cloudera Manager Admin Console, go to the Hue service status page.
9. Select Actions > Stop. Confirm you want to stop the service by clicking Stop.
10. Select Actions > Dump Database. Confirm you want to dump the database by clicking Dump Database.
11. Click the Configuration tab.
12. Select Scope > All.
13. Select Category > Advanced.
14. Set the Hue Service Environment Advanced Configuration Snippet (Safety Valve) property to
ORACLE_HOME=oracle_download_directory
LD_LIBRARY_PATH=$LD_LIBRARY_PATH:oracle_download_directory
15. Set the Hue Service Advanced Configuration Snippet (Safety Valve) for hue_safety_valve.ini property.
Note: If you set Hue Database Hostname, Hue Database Port, Hue Database Username, and
Hue Database Password at the service-level, under Service-Wide > Database, you can omit those
properties from the server-lever configuration above and avoid storing the Hue password as plain
text. In either case, set engine and name in the server-level safety-valve.
Add the following options (and modify accordingly for your setup):
[desktop]
[[database]]
host=localhost
Cloudera Administration | 187
Managing CDH and Managed Services
port=1521
engine=oracle
user=hue
password=secretpassword
name=<SID of the Oracle database, for example, 'XE'>
For CDH 5.1 and higher you can use an Oracle service name. To use the Oracle service name instead of the SID,
use the following configuration instead:
port=0
engine=oracle
user=hue
password=secretpassword
name=oracle.example.com:1521/orcl.example.com
The directive port=0 allows Hue to use a service name. The name string is the connect string, including hostname,
port, and service name.
To add support for a multithreaded environment, set the threaded option to true under the
[desktop]>[[database]] section.
options={"threaded":true}
16. Grant required permissions to the hue user in Oracle:
GRANT
GRANT
GRANT
GRANT
GRANT
GRANT
GRANT
GRANT
CREATE <sequence> TO <user>;
CREATE <session> TO <user>;
CREATE <table> TO <user>;
CREATE <view> TO <user>;
CREATE <procedure> TO <user>;
CREATE <trigger> TO <user>;
EXECUTE ON sys.dbms_crypto TO <user>;
EXECUTE ON SYS.DBMS_LOB TO <user>;
17. Go to the Hue Server instance in Cloudera Manager and select Actions > Synchronize Database.
18. Ensure you are connected to Oracle as the hue user, then run the following command to delete all data from
Oracle tables:
> set pagesize 100;
> SELECT 'DELETE FROM ' || table_name || ';' FROM user_tables;
19. Run the statements generated in the preceding step.
20. Commit your changes.
commit;
21. Load the data that you dumped. Go to the Hue Server instance and select Actions > Load Database. This step is
not necessary if you have a fresh Hue install with no data or if you don’t want to save the Hue data.
22. Start the Hue service.
Using an External Database for Hue Using the Command Line
The Hue server requires a SQL database to store small amounts of data such as user account information, job submissions,
and Hive queries. SQLite is the default embedded database. Hue also supports several types of external databases.
This page explains how to configure Hue with a selection of external Supported Databases.
Important: Cloudera strongly recommends an external database for clusters with multiple Hue users.
To configure Hue with any of the supported external databases, the high-level steps are:
188 | Cloudera Administration
Managing CDH and Managed Services
1.
2.
3.
4.
5.
Stop Hue service.
Backup default SQLite database.
Install database software and dependencies.
Create and configure database and load data.
Start Hue service.
See the tasks on this page for details. If you don't need to migrate a SQLite database, you can skip the steps on dumping
the database and editing the JSON objects.
Prerequisites
Before using an external database with Hue, install all of the support libraries required by your operating system. See
Development Preferences in the Hue documentation for the full list.
Embedded Database
By default, Hue is configured to use the embedded database, SQLite, and should require no configuration or management
by the administrator.
Inspecting the Embedded Hue Database
The default SQLite database used by Hue is located in /var/lib/hue/desktop.db. You can inspect this database
from the command line using the sqlite3 program. For example:
# sqlite3 /var/lib/hue/desktop.db
SQLite version 3.6.22
Enter ".help" for instructions
Enter SQL statements terminated with a ";"
sqlite> select username from auth_user;
admin
test
sample
sqlite>
Important: It is strongly recommended that you avoid making any modifications to the database
directly using sqlite3, though sqlite3 is useful for management or troubleshooting.
Backing up the Embedded Hue Database
If you use the default embedded SQLite database, copy the desktop.db file to another node for backup. Cloudera
recommends that you backup regularly, and also that you backup before upgrading to a new version of Hue.
External Database
Cloudera strongly recommends an external database for clusters with multiple Hue users, especially clusters in a
production environment. The default database, SQLite, cannot support large data migrations. Hue supports MariaDB,
MySQL, PostgreSQL, and Oracle. See Supported Databases for the supported versions.
In the instructions that follow, dumping the database and editing the JSON objects is only necessary if you have data
in SQLite that you need to migrate. If you do not need to migrate data from SQLite, you can skip those steps.
Configuring the Hue Server to Store Data in MariaDB
Note: Cloudera recommends InnoDB over MyISAM as the Hue MySQL engine. On CDH 5, Hue requires
InnoDB.
1. Shut down the Hue server if it is running.
Cloudera Administration | 189
Managing CDH and Managed Services
2. Open <some-temporary-file>.json and remove all JSON objects with useradmin.userprofile in the
model field. Here are some examples of JSON objects that should be deleted.
{
"pk": 1,
"model": "useradmin.userprofile",
"fields": {
"creation_method": "HUE",
"user": 1,
"home_directory": "/user/alice"
}
},
{
"pk": 2,
"model": "useradmin.userprofile",
"fields": {
"creation_method": "HUE",
"user": 1100714,
"home_directory": "/user/bob"
}
},
.....
3. Start the Hue server.
4. Install the MariaDB client developer package.
OS
Command
RHEL
$ sudo yum install mariadb-devel
SLES
$ sudo zypper install mariadb-devel
Ubuntu or Debian
$ sudo apt-get install libmariadbclient-dev
5. Install the MariaDB connector.
OS
Command
RHEL
$ sudo yum install mariadb-connector-java
SLES
$ sudo zypper install mariadb-connector-java
Ubuntu or Debian
$ sudo apt-get install libmariadb-java
6. Install and start MariaDB.
OS
Command
RHEL
$ sudo yum install mariadb-server
SLES
$ sudo zypper install
Ubuntu or Debian
$ sudo apt-get install mariadb-server
mariadb-server
$ sudo zypper install libmariadblclient18
7. Change the /etc/my.cnf file as follows:
[mysqld]
datadir=/var/lib/mysql
socket=/var/lib/mysql/mysql.sock
bind-address=<ip-address>
default-storage-engine=InnoDB
sql_mode=STRICT_ALL_TABLES
190 | Cloudera Administration
Managing CDH and Managed Services
8. Start the MariaDB daemon.
$ sudo service mariadb start
9. Configure MariaDB to use a strong password. In the following procedure, your current root password is blank.
Press the Enter key when you're prompted for the root password.
$ sudo /usr/bin/mysql_secure_installation
[...]
Enter current password for root (enter for none):
OK, successfully used password, moving on...
[...]
Set root password? [Y/n] y
New password:
Re-enter new password:
Remove anonymous users? [Y/n] Y
[...]
Disallow root login remotely? [Y/n] N
[...]
Remove test database and access to it [Y/n] Y
[...]
Reload privilege tables now? [Y/n] Y
All done!
10. Configure MariaDB to start at boot.
OS
Command
RHEL
$ sudo /sbin/chkconfig mariadb on
$ sudo /sbin/chkconfig --list mariadb
mysqld
0:off
1:off
2:on
6:off
SLES
$ sudo chkconfig --add mariadb
Ubuntu or Debian
$ sudo chkconfig mariadb on
3:on
4:on
5:on
11. Create the Hue database and grant privileges to a hue user to manage the database.
mysql> create database hue;
Query OK, 1 row affected (0.01 sec)
mysql> grant all on hue.* to 'hue'@'localhost' identified by '<secretpassword>';
Query OK, 0 rows affected (0.00 sec)
12. Open the Hue configuration file in a text editor.
13. Edit the Hue configuration file hue.ini. Directly below the [[database]] section under the [desktop] line,
add the following options (and modify accordingly for your setup):
host=localhost
port=3306
engine=mysql
user=hue
password=<secretpassword>
name=hue
14. As the hue user, load the existing data and create the necessary database tables using syncdb and migrate
commands. When running these commands, Hue will try to access a logs directory, located at
/opt/cloudera/parcels/CDH/lib/hue/logs, which might be missing. If that is the case, first create the
logs directory and give the hue user and group ownership of the directory.
Cloudera Administration | 191
Managing CDH and Managed Services
Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ sudo mkdir /opt/cloudera/parcels/CDH/lib/hue/logs
$ sudo chown hue:hue /opt/cloudera/parcels/CDH/lib/hue/logs
$ sudo -u hue <HUE_HOME>/build/env/bin/hue syncdb --noinput
$ sudo -u hue <HUE_HOME>/build/env/bin/hue migrate
$ mysql -u hue -p <secretpassword>
mysql > SHOW CREATE TABLE auth_permission;
15. (InnoDB only) Drop the foreign key.
mysql > ALTER TABLE auth_permission DROP FOREIGN KEY content_type_id_refs_id_XXXXXX;
16. Delete the rows in the django_content_type table.
mysql > DELETE FROM hue.django_content_type;
17. Load the data.
$ <HUE_HOME>/build/env/bin/hue loaddata <some-temporary-file>.json
18. (InnoDB only) Add the foreign key.
$ mysql -u hue -p <secretpassword>
mysql > ALTER TABLE auth_permission ADD FOREIGN KEY (`content_type_id`) REFERENCES
`django_content_type` (`id`);
Configuring the Hue Server to Store Data in MySQL
Note: Cloudera recommends InnoDB over MyISAM as the Hue MySQL engine. On CDH 5, Hue requires
InnoDB.
1. Shut down the Hue server if it is running.
2. Dump the existing database data to a text file. Note that using the .json extension is required.
Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ sudo -u hue <HUE_HOME>/build/env/bin/hue dumpdata > <some-temporary-file>.json
3. Open <some-temporary-file>.json and remove all JSON objects with useradmin.userprofile in the
model field. Here are some examples of JSON objects that should be deleted.
{
"pk": 1,
"model": "useradmin.userprofile",
"fields": {
"creation_method": "HUE",
"user": 1,
"home_directory": "/user/alice"
}
},
{
192 | Cloudera Administration
Managing CDH and Managed Services
"pk": 2,
"model": "useradmin.userprofile",
"fields": {
"creation_method": "HUE",
"user": 1100714,
"home_directory": "/user/bob"
}
},
.....
4. Start the Hue server.
5. Install the MySQL client developer package.
OS
Command
RHEL
$ sudo yum install mysql-devel
SLES
$ sudo zypper install mysql-devel
Ubuntu or Debian
$ sudo apt-get install libmysqlclient-dev
6. Install the MySQL connector.
OS
Command
RHEL
$ sudo yum install mysql-connector-java
SLES
$ sudo zypper install mysql-connector-java
Ubuntu or Debian
$ sudo apt-get install libmysql-java
7. Install and start MySQL.
OS
Command
RHEL
$ sudo yum install mysql-server
SLES
$ sudo zypper install mysql
$ sudo zypper install libmysqlclient_r15
Ubuntu or Debian
$ sudo apt-get install mysql-server
8. Change the /etc/my.cnf file as follows:
[mysqld]
datadir=/var/lib/mysql
socket=/var/lib/mysql/mysql.sock
bind-address=<ip-address>
default-storage-engine=InnoDB
sql_mode=STRICT_ALL_TABLES
9. Start the mysql daemon.
OS
Command
RHEL
$ sudo service mysqld start
SLES and Ubuntu or
Debian
$ sudo service mysql start
Cloudera Administration | 193
Managing CDH and Managed Services
10. Configure MySQL to use a strong password. In the following procedure, your current root password is blank.
Press the Enter key when you're prompted for the root password.
$ sudo /usr/bin/mysql_secure_installation
[...]
Enter current password for root (enter for none):
OK, successfully used password, moving on...
[...]
Set root password? [Y/n] y
New password:
Re-enter new password:
Remove anonymous users? [Y/n] Y
[...]
Disallow root login remotely? [Y/n] N
[...]
Remove test database and access to it [Y/n] Y
[...]
Reload privilege tables now? [Y/n] Y
All done!
11. Configure MySQL to start at boot.
OS
Command
RHEL
$ sudo /sbin/chkconfig mysqld on
$ sudo /sbin/chkconfig --list mysqld
mysqld
0:off
1:off
2:on
6:off
SLES
$ sudo chkconfig --add mysql
Ubuntu or Debian
$ sudo chkconfig mysql on
3:on
4:on
5:on
12. Create the Hue database and grant privileges to a hue user to manage the database.
mysql> create database hue;
Query OK, 1 row affected (0.01 sec)
mysql> grant all on hue.* to 'hue'@'localhost' identified by '<secretpassword>';
Query OK, 0 rows affected (0.00 sec)
13. Open the Hue configuration file in a text editor.
14. Edit the Hue configuration file hue.ini. Directly below the [[database]] section under the [desktop] line,
add the following options (and modify accordingly for your setup):
host=localhost
port=3306
engine=mysql
user=hue
password=<secretpassword>
name=hue
15. As the hue user, load the existing data and create the necessary database tables using syncdb and migrate
commands. When running these commands, Hue will try to access a logs directory, located at
/opt/cloudera/parcels/CDH/lib/hue/logs, which might be missing. If that is the case, first create the
logs directory and give the hue user and group ownership of the directory.
Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ sudo mkdir /opt/cloudera/parcels/CDH/lib/hue/logs
$ sudo chown hue:hue /opt/cloudera/parcels/CDH/lib/hue/logs
$ sudo -u hue <HUE_HOME>/build/env/bin/hue syncdb --noinput
194 | Cloudera Administration
Managing CDH and Managed Services
$ sudo -u hue <HUE_HOME>/build/env/bin/hue migrate
$ mysql -u hue -p <secretpassword>
mysql > SHOW CREATE TABLE auth_permission;
16. (InnoDB only) Drop the foreign key.
mysql > ALTER TABLE auth_permission DROP FOREIGN KEY content_type_id_refs_id_XXXXXX;
17. Delete the rows in the django_content_type table.
mysql > DELETE FROM hue.django_content_type;
18. Load the data.
$ <HUE_HOME>/build/env/bin/hue loaddata <some-temporary-file>.json
19. (InnoDB only) Add the foreign key.
$ mysql -u hue -p <secretpassword>
mysql > ALTER TABLE auth_permission ADD FOREIGN KEY (`content_type_id`) REFERENCES
`django_content_type` (`id`);
Configuring the Hue Server to Store Data in PostgreSQL
Warning: Hue requires PostgreSQL 8.4 or higher.
1. Shut down the Hue server if it is running.
2. Dump the existing database data to a text file. Note that using the .json extension is required.
Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ sudo -u hue <HUE_HOME>/build/env/bin/hue dumpdata > <some-temporary-file>.json
3. Open <some-temporary-file>.json and remove all JSON objects with useradmin.userprofile in the
model field. Here are some examples of JSON objects that should be deleted.
{
"pk": 1,
"model": "useradmin.userprofile",
"fields": {
"creation_method": "HUE",
"user": 1,
"home_directory": "/user/alice"
}
},
{
"pk": 2,
"model": "useradmin.userprofile",
"fields": {
"creation_method": "HUE",
"user": 1100714,
"home_directory": "/user/bob"
}
},
.....
Cloudera Administration | 195
Managing CDH and Managed Services
4. Install required packages.
OS
Command
RHEL
$ sudo yum install postgresql-devel gcc python-devel
SLES
$ sudo zypper install postgresql-devel gcc python-devel
Ubuntu or Debian
$ sudo apt-get install postgresql-devel gcc python-devel
5. Install the module that provides the connector to PostgreSQL.
sudo -u hue <HUE_HOME>/build/env/bin/pip install setuptools
sudo -u hue <HUE_HOME>/build/env/bin/pip install psycopg2
6. Install the PostgreSQL server.
OS
Command
RHEL
$ sudo yum install postgresql-server
SLES
$ sudo zypper install postgresql-server
Ubuntu or Debian
$ sudo apt-get install postgresql
7. Initialize the data directories:
$ service postgresql initdb
8. Configure client authentication.
a. Edit /var/lib/pgsql/data/pg_hba.conf.
b. Set the authentication methods for local to trust and for host to password and add the following line at
the end.
host hue hue 0.0.0.0/0 md5
9. Start the PostgreSQL server.
$ su - postgres
# /usr/bin/postgres -D /var/lib/pgsql/data > logfile 2>&1 &
10. Configure PostgreSQL to listen on all network interfaces.
Edit /var/lib/pgsql/data/postgresql.conf and set list_addresses:
listen_addresses = '0.0.0.0'
# Listen on all addresses
11. Create the hue database and grant privileges to a hue user to manage the database.
# psql -U postgres
postgres=# create database hue;
postgres=# \c hue;
You are now connected to database 'hue'.
postgres=# create user hue with password '<secretpassword>';
postgres=# grant all privileges on database hue to hue;
postgres=# \q
196 | Cloudera Administration
Managing CDH and Managed Services
12. Restart the PostgreSQL server.
$ sudo service postgresql restart
13. Verify connectivity.
psql -h localhost -U hue -d hue
Password for user hue: <secretpassword>
14. Configure the PostgreSQL server to start at boot.
OS
Command
RHEL
$ sudo /sbin/chkconfig postgresql on
$ sudo /sbin/chkconfig --list postgresql
postgresql
0:off
1:off
2:on
5:on
6:off
SLES
$ sudo chkconfig --add postgresql
Ubuntu or Debian
$ sudo chkconfig postgresql on
3:on
4:on
15. Open the Hue configuration file in a text editor.
16. Edit the Hue configuration file hue.ini. Directly below the [[database]] section under the [desktop] line,
add the following options (and modify accordingly for your setup):
host=localhost
port=5432
engine=postgresql_psycopg2
user=hue
password=<secretpassword>
name=hue
17. As the hue user, configure Hue to load the existing data and create the necessary database tables. You will need
to run both the migrate and syncdb commands. When running these commands, Hue will try to access a logs
directory, located at /opt/cloudera/parcels/CDH/lib/hue/logs, which might be missing. If that is the
case, first create the logs directory and give the hue user and group ownership of the directory.
$
$
$
$
sudo
sudo
sudo
sudo
mkdir /opt/cloudera/parcels/CDH/lib/hue/logs
chown hue:hue /opt/cloudera/parcels/CDH/lib/hue/logs
-u hue <HUE_HOME>/build/env/bin/hue syncdb --noinput
-u hue <HUE_HOME>/build/env/bin/hue migrate
18. Determine the foreign key ID.
bash# su - postgres
$ psql -h localhost -U hue -d hue
postgres=# \d auth_permission;
19. Drop the foreign key that you retrieved in the previous step.
postgres=# ALTER TABLE auth_permission DROP CONSTRAINT content_type_id_refs_id_<XXXXXX>;
20. Delete the rows in the django_content_type table.
postgres=# TRUNCATE django_content_type CASCADE;
21. Load the data.
$ sudo -u hue <HUE_HOME>/build/env/bin/hue loaddata <some-temporary-file>.json
Cloudera Administration | 197
Managing CDH and Managed Services
22. Add back the foreign key you dropped.
bash# su - postgres
$ psql -h localhost -U hue -d hue
postgres=# ALTER TABLE auth_permission ADD CONSTRAINT content_type_id_refs_id_<XXXXXX>
FOREIGN KEY (content_type_id) REFERENCES django_content_type(id) DEFERRABLE INITIALLY
DEFERRED;
Configuring the Hue Server to Store Data in Oracle
Important: Configure the database for character set AL32UTF8 and national character set UTF8.
1. Ensure Python 2.6 or higher is installed on the server Hue is running on.
2. Download the Oracle client libraries at Instant Client for Linux x86-64 Version 11.1.0.7.0, Basic and SDK (with
headers) zip files to the same directory.
3. Unzip the zip files.
4. Set environment variables to reference the libraries.
$ export ORACLE_HOME=<download directory>
$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$ORACLE_HOME
5. Create a symbolic link for the shared object:
$ cd $ORACLE_HOME
$ ln -sf libclntsh.so.11.1 libclntsh.so
6. Get a data dump by executing:
Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ <HUE_HOME>/build/env/bin/hue dumpdata > <some-temporary-file>.json --indent 2
7. Edit the Hue configuration file hue.ini. Directly below the [[database]] section under the [desktop] line,
add the following options (and modify accordingly for your setup):
host=localhost
port=1521
engine=oracle
user=hue
password=<secretpassword>
name=<SID of the Oracle database, for example, 'XE'>
To use the Oracle service name instead of the SID, use the following configuration instead:
port=0
engine=oracle
user=hue
password=password
name=oracle.example.com:1521/orcl.example.com
The directive port=0 allows Hue to use a service name. The name string is the connect string, including hostname,
port, and service name.
198 | Cloudera Administration
Managing CDH and Managed Services
To add support for a multithreaded environment, set the threaded option to true under the
[desktop]>[[database]] section.
options={'threaded':true}
8. Grant required permissions to the Hue user in Oracle:
GRANT
GRANT
GRANT
GRANT
GRANT
GRANT
GRANT
GRANT
CREATE <sequence> TO <user>;
CREATE <session> TO <user>;
CREATE <table> TO <user>;
CREATE <view> TO <user>;
CREATE <procedure> TO <user>;
CREATE <trigger> TO <user>;
EXECUTE ON sys.dbms_crypto TO <user>;
EXECUTE ON SYS.DBMS_LOB TO <user>;
9. As the hue user, configure Hue to load the existing data and create the necessary database tables. You will need
to run both the syncdb and migrate commands. When running these commands, Hue will try to access a logs
directory, located at /opt/cloudera/parcels/CDH/lib/hue/logs, which might be missing. If that is the
case, first create the logs directory and give the hue user and group ownership of the directory.
$
$
$
$
sudo
sudo
sudo
sudo
mkdir /opt/cloudera/parcels/CDH/lib/hue/logs
chown hue:hue /opt/cloudera/parcels/CDH/lib/hue/logs
-u hue <HUE_HOME>/build/env/bin/hue syncdb --noinput
-u hue <HUE_HOME>/build/env/bin/hue migrate
10. Ensure you are connected to Oracle as the hue user, then run the following command to delete all data from
Oracle tables:
SELECT 'DELETE FROM ' || '.' || table_name || ';' FROM user_tables;
11. Run the statements generated in the preceding step.
12. Commit your changes.
commit;
13. Load the data.
$ sudo -u hue <HUE_HOME>/build/env/bin/hue loaddata <some-temporary-file>.json
Managing Impala
This section explains how to configure Impala to accept connections from applications that use popular programming
APIs:
• Post-Installation Configuration for Impala on page 202
• Configuring Impala to Work with ODBC on page 204
• Configuring Impala to Work with JDBC on page 206
This type of configuration is especially useful when using Impala in combination with Business Intelligence tools, which
use these standard interfaces to query different kinds of database and Big Data systems.
You can also configure these other aspects of Impala:
• Overview of Impala Security
• Modifying Impala Startup Options
Cloudera Administration | 199
Managing CDH and Managed Services
The Impala Service
The Impala Service is the Cloudera Manager representation of the three daemons that make up the Impala interactive
SQL engine. Through the Impala Service page, you can monitor, start and stop, and configure all the related daemons
from a central page.
For general information about Impala and how to use it, especially for writing Impala SQL queries, see Impala Guide.
For information on features that support Impala resource management see Admission Control and Query Queuing on
page 255.
Installing Impala and Creating the Service
You can install Impala through the Cloudera Manager installation wizard, using either parcels or packages, and have
the service created and started as part of the Installation wizard. See Installing Impala.
If you elect not to include the Impala service using the Installation wizard, you can use the Add Service wizard to
perform the installation. The wizard will automatically configure and start the dependent services and the Impala
service. See Adding a Service on page 36 for instructions.
For general information about Impala and how to use it, see Impala Guide.
For information on features that support Impala resource management see Impala Resource Management on page
255.
Configuring the Impala Service
There are several types of configuration settings you may need to apply, depending on your situation.
Configuring Table Statistics
Configuring table statistics is highly recommended when using Impala. It allows Impala to make optimizations that can
result in significant (over 10x) performance improvement for some joins. If these are not available, Impala will still
function, but at lower performance.
The Impala implementation to compute table statistics is available in CDH 5.0.0 or higher and in Impala version 1.2.2
or higher. The Impala implementation of COMPUTE STATS requires no setup steps and is preferred over the Hive
implementation. See Overview of Table Statistics. If you are running an older version of Impala, follow the procedure
in Hive Table Statistics on page 170.
Using a Load Balancer with Impala
To configure a load balancer:
1.
2.
3.
4.
5.
Go to the Impala service.
Click the Configuration tab.
Select Scope > Impala Daemon
Select Category > All
Enter the hostname and port number of the load balancer in the Impala Daemons Load Balancer property in the
format hostname:port number.
Note:
When you set this property, Cloudera Manager regenerates the keytabs for Impala Daemon roles.
The principal in these keytabs contains the load balancer hostname.
If there is a Hue service that depends on this Impala service, it also uses the load balancer to
communicate with Impala.
6. Click Save Changes to commit the changes.
200 | Cloudera Administration
Managing CDH and Managed Services
Impala Web Servers
Enabling and Disabling Access to Impala Web Servers
Each of the Impala-related daemons includes a built-in web server that lets an administrator diagnose issues with each
daemon on a particular host, or perform other administrative actions such as cancelling a running query. By default,
these web servers are enabled. You might turn them off in a high-security configuration where it is not appropriate
for users to have access to this kind of monitoring information through a web interface. (To leave the web servers
enabled but control who can access their web pages, consult the Configuring Secure Access for Impala Web Servers
later in this section.)
• Impala Daemon
1.
2.
3.
4.
5.
6.
7.
Go to the Impala service.
Click the Configuration tab.
Select Scope > Impala Daemon
Select Category > Ports and Addresses.
Select or deselect Enable Impala Daemon Web Server.
Click Save Changes to commit the changes.
Restart the Impala service.
• Impala StateStore
1.
2.
3.
4.
5.
6.
7.
Go to the Impala service.
Click the Configuration tab.
Select Scope > Impala StateStore.
Select Category > All
Select or deselect Enable StateStore Web Server.
Click Save Changes to commit the changes.
Restart the Impala service.
• Impala Catalog Server
1.
2.
3.
4.
5.
6.
7.
Go to the Impala service.
Click the Configuration tab.
Select Scope > Impala Catalog Server.
Select Category > All
Check or uncheck Enable Catalog Server Web Server.
Click Save Changes to commit the changes.
Restart the Impala service.
Opening Impala Web Server UIs
• Impala StateStore
1. Go to the Impala service.
2. Select Web UI > Impala StateStore Web UI.
• Impala Daemon
1.
2.
3.
4.
Go to the Impala service.
Click the Instances tab.
Click an Impala Daemon instance.
Click Impala Daemon Web UI.
• Impala Catalog Server
1. Go to the Impala service.
Cloudera Administration | 201
Managing CDH and Managed Services
2. Select Web UI > Impala Catalog Web UI.
• Impala Llama ApplicationMaster
1.
2.
3.
4.
Go to the Impala service.
Click the Instances tab.
Click a Impala Llama ApplicationMaster instance.
Click Llama Web UI.
Configuring Secure Access for Impala Web Servers
Cloudera Manager supports two methods of authentication for secure access to the Impala Catalog Server, Daemon,
and StateStoreweb servers: password-based authentication and TLS/SSL certificate authentication. Both of these can
be configured through properties of the Impala Catalog Server, Daemon, and StateStore. Authentication for the three
types of daemons can be configured independently.
Configuring Password Authentication
1. Go to the Impala service.
2. Click the Configuration tab.
3. Search for "password" using the Search box within the Configuration page. This should display the password-related
properties (Username and Password properties) for the Impala Catalog Server, Daemon, and StateStore. If there
are multiple role groups configured for Impala Daemon instances, the search should display all of them.
4. Enter a username and password into these fields.
5. Click Save Changes to commit the changes.
6. Restart the Impala service.
Now when you access the Web UI for the Impala Catalog Server, Daemon, and StateStore, you are asked to log in
before access is granted.
Configuring TLS/SSL Certificate Authentication
1. Create or obtain an TLS/SSL certificate.
2. Place the certificate, in .pem format, on the hosts where the Impala Catalog Server and StateStore are running,
and on each host where an Impala Daemon is running. It can be placed in any location (path) you choose. If all
the Impala Daemons are members of the same role group, then the .pem file must have the same path on every
host.
3. Go to the Impala service page.
4. Click the Configuration tab.
5. Search for "certificate" using the Search box within the Configuration page. This should display the certificate file
location properties for the Impala Catalog Server, Daemon, and StateStore. If there are multiple role groups
configured for Impala Daemon instances, the search should display all of them.
6. In the property fields, enter the full path name to the certificate file.
7. Click Save Changes to commit the changes.
8. Restart the Impala service.
Important: If Cloudera Manager cannot find the .pem file on the host for a specific role instance,
that role will fail to start.
When you access the Web UI for the Impala Catalog Server, Daemon, and StateStore, https will be used.
Post-Installation Configuration for Impala
This section describes the mandatory and recommended configuration settings for Impala. If Impala is installed using
Cloudera Manager, some of these configurations are completed automatically; you must still configure short-circuit
202 | Cloudera Administration
Managing CDH and Managed Services
reads manually. If you installed Impala without Cloudera Manager, or if you want to customize your environment,
consider making the changes described in this topic.
In some cases, depending on the level of Impala, CDH, and Cloudera Manager, you might need to add particular
component configuration details in one of the free-form fields on the Impala configuration pages within Cloudera
Manager. In Cloudera Manager 4, these fields are labelled Safety Valve; in Cloudera Manager 5, they are called
Advanced Configuration Snippet.
• You must enable short-circuit reads, whether or not Impala was installed through Cloudera Manager. This setting
goes in the Impala configuration settings, not the Hadoop-wide settings.
• If you installed Impala in an environment that is not managed by Cloudera Manager, you must enable block location
tracking, and you can optionally enable native checksumming for optimal performance.
• If you deployed Impala using Cloudera Manager see Testing Impala Performance to confirm proper configuration.
Mandatory: Short-Circuit Reads
Enabling short-circuit reads allows Impala to read local data directly from the file system. This removes the need to
communicate through the DataNodes, improving performance. This setting also minimizes the number of additional
copies of data. Short-circuit reads requires libhadoop.so (the Hadoop Native Library) to be accessible to both the
server and the client. libhadoop.so is not available if you have installed from a tarball. You must install from an
.rpm, .deb, or parcel to use short-circuit local reads.
Note: If you use Cloudera Manager, you can enable short-circuit reads through a checkbox in the
user interface and that setting takes effect for Impala as well.
To configure DataNodes for short-circuit reads:
1. Copy the client core-site.xml and hdfs-site.xml configuration files from the Hadoop configuration directory
to the Impala configuration directory. The default Impala configuration location is /etc/impala/conf.
2. On all Impala nodes, configure the following properties in Impala's copy of hdfs-site.xml as shown:
<property>
<name>dfs.client.read.shortcircuit</name>
<value>true</value>
</property>
<property>
<name>dfs.domain.socket.path</name>
<value>/var/run/hdfs-sockets/dn</value>
</property>
<property>
<name>dfs.client.file-block-storage-locations.timeout.millis</name>
<value>10000</value>
</property>
3. If /var/run/hadoop-hdfs/ is group-writable, make sure its group is root.
Note: If you are also going to enable block location tracking, you can skip copying configuration
files and restarting DataNodes and go straight to Optional: Block Location Tracking. Configuring
short-circuit reads and block location tracking require the same process of copying files and
restarting services, so you can complete that process once when you have completed all
configuration changes. Whether you copy files and restart services now or during configuring
block location tracking, short-circuit reads are not enabled until you complete those final steps.
4. After applying these changes, restart all DataNodes.
Cloudera Administration | 203
Managing CDH and Managed Services
Mandatory: Block Location Tracking
Enabling block location metadata allows Impala to know which disk data blocks are located on, allowing better utilization
of the underlying disks. Impala will not start unless this setting is enabled.
To enable block location tracking:
1. For each DataNode, adding the following to the hdfs-site.xml file:
<property>
<name>dfs.datanode.hdfs-blocks-metadata.enabled</name>
<value>true</value>
</property>
2. Copy the client core-site.xml and hdfs-site.xml configuration files from the Hadoop configuration directory
to the Impala configuration directory. The default Impala configuration location is /etc/impala/conf.
3. After applying these changes, restart all DataNodes.
Optional: Native Checksumming
Enabling native checksumming causes Impala to use an optimized native library for computing checksums, if that library
is available.
To enable native checksumming:
If you installed CDH from packages, the native checksumming library is installed and setup correctly. In such a case,
no additional steps are required. Conversely, if you installed by other means, such as with tarballs, native checksumming
may not be available due to missing shared objects. Finding the message "Unable to load native-hadoop
library for your platform... using builtin-java classes where applicable" in the Impala logs
indicates native checksumming may be unavailable. To enable native checksumming, you must build and install
libhadoop.so (the Hadoop Native Library).
Configuring Impala to Work with ODBC
Third-party products can be designed to integrate with Impala using ODBC. For the best experience, ensure any
third-party product you intend to use is supported. Verifying support includes checking that the versions of Impala,
ODBC, the operating system, and the third-party product have all been approved for use together. Before configuring
your systems to use ODBC, download a connector. You may need to sign in and accept license agreements before
accessing the pages required for downloading ODBC connectors.
Downloading the ODBC Driver
Important: As of late 2015, most business intelligence applications are certified with the 2.x ODBC
drivers. Although the instructions on this page cover both the 2.x and 1.x drivers, expect to use the
2.x drivers exclusively for most ODBC applications connecting to Impala.
See the downloads page for a matrix of the certified driver version for different products. See the documentation page
for installation instructions.
Configuring the ODBC Port
Versions 2.5 and 2.0 of the Cloudera ODBC Connector, currently certified for some but not all BI applications, use the
HiveServer2 protocol, corresponding to Impala port 21050. Impala supports Kerberos authentication with all the
supported versions of the driver, and requires ODBC 2.05.13 for Impala or higher for LDAP username/password
authentication.
Version 1.x of the Cloudera ODBC Connector uses the original HiveServer1 protocol, corresponding to Impala port
21000.
Example of Setting Up an ODBC Application for Impala
To illustrate the outline of the setup process, here is a transcript of a session to set up all required drivers and a business
intelligence application that uses the ODBC driver, under Mac OS X. Each .dmg file runs a GUI-based installer, first for
204 | Cloudera Administration
Managing CDH and Managed Services
the underlying IODBC driver needed for non-Windows systems, then for the Cloudera ODBC Connector, and finally for
the BI tool itself.
$ ls -1
Cloudera-ODBC-Driver-for-Impala-Install-Guide.pdf
BI_Tool_Installer.dmg
iodbc-sdk-3.52.7-macosx-10.5.dmg
ClouderaImpalaODBC.dmg
$ open iodbc-sdk-3.52.7-macosx-10.dmg
Install the IODBC driver using its installer
$ open ClouderaImpalaODBC.dmg
Install the Cloudera ODBC Connector using its installer
$ installer_dir=$(pwd)
$ cd /opt/cloudera/impalaodbc
$ ls -1
Cloudera ODBC Driver for Impala Install Guide.pdf
Readme.txt
Setup
lib
ErrorMessages
Release Notes.txt
Tools
$ cd Setup
$ ls
odbc.ini
odbcinst.ini
$ cp odbc.ini ~/.odbc.ini
$ vi ~/.odbc.ini
$ cat ~/.odbc.ini
[ODBC]
# Specify any global ODBC configuration here such as ODBC tracing.
[ODBC Data Sources]
Sample Cloudera Impala DSN=Cloudera ODBC Driver for Impala
[Sample Cloudera Impala DSN]
# Description: DSN Description.
# This key is not necessary and is only to give a description of the data source.
Description=Cloudera ODBC Driver for Impala DSN
# Driver: The location where the ODBC driver is installed to.
Driver=/opt/cloudera/impalaodbc/lib/universal/libclouderaimpalaodbc.dylib
# The DriverUnicodeEncoding setting is only used for SimbaDM
# When set to 1, SimbaDM runs in UTF-16 mode.
# When set to 2, SimbaDM runs in UTF-8 mode.
#DriverUnicodeEncoding=2
# Values for HOST, PORT, KrbFQDN, and KrbServiceName should be set here.
# They can also be specified on the connection string.
HOST=hostname.sample.example.com
PORT=21050
Schema=default
# The authentication mechanism.
# 0 - No authentication (NOSASL)
# 1 - Kerberos authentication (SASL)
# 2 - Username authentication (SASL)
# 3 - Username/password authentication (SASL)
# 4 - Username/password authentication with SSL (SASL)
# 5 - No authentication with SSL (NOSASL)
# 6 - Username/password authentication (NOSASL)
AuthMech=0
# Kerberos related settings.
KrbFQDN=
KrbRealm=
KrbServiceName=
# Username/password authentication with SSL settings.
UID=
PWD
CAIssuedCertNamesMismatch=1
Cloudera Administration | 205
Managing CDH and Managed Services
TrustedCerts=/opt/cloudera/impalaodbc/lib/universal/cacerts.pem
# Specify the proxy user ID to use.
#DelegationUID=
# General settings
TSaslTransportBufSize=1000
RowsFetchedPerBlock=10000
SocketTimeout=0
StringColumnLength=32767
UseNativeQuery=0
$ pwd
/opt/cloudera/impalaodbc/Setup
$ cd $installer_dir
$ open BI_Tool_Installer.dmg
Install the BI tool using its installer
$ ls /Applications | grep BI_Tool
BI_Tool.app
$ open -a BI_Tool.app
In the BI tool, connect to a data source using port 21050
Notes about JDBC and ODBC Interaction with Impala SQL Features
Most Impala SQL features work equivalently through the impala-shell interpreter of the JDBC or ODBC APIs. The
following are some exceptions to keep in mind when switching between the interactive shell and applications using
the APIs:
Note: If your JDBC or ODBC application connects to Impala through a load balancer such as haproxy,
be cautious about reusing the connections. If the load balancer has set up connection timeout values,
either check the connection frequently so that it never sits idle longer than the load balancer timeout
value, or check the connection validity before using it and create a new one if the connection has
been closed.
• The Impala complex types (STRUCT, ARRAY, or MAP) are available in CDH 5.5 / Impala 2.3 and higher. To use these
types with JDBC requires version 2.5.28 or higher of the Cloudera JDBC Connector for Impala. To use these types
with ODBC requires version 2.5.30 or higher of the Cloudera ODBC Connector for Impala. Consider upgrading all
JDBC and ODBC drivers at the same time you upgrade from CDH 5.5 or higher.
• Although the result sets from queries involving complex types consist of all scalar values, the queries involve join
notation and column references that might not be understood by a particular JDBC or ODBC connector. Consider
defining a view that represents the flattened version of a table containing complex type columns, and pointing
the JDBC or ODBC application at the view. See Complex Types (CDH 5.5 or higher only) for details.
Configuring Impala to Work with JDBC
Impala supports the standard JDBC interface, allowing access from commercial Business Intelligence tools and custom
software written in Java or other programming languages. The JDBC driver allows you to access Impala from a Java
program that you write, or a Business Intelligence or similar tool that uses JDBC to communicate with various database
products.
Setting up a JDBC connection to Impala involves the following steps:
• Verifying the communication port where the Impala daemons in your cluster are listening for incoming JDBC
requests.
• Installing the JDBC driver on every system that runs the JDBC-enabled application.
• Specifying a connection string for the JDBC application to access one of the servers running the impalad daemon,
with the appropriate security settings.
Configuring the JDBC Port
The default port used by JDBC 2.0 and later (as well as ODBC 2.x) is 21050. Impala server accepts JDBC connections
through this same port 21050 by default. Make sure this port is available for communication with other hosts on your
network, for example, that it is not blocked by firewall software. If your JDBC client software connects to a different
206 | Cloudera Administration
Managing CDH and Managed Services
port, specify that alternative port number with the --hs2_port option when starting impalad. See Starting Impala
for details about Impala startup options. See Ports Used by Impala for information about all ports used for communication
between Impala and clients or between Impala components.
Choosing the JDBC Driver
In Impala 2.0 and later, you have the choice between the Cloudera JDBC Connector and the Hive 0.13 JDBC driver.
Cloudera recommends using the Cloudera JDBC Connector where practical.
If you are already using JDBC applications with an earlier Impala release, you must update your JDBC driver to one of
these choices, because the Hive 0.12 driver that was formerly the only choice is not compatible with Impala 2.0 and
later.
Both the Cloudera JDBC 2.5 Connector and the Hive JDBC driver provide a substantial speed increase for JDBC applications
with Impala 2.0 and higher, for queries that return large result sets.
Complex type considerations:
The Impala complex types (STRUCT, ARRAY, or MAP) are available in CDH 5.5 / Impala 2.3 and higher. To use these
types with JDBC requires version 2.5.28 or higher of the Cloudera JDBC Connector for Impala. To use these types with
ODBC requires version 2.5.30 or higher of the Cloudera ODBC Connector for Impala. Consider upgrading all JDBC and
ODBC drivers at the same time you upgrade from CDH 5.5 or higher.
Although the result sets from queries involving complex types consist of all scalar values, the queries involve join
notation and column references that might not be understood by a particular JDBC or ODBC connector. Consider
defining a view that represents the flattened version of a table containing complex type columns, and pointing the
JDBC or ODBC application at the view. See Complex Types (CDH 5.5 or higher only) for details.
Enabling Impala JDBC Support on Client Systems
Using the Cloudera JDBC Connector (recommended)
You download and install the Cloudera JDBC 2.5 connector on any Linux, Windows, or Mac system where you intend
to run JDBC-enabled applications. From the Cloudera Connectors download page, you choose the appropriate protocol
(JDBC or ODBC) and target product (Impala or Hive). The ease of downloading and installing on non-CDH systems makes
this connector a convenient choice for organizations with heterogeneous environments.
Using the Hive JDBC Driver
You install the Hive JDBC driver (hive-jdbc package) through the Linux package manager, on hosts within the CDH
cluster. The driver consists of several Java JAR files. The same driver can be used by Impala and Hive.
To get the JAR files, install the Hive JDBC driver on each CDH-enabled host in the cluster that will run JDBC applications.
Follow the instructions for CDH 5.
Note: The latest JDBC driver, corresponding to Hive 0.13, provides substantial performance
improvements for Impala queries that return large result sets. Impala 2.0 and later are compatible
with the Hive 0.13 driver. If you already have an older JDBC driver installed, and are running Impala
2.0 or higher, consider upgrading to the latest Hive JDBC driver for best performance with JDBC
applications.
If you are using JDBC-enabled applications on hosts outside the CDH cluster, you cannot use the CDH install procedure
on the non-CDH hosts. Install the JDBC driver on at least one CDH host using the preceding procedure. Then download
the JAR files to each client machine that will use JDBC with Impala:
commons-logging-X.X.X.jar
hadoop-common.jar
hive-common-X.XX.X-cdhX.X.X.jar
hive-jdbc-X.XX.X-cdhX.X.X.jar
hive-metastore-X.XX.X-cdhX.X.X.jar
hive-service-X.XX.X-cdhX.X.X.jar
httpclient-X.X.X.jar
Cloudera Administration | 207
Managing CDH and Managed Services
httpcore-X.X.X.jar
libfb303-X.X.X.jar
libthrift-X.X.X.jar
log4j-X.X.XX.jar
slf4j-api-X.X.X.jar
slf4j-logXjXX-X.X.X.jar
To enable JDBC support for Impala on the system where you run the JDBC application:
1. Download the JAR files listed above to each client machine.
Note: For Maven users, see this sample github page for an example of the dependencies you
could add to a pom file instead of downloading the individual JARs.
2. Store the JAR files in a location of your choosing, ideally a directory already referenced in your CLASSPATH setting.
For example:
• On Linux, you might use a location such as /opt/jars/.
• On Windows, you might use a subdirectory underneath C:\Program Files.
3. To successfully load the Impala JDBC driver, client programs must be able to locate the associated JAR files. This
often means setting the CLASSPATH for the client process to include the JARs. Consult the documentation for
your JDBC client for more details on how to install new JDBC drivers, but some examples of how to set CLASSPATH
variables include:
• On Linux, if you extracted the JARs to /opt/jars/, you might issue the following command to prepend the
JAR files path to an existing classpath:
export CLASSPATH=/opt/jars/*.jar:$CLASSPATH
• On Windows, use the System Properties control panel item to modify the Environment Variables for your
system. Modify the environment variables to include the path to which you extracted the files.
Note: If the existing CLASSPATH on your client machine refers to some older version of the
Hive JARs, ensure that the new JARs are the first ones listed. Either put the new JAR files
earlier in the listings, or delete the other references to Hive JAR files.
Establishing JDBC Connections
The JDBC driver class depends on which driver you select.
Note: If your JDBC or ODBC application connects to Impala through a load balancer such as haproxy,
be cautious about reusing the connections. If the load balancer has set up connection timeout values,
either check the connection frequently so that it never sits idle longer than the load balancer timeout
value, or check the connection validity before using it and create a new one if the connection has
been closed.
Using the Cloudera JDBC Connector (recommended)
Depending on the level of the JDBC API your application is targeting, you can use the following fully-qualified class
names (FQCNs):
• com.cloudera.impala.jdbc41.Driver
• com.cloudera.impala.jdbc41.DataSource
• com.cloudera.impala.jdbc4.Driver
208 | Cloudera Administration
Managing CDH and Managed Services
• com.cloudera.impala.jdbc4.DataSource
• com.cloudera.impala.jdbc3.Driver
• com.cloudera.impala.jdbc3.DataSource
The connection string has the following format:
jdbc:impala://Host:Port[/Schema];Property1=Value;Property2=Value;...
The port value is typically 21050 for Impala.
For full details about the classes and the connection string (especially the property values available for the connection
string), download the appropriate driver documentation for your platform from the Impala JDBC Connector download
page.
Using the Hive JDBC Driver
For example, with the Hive JDBC driver, the class name is org.apache.hive.jdbc.HiveDriver. Once you have
configured Impala to work with JDBC, you can establish connections between the two. To do so for a cluster that does
not use Kerberos authentication, use a connection string of the form jdbc:hive2://host:port/;auth=noSasl.
For example, you might use:
jdbc:hive2://myhost.example.com:21050/;auth=noSasl
To connect to an instance of Impala that requires Kerberos authentication, use a connection string of the form
jdbc:hive2://host:port/;principal=principal_name. The principal must be the same user principal you
used when starting Impala. For example, you might use:
jdbc:hive2://myhost.example.com:21050/;principal=impala/myhost.example.com@H2.EXAMPLE.COM
To connect to an instance of Impala that requires LDAP authentication, use a connection string of the form
jdbc:hive2://host:port/db_name;user=ldap_userid;password=ldap_password. For example, you might
use:
jdbc:hive2://myhost.example.com:21050/test_db;user=fred;password=xyz123
Note:
Currently, the Hive JDBC driver does not support connections that use both Kerberos authentication
and SSL encryption. To use both of these security features with Impala through a JDBC application,
use the Cloudera JDBC Connector as the JDBC driver.
Notes about JDBC and ODBC Interaction with Impala SQL Features
Most Impala SQL features work equivalently through the impala-shell interpreter of the JDBC or ODBC APIs. The
following are some exceptions to keep in mind when switching between the interactive shell and applications using
the APIs:
• Complex type considerations:
– Queries involving the complex types (ARRAY, STRUCT, and MAP) require notation that might not be available
in all levels of JDBC and ODBC drivers. If you have trouble querying such a table due to the driver level or
inability to edit the queries used by the application, you can create a view that exposes a “flattened” version
of the complex columns and point the application at the view. See Complex Types (CDH 5.5 or higher only)
for details.
– The complex types available in CDH 5.5 / Impala 2.3 and higher are supported by the JDBC getColumns()
API. Both MAP and ARRAY are reported as the JDBC SQL Type ARRAY, because this is the closest matching Java
SQL type. This behavior is consistent with Hive. STRUCT types are reported as the JDBC SQL Type STRUCT.
Cloudera Administration | 209
Managing CDH and Managed Services
To be consistent with Hive's behavior, the TYPE_NAME field is populated with the primitive type name for
scalar types, and with the full toSql() for complex types. The resulting type names are somewhat inconsistent,
because nested types are printed differently than top-level types. For example, the following list shows how
toSQL() for Impala types are translated to TYPE_NAME values:
DECIMAL(10,10)
CHAR(10)
VARCHAR(10)
ARRAY<DECIMAL(10,10)>
ARRAY<CHAR(10)>
ARRAY<VARCHAR(10)>
becomes
becomes
becomes
becomes
becomes
becomes
DECIMAL
CHAR
VARCHAR
ARRAY<DECIMAL(10,10)>
ARRAY<CHAR(10)>
ARRAY<VARCHAR(10)>
Managing Key-Value Store Indexer
The Key-Value Store Indexer service uses the Lily HBase Indexer Service to index the stream of records being added to
HBase tables. Indexing allows you to query data stored in HBase with the Solr service.
The Key-Value Store Indexer service is installed in the same parcel or package along with the CDH 5 or Solr service.
Adding the Key-Value Store Indexer Service
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
1. On the Home > Status tab, click
to the right of the cluster name and select Add a Service. A list of service types display. You can add one type of
service at a time.
2. Select the Key-Value Store Indexer service and click Continue.
3. Select the radio button next to the services on which the new service should depend. All services must depend
on the same ZooKeeper service. Click Continue.
4. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of the
hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of hosts to
which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple hosts,
you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts dialog.
The following shortcuts for specifying hostname patterns are supported:
• Range of hostnames (without the domain portion)
Range Definition
Matching Hosts
10.1.1.[1-4]
10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4
host[1-3].company.com
host1.company.com, host2.company.com, host3.company.com
host[07-10].company.com
host07.company.com, host08.company.com, host09.company.com,
host10.company.com
• IP addresses
• Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
5. Click Continue.
6. Review the configuration changes to be applied. Confirm the settings entered for file system paths. The file paths
required vary based on the services to be installed. If you chose to add the Sqoop service, indicate whether to use
the default Derby database or the embedded PostgreSQL database. If the latter, type the database name, host,
and user credentials that you specified when you created the database.
210 | Cloudera Administration
Managing CDH and Managed Services
Warning: Do not place DataNode data directories on NAS devices. When resizing an NAS, block
replicas can be deleted, which will result in reports of missing blocks.
Click Continue. The wizard starts the services.
7. Click Continue.
8. Click Finish.
Enabling Morphlines with Search and HBase Indexing
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Cloudera Morphlines is an open source framework that reduces the time and skills necessary to build or change Search
indexing applications. A morphline is a rich configuration file that simplifies defining an ETL transformation chain.
1.
2.
3.
4.
5.
Go to the Indexer service.
Click the Configuration tab.
Select Scope > All.
Select Category > Morphlines.
Create the necessary configuration files, and modify the content in the following properties:
• Morphlines File — Text that goes into the morphlines.conf used by HBase indexers. You should use
$ZK_HOST in this file instead of specifying a ZooKeeper quorum. Cloudera Manager automatically replaces
the $ZK_HOST variable with the correct value during the Solr configuration deployment.
• Custom MIME-types File — Text that goes verbatim into the custom-mimetypes.xml file used by HBase
Indexers with the detectMimeTypes command. See the Cloudera Morphlines Reference Guide for details
on this command.
• Grok Dictionary File — Text that goes verbatim into the grok-dictionary.conf file used by HBase Indexers
with the grok command. See the Cloudera Morphlines Reference Guide for details of this command.
See Extracting, Transforming, and Loading Data With Cloudera Morphlines for information about using morphlines
with Search and HBase.
Managing MapReduce and YARN
CDH supports two versions of the MapReduce computation framework: MRv1 and MRv2, which are implemented by
the MapReduce (MRv1) and YARN (MRv2) services. YARN is backwards-compatible with MapReduce. (All jobs that run
against MapReduce will also run in a YARN cluster).
The MRv2 YARN architecture splits the two primary responsibilities of the JobTracker — resource management and
job scheduling/monitoring — into separate daemons: a global ResourceManager (RM) and per-application
ApplicationMasters (AM). With MRv2, the ResourceManager (RM) and per-node NodeManagers (NM) form the
data-computation framework. The ResourceManager service effectively replaces the functions of the JobTracker, and
NodeManagers run on worker hosts instead of TaskTracker daemons. The per-application ApplicationMaster is, in
effect, a framework-specific library and negotiates resources from the ResourceManager and works with the
NodeManagers to execute and monitor the tasks. For details of this architecture, see Apache Hadoop NextGen
MapReduce (YARN).
• The Cloudera Manager Admin Console has different methods for displaying MapReduce and YARN job history.
See Monitoring MapReduce Jobs and Monitoring YARN Applications.
• For information on configuring the MapReduce and YARN services for high availability, see MapReduce (MRv1)
and YARN (MRv2) High Availability on page 315
• For information on configuring MapReduce and YARN resource management features, see Resource Management
on page 236.
Cloudera Administration | 211
Managing CDH and Managed Services
Defaults and Recommendations
• In a Cloudera Manager deployment of a CDH 4 cluster, the MapReduce service is the default MapReduce
computation framework.You can create a YARN service in a CDH 4 cluster, but it is not considered production
ready.
• In a Cloudera Manager deployment of a CDH 5 cluster, the YARN service is the default MapReduce computation
framework.In CDH 5, the MapReduce service has been deprecated. However, the MapReduce service is fully
supported for backward compatibility through the CDH 5 lifecycle.
• For production uses, Cloudera recommends that only one MapReduce framework should be running at any given
time. If development needs or other use case requires switching between MapReduce and YARN, both services
can be configured at the same time, but only one should be running (to fully optimize the hardware resources
available).
Migrating from MapReduce to YARN
Cloudera Manager provides a wizard described in Importing MapReduce Configurations to YARN on page 217 to easily
migrate MapReduce configurations to YARN. The wizard performs all the steps (Switching Between MapReduce and
YARN Services on page 213, Updating Dependent Services on page 213, and Configuring Alternatives Priority on page
213) on this page.
The Activity Monitor role collects information about activities run by the MapReduce service. If MapReduce is not
being used and the reporting data is no longer required, then the Activity Monitor role and database can be removed:
1. Do one of the following:
• Select Clusters > Cloudera Management Service > Cloudera Management Service.
• On the Status tab of the Home > Status tab, in Cloudera Management Service table, click the Cloudera
Management Service link.
2.
3.
4.
5.
Click the Instances tab.
Select checkbox for Activity Monitor, select Actions for Selected > Stop, and click Stop to confirm.
Select checkbox for Activity Monitor, select Actions for Selected > Delete, and click Delete to confirm.
Manage the Activity Monitor database. The example below is for a MySQL backend database:
a. Verify the Activity Monitor database:
mysql> show databases;
+--------------------+
| Database
|
+--------------------+
| amon
|
+--------------------+
b. Back up the database:
$ mysqldump -uroot -pcloudera amon > /safe_backup_directory/amon.sql
Drop the database:
mysql> drop database amon;
Once you have migrated to YARN and deleted the MapReduce service, you can remove local data from each TaskTracker
node. The mapred.local.dir parameter is a directory on the local filesystem of each TaskTracker that contains
temporary data for MapReduce. Once the service is stopped, you can remove this directory to free disk space on each
node.
For detailed information on migrating from MapReduce to YARN, see Migrating from MapReduce 1 (MRv1) to MapReduce
2 (MRv2, YARN).
212 | Cloudera Administration
Managing CDH and Managed Services
Switching Between MapReduce and YARN Services
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
MapReduce and YARN use separate sets of configuration files. No files are removed or altered when you change to a
different framework. To change from YARN to MapReduce (or vice versa):
1.
2.
3.
4.
5.
6.
7.
(Optional) Configure the new MapReduce or YARN service.
Update dependent services to use the chosen framework.
Configure the alternatives priority.
Redeploy the Oozie ShareLib.
Redeploy the client configuration.
Start the framework service to switch to.
(Optional) Stop the unused framework service to free up the resources it uses.
Updating Dependent Services
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
When you change the MapReduce framework, the dependent services that must be updated to use the new framework
are:
• Hive
• Sqoop 2
• Oozie
To update a service:
1.
2.
3.
4.
5.
6.
7.
Go to the service.
Click the Configuration tab.
Select Scope > service name(Service Wide).
Select Scope > All.
Locate the MapReduce Service property and select the YARN or MapReduce service.
Click Save Changes to commit the changes.
Select Actions > Restart.
The Hue service is automatically reconfigured to use the same framework as Oozie and Hive. This cannot be changed.
To update the Hue service:
1. Go to the Hue service.
2. Select Actions > Restart.
Configuring Alternatives Priority
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
The alternatives priority property determines which service—MapReduce or YARN—is used by clients to run MapReduce
jobs. The service with a higher value of the property is used. In CDH 4, the MapReduce service alternatives priority is
set to 92 and the YARN service is set to 91. In CDH 5, the values are reversed; the MapReduce service alternatives
priority is set to 91 and the YARN service is set to 92.
To configure the alternatives priority:
1.
2.
3.
4.
5.
6.
7.
8.
Go to the MapReduce or YARN service.
Click the Configuration tab.
Select Scope > Gateway Default Group.
Select Category > All.
Type Alternatives in Search box.
In the Alternatives Priority property, set the priority value.
Click Save Changes to commit the changes.
Redeploy the client configuration.
Cloudera Administration | 213
Managing CDH and Managed Services
Managing MapReduce
For an overview of computation frameworks, insight into their usage and restrictions, and examples of common tasks
they perform, see Managing MapReduce and YARN on page 211.
Configuring the MapReduce Scheduler
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
The MapReduce service is configured by default to use the FairScheduler. You can change the scheduler type to FIFO
or Capacity Scheduler. You can also modify the Fair Scheduler and Capacity Scheduler configuration. For further
information on schedulers, see Schedulers on page 236.
Configuring the Task Scheduler Type
1.
2.
3.
4.
5.
Go to the MapReduce service.
Click the Configuration tab.
Select Scope > JobTracker.
Select Category > Classes .
In the Task Scheduler property, select a scheduler.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
6. Click Save Changes to commit the changes.
7. Restart the JobTracker to apply the new configuration:
a. Click the Instances tab.
b. Click the JobTracker role.
c. Select Actions for Selected > Restart.
Modifying the Scheduler Configuration
1.
2.
3.
4.
5.
Go to the MapReduce service.
Click the Configuration tab.
Select Scope > JobTracker.
Select Category > Jobs.
Modify the configuration properties.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
6. Click Save Changes to commit the changes.
7. Restart the JobTracker to apply the new configuration:
a. Click the Instances tab.
b. Click the JobTracker role.
c. Select Actions for Selected > Restart.
Configuring the MapReduce Service to Save Job History
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Normally job history is saved on the host on which the JobTracker is running. You can configure JobTracker to write
information about every job that completes to a specified HDFS location. By default, the information is retained for 7
days.
Enabling Map Reduce Job History To Be Saved to HDFS
1. Create a folder in HDFS to contain the history information. When creating the folder, set the owner and group to
mapred:hadoop with permission setting 775.
214 | Cloudera Administration
Managing CDH and Managed Services
2.
3.
4.
5.
6.
Go to the MapReduce service.
Click the Configuration tab.
Select Scope > JobTracker.
Select Category > Paths.
Set the Completed Job History Location property to the location that you created in step 1.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
7. Click Save Changes.
8. Restart the MapReduce service.
Setting the Job History Retention Duration
1. Select the JobTracker Default Group category.
2. Set the Job History Files Maximum Age property (mapreduce.jobhistory.max-age-ms) to the length of time
(in milliseconds, seconds, minutes, or hours) that you want job history files to be kept.
3. Restart the MapReduce service.
The Job History Files Cleaner runs at regular intervals to check for job history files that are ready to be deleted. By
default, the interval is 24 hours. To change the frequency with which the Job History Files Cleaner runs:
1. Select the JobTracker Default Group category.
2. Set the Job History Files Cleaner Interval property (mapreduce.jobhistory.cleaner.interval) to the
desired frequency (in milliseconds, seconds, minutes, or hours).
3. Restart the MapReduce service.
Configuring Client Overrides
A configuration property qualified with (Client Override) is a server-side setting that ignores any value a client tries to
set for that property. It performs the same role as its unqualified counterpart, and applies the configuration to the
service with the setting <final>true</final>.
For example, if you set the Map task heap property to 1 GB in the job configuration code, but the service's heap property
qualified with (Client Override) is set to 500 MB, then 500 MB is applied.
Managing YARN
For an overview of computation frameworks, insight into their usage and restrictions, and examples of common tasks
they perform, see Managing MapReduce and YARN on page 211.
Adding the YARN Service
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
1. On the Home > Status tab, click
to the right of the cluster name and select Add a Service. A list of service types display. You can add one type of
service at a time.
2. Click the YARN (MR2 Included) radio button and click Continue.
3. Select the radio button next to the services on which the new service should depend. All services must depend
on the same ZooKeeper service. Click Continue.
4. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of the
hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of hosts to
which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple hosts,
you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts dialog.
The following shortcuts for specifying hostname patterns are supported:
Cloudera Administration | 215
Managing CDH and Managed Services
• Range of hostnames (without the domain portion)
Range Definition
Matching Hosts
10.1.1.[1-4]
10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4
host[1-3].company.com
host1.company.com, host2.company.com, host3.company.com
host[07-10].company.com
host07.company.com, host08.company.com, host09.company.com,
host10.company.com
• IP addresses
• Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
Configuring Memory Settings for YARN and MRv2
The memory configuration for YARN and MRv2 memory is important to get the best performance from your cluster.
Several different settings are involved. The table below shows the default settings, as well as the settings that Cloudera
recommends, for each configuration option. See Managing MapReduce and YARN on page 211 for more configuration
specifics; and, for detailed tuning advice with sample configurations, see Tuning YARN on page 281.
Table 4: YARN and MRv2 Memory Configuration
Cloudera Manager Property CDH Property Name
Name
Default Configuration
Cloudera Tuning Guidelines
Container Memory
Minimum
yarn.scheduler.
1 GB
minimum-allocation-mb
0
Container Memory
Maximum
yarn.scheduler.
64 GB
maximum-allocation-mb
amount of memory on
largest node
Container Memory
Increment
yarn.scheduler.
512 MB
increment-allocation-mb
Use a fairly large value, such
as 128 MB
Container Memory
yarn.nodemanager.
resource.memory-mb
8 GB
8 GB
Map Task Memory
mapreduce.map.memory.mb 1 GB
1 GB
Reduce Task Memory
mapreduce.reduce.memory.mb 1 GB
1 GB
Map Task Java Opts Base
mapreduce.map.java.opts -Djava.net.preferIPv4Stack=true -Djava.net.preferIPv4Stack=true
-Xmx768m
Reduce Task Java Opts Base mapreduce.reduce.java.opts -Djava.net.preferIPv4Stack=true -Djava.net.preferIPv4Stack=true
-Xmx768m
ApplicationMaster Memory yarn.app.mapreduce.
1 GB
ApplicationMaster Java Opts yarn.app.mapreduce.
am.command-opts
Base
-Djava.net.preferIPv4Stack=true -Djava.net.preferIPv4Stack=true
-Xmx768m
am.resource.mb
1 GB
Configuring Directories
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
Creating the Job History Directory
When adding the YARN service, the Add Service wizard automatically creates a job history directory. If you quit the
Add Service wizard or it does not finish, you can create the directory outside the wizard:
216 | Cloudera Administration
Managing CDH and Managed Services
1. Go to the YARN service.
2. Select Actions > Create Job History Dir.
3. Click Create Job History Dir again to confirm.
Creating the NodeManager Remote Application Log Directory
When adding the YARN service, the Add Service wizard automatically creates a remote application log directory. If you
quit the Add Service wizard or it does not finish, you can create the directory outside the wizard:
1. Go to the YARN service.
2. Select Actions > Create NodeManager Remote Application Log Directory.
3. Click Create NodeManager Remote Application Log Directory again to confirm.
Importing MapReduce Configurations to YARN
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
Warning: In addition to importing configuration settings, the import process:
• Configures services to use YARN as the MapReduce computation framework instead of MapReduce.
• Overwrites existing YARN configuration and role assignments.
When you upgrade from CDH 4 to CDH 5, you can import MapReduce configurations to YARN as part of the upgrade
wizard. If you do not import configurations during upgrade, you can manually import the configurations at a later time:
1. Go to the YARN service page.
2. Stop the YARN service.
3. Select Actions > Import MapReduce Configuration. The import wizard presents a warning letting you know that
it will import your configuration, restart the YARN service and its dependent services, and update the client
configuration.
4. Click Continue to proceed. The next page indicates some additional configuration required by YARN.
5. Verify or modify the configurations and click Continue. The Switch Cluster to MR2 step proceeds.
6. When all steps have been completed, click Finish.
7. (Optional) Remove the MapReduce service.
a. Click the Cloudera Manager logo to return to the Home page.
b. In the MapReduce row, right-click
and select Delete. Click Delete to confirm.
8. Recompile JARs used in MapReduce applications. For further information, see For MapReduce Programmers:
Writing and Running Jobs.
Configuring the YARN Scheduler
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
The YARN service is configured by default to use the FairScheduler. You can change the scheduler type to FIFO or
Capacity Scheduler. You can also modify the Fair Scheduler and Capacity Scheduler configuration. For further information
on schedulers, see Schedulers on page 236.
Configuring the Scheduler Type
1.
2.
3.
4.
5.
Go to the YARN service.
Click the Configuration tab.
Select Scope > ResourceManager.
Select Category > Main.
Select a scheduler class.
Cloudera Administration | 217
Managing CDH and Managed Services
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
6. Click Save Changes to commit the changes.
7. Restart the YARN service.
Modifying the Scheduler Configuration
1.
2.
3.
4.
5.
6.
Go to the YARN service.
Click the Configuration tab.
Click the ResourceManager Default Group category.
Select Scope > ResourceManager.
Type Scheduler in the Search box.
Locate a property and modify the configuration.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
7. Click Save Changes to commit the changes.
8. Restart the YARN service.
Dynamic Resource Management
In addition to the static resource management available to all services, the YARN service also supports dynamic
management of its static allocation. See Dynamic Resource Pools on page 243.
Configuring YARN for Long-running Applications
On a secure cluster, long-running applications such as Spark Streaming jobs will need additional configuration since
the default settings only allow the hdfs user's delegation tokens a maximum lifetime of 7 days, which is not always
sufficient. For instructions on how to work around this issue, see Configuring Spark on YARN for Long-Running
Applications.
Task Process Exit Codes
All YARN tasks on the NodeManager are executed in a JVM. When a task executes successfully, the exit code is 0. Exit
codes of 0 are not logged, as they are the expected result. Any non-zero exit code is logged as an error. The non-zero
exit code is reported by the NodeManager as an error in the child process. The NodeManager itself is not affected by
the error.
There are multiple reasons why the task JVM might exit with a non-zero code, though there is no exhaustive list. Exit
codes can be split into two categories:
1. Set by the JVM based on the OS signal received by the JVM
2. Directly set in the code
Signal-related exit codes When the OS sends a signal to the JVM, the JVM handles the signal, which could cause the
JVM to exit. Not all signals cause the JVM to exit. Exit codes for OS signals have a value between 128 and 160. Logs
show non-zero status codes without further explanation.
Two exit values that typically do not require investigation are 137 and 143. These values are logged when the JVM is
killed by the NodeManager or the OS. The NodeManager might kill a JVM due to task pre-emption (if that is configured)
or speculative execution. The OS might kill the JVM when the JVM exceeds system limits like CPU time. You should
investigate these codes if they appear frequently, as they might indicate a misconfiguration or a structural problem
with regard to resources.
Exit code 154 is used in RecoveredContainerLaunch#call to indicate containers that were lost between
NodeManager restarts without an exit code being recorded. This is usually a bug, and requires investigation.
Other exit codes The JVM might exit if there is an unrecoverable error while executing a task. The exit code and the
message logged should provide more detail. A Java stack trace might also be logged as part of the exit. These exits
should be investigated further to discover a root cause.
218 | Cloudera Administration
Managing CDH and Managed Services
In the case of a streaming MapReduce job, the exit code of the JVM is the same as the mapper or reducer in use. The
mapper or reducer can be a shell script or Python script. This means that the underlying script dictates the exit code:
in streaming jobs, you should take this into account during your investigation.
Managing Oozie
This section describes tasks for managing Oozie.
Configuring Oozie for High Availability
In CDH 5, you can configure multiple active Oozie servers against the same database. Oozie high availability is
“active-active” or “hot-hot” so that both Oozie servers are active at the same time, with no failover. High availability
for Oozie is supported in both MRv1 and MRv2 (YARN).
Requirements
The requirements for Oozie high availability are:
• Multiple active Oozie servers, preferably identically configured.
• JDBC JAR in the same location across all Oozie hosts (for example, /var/lib/oozie/).
• External database that supports multiple concurrent connections, preferably with HA support. The default Derby
database does not support multiple concurrent connections.
• ZooKeeper ensemble with distributed locks to control database access, and service discovery for log aggregation.
• Load balancer (preferably with HA support, for example HAProxy), virtual IP, or round-robin DNS to provide a
single entry point (of the multiple active servers), and for callbacks from the Application Master or JobTracker.
For information on setting up TLS/SSL communication with Oozie HA enabled, see Additional Considerations when
Configuring TLS/SSL for Oozie HA.
Configuring Oozie High Availability Using Cloudera Manager
Minimum Required Role: Full Administrator
Important: Enabling or disabling high availability makes the previous monitoring history unavailable.
Enabling Oozie High Availability
1. Ensure that the requirements are satisfied.
2. In the Cloudera Manager Admin Console, go to the Oozie service.
3. Select Actions > Enable High Availability to see eligible Oozie server hosts. The host running the current Oozie
server is not eligible.
4. Select the host on which to install an additional Oozie server and click Continue.
5. Specify the host and port of the Oozie load balancer, and click Continue. Cloudera Manager stops the Oozie servers,
adds another Oozie server, initializes the Oozie server High Availability state in ZooKeeper, configures Hue to
reference the Oozie load balancer, and restarts the Oozie servers and dependent services.
Disabling Oozie High Availability
1. In the Cloudera Manager Admin Console, go to the Oozie service.
2. Select Actions > Disable High Availability to see all hosts currently running Oozie servers.
3. Select the one host to run the Oozie server and click Continue. Cloudera Manager stops the Oozie service, removes
the additional Oozie servers, configures Hue to reference the Oozie service, and restarts the Oozie service and
dependent services.
Configuring Oozie High Availability Using the Command Line
For installation and configuration instructions for configuring Oozie HA using the command line, see
https://archive.cloudera.com/cdh5/cdh/5/oozie.
Cloudera Administration | 219
Managing CDH and Managed Services
Adding the Oozie Service Using Cloudera Manager
The Oozie service can be automatically installed and started during your installation of CDH with Cloudera Manager.
You can also install Oozie manually with the Add Service wizard in Cloudera Manager. The wizard configures and starts
Oozie and its dependent services. See Adding a Service on page 36 for instructions.
Note: If your instance of Cloudera Manager uses an external database, you must also configure Oozie
with an external database. See Configuring an External Database for Oozie.
Redeploying the Oozie ShareLib
Some Oozie actions – specifically DistCp, Streaming, Pig, Sqoop, and Hive – require external JAR files in order to run.
Instead of having to keep these JAR files in each workflow's lib folder, or forcing you to manually manage them using
the oozie.libpath property on every workflow using one of these actions, Oozie provides the ShareLib. The ShareLib
behaves very similarly to oozie.libpath, except that it’s specific to the aforementioned actions and their required
JARs.
Redeploying the Oozie ShareLib Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
When you upgrade CDH or switch between MapReduce and YARN computation frameworks, redeploy the Oozie
ShareLib as follows:
1.
2.
3.
4.
Go to the Oozie service.
Select Actions > Stop.
Select Actions > Install Oozie ShareLib.
Select Actions > Start.
Redeploying the Oozie ShareLib Using the Command Line
See Installing the Oozie Shared Library in Hadoop HDFS.
Configuring Oozie Data Purge Settings Using Cloudera Manager
All Oozie workflows older than 30 days are purged from the database by default. However, actions associated with
long-running coordinators do not purge until the coordinators complete. If, for example, you schedule a coordinator
to run for a year, all those actions remain in the database for the year.
You can change your Oozie configuration to control when data is purged to improve performance, reduce database
disk usage, or keep the history for a longer period of time. Limiting the size of the Oozie database can also improve
performance during upgrades.
1.
2.
3.
4.
In the Cloudera Manager Admin Console, go to the Oozie service.
Click the Configuration tab.
Type purge in the Search box.
Set the following properties as required for your environment:
• Enable Purge for Long-Running Coordinator Jobs
Select this property to enable purging of long-running coordinator jobs for which the workflow jobs are older
than the value you set for the Days to Keep Completed Workflow Jobs property.
• Days to Keep Completed Workflow Jobs
• Days to Keep Completed Coordinator Jobs
• Days to Keep Completed Bundle Jobs
5. Click Save Changes to commit the changes.
6. Select Actions > Restart to restart the Oozie Service.
220 | Cloudera Administration
Managing CDH and Managed Services
Adding Schema to Oozie Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
This page explains how to manually add a schema (official or custom) with Cloudera Manager.
Cloudera Manager 5 automatically configures Oozie with all available official schemas, and corresponding tables, per
the latest CDH 5.x release. For all versions of CDH 4.x, Cloudera Manager configures Oozie with the CDH 4.0.0 schema,
even if you are using a higher version of CDH 4.x.
1.
2.
3.
4.
5.
In the Cloudera Manager Admin Console, go to the Oozie service.
Click the Configuration tab.
Select Scope > Oozie Server.
Select Category > Advanced.
Locate the Oozie SchemaService Workflow Extension Schemas property or search for it by typing its name in the
Search box.
6. Enter the desired schema from Table 5: Oozie Schema - CDH 5 on page 221 or Table 6: Oozie Schema - CDH 4 on
page 222, appending .xsd to each entry.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
7. Click Save Changes to commit the changes.
8. Restart the Oozie service.
Note: Releases are only included in the following tables if a schema was added or removed. If a
release is not in the table, it has the same set of schemas as the previous release that is in the table.
Table 5: Oozie Schema - CDH 5
CDH 5.5.0
distcp
CDH 5.4.0
CDH 5.2.0
CDH 5.1.0
CDH 5.0.1
CDH 5.0.0
distcp-action-0.1 distcp-action-0.1 distcp-action-0.1 distcp-action-0.1 distcp-action-0.1 distcp-action-0.1
distcp-action-0.2 distcp-action-0.2 distcp-action-0.2 distcp-action-0.2 distcp-action-0.2 distcp-action-0.2
email
email-action-0.1 email-action-0.1 email-action-0.1 email-action-0.1 email-action-0.1 email-action-0.1
email-action-0.2 email-action-0.2 email-action-0.2 email-action-0.2
hive
hive-action-0.2 hive-action-0.2 hive-action-0.2 hive-action-0.2 hive-action-0.2 hive-action-0.2
hive-action-0.3 hive-action-0.3 hive-action-0.3 hive-action-0.3 hive-action-0.3 hive-action-0.3
hive-action-0.4 hive-action-0.4 hive-action-0.4 hive-action-0.4 hive-action-0.4 hive-action-0.4
hive-action-0.5 hive-action-0.5 hive-action-0.5 hive-action-0.5 hive-action-0.5 hive-action-0.5
hive-action-0.6
HiveServer2
hive2-action-0.1 hive2-action-0.1 hive2-action-0.1
hive2-action-0.2
oozie-bundle
hive2-action-0.1 hive2-action-0.1 hive2-action-0.1 oozie-bundle-0.1 oozie-bundle-0.1 oozie-bundle-0.1
oozie-bundle-0.2 oozie-bundle-0.2 oozie-bundle-0.2
oozie-coordinator oozie-coordinator-0.1 oozie-coordinator-0.1 oozie-coordinator-0.1 oozie-coordinator-0.1 oozie-coordinator-0.1 oozie-coordinator-0.1
oozie-coordinator-0.2 oozie-coordinator-0.2 oozie-coordinator-0.2 oozie-coordinator-0.2 oozie-coordinator-0.2 oozie-coordinator-0.2
oozie-coordinator-0.3 oozie-coordinator-0.3 oozie-coordinator-0.3 oozie-coordinator-0.3 oozie-coordinator-0.3 oozie-coordinator-0.3
Cloudera Administration | 221
Managing CDH and Managed Services
CDH 5.5.0
CDH 5.4.0
CDH 5.2.0
CDH 5.1.0
CDH 5.0.1
CDH 5.0.0
oozie-coordinator-0.4 oozie-coordinator-0.4 oozie-coordinator-0.4 oozie-coordinator-0.4 oozie-coordinator-0.4 oozie-coordinator-0.4
oozie-sla
oozie-sla-0.1
oozie-sla-0.1
oozie-sla-0.1
oozie-sla-0.1
oozie-sla-0.1
oozie-sla-0.1
oozie-sla-0.2
oozie-sla-0.2
oozie-sla-0.2
oozie-sla-0.2
oozie-sla-0.2
oozie-sla-0.2
oozie-workflow oozie-workflow-0.1 oozie-workflow-0.1 oozie-workflow-0.1 oozie-workflow-0.1 oozie-workflow-0.1 oozie-workflow-0.1
oozie-workflow-0.2 oozie-workflow-0.2 oozie-workflow-0.2 oozie-workflow-0.2 oozie-workflow-0.2 oozie-workflow-0.2
oozie-workflow-0.2.5 oozie-workflow-0.2.5 oozie-workflow-0.2.5 oozie-workflow-0.2.5 oozie-workflow-0.2.5 oozie-workflow-0.2.5
oozie-workflow-0.3 oozie-workflow-0.3 oozie-workflow-0.3 oozie-workflow-0.3 oozie-workflow-0.3 oozie-workflow-0.3
oozie-workflow-0.4 oozie-workflow-0.4 oozie-workflow-0.4 oozie-workflow-0.4 oozie-workflow-0.4 oozie-workflow-0.4
oozie-workflow-0.4.5 oozie-workflow-0.4.5 oozie-workflow-0.4.5 oozie-workflow-0.4.5 oozie-workflow-0.4.5 oozie-workflow-0.5
oozie-workflow-0.5 oozie-workflow-0.5 oozie-workflow-0.5 oozie-workflow-0.5 oozie-workflow-0.5
shell
shell-action-0.1 shell-action-0.1 shell-action-0.1 shell-action-0.1 shell-action-0.1 shell-action-0.1
shell-action-0.2 shell-action-0.2 shell-action-0.2 shell-action-0.2 shell-action-0.2 shell-action-0.2
shell-action-0.3 shell-action-0.3 shell-action-0.3 shell-action-0.3 shell-action-0.3 shell-action-0.3
spark
spark-action-0.1 spark-action-0.1
sqoop
sqoop-action-0.2 sqoop-action-0.2 sqoop-action-0.2 sqoop-action-0.2 sqoop-action-0.2 sqoop-action-0.2
sqoop-action-0.3 sqoop-action-0.3 sqoop-action-0.3 sqoop-action-0.3 sqoop-action-0.3 sqoop-action-0.3
sqoop-action-0.4 sqoop-action-0.4 sqoop-action-0.4 sqoop-action-0.4 sqoop-action-0.4 sqoop-action-0.4
ssh
ssh-action-0.1
ssh-action-0.1
ssh-action-0.1
ssh-action-0.1
ssh-action-0.1
ssh-action-0.1
ssh-action-0.2
ssh-action-0.2
ssh-action-0.2
ssh-action-0.2
ssh-action-0.2
ssh-action-0.2
Table 6: Oozie Schema - CDH 4
CDH 4.3.0
CDH 4.2.0
CDH 4.1.0
CDH 4.0.0
distcp-action-0.1
distcp-action-0.1
distcp-action-0.1
distcp-action-0.1
distcp-action-0.2
distcp-action-0.2
email
email-action-0.1
email-action-0.1
email-action-0.1
email-action-0.1
hive
hive-action-0.2
hive-action-0.2
hive-action-0.2
hive-action-0.2
hive-action-0.3
hive-action-0.3
hive-action-0.3
hive-action-0.4
hive-action-0.4
hive-action-0.4
oozie-bundle-0.1
oozie-bundle-0.1
oozie-bundle-0.1
oozie-bundle-0.2
oozie-bundle-0.2
oozie-bundle-0.2
distcp
hive-action-0.5
oozie-bundle
oozie-coordinator
oozie-bundle-0.1
oozie-coordinator-0.1 oozie-coordinator-0.1 oozie-coordinator-0.1 oozie-coordinator-0.1
oozie-coordinator-0.2 oozie-coordinator-0.2 oozie-coordinator-0.2 oozie-coordinator-0.2
oozie-coordinator-0.3 oozie-coordinator-0.3 oozie-coordinator-0.3 oozie-coordinator-0.3
222 | Cloudera Administration
Managing CDH and Managed Services
CDH 4.3.0
CDH 4.2.0
CDH 4.1.0
CDH 4.0.0
oozie-coordinator-0.4 oozie-coordinator-0.4 oozie-coordinator-0.4
oozie-sla
oozie-sla-0.1
oozie-sla-0.1
oozie-sla-0.1
oozie-sla-0.1
oozie-workflow
oozie-workflow-0.1
oozie-workflow-0.1
oozie-workflow-0.1
oozie-workflow-0.1
oozie-workflow-0.2
oozie-workflow-0.2
oozie-workflow-0.2
oozie-workflow-0.2
oozie-workflow-0.2.5 oozie-workflow-0.2.5 oozie-workflow-0.2.5 oozie-workflow-0.2.5
oozie-workflow-0.3
oozie-workflow-0.3
oozie-workflow-0.3
oozie-workflow-0.4
oozie-workflow-0.4
oozie-workflow-0.4
shell-action-0.1
shell-action-0.1
shell-action-0.1
shell-action-0.2
shell-action-0.2
shell-action-0.2
shell-action-0.3
shell-action-0.3
shell-action-0.3
sqoop-action-0.2
sqoop-action-0.2
sqoop-action-0.2
sqoop-action-0.3
sqoop-action-0.3
sqoop-action-0.3
sqoop-action-0.4
sqoop-action-0.4
sqoop-action-0.4
ssh-action-0.1
ssh-action-0.1
ssh-action-0.1
oozie-workflow-0.3
oozie-workflow-0.4.5
shell
sqoop
ssh
shell-action-0.1
sqoop-action-0.2
ssh-action-0.1
ssh-action-0.2
Enabling the Oozie Web Console
Enabling the Oozie Web Console Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
1.
2.
3.
4.
5.
6.
7.
Download ext-2.2. Extract the contents of the file to /var/lib/oozie/ on the same host as the Oozie Server.
In the Cloudera Manager Admin Console, go to the Oozie service.
Click the Configuration tab.
Locate the Enable Oozie server web console property or search for it by typing its name in the Search box.
Select Enable Oozie server web console.
Click Save Changes to commit the changes.
Restart the Oozie service.
Enabling the Oozie Web Console Using the Command Line
See Enabling the Oozie Web Console.
Setting the Oozie Database Timezone
We recommended that you set the timezone in the Oozie database to GMT. Databases do not handle Daylight Saving
Time (DST) shifts correctly. There might be problems if you run any Coordinators with actions scheduled to materialize
during the one-hour period that gets lost in DST.
• To set the timezone in Derby, add the following to CATALINA_OPTS in the oozie-env.sh file:
-Duser.timezone=GMT
Cloudera Administration | 223
Managing CDH and Managed Services
• To set the timezone just for Oozie in MySQL, add the following argument to
oozie.service.JPAService.jdbc.url:
useLegacyDatetimeCode=false&serverTimezone=GMT
Important: Changing the timezone on an existing Oozie database while Coordinators are already
running might cause Coordinators to shift by the offset of their timezone from GMT one time after
you make this change.
For more information about how to set your database's timezone, see your database's documentation.
Scheduling in Oozie Using Cron-like Syntax
Most Linux distributions include the cron utility, which is used for scheduling time-based jobs. For example, you might
want cron to run a script that deletes your Internet history once a week. This topic explains how to schedule Oozie
using Cron-like syntax.
Location
Set the scheduling information in the frequency attribute of the coordinator.xml file. A simple file looks like the
following example. The frequency attribute and scheduling information appear in bold.
<coordinator-app name="MY_APP" frequency="30 14 * *
*" start="2009-01-01T05:00Z" end="2009-01-01T06:00Z" timezone="UTC"
xmlns="uri:oozie:coordinator:0.5">
<action>
<workflow>
<app-path>hdfs://localhost:8020/tmp/workflows</app-path>
</workflow>
</action>
</coordinator-app>
Important: Before CDH 5 Oozie used fixed-frequency scheduling. You could only schedule according
to a set amount of minutes or a set time configured in an EL (Expression Language) function. The
cron-like syntax allows more flexibility.
Syntax and Structure
The cron-like syntax used by Oozie is a string with five space-separated fields:
•
•
•
•
•
minute
hour
day-of-month
month
day-of-week
The structure takes the form of * * * * *. For example, 30 14 * * * means that the job runs at at 2:30 p.m.
everyday. The minute field is set to 30, the hour field is set to 14, and the remaining fields are set to *.
Allowed Values and Special Characters
The following table describes special characters allowed and indicates in which fields they can be used.
Table 7: Special Characters
Character
Fields Allowed
Description
* (asterisk)
All
Match all values.
224 | Cloudera Administration
Managing CDH and Managed Services
Character
Fields Allowed
Description
, (comma)
All
Specify multiple values.
- (dash)
All
Specify a range.
/ (forward slash)
All
Specify an increment.
? (question mark)
Day-of-month, day-of-week Indicate no specific value (for example, if you want to specify
one but not the other).
L
Day-of-month, day-of-week Indicate the last day of the month or the last day of the week
(Saturday). In the day-of-week field, 6L indicates the last Friday
of the month.
W
Day-of-month
Indicate the nearest weekday to the given day.
# (pound sign)
Day-of-week
Indicate the nth day of the month
The following table summarizes the valid values for each field.
Field
Allowed Values
Allowed Special Characters
Minute
0-59
, - * /
Hour
0-23
, - * /
Day-of-month
0-31
, - * ? / L W
Month
1-12 or JAN-DEC
, - * /
Day-of-week
1-7 or SUN-SAT
, - * ? / L #
For more information about Oozie cron-like syntax, see Cron syntax in coordinator frequency.
Important: Some cron implementations accept 0-6 as the range for days of the week. Oozie accepts
1-7 instead.
Scheduling Examples
The following examples show cron scheduling in Oozie. Oozie’s processing time zone is UTC. If you are in a different
time zone, add to or subtract from the appropriate offset in these examples.
Run at the 30th minute of every hour
Set the minute field to 30 and the remaining fields to * so they match every value.
frequency="30 * * * *"
Run at 2:30 p.m. every day
Set the minute field to 30, the hour field to 14, and the remaining fields to *.
frequency="30 14 * * *"
Run at 2:30 p.m. every day in February
Set the minute field to 30, the hour field to 14, the day-of-month field to *, the month field to 2 (February), and
the day-of-week field to *.
frequency="30 14 * 2 *"
Cloudera Administration | 225
Managing CDH and Managed Services
Run every 20 minutes between 5:00-10:00 a.m. and between 12:00-2:00 p.m. on the fifth day of each month
Set the minute field to 0/20, the hour field to 5-9,12-14, the day-of-month field to 0/5, and the remaining fields
to *.
frequency="0/20 5-9,12-14 0/5 * *"
Run every Monday at 5:00 a.m.
Set the minute field to 0, the hour field to 5, the day-of-month field to ?, the month field to *, and the day-of-week
field to MON.
frequency="0 5 ? * MON"
Note: If the ? was set to *, this expression would run the job every day at 5:00 a.m., not just
Mondays.
Run on the last day of every month at 5:00 a.m.
Set the minute field to 0, the hour field to 5, the day-of-month field to L, the month field to *, and the day-of-week
field to ?.
frequency="0 5 L * ?"
Run at 5:00 a.m. on the weekday closest to the 15th day of each month
Set the minute field to 0, the hour field to 5, the day-of-month field to 15W, the month field to *, and the day-of-week
field to ?.
frequency="0 5 15W * ?"
Run every 33 minutes from 9:00-3:00 p.m. on the first Monday of every month
Set the minute field to 0/33, the hour field to 9-14, the day-of-week field to 2#1 (the first Monday), and the
remaining fields to *.
frequency="0/33 9-14 ? * 2#1"
Run every hour from 9:00 a.m.-5:00 p.m. on weekdays
Set the minute field to 0, the hour field to 9-17, the day-of-month field to ?, the month field to *, and the
day-of-week field to 2-6.
frequency="0 9-17 ? * 2-6"
Run on the second-to-last day of every month
Set the minute field to 0, the hour field to 0, the day-of-month field to L-1, the month field to *, and the day-of-week
field to ?.
frequency="0 0 L-1 * ?"
Note: “L-1 means the second-to-last day of the month.
Oozie uses Quartz, a job scheduler library, to parse the cron syntax. For more examples, go to the CronTrigger Tutorial
on the Quartz website. Quartz has two fields (second and year) that Oozie does not support.
226 | Cloudera Administration
Managing CDH and Managed Services
Managing Solr
You can install the Solr service through the Cloudera Manager installation wizard, using either parcels or packages.
See Installing Search.
You can elect to have the service created and started as part of the Installation wizard. If you elect not to create the
service using the Installation wizard, you can use the Add Service wizard to perform the installation. The wizard will
automatically configure and start the dependent services and the Solr service. See Adding a Service on page 36 for
instructions.
For further information on the Solr service, see Cloudera Search Guide.
The following sections describe how to configure other CDH components to work with the Solr service.
Configuring the Flume Morphline Solr Sink for Use with the Solr Service
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
To use a Flume Morphline Solr sink, the Flume service must be running on your cluster. See the Flume Near Real-Time
Indexing Reference (CDH 5) for information about the Flume Morphline Solr Sink and Managing Flume on page 78.
1.
2.
3.
4.
5.
Go to the Flume service.
Click the Configuration tab.
Select Scope > Agent
Select Category > Flume-NG Solr Sink.
Edit the following settings, which are templates that you must modify for your deployment:
• Morphlines File (morphlines.conf) - Configures Morphlines for Flume agents. You must use $ZK_HOST in
this field instead of specifying a ZooKeeper quorum. Cloudera Manager automatically replaces the $ZK_HOST
variable with the correct value during the Flume configuration deployment.
• Custom MIME-types File (custom-mimetypes.xml) - Configuration for the detectMimeTypes command.
See the Cloudera Morphlines Reference Guide for details on this command.
• Grok Dictionary File (grok-dictionary.conf) - Configuration for the grok command. See the Cloudera
Morphlines Reference Guide for details on this command.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
Once configuration is complete, Cloudera Manager automatically deploys the required files to the Flume agent process
directory when it starts the Flume agent. Therefore, you can reference the files in the Flume agent configuration using
their relative path names. For example, you can use the name morphlines.conf to refer to the location of the
Morphlines configuration file.
Deploying Solr with Hue
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
In CDH 4.3 and lower, in order to use Solr with Hue, you must update the URL for the Solr Server in the Hue Server
advanced configuration snippet.
1. Go to the Hue service.
2. Click the Configuration tab.
3. Type the word "snippet" in the Search box.
A set of Hue advanced configuration snippet properties displays.
4. Add information about your Solr host to the Hue Server Configuration Advanced Configuration Snippet for
hue_safety_valve_server.ini property. For example, if your hostname is SOLR_HOST, you might add the following:
[search]
## URL of the Solr Server
solr_url=http://SOLR_HOST:8983/solr
Cloudera Administration | 227
Managing CDH and Managed Services
5. Click Save Changes to save your advanced configuration snippet changes.
6. Restart the Hue Service.
Important: If you are using parcels with CDH 4.3, you must register the "hue-search" application
manually or access will fail. You do not need to do this if you are using CDH 4.4 or higher.
1. Stop the Hue service.
2. From the command line do the following:
a.
cd /opt/cloudera/parcels/CDH 4.3.0-1.cdh4.3.0.pXXX/share/hue
(Substitute your own local repository path for the /opt/cloudera/parcels/... if yours
is different, and specify the appropriate name of the CDH 4.3 parcel that exists in your
repository.)
b.
./build/env/bin/python ./tools/app_reg/app_reg.py
--install
/opt/cloudera/parcels/SOLR-0.9.0-1.cdh4.3.0.pXXX/share/hue/apps/search
c.
sed -i 's/\.\/apps/..\/..\/..\/..\/..\/apps/g'
./build/env/lib/python2.X/site-packages/hue.pth
where python2.X should be the version you are using (for example, python2.4).
3. Start the Hue service.
Using a Load Balancer with Solr
To configure a load balancer:
1.
2.
3.
4.
5.
Go to the Solr service.
Click the Configuration tab.
Select Scope > Solr
Select Category > All
Enter the hostname and port number of the load balancer in the Solr Load Balancer property in the format
hostname:port number.
Note:
When you set this property, Cloudera Manager regenerates the keytabs for Solr roles. The principal
in these keytabs contains the load balancer hostname.
If there is a Hue service that depends on this Solr service, it also uses the load balancer to
communicate with Solr.
6. Click Save Changes to commit the changes.
Migrating Solr Replicas
When you replace a host, migrating replicas on that host to the new host, instead of depending on failure recovery,
can help ensure optimal performance.
Where possible, the Solr service routes requests to the proper host. Both ADDREPLICA and DELETEREPLICA calls
can be sent to any host in the cluster.
• For adding replicas, the node parameter ensures the new replica is created on the intended host. If no host is
specified, Solr selects a host with relatively fewer replicas.
• For deleting replicas, the request is routed to the host that hosts the replica to be deleted.
228 | Cloudera Administration
Managing CDH and Managed Services
Adding replicas can be resource intensive. For best results, add replicas when the system is not under heavy load. For
example, do not add additional replicas when heavy indexing is occurring or when MapReduceIndexerTool jobs are
running.
Cloudera recommends using API calls to create and unload cores. Do not use the Cloudera Manager Admin Console
or the Solr Admin UI for these tasks.
This procedure uses the following names:
• Host names:
– origin at the IP address 192.0.2.1.
– destination at the IP address 192.0.2.2.
• Collection name email
• Replicas:
– The original replica email_shard1_replica2, which is on origin.
– The new replica email_shard1_replica3, which will be on destination.
To migrate a replica to a new host
1. Create the new replica on destination server using the ADDREPLICA API:
http://example.com:8983/solr/admin/collections?action=ADDREPLICA&collection=email&shard=email_shard1&node=192.0.2.2:7542_solr
2. Verify that the replica creation succeeds and moves from recovery state to ACTIVE. You can check the replica
status in the Cloud view, which can be found at a URL similar to:
http://destination.example.com:8983/solr/#/~cloud.
Note: Do not delete the original replica until the new one is in the ACTIVE state. When the newly
added replica is listed as ACTIVE, the index has been fully replicated to the newly added replica.
The total time to replicate an index varies according to factors such as network bandwidth and
the size of the index. Replication times on the scale of hours are not uncommon and do not
necessarily indicate a problem.
3. Use the CLUSTERSTATUS API to retrieve information about the cluster, including current cluster status:
http://example.com:8983/solr/admin/collections?action=clusterstatus&wt=json
Review the returned information to find the correct replica to remove.
4. Delete the old replica on origin server using the DELETEREPLICA API:
http://example.com:8983/solr/admin/collections?action=DELETEREPLICA&collection=email&shard=shard1&replica=core_node2
The DELTEREPLICA call removes the datadir.
Managing Spark
Apache Spark is a general framework for distributed computing that offers high performance for both batch and
interactive processing.
To run applications distributed across a cluster, Spark requires a cluster manager. Cloudera supports two cluster
managers: YARN and Spark Standalone. When run on YARN, Spark application processes are managed by the YARN
ResourceManager and NodeManager roles. When run on Spark Standalone, Spark application processes are managed
by Spark Master and Worker roles.
In CDH 5, Cloudera recommends running Spark applications on a YARN cluster manager instead of on a Spark Standalone
cluster manager, for the following benefits:
Cloudera Administration | 229
Managing CDH and Managed Services
• You can dynamically share and centrally configure the same pool of cluster resources among all frameworks that
run on YARN.
• You can use all the features of YARN schedulers for categorizing, isolating, and prioritizing workloads.
• You choose the number of executors to use; in contrast, Spark Standalone requires each application to run an
executor on every host in the cluster.
• Spark can run against Kerberos-enabled Hadoop clusters and use secure authentication between its processes.
Related Information
•
•
•
•
•
•
Spark Guide
Monitoring Spark Applications
Tuning Spark Applications on page 274
Spark Authentication
Cloudera Spark forum
Apache Spark documentation
This section describes how to manage Spark services.
Managing Spark Using Cloudera Manager
Spark is available as two services: Spark and Spark (Standalone).
In Cloudera Manager 5.1 and lower, the Spark service runs a Spark Standalone cluster, which has Master and Worker
roles.
In Cloudera Manager 5.2 and higher, the service that runs a Spark Standalone cluster has been renamed Spark
(Standalone), and the Spark service runs Spark as a YARN application with only gateway roles. Both services have a
Spark History Server role.
You can install, add, and start Spark through the Cloudera Manager Installation wizard using parcels. For more
information, see Installing Spark.
If you do not add the Spark service using the Installation wizard, you can use the Add Service wizard to create the
service. The wizard automatically configures dependent services and the Spark service. For instructions, see Adding a
Service on page 36.
When you upgrade from Cloudera Manager 5.1 or lower to Cloudera 5.2 or higher, Cloudera Manager does not migrate
an existing Spark service, which runs Spark Standalone, to a Spark on YARN service.
For information on Spark applications, see Spark Application Overview.
How Spark Configurations are Propagated to Spark Clients
Because the Spark service does not have worker roles, another mechanism is needed to enable the propagation of
client configurations to the other hosts in your cluster. In Cloudera Manager gateway roles fulfill this function. Whether
you add a Spark service at installation time or at a later time, ensure that you assign the gateway roles to hosts in the
cluster. If you do not have gateway roles, client configurations are not deployed.
Managing Spark Standalone Using the Command Line
Important: This item is deprecated and will be removed in a future release. Cloudera supports items
that are deprecated until they are removed. For more information about deprecated and removed
items, see Deprecated Items.
This section describes how to configure and start Spark Standalone services.
For information on installing Spark using the command line, see Spark Installation. For information on configuring and
starting the Spark History Server, see Configuring and Running the Spark History Server Using the Command Line on
page 232.
For information on Spark applications, see Spark Application Overview.
230 | Cloudera Administration
Managing CDH and Managed Services
Configuring Spark Standalone
Before running Spark Standalone, do the following on every host in the cluster:
• Edit /etc/spark/conf/spark-env.sh and change hostname in the last line to the name of the host where
the Spark Master will run:
###
### === IMPORTANT ===
### Change the following to specify the Master host
###
export STANDALONE_SPARK_MASTER_HOST=`hostname`
• Optionally, edit other configuration options:
– SPARK_MASTER_PORT / SPARK_MASTER_WEBUI_PORT and SPARK_WORKER_PORT /
SPARK_WORKER_WEBUI_PORT, to use non-default ports
– SPARK_WORKER_CORES, to set the number of cores to use on this machine
– SPARK_WORKER_MEMORY, to set how much memory to use (for example: 1000 MB, 2 GB)
– SPARK_WORKER_INSTANCE, to set the number of worker processes per node
– SPARK_WORKER_DIR, to set the working directory of worker processes
Starting and Stopping Spark Standalone Clusters
To start Spark Standalone clusters:
1. On one host in the cluster, start the Spark Master:
$ sudo service spark-master start
You can access the Spark Master UI at spark_master:18080.
2. On all the other hosts, start the workers:
$ sudo service spark-worker start
To stop Spark, use the following commands on the appropriate hosts:
$ sudo service spark-worker stop
$ sudo service spark-master stop
Service logs are stored in /var/log/spark.
Managing the Spark History Server
The Spark History Server displays information about the history of completed Spark applications. For further information,
see Monitoring Spark Applications.
For instructions for configuring the Spark History Server to use Kerberos, see Spark Authentication.
Adding the Spark History Server Using Cloudera Manager
By default, the Spark (Standalone) service does not include a History Server. To configure applications to store history,
on Spark clients, set spark.eventLog.enabled to true before starting the application.
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
To add the History Server:
1.
2.
3.
4.
Go to the Spark service.
Click the Instances tab.
Click the Add Role Instances button.
Select a host in the column under History Server, and then click OK.
Cloudera Administration | 231
Managing CDH and Managed Services
5.
6.
7.
8.
Click Continue.
Check the checkbox next to the History Server role.
Select Actions for Selected > Start and click Start.
Click Close when the action completes.
Configuring and Running the Spark History Server Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
1. Create the /user/spark/applicationHistory/ directory in HDFS and set ownership and permissions as
follows:
$
$
$
$
sudo
sudo
sudo
sudo
-u
-u
-u
-u
hdfs
hdfs
hdfs
hdfs
hadoop
hadoop
hadoop
hadoop
fs
fs
fs
fs
-mkdir
-mkdir
-chown
-chmod
/user/spark
/user/spark/applicationHistory
-R spark:spark /user/spark
1777 /user/spark/applicationHistory
2. On hosts from which you will launch Spark jobs, do the following:
a. Create /etc/spark/conf/spark-defaults.conf:
cp /etc/spark/conf/spark-defaults.conf.template /etc/spark/conf/spark-defaults.conf
b. Add the following to /etc/spark/conf/spark-defaults.conf:
spark.eventLog.dir=hdfs://namenode_host:namenode_port/user/spark/applicationHistory
spark.eventLog.enabled=true
or
spark.eventLog.dir=hdfs://name_service_id/user/spark/applicationHistory
spark.eventLog.enabled=true
c. On one host, start the History Server:
$ sudo service spark-history-server start
To link the YARN ResourceManager directly to the Spark History Server, set the spark.yarn.historyServer.address
property in /etc/spark/conf/spark-defaults.conf:
spark.yarn.historyServer.address=http://spark_history_server:history_port
By default, history_port is 18088. This causes Spark applications to write their history to the directory that the History
Server reads.
Managing the Sqoop 1 Client
The Sqoop 1 client allows you to create a Sqoop 1 gateway and deploy the client configuration.
Installing JDBC Drivers
Sqoop 1 does not ship with third-party JDBC drivers; you must download them separately. For information on
downloading and saving the drivers, see (CDH 4) Installing JDBC Drivers and (CDH 5) Installing JDBC Drivers. Ensure
that you do not save JARs in the CDH parcel directory /opt/cloudera/parcels/CDH, because this directory is
overwritten when you upgrade CDH.
232 | Cloudera Administration
Managing CDH and Managed Services
Adding the Sqoop 1 Client
Minimum Required Role: Full Administrator
The Sqoop 1 client packages are installed by the Installation wizard. However, the client configuration is not deployed.
To create a Sqoop 1 gateway and deploy the client configuration:
1. On the Home > Status tab, click
to the right of the cluster name and select Add a Service. A list of service types display. You can add one type of
service at a time.
2. Select the Sqoop 1 Client service and click Continue.
3. Select the radio button next to the services on which the new service should depend. All services must depend
on the same ZooKeeper service. Click Continue.
4. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of the
hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of hosts to
which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple hosts,
you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts dialog.
The following shortcuts for specifying hostname patterns are supported:
• Range of hostnames (without the domain portion)
Range Definition
Matching Hosts
10.1.1.[1-4]
10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4
host[1-3].company.com
host1.company.com, host2.company.com, host3.company.com
host[07-10].company.com
host07.company.com, host08.company.com, host09.company.com,
host10.company.com
• IP addresses
• Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
5. Click Continue. The client configuration deployment command runs.
6. Click Continue and click Finish.
Managing Sqoop 2
Cloudera Manager can install the Sqoop 2 service as part of the CDH installation.
Note: Sqoop 2 is being deprecated. Cloudera recommends you use Sqoop 1.
You can elect to have the service created and started as part of the Installation wizard if you choose to add it in Custom
Services. If you elect not to create the service using the Installation wizard, you can use the Add Service wizard to
perform the installation. The wizard will automatically configure and start the dependent services and the Sqoop 2
service. See Adding a Service on page 36 for instructions.
Installing JDBC Drivers
The Sqoop 2 service does not ship with third-party JDBC drivers; you must download them separately. For information
on downloading and saving the drivers, see (CDH 4) Configuring Sqoop 2 and (CDH 5) Configuring Sqoop 2. Ensure that
you do not save JARs in the CDH parcel directory /opt/cloudera/parcels/CDH, because this directory is overwritten
when you upgrade CDH.
Cloudera Administration | 233
Managing CDH and Managed Services
Managing ZooKeeper
Minimum Required Role: Full Administrator
When adding the ZooKeeper service, the Add Service wizard automatically initializes the data directories. If you quit
the Add Service wizard or it does not finish successfully, you can initialize the directories outside the wizard by doing
these steps:
1. Go to the ZooKeeper service.
2. Select Actions > Initialize.
3. Click Initialize again to confirm.
Note: If the data directories are not initialized, the ZooKeeper servers cannot be started.
In a production environment, you should deploy ZooKeeper as an ensemble with an odd number of servers. As long
as a majority of the servers in the ensemble are available, the ZooKeeper service will be available. The minimum
recommended ensemble size is three ZooKeeper servers, and Cloudera recommends that each server run on a separate
machine. In addition, the ZooKeeper server process should have its own dedicated disk storage if possible.
Configuring Services to Use the GPL Extras Parcel
After you install the GPL Extras parcel, reconfigure and restart services that need to use LZO functionality. Any service
that does not require the use of LZO need not be configured.
HDFS
1.
2.
3.
4.
5.
Go to the HDFS service.
Click the Configuration tab.
Search for the io.compression.codecs property.
In the Compression Codecs property, click in the field, then click the + sign to open a new value field.
Add the following two codecs:
• com.hadoop.compression.lzo.LzoCodec
• com.hadoop.compression.lzo.LzopCodec
6. Save your configuration changes.
7. Restart HDFS.
8. Redeploy the HDFS client configuration.
Oozie
1. Go to /var/lib/oozie on each Oozie server and even if the LZO JAR is present, symlink the Hadoop LZO JAR:
• CDH 5 - /opt/cloudera/parcels/GPLEXTRAS/lib/hadoop/lib/hadoop-lzo.jar
• CDH 4 - /opt/cloudera/parcels/HADOOP_LZO/lib/hadoop/lib/hadoop-lzo.jar
2. Restart Oozie.
HBase
Restart HBase.
Impala
Restart Impala.
Hive
Restart the Hive server.
234 | Cloudera Administration
Managing CDH and Managed Services
Sqoop 2
1. Add the following entries to the Sqoop Service Environment Advanced Configuration Snippet:
• HADOOP_CLASSPATH=$HADOOP_CLASSPATH:/opt/cloudera/parcels/GPLEXTRAS/lib/hadoop/lib/*
• JAVA_LIBRARY_PATH=$JAVA_LIBRARY_PATH:/opt/cloudera/parcels/GPLEXTRAS/lib/hadoop/lib/native
2. Restart the Sqoop service.
Cloudera Administration | 235
Resource Management
Resource Management
Resource management helps ensure predictable behavior by defining the impact of different services on cluster
resources. The goals of resource management features are to:
• Guarantee completion in a reasonable time frame for critical workloads
• Support reasonable cluster scheduling between groups of users based on fair allocation of resources per group
• Prevent users from depriving other users access to the cluster
Schedulers
A scheduler is responsible for deciding which tasks get to run and where and when to run them. The MapReduce and
YARN computation frameworks support the following schedulers:
• FIFO - Allocates resources based on arrival time.
• Fair - Allocates resources to weighted pools, with fair sharing within each pool.
– CDH 4 Fair Scheduler
– CDH 5 Fair Scheduler
• Capacity - Allocates resources to pools, with FIFO scheduling within each pool.
– CDH 4 Capacity Scheduler
– CDH 5 Capacity Scheduler
The scheduler defaults for MapReduce and YARN are:
• MapReduce - Cloudera Manager, CDH 5, and CDH 4 set the default scheduler to FIFO. FIFO is set as the default
for backward-compatibility purposes, but Cloudera recommends Fair Scheduler because Impala and Llama are
optimized for it. Capacity Scheduler is also available.
If you are running CDH 4, you can specify how MapReduce jobs share resources by configuring the MapReduce
scheduler.
• YARN - Cloudera Manager, CDH 5, and CDH 4 set the default to Fair Scheduler. Cloudera recommends Fair Scheduler
because Impala and Llama are optimized for it. FIFO and Capacity Scheduler are also available.
In YARN, the scheduler is responsible for allocating resources to the various running applications subject to familiar
constraints of capacities, queues, and so on. The scheduler performs its scheduling function based the resource
requirements of the applications; it does so based on the abstract notion of a resource container that incorporates
elements such as memory, CPU, disk, network, and so on.
The YARN scheduler has a pluggable policy plug-in, which is responsible for partitioning the cluster resources
among the various queues, applications, and so on.
If you are running CDH 5, you can specify how YARN applications share resources by manually configuring the
YARN scheduler. Alternatively you can use Cloudera Manager dynamic allocation features to manage the scheduler
configuration.
Cloudera Manager Resource Management
Cloudera Manager provides two methods for allocating cluster resources to services: static and dynamic.
236 | Cloudera Administration
Resource Management
Static Allocation
Cloudera Manager 4 introduced the ability to partition resources across HBase, HDFS, Impala, MapReduce, and YARN
services by allowing you to set configuration properties that were enforced by Linux control groups (Linux cgroups).
With Cloudera Manager 5, the ability to statically allocate resources using cgroups is configurable through a single
static service pool wizard. You allocate services a percentage of total resources and the wizard configures the cgroups.
For example, the following figure illustrates static pools for HBase, HDFS, Impala, and YARN services that are respectively
assigned 20%, 30%, 20%, and 30% of cluster resources.
Dynamic Allocation
Cloudera Manager allows you to manage mechanisms for dynamically apportioning resources statically allocated to
YARN and Impala using dynamic resource pools.
Depending on the version of CDH you are using, dynamic resource pools in Cloudera Manager support the following
resource management (RM) scenarios:
• (CDH 5) YARN Independent RM - YARN manages the virtual cores, memory, running applications, and scheduling
policy for each pool. In the preceding diagram, three dynamic resource pools - Dev, Product, and Mktg with weights
3, 2, and 1 respectively - are defined for YARN. If an application starts and is assigned to the Product pool, and
other applications are using the Dev and Mktg pools, the Product resource pool will receive 30% x 2/6 (or 10%)
of the total cluster resources. If there are no applications using the Dev and Mktg pools, the YARN Product pool
will be allocated 30% of the cluster resources.
• (CDH 5) YARN and Impala Independent RM - YARN manages the virtual cores, memory, running applications, and
scheduling policy for each pool; Impala manages memory for pools running queries and limits the number of
running and queued queries in each pool.
• (CDH 5 and CDH 4) Impala Independent RM - Impala manages memory for pools running queries and limits the
number of running and queued queries in each pool.
• (CDH 5) YARN and Impala Integrated RM -
Cloudera Administration | 237
Resource Management
Note: Though Impala can be used together with YARN via simple configuration of Static Service
Pools in Cloudera Manager, the use of the general-purpose component Llama for integrated
resource management within YARN is no longer supported with CDH 5.5 / Impala 2.3 and higher.
YARN manages memory for pools running Impala queries; Impala limits the number of running and queued queries
in each pool. In the YARN and Impala integrated RM scenario, Impala services can reserve resources through
YARN, effectively sharing the static YARN service pool and resource pools with YARN applications. The integrated
resource management scenario, where both YARN and Impala use the YARN resource management framework,
require the Impala Llama role.
In the following figure, the YARN and Impala services have a 50% static share which is subdivided among the
original resource pools with an additional resource pool designated for the Impala service. If YARN applications
are using all the original pools, and Impala uses its designated resource pool, Impala queries will have the same
resource allocation 50% x 4/8 = 25% as in the first scenario. However, when YARN applications are not using the
original pools, Impala queries will have access to 50% of the cluster resources.
The scenarios where YARN manages resources, whether for independent RM or integrated RM, map to the YARN
scheduler configuration. The scenarios where Impala independently manages resources employ the Impala admission
control feature.
To submit a YARN application to a specific resource pool, specify the mapreduce.job.queuename property. The
YARN application's queue property is mapped to a resource pool. To submit an Impala query to a specific resource
pool, specify the REQUEST_POOL option.
For details on how to configure specific resource management features, see the following topics:
Linux Control Groups (cgroups)
Minimum Required Role: Full Administrator
Cloudera Manager supports the Linux control groups (cgroups) kernel feature. With cgroups, administrators can impose
per-resource restrictions and limits on services and roles. This provides the ability to allocate resources using cgroups
238 | Cloudera Administration
Resource Management
to enable isolation of compute frameworks from one another. Resource allocation is implemented by setting properties
for the services and roles.
Linux Distribution Support
Cgroups are a feature of the Linux kernel, and as such, support depends on the host's Linux distribution and version
as shown in the following tables. If a distribution lacks support for a given parameter, changes to the parameter have
no effect.
Table 8: RHEL-compatible
Distribution
CPU Shares
I/O Weight
Memory Soft Limit Memory Hard Limit
CPU Shares
I/O Weight
Memory Soft Limit Memory Hard Limit
CPU Shares
I/O Weight
Memory Soft Limit Memory Hard Limit
CPU Shares
I/O Weight
Memory Soft Limit Memory Hard Limit
Red Hat Enterprise Linux,
CentOS, and Oracle Enterprise
Linux 7
Red Hat Enterprise Linux,
CentOS, and Oracle Enterprise
Linux 6
Red Hat Enterprise Linux,
CentOS, and Oracle Enterprise
Linux 5
Table 9: SLES
Distribution
SUSE Linux Enterprise Server
11
Table 10: Ubuntu
Distribution
Ubuntu 14.04 LTS
Ubuntu 12.04 LTS
Ubuntu 10.04 LTS
Table 11: Debian
Distribution
Debian 7.1
Debian 7.0
Debian 6.0
The exact level of support can be found in the Cloudera Manager Agent log file, shortly after the Agent has started.
See Viewing the Cloudera Manager Server Log to find the Agent log. In the log file, look for an entry like this:
Found cgroups capabilities: {
'has_memory': True,
'default_memory_limit_in_bytes': 9223372036854775807,
'writable_cgroup_dot_procs': True,
'has_cpu': True,
'default_blkio_weight': 1000,
Cloudera Administration | 239
Resource Management
'default_cpu_shares': 1024,
'has_blkio': True}
The has_cpu and similar entries correspond directly to support for the CPU, I/O, and memory parameters.
Further Reading
•
•
•
•
•
http://www.kernel.org/doc/Documentation/cgroups/cgroups.txt
http://www.kernel.org/doc/Documentation/cgroups/blkio-controller.txt
http://www.kernel.org/doc/Documentation/cgroups/memory.txt
MANAGING SYSTEM RESOURCES ON RED HAT ENTERPRISE LINUX 6
MANAGING SYSTEM RESOURCES ON RED HAT ENTERPRISE LINUX 7
Resource Management with Control Groups
To use cgroups, you must enable cgroup-based resource management under the host resource management
configuration properties. However, if you configure static service pools, this property is set as part of that process.
Enabling Resource Management
Cgroups-based resource management can be enabled for all hosts, or on a per-host basis.
1. If you have upgraded from a version of Cloudera Manager older than Cloudera Manager 4.5, restart every Cloudera
Manager Agent before using cgroups-based resource management:
a. Stop all services, including the Cloudera Management Service.
b. On each cluster host, run as root:
• RHEL-compatible 7 and higher:
$ sudo service cloudera-scm-agent next_stop_hard
$ sudo service cloudera-scm-agent restart
• All other Linux distributions:
$ sudo service cloudera-scm-agent hard_restart
c. Start all services.
2.
3.
4.
5.
6.
7.
Click the Hosts tab.
Optionally click the link of the host where you want to enable cgroups.
Click the Configuration tab.
Select Category > Resource Management.
Select Enable Cgroup-based Resource Management.
Restart all roles on the host or hosts.
Limitations
• Role group and role instance override cgroup-based resource management parameters must be saved one at a
time. Otherwise some of the changes that should be reflected dynamically will be ignored.
• The role group abstraction is an imperfect fit for resource management parameters, where the goal is often to
take a numeric value for a host resource and distribute it amongst running roles. The role group represents a
"horizontal" slice: the same role across a set of hosts. However, the cluster is often viewed in terms of "vertical"
slices, each being a combination of worker roles (such as TaskTracker, DataNode, RegionServer, Impala Daemon,
and so on). Nothing in Cloudera Manager guarantees that these disparate horizontal slices are "aligned" (meaning,
that the role assignment is identical across hosts). If they are unaligned, some of the role group values will be
incorrect on unaligned hosts. For example a host whose role groups have been configured with memory limits
but that's missing a role will probably have unassigned memory.
240 | Cloudera Administration
Resource Management
Configuring Resource Parameters
After enabling cgroups, you can restrict and limit the resource consumption of roles (or role groups) on a per-resource
basis. All of these parameters can be found in the Cloudera Manager Admin Console, under the Resource Management
category:
• CPU Shares - The more CPU shares given to a role, the larger its share of the CPU when under contention. Until
processes on the host (including both roles managed by Cloudera Manager and other system processes) are
contending for all of the CPUs, this will have no effect. When there is contention, those processes with higher
CPU shares will be given more CPU time. The effect is linear: a process with 4 CPU shares will be given roughly
twice as much CPU time as a process with 2 CPU shares.
Updates to this parameter will be dynamically reflected in the running role.
• I/O Weight - The greater the I/O weight, the higher priority will be given to I/O requests made by the role when
I/O is under contention (either by roles managed by Cloudera Manager or by other system processes).
This only affects read requests; write requests remain unprioritized. The Linux I/O scheduler controls when buffered
writes are flushed to disk, based on time and quantity thresholds. It continually flushes buffered writes from
multiple sources, not certain prioritized processes.
Updates to this parameter will be dynamically reflected in the running role.
• Memory Soft Limit - When the limit is reached, the kernel will reclaim pages charged to the process if and only if
the host is facing memory pressure. If reclaiming fails, the kernel may kill the process. Both anonymous as well
as page cache pages contribute to the limit.
After updating this parameter, the role must be restarted before changes take effect.
• Memory Hard Limit - When a role's resident set size (RSS) exceeds the value of this parameter, the kernel will
swap out some of the role's memory. If it's unable to do so, it will kill the process. Note that the kernel measures
memory consumption in a manner that doesn't necessarily match what the top or ps report for RSS, so expect
that this limit is a rough approximation.
After updating this parameter, the role must be restarted before changes take effect.
Example: Protecting Production MapReduce Jobs from Impala Queries
Suppose you have MapReduce deployed in production and want to roll out Impala without affecting production
MapReduce jobs. For simplicity, we will make the following assumptions:
•
•
•
•
•
•
The cluster is using homogenous hardware
Each worker host has two cores
Each worker host has 8 GB of RAM
Each worker host is running a DataNode, TaskTracker, and an Impala Daemon
Each role type is in a single role group
Cgroups-based resource management has been enabled on all hosts
Action
Procedure
CPU
1. Leave DataNode and TaskTracker role group CPU shares at 1024.
2. Set Impala Daemon role group's CPU shares to 256.
3. The TaskTracker role group should be configured with a Maximum Number of Simultaneous
Map Tasks of 2 and a Maximum Number of Simultaneous Reduce Tasks of 1. This yields an
upper bound of three MapReduce tasks at any given time; this is an important detail for
memory sizing.
Memory
1.
2.
3.
4.
Set Impala Daemon role group memory limit to 1024 MB.
Leave DataNode maximum Java heap size at 1 GB.
Leave TaskTracker maximum Java heap size at 1 GB.
Leave MapReduce Child Java Maximum Heap Size for Gateway at 1 GB.
Cloudera Administration | 241
Resource Management
Action
Procedure
5. Leave cgroups hard memory limits alone. We'll rely on "cooperative" memory limits
exclusively, as they yield a nicer user experience than the cgroups-based hard memory limits.
I/O
1. Leave DataNode and TaskTracker role group I/O weight at 500.
2. Impala Daemon role group I/O weight is set to 125.
When you're done with configuration, restart all services for these changes to take effect. The results are:
1. When MapReduce jobs are running, all Impala queries together will consume up to a fifth of the cluster's CPU
resources.
2. Individual Impala Daemons won't consume more than 1 GB of RAM. If this figure is exceeded, new queries will
be cancelled.
3. DataNodes and TaskTrackers can consume up to 1 GB of RAM each.
4. We expect up to 3 MapReduce tasks at a given time, each with a maximum heap size of 1 GB of RAM. That's up
to 3 GB for MapReduce tasks.
5. The remainder of each host's available RAM (6 GB) is reserved for other host processes.
6. When MapReduce jobs are running, read requests issued by Impala queries will receive a fifth of the priority of
either HDFS read requests or MapReduce read requests.
Static Service Pools
Static service pools isolate the services in your cluster from one another, so that load on one service has a bounded
impact on other services. Services are allocated a static percentage of total resources—CPU, memory, and I/O
weight—which are not shared with other services. When you configure static service pools, Cloudera Manager computes
recommended memory, CPU, and I/O configurations for the worker roles of the services that correspond to the
percentage assigned to each service. Static service pools are implemented per role group within a cluster, using Linux
control groups (cgroups) and cooperative memory limits (for example, Java maximum heap sizes). Static service pools
can be used to control access to resources by HBase, HDFS, Impala, MapReduce, Solr, Spark, YARN, and add-on services.
Static service pools are not enabled by default.
Note:
• I/O allocation only works when short-circuit reads are enabled.
• I/O allocation does not handle write side I/O because cgroups in the Linux kernel do not currently
support buffered writes.
Viewing Static Service Pool Status
Select Clusters > Cluster name > Resource Management > Static Service Pools.If the cluster has a YARN service, the
Static Service Pools Status tab displays and shows whether resource management is enabled for the cluster, and the
currently configured service pools.
See Monitoring Static Service Pools for more information.
Enabling and Configuring Static Service Pools
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
1. Select Clusters > Cluster name > Resource Management > Static Service Pools.
2. Click the Configuration tab. The Step 1 of 4: Basic Allocation Setup page displays. In each field in the basic allocation
table, enter the percentage of resources to give to each service. The total must add up to 100%. In CDH 5 clusters,
if you enable integrated resource management, the Impala service shares the YARN service pool, rather than use
its own static pool. In this case, you cannot specify a percentage for the Impala service. Click Continue to proceed.
242 | Cloudera Administration
Resource Management
3. Step 2: Review Changes - The allocation of resources for each resource type and role displays with the new values
as well as the values previously in effect. The values for each role are set by role group; if there is more than one
role group for a given role type (for example, for RegionServers or DataNodes) then resources will be allocated
separately for the hosts in each role group. Take note of changed settings. If you have previously customized these
settings, check these over carefully.
• Click the to the right of each percentage to display the allocations for a single service. Click
of the Total (100%) to view all the allocations in a single page.
• Click the Back button to go to the previous page and change your allocations.
to the right
When you are satisfied with the allocations, click Continue.
4. Step 3 of 4: Restart Services - To apply the new allocation percentages, click Restart Now to restart the cluster.
To skip this step, click Restart Later. If HDFS High Availability is enabled, you will have the option to choose a
rolling restart.
5. Step 4 of 4: Progress displays the status of the restart commands. Click Finished after the restart commands
complete.
Disabling Static Service Pools
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
To disable static service pools, disable cgroup-based resource management for all hosts in all clusters:
1.
2.
3.
4.
5.
6.
In the main navigation bar, click Hosts.
Click the Configuration tab.
Select Scope > Resource Management.
Deselect the Enable Cgroup-based Resource Management property.
Click Save Changes.
Restart all services.
Static resource management is disabled, but the percentages you set when you configured the pools, and all the
changed settings (for example, heap sizes), are retained by the services. The percentages and settings will also be used
when you re-enable static service pools. If you want to revert to the settings you had before static service pools were
enabled, follow the procedures in Viewing and Reverting Configuration Changes on page 31.
Dynamic Resource Pools
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
A dynamic resource pool is a named configuration of resources and a policy for scheduling the resources among YARN
applications and Impala queries running in the pool. Dynamic resource pools allow you to schedule and allocate
resources to YARN applications and Impala queries based on a user's access to specific pools and the resources available
to those pools. If a pool's allocation is not in use it can be given to other pools. Otherwise, a pool receives a share of
resources in accordance with the pool's weight. Dynamic resource pools have ACLs that restrict who can submit work
to and administer them.
A configuration set defines the allocation of resources across pools that may be active at a given time. For example,
you can define "weekday" and "weekend" configuration sets, which define different resource pool configurations for
different days of the week.
A scheduling rule defines when a configuration set is active. The configuration set is updated in affected services every
hour.
Resource pools can be nested, with sub-pools restricted by the settings of their parent pool.
The resources available for sharing are subject to the allocations made for each service if static service pools (cgroups)
are being enforced. For example, if the static pool for YARN is 75% of the total cluster resources, then resource pools
will use only that 75% of resources.
Cloudera Administration | 243
Resource Management
Managing Dynamic Resource Pools
After you create or edit a resource pool,
displays while the settings are propagated to the service configuration files. You can also manually refresh the files.
Viewing Dynamic Resource Pool Configuration
Depending on which resource management scenario described in Cloudera Manager Resource Management on page
236 is in effect, the dynamic resource pool configuration overview displays the following information:
• YARN Independent RM - Weight, Virtual Cores, Min and Max Memory, Max Running Apps, and Scheduling Policy
• YARN and Impala Integrated RM
– YARN - Weight, Virtual Cores, Min and Max Memory, Max Running Apps, and Scheduling Policy
– Impala - Max Running Queries and Max Queued Queries
• YARN and Impala Independent RM
– YARN - Weight, Virtual Cores, Min and Max Memory, Max Running Apps, and Scheduling Policy
– Impala - Max Memory, Max Running Queries, and Max Queued Queries
• Impala Independent RM - Max Memory, Max Running Queries, and Max Queued Queries
To view dynamic resource pool configuration:
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled for
dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
Enabling and Disabling Dynamic Resource Pools for Impala
By default dynamic resource pools for Impala are disabled. If dynamic resource pools are disabled, the Impala section
will not appear in the Dynamic Resource Pools tab or in the resource pool dialogs within that page. To modify the
Impala dynamic resource pool setting:
1.
2.
3.
4.
5.
6.
Go to the Impala service.
Click the Configuration tab.
Select Category > Admission Control.
Select or deselect the Enable Dynamic Resource Pools checkbox.
Click Save Changes to commit the changes.
Restart the Impala service.
Creating a Dynamic Resource Pool
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled for
dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click Add Resource Pool. The Add dialog box displays showing the General tab.
4. Specify a name and resource limits for the pool:
• In the Resource Pool Name field, specify the pool name. Enter a unique name containing only alphanumeric
characters. If referencing a user or group name that contains a ".", replace the "." with "_dot_".
• Specify the policy for scheduling resources among applications running in the pool:
244 | Cloudera Administration
Resource Management
– Dominant Resource Fairness (DRF) (default) - An extension of fair scheduling for more than one
resource—it determines resource shares (CPU, memory) for a job separately based on the availability
of those resources and the needs of the job.
– Fair Scheduler (FAIR) - Determines resource shares based on memory.
– First-In, First-Out (FIFO) - Determines resource shares based on when the job was added.
• If you have enabled Fair Scheduler preemption, optionally set a preemption timeout to specify how long a
job in this pool must wait before it can preempt resources from jobs in other pools. To enable preemption,
click the Fair Scheduler Preemption link or follow the procedure in Enabling Preemption on page 247.
5. Do one or more of the following:
• Click the YARN tab.
1. Click a configuration set.
2. Specify a weight that indicates that pool's share of resources relative to other pools, minimum and
maximums for virtual cores and memory, and a limit on the number of applications that can run
simultaneously in the pool.
• Click the Impala tab.
1. Click a configuration set.
2. Specify the maximum number of concurrently running and queued queries in the pool.
6. If you have enabled ACLs and specified users or groups, optionally click the Submission and Administration Access
Control tabs to specify which users and groups can submit applications and which users can view all and kill
applications. The default is that anyone can submit, view all, and kill applications. To restrict either of these
permissions, select the Allow these users and groups radio button and provide a comma-delimited list of users
and groups in the Users and Groups fields respectively. Click OK.
Adding Sub-Pools
Pools can be nested as sub-pools. They share among their siblings the resources of the parent pool. Each sub-pool can
have its own resource restrictions; if those restrictions fall within the configuration of the parent pool, then the limits
for the sub-pool take effect. If the limits for the sub-pool exceed those of the parent, then the parent limits take effect.
Once you create sub-pools, jobs cannot be submitted to the parent pool; they must be submitted to a sub-pool.
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled for
dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3.
Click
at the right of a resource pool row and select Add Sub Pool. Configure sub-pool properties.
4. Click OK.
Configuring Default Scheduler Properties
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled for
dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click the Default Settings button.
4. Specify the default scheduling policy, maximum applications, and preemption timeout properties.
5. Click OK.
Cloudera Administration | 245
Resource Management
Editing Dynamic Resource Pools
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled for
dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click Edit at the right of a resource pool row. Edit the properties and click OK.
4. If you have enabled ACLs and specified users or groups, optionally click the Submission and Administration Access
Control tabs to specify which users and groups can submit applications and which users can view all and kill
applications. The default is that anyone can submit, view all, and kill applications. To restrict either of these
permissions, select the Allow these users and groups radio button and provide a comma-delimited list of users
and groups in the Users and Groups fields respectively. Click OK.
Refreshing Dynamic Resource Pool Configuration Files
After updating resource pool settings, you can refresh service configuration files as follows:
1. On the Home > Status tab, select Clusters > Cluster name > Refresh Dynamic Resource Pools.
YARN Pool Status and Configuration Options
Viewing Dynamic Resource Pool Status
Select Clusters > ClusterName > Dynamic Resource Pools. The Status tab displays the YARN resource pools currently
in use for the cluster. See Monitoring Dynamic Resource Pools for more information.
Setting User Limits
Pool properties determine the maximum number of applications that can run in a pool. To limit the number of
applications specific users can run at the same time in a pool:
1.
2.
3.
4.
5.
Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools.
Click the Configuration tab.
Click the User Limits tab. The table displays a list of users and the maximum number of jobs each user can submit.
Click Add User Limit.
Specify a username. Enter a unique name containing only alphanumeric characters. If referencing a user or group
name that contains a ".", replace the "." with "_dot_".
6. Specify the maximum number of running applications.
7. Click OK.
Enabling ACLs
To specify whether ACLs are checked:
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools.
2. Click the Configuration tab.
3. Click Other Settings.
4. In the Enable ResourceManager ACLs property, click . The YARN service configuration page displays.
5. Select the checkbox.
6. Click Save Changes to commit the changes.
7. Click to invoke the cluster restart wizard.
8. Click Restart Cluster.
9. Click Restart Now.
10. Click Finish.
246 | Cloudera Administration
Resource Management
Configuring ACLs
To configure which users and groups can submit and kill YARN applications in any resource pool:
1. Enable ACLs.
2. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools.
3. Click the Configuration tab.
4. Click Other Settings.
5. In the Admin ACL property, click . The YARN service configuration page displays.
6. Specify which users and groups can submit and kill applications.
7. Click Save Changes to commit the changes.
8. Click to invoke the cluster restart wizard.
9. Click Restart Cluster.
10. Click Restart Now.
11. Click Finish.
Enabling Preemption
You can enable the Fair Scheduler to preempt applications in other pools if a pool's minimum share is not met for some
period of time. When you create a pool you can specify how long a pool must wait before other applications are
preempted.
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools.
2. Click the Configuration tab.
3. Click the User Limits tab. The table shows you a list of users and the maximum number of jobs each user can
submit.
4. Click Other Settings.
5. In the Fair Scheduler Preemption, click . The YARN service configuration page displays.
6. Select the checkbox.
7. Click Save Changes to commit the changes.
8. Click to invoke the cluster restart wizard.
9. Click Restart Cluster.
10. Click Restart Now.
11. Click Finish.
Assigning Applications and Queries to Resource Pools
To submit a YARN application to a specific resource pool, specify the mapreduce.job.queuename property. The
YARN application's queue property is mapped to a resource pool. To submit an Impala query to a specific resource
pool, specify the REQUEST_POOL option.
If a specific pool is not requested, Cloudera Manager provides many options for determining how YARN applications
and Impala queries are placed in resource pools. You can specify basic rules that place applications and queries in pools
based on runtime configurations or the name of the user running the application or query or select an advanced option
that allows you to specify a set of ordered rules for placing applications and queries in pools.
Enabling and Disabling Undeclared Pools
If you do not specify a pool with a job or query property, by default YARN and Impala create a pool "on-the-fly" with
the name of the user that submitted the request and assigns it to that resource pool. For YARN, you can change this
behavior so that the default pool is used instead:
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled for
dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
Cloudera Administration | 247
Resource Management
3. Click the Placement Rules tab.
4. Click Basic radio button.
5. Click the Allow Undeclared Pools property.
6. Select or deselect the Allow Undeclared Pools checkbox.
7. Click Save Changes to commit the changes.
8. Click to invoke the cluster restart wizard.
9. Click Restart Cluster.
10. Click Restart Now.
11. Click Finish.
Note: YARN and Impala pools created "on-the-fly" are deleted when you restart the YARN and Impala
services.
Enabling and Disabling the Default Pool
If an application specifies a pool that has not been explicitly configured or is assigned to a pool with the name of user
according to the Fair Scheduler User As Default Queue property, by default YARN creates the pool at runtime with
default settings. To change the behavior so that under these circumstances the default pool is used instead:
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled for
dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click the Placement Rules tab.
4. Click Basic radio button.
5. Click the Fair Scheduler User As Default Queue property.
6. Select or deselect the checkbox.
7. Click Save Changes to commit the changes.
8. Click to invoke the cluster restart wizard.
9. Click Restart Cluster.
10. Click Restart Now.
11. Click Finish.
Specifying Advanced Placement Rules and Rule Ordering
You use placement rules to indicate whether applications are placed in specified pools, pools named by a user or group,
or the default pool. To configure and order a set of rules:
1. Select the Advanced radio button on the Placement Rules tab.
2. Click to add a new rule row and to remove a rule row.
3.
In each row, click
and select a rule. The available rules are:
• specified pool; create the pool if it doesn't exist (default 1st)
• root.<username> pool; create the pool if it doesn't exist (default 2nd) - the application or query is placed into
a pool with the name of the user who submitted it.
• specified pool only if the poll exists
• root.<username> pool only if the pool exists
• root.<primaryGroupName> pool; create the pool if it doesn't exist - the application or query is placed into a
pool with the name of the primary group of the user who submitted it.
• root.<primaryGroupName> pool only if the pool exists
• root.<secondaryGroupName> pool only if one of these pools exists - the application or query is placed into
a pool with a name that matches a secondary group of the user who submitted it.
248 | Cloudera Administration
Resource Management
• default pool; create the pool if it doesn't exist
For more information about these rules, see the description of the queuePlacementPolicy element in Allocation
File Format. Reorder rules by selecting different rules for existing rule rows. If a rule is always satisfied, subsequent
rules are not evaluated and appear disabled.
4. Click Save. The Fair Scheduler allocation file (by default, fair-scheduler.xml) is updated.
Configuration Sets
A configuration set defines the allocation of resources across pools that may be active at a given time. For example,
you can define "weekday" and "weekend" configuration sets, which define different resource pool configurations for
different days of the week.
Creating a Configuration Set
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled for
dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click the Scheduling Rules tab.
4. Click Add Scheduling Rule.
5. In the Configuration Set field, select the Create New radio button.
6.
7. Click Add Configuration Set. The Add Configuration Set dialog displays.
a. Type a name in the Name field and select the configuration set to clone from the Clone from Configuration
Set drop-down.
b. Click OK. The configuration set is added to and selected in the Configuration Sets drop-down.
8. For each resource pool, click Edit.
a. Select a resource pool configuration set name.
b. Edit the pool properties and click OK.
9. Define one or more scheduling rules to specify when the configuration set is active.
Example Configuration Sets
The weekday configuration set assigns the production pool four times the resources of the development pool:
The weekend configuration set assigns the production and development pools an equal share of the resources:
Cloudera Administration | 249
Resource Management
The default configuration set assigns the production pool twice the resources of the development pool:
See example scheduling rules for these configuration sets.
Viewing the Properties of a Configuration Set
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled for
dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. In the Configuration Sets drop-down, select a configuration set. The properties of each pool for that configuration
set display.
Deleting a Configuration Set
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled for
dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. In the Configuration Sets drop-down, select a configuration set. The properties of each pool for that configuration
set display.
4. Click Delete.
Scheduling Rules
A scheduling rule defines when a configuration set is active. The configuration set is updated in affected services every
hour.
Example Scheduling Rules
Consider the example weekday and weekend configuration sets. To specify that the weekday configuration set is active
every weekday, weekend configuration set is active on the weekend (weekly on Saturday and Sunday), and the default
configuration set is active all other times, define the following rules:
250 | Cloudera Administration
Resource Management
Adding a Scheduling Rule
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled for
dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click the Scheduling Rules tab.
4. Click Add Scheduling Rule.
5. In the Configuration Set drop-down, select a configuration set.
6. Choose whether the rule should repeat, the repeat frequency, and if the frequency is weekly, the repeat day or
days.
7. If the schedule is not repeating, click the left side of the on field to display a drop-down calendar where you set
the starting date and time. When you specify the date and time, a default time window of two hours is set in the
right side of the on field. Click the right side to adjust the date and time.
8. Click OK.
Editing a Scheduling Rule
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled for
dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click Scheduling Rules.
4. Click Edit at the right of a rule.
5. Edit the rule as desired.
6. Click OK.
Deleting a Scheduling Rule
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled for
dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click Scheduling Rules.
4.
Click
at the right of a rule and select Delete.
5. Click OK.
Managing Impala Admission Control
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
Admission control is an Impala feature that imposes limits on concurrent SQL queries, to avoid resource usage spikes
and out-of-memory conditions on busy CDH clusters. It is a form of “throttling”. New queries are accepted and executed
Cloudera Administration | 251
Resource Management
until certain conditions are met, such as too many queries or too much total memory used across the cluster. When
one of these thresholds is reached, incoming queries wait to begin execution. These queries are queued and are
admitted (that is, begin executing) when the resources become available.
For further information on Impala admission control, see Admission Control and Query Queuing on page 255.
Enabling and Disabling Impala Admission Control Using Cloudera Manager
1.
2.
3.
4.
5.
6.
Go to the Impala service.
Click the Configuration tab.
Select Category > Admission Control.
Select or deselect the Enable Impala Admission Control checkbox.
Click Save Changes to commit the changes.
Restart the Impala service.
Configuring Impala Admission Control Using Cloudera Manager
1.
2.
3.
4.
Go to the Impala service.
Click the Configuration tab.
Select Category > Admission Control.
Configure admission control properties:
default_pool_max_queued
Purpose: Maximum number of requests allowed to be queued before rejecting requests. Because this limit applies
cluster-wide, but each Impala node makes independent decisions to run queries immediately or queue them, it
is a soft limit; the overall number of queued queries might be slightly higher during times of heavy load. A negative
value or 0 indicates requests are always rejected once the maximum concurrent requests are executing. Ignored
if fair_scheduler_config_path and llama_site_path are set.
Type: int64
Default: 200
default_pool_max_requests
Purpose: Maximum number of concurrent outstanding requests allowed to run before incoming requests are
queued. Because this limit applies cluster-wide, but each Impala node makes independent decisions to run queries
immediately or queue them, it is a soft limit; the overall number of concurrent queries might be slightly higher
during times of heavy load. A negative value indicates no limit. Ignored if fair_scheduler_config_path and
llama_site_path are set.
Type: int64
Default: 200
default_pool_mem_limit
Purpose: Maximum amount of memory (across the entire cluster) that all outstanding requests in this pool can
use before new requests to this pool are queued. Specified in bytes, megabytes, or gigabytes by a number followed
by the suffix b (optional), m, or g, either uppercase or lowercase. You can specify floating-point values for megabytes
and gigabytes, to represent fractional numbers such as 1.5. You can also specify it as a percentage of the physical
memory by specifying the suffix %. 0 or no setting indicates no limit. Defaults to bytes if no unit is given. Because
this limit applies cluster-wide, but each Impala node makes independent decisions to run queries immediately
or queue them, it is a soft limit; the overall memory used by concurrent queries might be slightly higher during
times of heavy load. Ignored if fair_scheduler_config_path and llama_site_path are set.
Note: Impala relies on the statistics produced by the COMPUTE STATS statement to estimate
memory usage for each query. See COMPUTE STATS Statement for guidelines about how and
when to use this statement.
252 | Cloudera Administration
Resource Management
Type: string
Default: "" (empty string, meaning unlimited)
disable_admission_control
Purpose: Turns off the admission control feature entirely, regardless of other configuration option settings.
Type: Boolean
Default: true
disable_pool_max_requests
Purpose: Disables all per-pool limits on the maximum number of running requests.
Type: Boolean
Default: false
disable_pool_mem_limits
Purpose: Disables all per-pool mem limits.
Type: Boolean
Default: false
fair_scheduler_allocation_path
Purpose: Path to the fair scheduler allocation file (fair-scheduler.xml).
Type: string
Default: "" (empty string)
Usage notes: Admission control only uses a small subset of the settings that can go in this file, as described below.
For details about all the Fair Scheduler configuration settings, see the Apache wiki.
llama_site_path
Purpose: Path to the Llama configuration file (llama-site.xml). If set, fair_scheduler_allocation_path
must also be set.
Type: string
Default: "" (empty string)
Usage notes: Admission control only uses a small subset of the settings that can go in this file, as described below.
For details about all the Llama configuration settings, see the documentation on Github.
queue_wait_timeout_ms
Purpose: Maximum amount of time (in milliseconds) that a request waits to be admitted before timing out.
Type: int64
Default: 60000
5. Click Save Changes to commit the changes.
6. Restart the Impala service.
Managing the Impala Llama ApplicationMaster
Note: Though Impala can be used together with YARN using simple configuration of static service
pools in Cloudera Manager, the use of the general-purpose component Llama for integrated resource
management within YARN is no longer supported with CDH 5.5 / Impala 2.3 and higher.
Cloudera Administration | 253
Resource Management
The Impala Llama ApplicationMaster (Llama) reserves and releases YARN-managed resources for Impala, thus reducing
resource management overhead when performing Impala queries. Llama is used when you want to enable integrated
resource management.
By default, YARN allocates resources bit-by-bit as needed by MapReduce jobs. Impala needs all resources available at
the same time, so that intermediate results can be exchanged between cluster nodes, and queries do not stall partway
through waiting for new resources to be allocated. Llama is the intermediary process that ensures all requested
resources are available before each Impala query actually begins.
For more information about Llama, see Llama - Low Latency Application MAster.
For information on enabling Llama high availability, see Llama High Availability on page 351.
Enabling Integrated Resource Management Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
The Enable Integrated Resource Management wizard enables cgroups for the all the hosts in the cluster running Impala
and YARN, adds one or more Llama roles to the Impala service, and configures the Impala and YARN services.
1. Start the wizard using one of the following paths:
• Cluster-level
1. Select Clusters > Cluster name > Dynamic Resource Pools.
2. In the Status section, click Enable.
• Service-level
1. Go to the Impala service.
2. Select Actions > Enable Integrated Resource Management.
The Enable Integrated Resource Management wizard starts and displays information about resource management
options and the actions performed by the wizard.
2. Click Continue.
3. Leave the Enable Cgroup-based Resource Management checkbox checked and click Continue.
4. Click the Impala Llama ApplicationMaster Hosts field to display a dialog for choosing Llama hosts.
The following shortcuts for specifying hostname patterns are supported:
• Range of hostnames (without the domain portion)
Range Definition
Matching Hosts
10.1.1.[1-4]
10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4
host[1-3].company.com
host1.company.com, host2.company.com, host3.company.com
host[07-10].company.com
host07.company.com, host08.company.com, host09.company.com,
host10.company.com
• IP addresses
• Rack name
5.
6.
7.
8.
Specify or select one or more hosts and click OK.
Click Continue. A progress screen displays with a summary of the wizard actions.
Click Continue.
Click Restart Now to restart the cluster and apply the configuration changes or click leave this wizard to restart
at a later time.
9. Click Finish.
254 | Cloudera Administration
Resource Management
Disabling Integrated Resource Management Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
The Enable Integrated Resource Management wizard enables cgroups for the all the hosts in the cluster running Impala
and YARN, adds one or more Llama roles to the Impala service, and configures the Impala and YARN services.
1. Start the wizard using one of the following paths:
• Cluster-level
1. Select Clusters > Cluster name > Dynamic Resource Pools.
2. In the Status section, click Disable.
• Service-level
1. Go to the Impala service.
2. Select Actions > Disable Integrated Resource Management.
The Disable Integrated Resource Management wizard starts and displays information about resource management
options and the actions performed by the wizard.
Configuring Llama Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
1.
2.
3.
4.
5.
6.
Go to the Impala service.
Click the Configuration tab.
Select Scope > Impala Llama ApplicationMaster.
Edit configuration properties.
Click Save Changes to commit the changes.
Restart the Llama role.
Impala Resource Management
Impala supports two types of resource management: independent and integrated. Independent resource management
is supported for CDH 4 and CDH 5 and is implemented by admission control. Integrated resource management is
supported for CDH 5 and is implemented by YARN and Llama.
Note: Though Impala can be used together with YARN via simple configuration of Static Service Pools
in Cloudera Manager, the use of the general-purpose component Llama for integrated resource
management within YARN is no longer supported with CDH 5.5 / Impala 2.3 and higher.
Admission Control and Query Queuing
Admission control is an Impala feature that imposes limits on concurrent SQL queries, to avoid resource usage spikes
and out-of-memory conditions on busy CDH clusters. It is a form of “throttling”. New queries are accepted and executed
until certain conditions are met, such as too many queries or too much total memory used across the cluster. When
one of these thresholds is reached, incoming queries wait to begin execution. These queries are queued and are
admitted (that is, begin executing) when the resources become available.
In addition to the threshold values for currently executing queries, you can place limits on the maximum number of
queries that are queued (waiting) and a limit on the amount of time they might wait before returning with an error.
These queue settings let you ensure that queries do not wait indefinitely, so that you can detect and correct “starvation”
scenarios.
Enable this feature if your cluster is underutilized at some times and overutilized at others. Overutilization is indicated
by performance bottlenecks and queries being cancelled due to out-of-memory conditions, when those same queries
Cloudera Administration | 255
Resource Management
are successful and perform well during times with less concurrent load. Admission control works as a safeguard to
avoid out-of-memory conditions during heavy concurrent usage.
Important:
• Cloudera strongly recommends you upgrade to CDH 5 or higher to use admission control. In CDH
4, admission control will only work if you do not have Hue deployed; unclosed Hue queries will
accumulate and exceed the queue size limit. On CDH 4, to use admission control, you must
explicitly enable it by specifying --disable_admission_control=false in the impalad
command-line options.
• Use the COMPUTE STATS statement for large tables involved in join queries, and follow other
steps from Tuning Impala for Performance to tune your queries. Although COMPUTE STATS is
an important statement to help optimize query performance, it is especially important when
admission control is enabled:
– When queries complete quickly and are tuned for optimal memory usage, there is less chance
of performance or capacity problems during times of heavy load.
– The admission control feature also relies on the statistics produced by the COMPUTE STATS
statement to generate accurate estimates of memory usage for complex queries. If the
estimates are inaccurate due to missing statistics, Impala might hold back queries
unnecessarily even though there is sufficient memory to run them, or might allow queries
to run that end up exceeding the memory limit and being cancelled.
Overview of Impala Admission Control
On a busy CDH cluster, you might find there is an optimal number of Impala queries that run concurrently. Because
Impala queries are typically I/O-intensive, you might not find any throughput benefit in running more concurrent
queries when the I/O capacity is fully utilized. Because Impala by default cancels queries that exceed the specified
memory limit, running multiple large-scale queries at once can result in having to re-run some queries that are cancelled.
The admission control feature lets you set a cluster-wide upper limit on the number of concurrent Impala queries and
on the memory used by those queries. Any additional queries are queued until the earlier ones finish, rather than
being cancelled or running slowly and causing contention. As other queries finish, the queued queries are allowed to
proceed.
For details on the internal workings of admission control, see How Impala Schedules and Enforces Limits on Concurrent
Queries on page 257.
How Impala Admission Control Relates to YARN
The admission control feature is similar in some ways to the YARN resource management framework, and they can be
used separately or together. This section describes some similarities and differences, to help you decide when to use
one, the other, or both together.
Admission control is a lightweight, decentralized system that is suitable for workloads consisting primarily of Impala
queries and other SQL statements. It sets “soft” limits that smooth out Impala memory usage during times of heavy
load, rather than taking an all-or-nothing approach that cancels jobs that are too resource-intensive.
Because the admission control system is not aware of other Hadoop workloads such as MapReduce jobs, you might
use YARN with static service pools on heterogeneous CDH 5 clusters where resources are shared between Impala and
other Hadoop components. Devote a percentage of cluster resources to Impala, allocate another percentage for
MapReduce and other batch-style workloads; let admission control handle the concurrency and memory usage for the
Impala work within the cluster, and let YARN manage the remainder of work within the cluster.
The Impala admission control feature uses the same configuration mechanism as the YARN resource manager to map
users to pools and authenticate them.
For full details about using Impala with YARN, see Integrated Resource Management with YARN on page 263.
256 | Cloudera Administration
Resource Management
How Impala Schedules and Enforces Limits on Concurrent Queries
The admission control system is decentralized, embedded in each Impala daemon and communicating through the
statestore mechanism. Although the limits you set for memory usage and number of concurrent queries apply
cluster-wide, each Impala daemon makes its own decisions about whether to allow each query to run immediately or
to queue it for a less-busy time. These decisions are fast, meaning the admission control mechanism is low-overhead,
but might be imprecise during times of heavy load. There could be times when the query queue contained more queries
than the specified limit, or when the estimated of memory usage for a query is not exact and the overall memory usage
exceeds the specified limit. Thus, you typically err on the high side for the size of the queue, because there is not a big
penalty for having a large number of queued queries; and you typically err on the low side for the memory limit, to
leave some headroom for queries to use more memory than expected, without being cancelled as a result.
At any time, the set of queued queries could include queries submitted through multiple different Impala daemon
hosts. All the queries submitted through a particular host will be executed in order, so a CREATE TABLE followed by
an INSERT on the same table would succeed. Queries submitted through different hosts are not guaranteed to be
executed in the order they were received. Therefore, if you are using load-balancing or other round-robin scheduling
where different statements are submitted through different hosts, set up all table structures ahead of time so that
the statements controlled by the queuing system are primarily queries, where order is not significant. Or, if a sequence
of statements needs to happen in strict order (such as an INSERT followed by a SELECT), submit all those statements
through a single session, while connected to the same Impala daemon host.
The limit on the number of concurrent queries is a “soft” one, To achieve high throughput, Impala makes quick decisions
at the host level about which queued queries to dispatch. Therefore, Impala might slightly exceed the limit from time
to time.
To avoid a large backlog of queued requests, you can also set an upper limit on the size of the queue for queries that
are delayed. When the number of queued queries exceeds this limit, further queries are cancelled rather than being
queued. You can also configure a timeout period, after which queued queries are cancelled, to avoid indefinite waits.
If a cluster reaches this state where queries are cancelled due to too many concurrent requests or long waits for query
execution to begin, that is a signal for an administrator to take action, either by provisioning more resources, scheduling
work on the cluster to smooth out the load, or by doing Impala performance tuning to enable higher throughput.
How Admission Control works with Impala Clients (JDBC, ODBC, HiveServer2)
Most aspects of admission control work transparently with client interfaces such as JDBC and ODBC:
• If a SQL statement is put into a queue rather than running immediately, the API call blocks until the statement is
dequeued and begins execution. At that point, the client program can request to fetch results, which might also
block until results become available.
• If a SQL statement is cancelled because it has been queued for too long or because it exceeded the memory limit
during execution, the error is returned to the client program with a descriptive error message.
If you want to submit queries to different resource pools through the REQUEST_POOL query option, as described in
REQUEST_POOL Query Option, In Impala 2.0 and higher you can change that query option through a SQL SET statement
that you submit from the client application, in the same session. Prior to Impala 2.0, that option was only settable for
a session through the impala-shell SET command, or cluster-wide through an impalad startup option.
Admission control has the following limitations or special behavior when used with JDBC or ODBC applications:
• The MEM_LIMIT query option, sometimes useful to work around problems caused by inaccurate memory estimates
for complicated queries, is only settable through the impala-shell interpreter and cannot be used directly
through JDBC or ODBC applications.
• Admission control does not use the other resource-related query options, RESERVATION_REQUEST_TIMEOUT or
V_CPU_CORES. Those query options only apply to the YARN resource management framework.
Configuring Admission Control
The configuration options for admission control range from the simple (a single resource pool with a single set of
options) to the complex (multiple resource pools with different options, each pool handling queries for a different set
of users and groups). You can configure the settings through the Cloudera Manager user interface, or on a system
without Cloudera Manager by editing configuration files or through startup options to the impalad daemon.
Cloudera Administration | 257
Resource Management
Impala Service Flags for Admission Control (Advanced)
The following Impala configuration options let you adjust the settings of the admission control feature. When supplying
the options on the command line, prepend the option name with --.
default_pool_max_queued
Purpose: Maximum number of requests allowed to be queued before rejecting requests. Because this limit applies
cluster-wide, but each Impala node makes independent decisions to run queries immediately or queue them, it is
a soft limit; the overall number of queued queries might be slightly higher during times of heavy load. A negative
value or 0 indicates requests are always rejected once the maximum concurrent requests are executing. Ignored if
fair_scheduler_config_path and llama_site_path are set.
Type: int64
Default: 200
default_pool_max_requests
Purpose: Maximum number of concurrent outstanding requests allowed to run before incoming requests are
queued. Because this limit applies cluster-wide, but each Impala node makes independent decisions to run queries
immediately or queue them, it is a soft limit; the overall number of concurrent queries might be slightly higher
during times of heavy load. A negative value indicates no limit. Ignored if fair_scheduler_config_path and
llama_site_path are set.
Type: int64
Default: 200
default_pool_mem_limit
Purpose: Maximum amount of memory (across the entire cluster) that all outstanding requests in this pool can use
before new requests to this pool are queued. Specified in bytes, megabytes, or gigabytes by a number followed by
the suffix b (optional), m, or g, either uppercase or lowercase. You can specify floating-point values for megabytes
and gigabytes, to represent fractional numbers such as 1.5. You can also specify it as a percentage of the physical
memory by specifying the suffix %. 0 or no setting indicates no limit. Defaults to bytes if no unit is given. Because
this limit applies cluster-wide, but each Impala node makes independent decisions to run queries immediately or
queue them, it is a soft limit; the overall memory used by concurrent queries might be slightly higher during times
of heavy load. Ignored if fair_scheduler_config_path and llama_site_path are set.
Note: Impala relies on the statistics produced by the COMPUTE STATS statement to estimate
memory usage for each query. See COMPUTE STATS Statement for guidelines about how and when
to use this statement.
Type: string
Default: "" (empty string, meaning unlimited)
disable_admission_control
Purpose: Turns off the admission control feature entirely, regardless of other configuration option settings.
Type: Boolean
Default: true
disable_pool_max_requests
Purpose: Disables all per-pool limits on the maximum number of running requests.
Type: Boolean
Default: false
disable_pool_mem_limits
Purpose: Disables all per-pool mem limits.
258 | Cloudera Administration
Resource Management
Type: Boolean
Default: false
fair_scheduler_allocation_path
Purpose: Path to the fair scheduler allocation file (fair-scheduler.xml).
Type: string
Default: "" (empty string)
Usage notes: Admission control only uses a small subset of the settings that can go in this file, as described below.
For details about all the Fair Scheduler configuration settings, see the Apache wiki.
llama_site_path
Purpose: Path to the Llama configuration file (llama-site.xml). If set, fair_scheduler_allocation_path
must also be set.
Type: string
Default: "" (empty string)
Usage notes: Admission control only uses a small subset of the settings that can go in this file, as described below.
For details about all the Llama configuration settings, see the documentation on Github.
queue_wait_timeout_ms
Purpose: Maximum amount of time (in milliseconds) that a request waits to be admitted before timing out.
Type: int64
Default: 60000
Configuring Admission Control Using Cloudera Manager
In Cloudera Manager, you can configure pools to manage queued Impala queries, and the options for the limit on
number of concurrent queries and how to handle queries that exceed the limit. For details, see Managing Resources
with Cloudera Manager.
See Examples of Admission Control Configurations on page 260 for a sample setup for admission control under Cloudera
Manager.
Configuring Admission Control Using the Command Line
If you do not use Cloudera Manager, you use a combination of startup options for the Impala daemon, and optionally
editing or manually constructing the configuration files fair-scheduler.xml and llama-site.xml.
For a straightforward configuration using a single resource pool named default, you can specify configuration options
on the command line and skip the fair-scheduler.xml and llama-site.xml configuration files.
For an advanced configuration with multiple resource pools using different settings, set up the fair-scheduler.xml
and llama-site.xml configuration files manually. Provide the paths to each one using the Impala daemon
command-line options, --fair_scheduler_allocation_path and --llama_site_path respectively.
The Impala admission control feature only uses the Fair Scheduler configuration settings to determine how to map
users and groups to different resource pools. For example, you might set up different resource pools with separate
memory limits, and maximum number of concurrent and queued queries, for different categories of users within your
organization. For details about all the Fair Scheduler configuration settings, see the Apache wiki.
The Impala admission control feature only uses a small subset of possible settings from the llama-site.xml
configuration file:
llama.am.throttling.maximum.placed.reservations.queue_name
llama.am.throttling.maximum.queued.reservations.queue_name
For details about all the Llama configuration settings, see Llama Default Configuration.
Cloudera Administration | 259
Resource Management
See Example Admission Control Configurations Using Configuration Files on page 261 for sample configuration files for
admission control using multiple resource pools, without Cloudera Manager.
Examples of Admission Control Configurations
Example Admission Control Configurations Using Cloudera Manager
For full instructions about configuring dynamic resource pools through Cloudera Manager, see Dynamic Resource Pools
on page 243. The following examples demonstrate some important points related to the Impala admission control
feature.
The following figure shows a sample of the Dynamic Resource Pools page in Cloudera Manager, accessed through the
Clusters > Cluster name > Resource Management > Dynamic Resource Pools menu choice and then the Configuration
tab. Numbers from all the resource pools are combined into the topmost root pool. The default pool is for users
who are not assigned any other pool by the user-to-pool mapping settings. The development and production pools
show how you can set different limits for different classes of users, for total memory, number of concurrent queries,
and number of queries that can be queued.
Figure 1: Sample Settings for Cloudera Manager Dynamic Resource Pools Page
The following figure shows a sample of the Placement Rules page in Cloudera Manager, accessed through the Clusters >
Cluster name > Resource Management > Dynamic Resource Pools menu choice and then the Configuration > Placement
Rules tabs. The settings demonstrate a reasonable configuration of a pool named default to service all requests
where the specified resource pool does not exist, is not explicitly set, or the user or group is not authorized for the
specified pool.
260 | Cloudera Administration
Resource Management
Figure 2: Sample Settings for Cloudera Manager Placement Rules Page
Example Admission Control Configurations Using Configuration Files
For clusters not managed by Cloudera Manager, here are sample fair-scheduler.xml and llama-site.xml files
that define resource pools equivalent to the ones in the preceding Cloudera Manager dialog. These sample files are
stripped down: in a real deployment they might contain other settings for use with various aspects of the YARN and
Llama components. The settings shown here are the significant ones for the Impala admission control feature.
fair-scheduler.xml:
Although Impala does not use the vcores value, you must still specify it to satisfy YARN requirements for the file
contents.
Each <aclSubmitApps> tag (other than the one for root) contains a comma-separated list of users, then a space,
then a comma-separated list of groups; these are the users and groups allowed to submit Impala statements to the
corresponding resource pool.
If you leave the <aclSubmitApps> element empty for a pool, nobody can submit directly to that pool; child pools
can specify their own <aclSubmitApps> values to authorize users and groups to submit to those pools.
<allocations>
<queue name="root">
<aclSubmitApps> </aclSubmitApps>
<queue name="default">
<maxResources>50000 mb, 0 vcores</maxResources>
<aclSubmitApps>*</aclSubmitApps>
</queue>
<queue name="development">
<maxResources>200000 mb, 0 vcores</maxResources>
<aclSubmitApps>user1,user2 dev,ops,admin</aclSubmitApps>
</queue>
<queue name="production">
<maxResources>1000000 mb, 0 vcores</maxResources>
<aclSubmitApps> ops,admin</aclSubmitApps>
</queue>
</queue>
Cloudera Administration | 261
Resource Management
<queuePlacementPolicy>
<rule name="specified" create="false"/>
<rule name="default" />
</queuePlacementPolicy>
</allocations>
llama-site.xml:
<?xml version="1.0" encoding="UTF-8"?>
<configuration>
<property>
<name>llama.am.throttling.maximum.placed.reservations.root.default</name>
<value>10</value>
</property>
<property>
<name>llama.am.throttling.maximum.queued.reservations.root.default</name>
<value>50</value>
</property>
<property>
<name>llama.am.throttling.maximum.placed.reservations.root.development</name>
<value>50</value>
</property>
<property>
<name>llama.am.throttling.maximum.queued.reservations.root.development</name>
<value>100</value>
</property>
<property>
<name>llama.am.throttling.maximum.placed.reservations.root.production</name>
<value>100</value>
</property>
<property>
<name>llama.am.throttling.maximum.queued.reservations.root.production</name>
<value>200</value>
</property>
</configuration>
Guidelines for Using Admission Control
To see how admission control works for particular queries, examine the profile output for the query. This information
is available through the PROFILE statement in impala-shell immediately after running a query in the shell, on the
queries page of the Impala debug web UI, or in the Impala log file (basic information at log level 1, more detailed
information at log level 2). The profile output contains details about the admission decision, such as whether the query
was queued or not and which resource pool it was assigned to. It also includes the estimated and actual memory usage
for the query, so you can fine-tune the configuration for the memory limits of the resource pools.
Where practical, use Cloudera Manager to configure the admission control parameters. The Cloudera Manager GUI is
much simpler than editing the configuration files directly.
Remember that the limits imposed by admission control are “soft” limits. Although the limits you specify for number
of concurrent queries and amount of memory apply cluster-wide, the decentralized nature of this mechanism means
that each Impala node makes its own decisions about whether to allow queries to run immediately or to queue them.
These decisions rely on information passed back and forth between nodes by the statestore service. If a sudden surge
in requests causes more queries than anticipated to run concurrently, then as a fallback, the overall Impala memory
limit and the Linux cgroups mechanism serve as hard limits to prevent overallocation of memory, by cancelling queries
if necessary.
If you have trouble getting a query to run because its estimated memory usage is too high, you can override the estimate
by setting the MEM_LIMIT query option in impala-shell, then issuing the query through the shell in the same session.
The MEM_LIMIT value is treated as the estimated amount of memory, overriding the estimate that Impala would
generate based on table and column statistics. This value is used only for making admission control decisions, and is
not pre-allocated by the query.
In impala-shell, you can also specify which resource pool to direct queries to by setting the REQUEST_POOL query
option. (This option was named YARN_POOL during the CDH 5 beta period.)
262 | Cloudera Administration
Resource Management
The statements affected by the admission control feature are primarily queries, but also include statements that write
data such as INSERT and CREATE TABLE AS SELECT. Most write operations in Impala are not resource-intensive,
but inserting into a Parquet table can require substantial memory due to buffering 1 GB of data before writing out
each Parquet data block. See Loading Data into Parquet Tables for instructions about inserting data efficiently into
Parquet tables.
Although admission control does not scrutinize memory usage for other kinds of DDL statements, if a query is queued
due to a limit on concurrent queries or memory usage, subsequent statements in the same session are also queued
so that they are processed in the correct order:
-- This query could be queued to avoid out-of-memory at times of heavy load.
select * from huge_table join enormous_table using (id);
-- If so, this subsequent statement in the same session is also queued
-- until the previous statement completes.
drop table huge_table;
If you set up different resource pools for different users and groups, consider reusing any classifications and hierarchy
you developed for use with Sentry security. See Enabling Sentry Authorization for Impala for details.
For details about all the Fair Scheduler configuration settings, see Fair Scheduler Configuration, in particular the tags
such as <queue> and <aclSubmitApps> to map users and groups to particular resource pools (queues).
Integrated Resource Management with YARN
You can limit the CPU and memory resources used by Impala, to manage and prioritize workloads on clusters that run
jobs from many Hadoop components.
Requests from Impala to YARN go through an intermediary service called Llama. When the resource requests are
granted, Impala starts the query and places all relevant execution threads into the cgroup containers and sets up the
memory limit on each host. If sufficient resources are not available, the Impala query waits until other jobs complete
and the resources are freed. During query processing, as the need for additional resources arises, Llama can “expand”
already-requested resources, to avoid over-allocating at the start of the query.
After a query is finished, Llama caches the resources (for example, leaving memory allocated) in case they are needed
for subsequent Impala queries. This caching mechanism avoids the latency involved in making a whole new set of
resource requests for each query. If the resources are needed by YARN for other types of jobs, Llama returns them.
While the delays to wait for resources might make individual queries seem less responsive on a heavily loaded cluster,
the resource management feature makes the overall performance of the cluster smoother and more predictable,
without sudden spikes in utilization due to memory paging, CPUs pegged at 100%, and so on.
Note: Though Impala can be used together with YARN via simple configuration of Static Service Pools
in Cloudera Manager, the use of the general-purpose component Llama for integrated resource
management within YARN is no longer supported with CDH 5.5 / Impala 2.3 and higher.
The Llama Daemon
Llama is a system that mediates resource management between Impala and Hadoop YARN. Llama enables Impala to
reserve, use, and release resource allocations in a Hadoop cluster. Llama is only required if resource management is
enabled in Impala.
By default, YARN allocates resources bit-by-bit as needed by MapReduce jobs. Impala needs all resources available at
the same time, so that intermediate results can be exchanged between cluster nodes, and queries do not stall partway
through waiting for new resources to be allocated. Llama is the intermediary process that ensures all requested
resources are available before each Impala query actually begins.
For management through Cloudera Manager, see The Impala Llama ApplicationMaster.
How Resource Limits Are Enforced
• If Cloudera Manager Static Partitioning is used, it creates a cgroup in which Impala runs. This cgroup limits CPU,
network, and IO according to the static partitioning policy.
Cloudera Administration | 263
Resource Management
• Limits on memory usage are enforced by Impala's process memory limit (the MEM_LIMIT query option setting).
The admission control feature checks this setting to decide how many queries can be safely run at the same time.
Then the Impala daemon enforces the limit by activating the spill-to-disk mechanism when necessary, or cancelling
a query altogether if the limit is exceeded at runtime.
impala-shell Query Options for Resource Management
Before issuing SQL statements through the impala-shell interpreter, you can use the SET command to configure
the following parameters related to resource management:
• EXPLAIN_LEVEL Query Option
• MEM_LIMIT Query Option
Limitations of Resource Management for Impala
The MEM_LIMIT query option, and the other resource-related query options, are settable through the ODBC or JDBC
interfaces in Impala 2.0 and higher. This is a former limitation that is now lifted.
264 | Cloudera Administration
Performance Management
Performance Management
This section describes mechanisms and best practices for improving performance.
Related Information
• Tuning Impala for Performance
• Tuning YARN on page 281
Optimizing Performance in CDH
This section provides solutions to some performance problems, and describes configuration best practices.
Important: Work with your network administrators and hardware vendors to ensure that you have
the proper NIC firmware, drivers, and configurations in place and that your network performs properly.
Cloudera recognizes that network setup and upgrade are challenging problems, and will do its best
to share useful experiences.
Disabling Transparent Hugepage Compaction
Most Linux platforms supported by CDH 5 include a feature called transparent hugepage compaction which interacts
poorly with Hadoop workloads and can seriously degrade performance.
Symptom: top and other system monitoring tools show a large percentage of the CPU usage classified as "system
CPU". If system CPU usage is 30% or more of the total CPU usage, your system may be experiencing this issue.
What to do:
Note: In the following instructions, defrag_file_pathname depends on your operating system:
• Red Hat/CentOS: /sys/kernel/mm/redhat_transparent_hugepage/defrag
• Ubuntu/Debian, OEL, SLES: /sys/kernel/mm/transparent_hugepage/defrag
1. To see whether transparent hugepage compaction is enabled, run the following command and check the output:
$ cat defrag_file_pathname
• [always] never means that transparent hugepage compaction is enabled.
• always [never] means that transparent hugepage compaction is disabled.
2. To disable transparent hugepage compaction, add the following command to /etc/rc.local:
echo never > defrag_file_pathname
You can also disable transparent hugepage compaction interactively (but remember this will not survive a reboot).
To disable transparent hugepage compaction temporarily as root:
# echo 'never' > defrag_file_pathname
To disable transparent hugepage compaction temporarily using sudo:
$ sudo sh -c "echo 'never' > defrag_file_pathname"
Cloudera Administration | 265
Performance Management
Setting the vm.swappiness Linux Kernel Parameter
The Linux kernel parameter, vm.swappiness, is a value from 0-100 that controls the swapping of application data
(as anonymous pages) from physical memory to virtual memory on disk. The higher the value, the more aggressively
inactive processes are swapped out from physical memory. The lower the value, the less they are swapped, forcing
filesystem buffers to be emptied.
On most systems, vm.swappiness is set to 60 by default. This is not suitable for Hadoop clusters because processes
are sometimes swapped even when enough memory is available. This can cause lengthy garbage collection pauses for
important system daemons, affecting stability and performance.
Cloudera recommends that you set vm.swappiness to a value between 1 and 10, preferably 1, for minimum swapping.
To view your current setting for vm.swappiness, run:
cat /proc/sys/vm/swappiness
To set vm.swappiness to 1, run:
sudo sysctl -w vm.swappiness=1
Note: Cloudera previously recommended setting vm.swappiness to 0. However, a change in Linux
kernel 3.5-rc1 (fe35004f), can lead to frequent out of memory (OOM) errors. For details, see Evan
Klitzke's blog post. This commit was backported to RHEL / CentOS 6.4 and Ubuntu 12.04 LTS (Long
Term Support).
Improving Performance in Shuffle Handler and IFile Reader
The MapReduce shuffle handler and IFile reader use native Linux calls, (posix_fadvise(2) and sync_data_range),
on Linux systems with Hadoop native libraries installed.
Shuffle Handler
You can improve MapReduce shuffle handler performance by enabling shuffle readahead. This causes the TaskTracker
or Node Manager to pre-fetch map output before sending it over the socket to the reducer.
• To enable this feature for YARN, set mapreduce.shuffle.manage.os.cache, to true (default). To further
tune performance, adjust the value of mapreduce.shuffle.readahead.bytes. The default value is 4 MB.
• To enable this feature for MapReduce, set the mapred.tasktracker.shuffle.fadvise to true (default).
To further tune performance, adjust the value of mapred.tasktracker.shuffle.readahead.bytes. The
default value is 4 MB.
IFile Reader
Enabling IFile readahead increases the performance of merge operations. To enable this feature for either MRv1 or
YARN, set mapreduce.ifile.readahead to true (default). To further tune the performance, adjust the value of
mapreduce.ifile.readahead.bytes. The default value is 4MB.
Best Practices for MapReduce Configuration
The configuration settings described below can reduce inherent latencies in MapReduce execution. You set these
values in mapred-site.xml.
Send a heartbeat as soon as a task finishes
Set mapreduce.tasktracker.outofband.heartbeat to true for TaskTracker to send an out-of-band heartbeat
on task completion to reduce latency. The default value is false:
<property>
<name>mapreduce.tasktracker.outofband.heartbeat</name>
266 | Cloudera Administration
Performance Management
<value>true</value>
</property>
Reduce the interval for JobClient status reports on single node systems
The jobclient.progress.monitor.poll.interval property defines the interval (in milliseconds) at which
JobClient reports status to the console and checks for job completion. The default value is 1000 milliseconds; you may
want to set this to a lower value to make tests run faster on a single-node cluster. Adjusting this value on a large
production cluster may lead to unwanted client-server traffic.
<property>
<name>jobclient.progress.monitor.poll.interval</name>
<value>10</value>
</property>
Tune the JobTracker heartbeat interval
Tuning the minimum interval for the TaskTracker-to-JobTracker heartbeat to a smaller value may improve MapReduce
performance on small clusters.
<property>
<name>mapreduce.jobtracker.heartbeat.interval.min</name>
<value>10</value>
</property>
Start MapReduce JVMs immediately
The mapred.reduce.slowstart.completed.maps property specifies the proportion of Map tasks in a job that
must be completed before any Reduce tasks are scheduled. For small jobs that require fast turnaround, setting this
value to 0 can improve performance; larger values (as high as 50%) may be appropriate for larger jobs.
<property>
<name>mapred.reduce.slowstart.completed.maps</name>
<value>0</value>
</property>
Tips and Best Practices for Jobs
This section describes changes you can make at the job level.
Use the Distributed Cache to Transfer the Job JAR
Use the distributed cache to transfer the job JAR rather than using the JobConf(Class) constructor and the
JobConf.setJar() and JobConf.setJarByClass() methods.
To add JARs to the classpath, use -libjars jar1,jar2. This copies the local JAR files to HDFS and uses the distributed
cache mechanism to ensure they are available on the task nodes and added to the task classpath.
The advantage of this, over JobConf.setJar, is that if the JAR is on a task node, it does not need to be copied again
if a second task from the same job runs on that node, though it will still need to be copied from the launch machine
to HDFS.
Note: -libjars works only if your MapReduce driver uses ToolRunner. If it does not, you would
need to use the DistributedCache APIs (Cloudera does not recommend this).
For more information, see item 1 in the blog post How to Include Third-Party Libraries in Your MapReduce Job.
Changing the Logging Level on a Job (MRv1)
You can change the logging level for an individual job. You do this by setting the following properties in the job
configuration (JobConf):
• mapreduce.map.log.level
Cloudera Administration | 267
Performance Management
• mapreduce.reduce.log.level
Valid values are NONE, INFO, WARN, DEBUG, TRACE, and ALL.
Example:
JobConf conf = new JobConf();
...
conf.set("mapreduce.map.log.level", "DEBUG");
conf.set("mapreduce.reduce.log.level", "TRACE");
...
Choosing a Data Compression Format
Whether to compress your data and which compression formats to use can have a significant impact on performance.
Two of the most important places to consider data compression are in terms of MapReduce jobs and data stored in
HBase. For the most part, the principles are similar for each.
General Guidelines
• You need to balance the processing capacity required to compress and uncompress the data, the disk IO required
to read and write the data, and the network bandwidth required to send the data across the network. The correct
balance of these factors depends upon the characteristics of your cluster and your data, as well as your usage
patterns.
• Compression is not recommended if your data is already compressed (such as images in JPEG format). In fact, the
resulting file can actually be larger than the original.
• GZIP compression uses more CPU resources than Snappy or LZO, but provides a higher compression ratio. GZip
is often a good choice for cold data, which is accessed infrequently. Snappy or LZO are a better choice for hot
data, which is accessed frequently.
• BZip2 can also produce more compression than GZip for some types of files, at the cost of some speed when
compressing and decompressing. HBase does not support BZip2 compression.
• Snappy often performs better than LZO. It is worth running tests to see if you detect a significant difference.
• For MapReduce, if you need your compressed data to be splittable, BZip2 and LZO formats can be split. Snappy
and GZip blocks are not splittable, but files with Snappy blocks inside a container file format such as SequenceFile
or Avro can be split. Snappy is intended to be used with a container format, like SequenceFiles or Avro data files,
rather than being used directly on plain text, for example, since the latter is not splittable and cannot be processed
in parallel using MapReduce. Splittability is not relevant to HBase data.
• For MapReduce, you can compress either the intermediate data, the output, or both. Adjust the parameters you
provide for the MapReduce job accordingly. The following examples compress both the intermediate data and
the output. MR2 is shown first, followed by MR1.
– MRv2
hadoop jar hadoop-examples-.jar sort "-Dmapreduce.compress.map.output=true"
"-Dmapreduce.map.output.compression.codec=org.apache.hadoop.io.compress.GzipCodec"
"-Dmapreduce.output.compress=true"
"-Dmapreduce.output.compression.codec=org.apache.hadoop.io.compress.GzipCodec" -outKey
org.apache.hadoop.io.Text -outValue org.apache.hadoop.io.Text input output
– MRv1
hadoop jar hadoop-examples-.jar sort "-Dmapred.compress.map.output=true"
"-Dmapred.map.output.compression.codec=org.apache.hadoop.io.compress.GzipCodec"
"-Dmapred.output.compress=true"
"-Dmapred.output.compression.codec=org.apache.hadoop.io.compress.GzipCodec" -outKey
org.apache.hadoop.io.Text -outValue org.apache.hadoop.io.Text input output
268 | Cloudera Administration
Performance Management
Configuring Data Compression Using Cloudera Manager
To configure support for LZO using Cloudera Manager, you must install the GPL Extras parcel, then configure services
to use it. See Installing the GPL Extras Parcel and Configuring Services to Use the GPL Extras Parcel on page 234.
Configuring Data Compression Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
To configure support for LZO in CDH, see Step 5: (Optional) Install LZO and Configuring LZO. Snappy support is included
in CDH.
To use Snappy in a MapReduce job, see Using Snappy for MapReduce Compression. Use the same method for LZO,
with the codec com.hadoop.compression.lzo.LzopCodec instead.
Further Reading
For more information about compression algorithms in Hadoop, see Hadoop: The Definitive Guide by Tom White.
Tuning the Solr Server
Solr performance tuning is a complex task. The following sections provide more details.
Tuning to Complete During Setup
Some tuning is best completed during the setup of you system or may require some re-indexing.
Configuring Lucene Version Requirements
You can configure Solr to use a specific version of Lucene. This can help ensure that the Lucene version that Search
uses includes the latest features and bug fixes. At the time that a version of Solr ships, Solr is typically configured to
use the appropriate Lucene version, in which case there is no need to change this setting. If a subsequent Lucene
update occurs, you can configure the Lucene version requirements by directly editing the luceneMatchVersion
element in the solrconfig.xml file. Versions are typically of the form x.y, such as 4.4. For example, to specify
version 4.4, you would ensure the following setting exists in solrconfig.xml:
<luceneMatchVersion>4.4</luceneMatchVersion>
Designing the Schema
When constructing a schema, use data types that most accurately describe the data that the fields will contain. For
example:
• Use the tdate type for dates. Do this instead of representing dates as strings.
• Consider using the text type that applies to your language, instead of using String. For example, you might use
text_en. Text types support returning results for subsets of an entry. For example, querying on "john" would
find "John Smith", whereas with the string type, only exact matches are returned.
• For IDs, use the string type.
General Tuning
The following tuning categories can be completed at any time. It is less important to implement these changes before
beginning to use your system.
Cloudera Administration | 269
Performance Management
General Tips
• Enabling multi-threaded faceting can provide better performance for field faceting. When multi-threaded faceting
is enabled, field faceting tasks are completed in a parallel with a thread working on every field faceting task
simultaneously. Performance improvements do not occur in all cases, but improvements are likely when all of the
following are true:
– The system uses highly concurrent hardware.
– Faceting operations apply to large data sets over multiple fields.
– There is not an unusually high number of queries occurring simultaneously on the system. Systems that are
lightly loaded or that are mainly engaged with ingestion and indexing may be helped by multi-threaded
faceting; for example, a system ingesting articles and being queried by a researcher. Systems heavily loaded
by user queries are less likely to be helped by multi-threaded faceting; for example, an e-commerce site with
heavy user-traffic.
Note: Multi-threaded faceting only applies to field faceting and not to query faceting.
• Field faceting identifies the number of unique entries for a field. For example, multi-threaded
faceting could be used to simultaneously facet for the number of unique entries for the
fields, "color" and "size". In such a case, there would be two threads, and each thread would
work on faceting one of the two fields.
• Query faceting identifies the number of unique entries that match a query for a field. For
example, query faceting could be used to find the number of unique entries in the "size"
field are between 1 and 5. Multi-threaded faceting does not apply to these operations.
To enable multi-threaded faceting, add facet-threads to queries. For example, to use up to 1000 threads, you
might use a query as follows:
http://localhost:8983/solr/collection1/select?q=*:*&facet=true&fl=id&facet.field=f0_ws&facet.threads=1000
If facet-threads is omitted or set to 0, faceting is single-threaded. If facet-threads is set to a negative value,
such as -1, multi-threaded faceting will use as many threads as there are fields to facet up to the maximum number
of threads possible on the system.
• If your environment does not require Near Real Time (NRT), turn off soft auto-commit in solrconfig.xml.
• In most cases, do not change the default batch size setting of 1000. If you are working with especially large
documents, you may consider decreasing the batch size.
• To help identify any garbage collector (GC) issues, enable GC logging in production. The overhead is low and the
JVM supports GC log rolling as of 1.6.0_34.
– The minimum recommended GC logging flags are: -XX:+PrintGCTimeStamps -XX:+PrintGCDateStamps
-XX:+PrintGCDetails.
– To rotate the GC logs: -Xloggc: -XX:+UseGCLogFileRotation -XX:NumberOfGCLogFiles=
-XX:GCLogFileSize=.
Solr and HDFS - the Block Cache
Cloudera Search enables Solr to store indexes in an HDFS filesystem. To maintain performance, an HDFS block cache
has been implemented using Least Recently Used (LRU) semantics. This enables Solr to cache HDFS index files on read
and write, storing the portions of the file in JVM "direct memory" (meaning off heap) by default or optionally in the
JVM heap.
Batch jobs typically do not make use of the cache, while Solr servers (when serving queries or indexing documents)
should. When running indexing using MapReduce, the MR jobs themselves do not make use of the block cache. Block
caching is turned off by default and should be left disabled.
Tuning of this cache is complex and best practices are continually being refined. In general, allocate a cache that is
about 10-20% of the amount of memory available on the system. For example, when running HDFS and Solr on a host
270 | Cloudera Administration
Performance Management
with 50 GB of memory, typically allocate 5-10 GB of memory using solr.hdfs.blockcache.slab.count. As index
sizes grow you may need to tune this parameter to maintain optimal performance.
Note: Block cache metrics are currently unavailable.
Configuration
The following parameters control caching. They can be configured at the Solr process level by setting the respective
system property or by editing the solrconfig.xml directly.
Parameter
Default
Description
solr.hdfs.blockcache.enabled
true
Enable the block cache.
solr.hdfs.blockcache.read.enabled
true
Enable the read cache.
solr.hdfs.blockcache.write.enabled
false
Enable the write cache.
solr.hdfs.blockcache.direct.memory.allocation true
Enable direct memory allocation. If this is
false, heap is used.
solr.hdfs.blockcache.slab.count
1
Number of memory slabs to allocate. Each
slab is 128 MB in size.
solr.hdfs.blockcache.global
true
If enabled, one HDFS block cache is used
for each collection on a host. If
blockcache.global is disabled, each
SolrCore on a host creates its own private
HDFS block cache. Enabling this parameter
simplifies managing HDFS block cache
memory.
Note:
Increasing the direct memory cache size may make it necessary to increase the maximum direct
memory size allowed by the JVM. Each Solr slab allocates the slab's memory, which is 128 MB by
default, as well as allocating some additional direct memory overhead. Therefore, ensure that the
MaxDirectMemorySize is set comfortably above the value expected for slabs alone. The amount of
additional memory required varies according to multiple factors, but for most cases, setting
MaxDirectMemorySize to at least 20-30% more than the total memory configured for slabs is
sufficient. Setting the MaxDirectMemorySize to the number of slabs multiplied by the slab size does
not provide enough memory.
To set MaxDirectMemorySize using Cloudera Manager
1.
2.
3.
4.
5.
Go to the Solr service.
Click the Configuration tab.
In the Search box, type Java Direct Memory Size of Solr Server in Bytes.
Set the new direct memory value.
Restart Solr servers after editing the parameter.
Solr HDFS optimizes caching when performing NRT indexing using Lucene's NRTCachingDirectory.
Lucene caches a newly created segment if both of the following conditions are true:
• The segment is the result of a flush or a merge and the estimated size of the merged segment is <=
solr.hdfs.nrtcachingdirectory.maxmergesizemb.
• The total cached bytes is <= solr.hdfs.nrtcachingdirectory.maxcachedmb.
Cloudera Administration | 271
Performance Management
The following parameters control NRT caching behavior:
Parameter
Default
Description
solr.hdfs.nrtcachingdirectory.enable
true
Whether to enable the
NRTCachingDirectory.
solr.hdfs.nrtcachingdirectory.maxcachedmb
192
Size of the cache in megabytes.
solr.hdfs.nrtcachingdirectory.maxmergesizemb 16
Maximum segment size to cache.
Here is an example of solrconfig.xml with defaults:
<directoryFactory name="DirectoryFactory">
<bool name="solr.hdfs.blockcache.enabled">${solr.hdfs.blockcache.enabled:true}</bool>
<int name="solr.hdfs.blockcache.slab.count">${solr.hdfs.blockcache.slab.count:1}</int>
<bool
name="solr.hdfs.blockcache.direct.memory.allocation">${solr.hdfs.blockcache.direct.memory.allocation:true}</bool>
<int
name="solr.hdfs.blockcache.blocksperbank">${solr.hdfs.blockcache.blocksperbank:16384}</int>
<bool
name="solr.hdfs.blockcache.read.enabled">${solr.hdfs.blockcache.read.enabled:true}</bool>
<bool
name="solr.hdfs.blockcache.write.enabled">${solr.hdfs.blockcache.write.enabled:true}</bool>
<bool
name="solr.hdfs.nrtcachingdirectory.enable">${solr.hdfs.nrtcachingdirectory.enable:true}</bool>
<int
name="solr.hdfs.nrtcachingdirectory.maxmergesizemb">${solr.hdfs.nrtcachingdirectory.maxmergesizemb:16}</int>
<int
name="solr.hdfs.nrtcachingdirectory.maxcachedmb">${solr.hdfs.nrtcachingdirectory.maxcachedmb:192}</int>
</directoryFactory>
The following example illustrates passing Java options by editing the /etc/default/solr or
/opt/cloudera/parcels/CDH-*/etc/default/solr configuration file:
CATALINA_OPTS="-Xmx10g -XX:MaxDirectMemorySize=20g -XX:+UseLargePages
-Dsolr.hdfs.blockcache.slab.count=100"
For better performance, Cloudera recommends setting the Linux swap space on all Solr server hosts as shown below:
# minimize swappiness
sudo sysctl vm.swappiness=1
sudo bash -c 'echo "vm.swappiness=1">> /etc/sysctl.conf'
# disable swap space until next reboot:
sudo /sbin/swapoff -a
Note: Cloudera previously recommended setting vm.swappiness to 0. However, a change in Linux
kernel 3.5-rc1 (fe35004f), can lead to frequent out of memory (OOM) errors. For details, see Evan
Klitzke's blog post. This commit was backported to RHEL / CentOS 6.4 and Ubuntu 12.04 LTS (Long
Term Support).
Threads
Configure the Tomcat server to have more threads per Solr instance. Note that this is only effective if your hardware
is sufficiently powerful to accommodate the increased threads. 10,000 threads is a reasonable number to try in many
cases.
272 | Cloudera Administration
Performance Management
To change the maximum number of threads, add a maxThreads element to the Connector definition in the Tomcat
server's server.xml configuration file. For example, if you installed Search for CDH 5 using parcels installation, you
would modify the Connector definition in the <parcel
path>/CDH/etc/solr/tomcat-conf.dist/conf/server.xml file so this:
<Connector port="${solr.port}" protocol="HTTP/1.1"
connectionTimeout="20000"
redirectPort="8443" />
Becomes this:
<Connector port="${solr.port}" protocol="HTTP/1.1"
maxThreads="10000"
connectionTimeout="20000"
redirectPort="8443" />
Garbage Collection
Choose different garbage collection options for best performance in different environments. Some garbage collection
options typically chosen include:
• Concurrent low pause collector: Use this collector in most cases. This collector attempts to minimize "Stop the
World" events. Avoiding these events can reduce connection timeouts, such as with ZooKeeper, and may improve
user experience. This collector is enabled using -XX:+UseConcMarkSweepGC.
• Throughput collector: Consider this collector if raw throughput is more important than user experience. This
collector typically uses more "Stop the World" events so this may negatively affect user experience and connection
timeouts such as ZooKeeper heartbeats. This collector is enabled using -XX:+UseParallelGC. If UseParallelGC
"Stop the World" events create problems, such as ZooKeeper timeouts, consider using the UseParNewGC collector
as an alternative collector with similar throughput benefits.
You can also affect garbage collection behavior by increasing the Eden space to accommodate new objects. With
additional Eden space, garbage collection does not need to run as frequently on new objects.
Replicated Data
You can adjust the degree to which different data is replicated.
Replicas
If you have sufficient additional hardware, add more replicas for a linear boost of query throughput. Note that adding
replicas may slow write performance on the first replica, but otherwise this should have minimal negative consequences.
Transaction Log Replication
Beginning with CDH 5.4.1, Search for CDH supports configurable transaction log replication levels for replication logs
stored in HDFS.
Configure the replication factor by modifying the tlogDfsReplication setting in solrconfig.xml. The
tlogDfsReplication is a new setting in the updateLog settings area. An excerpt of the solrconfig.xml file
where the transaction log replication factor is set is as follows:
<updateHandler class="solr.DirectUpdateHandler2">
<!-- Enables a transaction log, used for real-time get,
and solr cloud replica recovery. The log can grow
uncommitted changes to the index, so use of a hard
is recommended (see below).
"dir" - the target directory for transaction logs,
solr data directory. -->
<updateLog>
<str name="dir">${solr.ulog.dir:}</str>
<int name="tlogDfsReplication">3</int>
</updateLog>
durability, and
as big as
autoCommit
defaults to the
Cloudera Administration | 273
Performance Management
You might want to increase the replication level from the default level of 1 to some higher value such as 3. Increasing
the transaction log replication level can:
• Reduce the chance of data loss, especially when the system is otherwise configured to have single replicas of
shards. For example, having single replicas of shards is reasonable when autoAddReplicas is enabled, but
without additional transaction log replicas, the risk of data loss during a node failure would increase.
• Facilitate rolling upgrade of HDFS while Search is running. If you have multiple copies of the log, when a node with
the transaction log becomes unavailable during the rolling upgrade process, another copy of the log can continue
to collect transactions.
• Facilitate HDFS write lease recovery.
Initial testing shows no significant performance regression for common use cases.
Shards
In some cases, oversharding can help improve performance including intake speed. If your environment includes
massively parallel hardware and you want to use these available resources, consider oversharding. You might increase
the number of replicas per host from 1 to 2 or 3. Making such changes creates complex interactions, so you should
continue to monitor your system's performance to ensure that the benefits of oversharding do not outweigh the costs.
Commits
Changing commit values may improve performance in some situation. These changes result in tradeoffs and may not
be beneficial in all cases.
• For hard commit values, the default value of 60000 (60 seconds) is typically effective, though changing this value
to 120 seconds may improve performance in some cases. Note that setting this value to higher values, such as
600 seconds may result in undesirable performance tradeoffs.
• Consider increasing the auto-commit value from 15000 (15 seconds) to 120000 (120 seconds).
• Enable soft commits and set the value to the largest value that meets your requirements. The default value of
1000 (1 second) is too aggressive for some environments.
Other Resources
• General information on Solr caching is available on the SolrCaching page on the Solr Wiki.
• Information on issues that influence performance is available on the SolrPerformanceFactors page on the Solr
Wiki.
• Resource Management describes how to use Cloudera Manager to manage resources, for example with Linux
cgroups.
• For information on improving querying performance, see How to make searching faster.
• For information on improving indexing performance, see How to make indexing faster.
Tuning Spark Applications
This topic describes various aspects in tuning Spark applications. During tuning you should monitor application behavior
to determine the effect of tuning actions.
For information on monitoring Spark applications, see Monitoring Spark Applications.
Shuffle Overview
A Spark dataset comprises a fixed number of partitions, each of which comprises a number of records. For the datasets
returned by narrow transformations, such as map and filter, the records required to compute the records in a single
partition reside in a single partition in the parent dataset. Each object is only dependent on a single object in the parent.
Operations such as coalesce can result in a task processing multiple input partitions, but the transformation is still
considered narrow because the input records used to compute any single output record can still only reside in a limited
subset of the partitions.
274 | Cloudera Administration
Performance Management
Spark also supports transformations with wide dependencies, such as groupByKey and reduceByKey. In these
dependencies, the data required to compute the records in a single partition can reside in many partitions of the parent
dataset. To perform these transformations, all of the tuples with the same key must end up in the same partition,
processed by the same task. To satisfy this requirement, Spark performs a shuffle, which transfers data around the
cluster and results in a new stage with a new set of partitions.
For example, consider the following code:
sc.textFile("someFile.txt").map(mapFunc).flatMap(flatMapFunc).filter(filterFunc).count()
It runs a single action, count, which depends on a sequence of three transformations on a dataset derived from a text
file. This code runs in a single stage, because none of the outputs of these three transformations depend on data that
comes from different partitions than their inputs.
In contrast, this Scala code finds how many times each character appears in all the words that appear more than 1,000
times in a text file:
val tokenized = sc.textFile(args(0)).flatMap(_.split(' '))
val wordCounts = tokenized.map((_, 1)).reduceByKey(_ + _)
val filtered = wordCounts.filter(_._2 >= 1000)
val charCounts = filtered.flatMap(_._1.toCharArray).map((_, 1)).reduceByKey(_ + _)
charCounts.collect()
This example has three stages. The two reduceByKey transformations each trigger stage boundaries, because computing
their outputs requires repartitioning the data by keys.
A final example is this more complicated transformation graph, which includes a join transformation with multiple
dependencies:
The pink boxes show the resulting stage graph used to run it:
At each stage boundary, data is written to disk by tasks in the parent stages and then fetched over the network by
tasks in the child stage. Because they incur high disk and network I/O, stage boundaries can be expensive and should
be avoided when possible. The number of data partitions in a parent stage may be different than the number of
partitions in a child stage. Transformations that can trigger a stage boundary typically accept a numPartitions
argument, which specifies into how many partitions to split the data in the child stage. Just as the number of reducers
Cloudera Administration | 275
Performance Management
is an important parameter in MapReduce jobs, the number of partitions at stage boundaries can determine an
application's performance. Tuning the Number of Partitions on page 279 describes how to tune this number.
Choosing Transformations to Minimize Shuffles
You can usually choose from many arrangements of actions and transformations that produce the same results.
However, not all these arrangements result in the same performance. Avoiding common pitfalls and picking the right
arrangement can significantly improve an application's performance.
When choosing an arrangement of transformations, minimize the number of shuffles and the amount of data shuffled.
Shuffles are expensive operations; all shuffle data must be written to disk and then transferred over the network.
repartition , join, cogroup, and any of the *By or *ByKey transformations can result in shuffles. Not all these
transformations are equal, however, and you should avoid the following patterns:
• groupByKey when performing an associative reductive operation. For example,
rdd.groupByKey().mapValues(_.sum) produces the same result as rdd.reduceByKey(_ + _). However,
the former transfers the entire dataset across the network, while the latter computes local sums for each key in
each partition and combines those local sums into larger sums after shuffling.
• reduceByKey when the input and output value types are different. For example, consider writing a transformation
that finds all the unique strings corresponding to each key. You could use map to transform each element into a
Set and then combine the Sets with reduceByKey:
rdd.map(kv => (kv._1, new Set[String]() + kv._2)).reduceByKey(_ ++ _)
This results in unnecessary object creation because a new set must be allocated for each record.
Instead, use aggregateByKey, which performs the map-side aggregation more efficiently:
val zero = new collection.mutable.Set[String]()
rdd.aggregateByKey(zero)((set, v) => set += v,(set1, set2) => set1 ++= set2)
• flatMap-join-groupBy. When two datasets are already grouped by key and you want to join them and keep
them grouped, use cogroup. This avoids the overhead associated with unpacking and repacking the groups.
When Shuffles Do Not Occur
In some circumstances, the transformations described previously do not result in shuffles. Spark does not shuffle when
a previous transformation has already partitioned the data according to the same partitioner. Consider the following
flow:
rdd1 = someRdd.reduceByKey(...)
rdd2 = someOtherRdd.reduceByKey(...)
rdd3 = rdd1.join(rdd2)
Because no partitioner is passed to reduceByKey, the default partitioner is used, resulting in rdd1 and rdd2 both
being hash-partitioned. These two reduceByKey transformations result in two shuffles. If the datasets have the same
number of partitions, a join requires no additional shuffling. Because the datasets are partitioned identically, the set
of keys in any single partition of rdd1 can only occur in a single partition of rdd2. Therefore, the contents of any single
output partition of rdd3 depends only on the contents of a single partition in rdd1 and single partition in rdd2, and
a third shuffle is not required.
For example, if someRdd has four partitions, someOtherRdd has two partitions, and both the reduceByKeys use
three partitions, the set of tasks that run would look like this:
276 | Cloudera Administration
Performance Management
If rdd1 and rdd2 use different partitioners or use the default (hash) partitioner with different numbers of partitions,
only one of the datasets (the one with the fewer number of partitions) needs to be reshuffled for the join:
To avoid shuffles when joining two datasets, you can use broadcast variables. When one of the datasets is small enough
to fit in memory in a single executor, it can be loaded into a hash table on the driver and then broadcast to every
executor. A map transformation can then reference the hash table to do lookups.
When to Add a Shuffle Transformation
The rule of minimizing the number of shuffles has some exceptions.
An extra shuffle can be advantageous when it increases parallelism. For example, if your data arrives in a few large
unsplittable files, the partitioning dictated by the InputFormat might place large numbers of records in each partition,
Cloudera Administration | 277
Performance Management
while not generating enough partitions to use all available cores. In this case, invoking repartition with a high number
of partitions (which triggers a shuffle) after loading the data allows the transformations that follow to use more of the
cluster's CPU.
Another example arises when using the reduce or aggregate action to aggregate data into the driver. When
aggregating over a high number of partitions, the computation can quickly become bottlenecked on a single thread in
the driver merging all the results together. To lighten the load on the driver, first use reduceByKey or aggregateByKey
to perform a round of distributed aggregation that divides the dataset into a smaller number of partitions. The values
in each partition are merged with each other in parallel, before being sent to the driver for a final round of aggregation.
See treeReduce and treeAggregate for examples of how to do that.
This method is especially useful when the aggregation is already grouped by a key. For example, consider an application
that counts the occurrences of each word in a corpus and pulls the results into the driver as a map. One approach,
which can be accomplished with the aggregate action, is to compute a local map at each partition and then merge
the maps at the driver. The alternative approach, which can be accomplished with aggregateByKey, is to perform
the count in a fully distributed way, and then simply collectAsMap the results to the driver.
Secondary Sort
The repartitionAndSortWithinPartitions transformation repartitions the dataset according to a partitioner
and, within each resulting partition, sorts records by their keys. This transformation pushes sorting down into the
shuffle machinery, where large amounts of data can be spilled efficiently and sorting can be combined with other
operations.
For example, Apache Hive on Spark uses this transformation inside its join implementation. It also acts as a vital
building block in the secondary sort pattern, in which you group records by key and then, when iterating over the
values that correspond to a key, have them appear in a particular order. This scenario occurs in algorithms that need
to group events by user and then analyze the events for each user, based on the time they occurred.
Tuning Resource Allocation
For background information on how Spark applications use the YARN cluster manager, see Running Spark Applications
on YARN.
The two main resources that Spark and YARN manage are CPU and memory. Disk and network I/O affect Spark
performance as well, but neither Spark nor YARN actively manage them.
Every Spark executor in an application has the same fixed number of cores and same fixed heap size. Specify the number
of cores with the --executor-cores command-line flag, or by setting the spark.executor.cores property.
Similarly, control the heap size with the --executor-memory flag or the spark.executor.memory property. The
cores property controls the number of concurrent tasks an executor can run. For example, set --executor-cores
5 for each executor to run a maximum of five tasks at the same time. The memory property controls the amount of
data Spark can cache, as well as the maximum sizes of the shuffle data structures used for grouping, aggregations, and
joins.
Starting with CDH 5.5 dynamic allocation, which adds and removes executors dynamically, is enabled. To explicitly
control the number of executors, you can override dynamic allocation by setting the --num-executors command-line
flag or spark.executor.instances configuration property.
Consider also how the resources requested by Spark fit into resources YARN has available. The relevant YARN properties
are:
• yarn.hostmanager.resource.memory-mb controls the maximum sum of memory used by the containers on
each host.
• yarn.hostmanager.resource.cpu-vcores controls the maximum sum of cores used by the containers on
each host.
Requesting five executor cores results in a request to YARN for five cores. The memory requested from YARN is more
complex for two reasons:
278 | Cloudera Administration
Performance Management
• The --executor-memory/spark.executor.memory property controls the executor heap size, but JVMs can
also use some memory off heap, for example for interned Strings and direct byte buffers. The value of the
spark.yarn.executor.memoryOverhead property is added to the executor memory to determine the full
memory request to YARN for each executor. It defaults to max(384, .07 * spark.executor.memory).
• YARN may round the requested memory up slightly. The yarn.scheduler.minimum-allocation-mb and
yarn.scheduler.increment-allocation-mb properties control the minimum and increment request values,
respectively.
The following diagram (not to scale with defaults) shows the hierarchy of memory properties in Spark and YARN:
Keep the following in mind when sizing Spark executors:
• The ApplicationMaster, which is a non-executor container that can request containers from YARN, requires memory
and CPU that must be accounted for. In client deployment mode, they default to 1024 MB and one core. In cluster
deployment mode, the ApplicationMaster runs the driver, so consider bolstering its resources with the
--driver-memory and --driver-cores flags.
• Running executors with too much memory often results in excessive garbage-collection delays. For a single
executor, use 64 GB as an upper limit.
• The HDFS client has difficulty processing many concurrent threads. At most, five tasks per executor can achieve
full write throughput, so keep the number of cores per executor below that number.
• Running tiny executors (with a single core and just enough memory needed to run a single task, for example)
offsets the benefits of running multiple tasks in a single JVM. For example, broadcast variables must be replicated
once on each executor, so many small executors results in many more copies of the data.
Resource Tuning Example
Consider a cluster with six hosts running NodeManagers, each equipped with 16 cores and 64 GB of memory.
The NodeManager capacities, yarn.nodemanager.resource.memory-mb and
yarn.nodemanager.resource.cpu-vcores, should be set to 63 * 1024 = 64512 (megabytes) and 15, respectively.
Avoid allocating 100% of the resources to YARN containers because the host needs some resources to run the OS and
Hadoop daemons. In this case, leave one GB and one core for these system processes. Cloudera Manager accounts
for these and configures these YARN properties automatically.
You might consider using --num-executors 6 --executor-cores 15 --executor-memory 63G. However,
this approach does not work:
• 63 GB plus the executor memory overhead does not fit within the 63 GB capacity of the NodeManagers.
• The ApplicationMaster uses a core on one of the hosts, so there is no room for a 15-core executor on that host.
• 15 cores per executor can lead to bad HDFS I/O throughput.
Instead, use --num-executors 17 --executor-cores 5 --executor-memory 19G:
• This results in three executors on all hosts except for the one with the ApplicationMaster, which has two executors.
• --executor-memory is computed as (63/3 executors per host) = 21. 21 * 0.07 = 1.47. 21 - 1.47 ~ 19.
Tuning the Number of Partitions
Spark has limited capacity to determine optimal parallelism. Every Spark stage has a number of tasks, each of which
processes data sequentially. The number of tasks per stage is the most important parameter in determining performance.
Cloudera Administration | 279
Performance Management
As described in Spark Execution Model, Spark groups datasets into stages. The number of tasks in a stage is the same
as the number of partitions in the last dataset in the stage. The number of partitions in a dataset is the same as the
number of partitions in the datasets on which it depends, with the following exceptions:
• The coalesce transformation creates a dataset with fewer partitions than its parent dataset.
• The union transformation creates a dataset with the sum of its parents' number of partitions.
• The cartesian transformation creates a dataset with the product of its parents' number of partitions.
Datasets with no parents, such as those produced by textFile or hadoopFile, have their partitions determined by
the underlying MapReduce InputFormat used. Typically, there is a partition for each HDFS block being read. The
number of partitions for datasets produced by parallelize are specified in the method, or
spark.default.parallelism if not specified. To determine the number of partitions in an dataset, call
rdd.partitions().size().
If the number of tasks is smaller than number of slots available to run them, CPU usage is suboptimal. In addition, more
memory is used by any aggregation operations that occur in each task. In join, cogroup, or *ByKey operations,
objects are held in in hashmaps or in-memory buffers to group or sort. join, cogroup, and groupByKey use these
data structures in the tasks for the stages that are on the fetching side of the shuffles they trigger. reduceByKey and
aggregateByKey use data structures in the tasks for the stages on both sides of the shuffles they trigger. If the records
in these aggregation operations exceed memory, the following issues can occur:
• Holding a high number records in these data structures increases garbage collection, which can lead to pauses in
computation.
• Spark spills them to disk, causing disk I/O and sorting that leads to job stalls.
To increase the number of partitions if the stage is reading from Hadoop:
• Use the repartition transformation, which triggers a shuffle.
• Configure your InputFormat to create more splits.
• Write the input data to HDFS with a smaller block size.
If the stage is receiving input from another stage, the transformation that triggered the stage boundary accepts a
numPartitions argument:
val rdd2 = rdd1.reduceByKey(_ + _, numPartitions = X)
Determining the optimal value for X requires experimentation. Find the number of partitions in the parent dataset,
and then multiply that by 1.5 until performance stops improving.
You can also calculate X in a more formulaic way, but some quantities in the formula are difficult to calculate. The main
goal is to run enough tasks so that the data destined for each task fits in the memory available to that task. The memory
available to each task is:
(spark.executor.memory * spark.shuffle.memoryFraction * spark.shuffle.safetyFraction)/
spark.executor.cores
memoryFraction and safetyFraction default to 0.2 and 0.8 respectively.
The in-memory size of the total shuffle data is more difficult to determine. The closest heuristic is to find the ratio
between shuffle spill memory and the shuffle spill disk for a stage that ran. Then, multiply the total shuffle write by
this number. However, this can be compounded if the stage is performing a reduction:
(observed shuffle write) * (observed shuffle spill memory) * (spark.executor.cores)/
(observed shuffle spill disk) * (spark.executor.memory) * (spark.shuffle.memoryFraction)
* (spark.shuffle.safetyFraction)
Then, round up slightly, because too many partitions is usually better than too few.
When in doubt, err on the side of a larger number of tasks (and thus partitions). This contrasts with recommendations
for MapReduce, which unlike Spark, has a high startup overhead for tasks.
280 | Cloudera Administration
Performance Management
Reducing the Size of Data Structures
Data flows through Spark in the form of records. A record has two representations: a deserialized Java object
representation and a serialized binary representation. In general, Spark uses the deserialized representation for records
in memory and the serialized representation for records stored on disk or transferred over the network. For sort-based
shuffles, in-memory shuffle data is stored in serialized form.
The spark.serializer property controls the serializer used to convert between these two representations. Cloudera
recommends using the Kryo serializer, org.apache.spark.serializer.KryoSerializer.
The footprint of your records in these two representations has a significant impact on Spark performance. Review the
data types that are passed and look for places to reduce their size. Large deserialized objects result in Spark spilling
data to disk more often and reduces the number of deserialized records Spark can cache (for example, at the MEMORY
storage level). The Apache Spark tuning guide describes how to reduce the size of such objects. Large serialized objects
result in greater disk and network I/O, as well as reduce the number of serialized records Spark can cache (for example,
at the MEMORY_SER storage level.) Make sure to register any custom classes you use with the
SparkConf#registerKryoClasses API.
Choosing Data Formats
When storing data on disk, use an extensible binary format like Avro, Parquet, Thrift, or Protobuf and store in a sequence
file.
Tuning YARN
This topic applies to YARN clusters only, and describes how to tune and optimize YARN for your cluster.
Note: Download the Cloudera YARN tuning spreadsheet to help calculate YARN configurations. For
a short video overview, see Tuning YARN Applications.
Overview
This overview provides an abstract description of a YARN cluster and the goals of YARN tuning.
A YARN cluster is composed of host machines. Hosts
provide memory and CPU resources. A vcore, or virtual
core, is a usage share of a host CPU.
Cloudera Administration | 281
Performance Management
Tuning YARN consists primarily of optimally defining
containers on your worker hosts. You can think of a
container as a rectangular graph consisting of memory
and vcores. Containers perform tasks.
Some tasks use a great deal of memory, with minimal
processing on a large volume of data.
282 | Cloudera Administration
Performance Management
Other tasks require a great deal of processing power, but
use less memory. For example, a Monte Carlo Simulation
that evaluates many possible "what if?" scenarios uses a
great deal of processing power on a relatively small
dataset.
The YARN Resource Manager allocates memory and vcores
to use all available resources in the most efficient way
possible. Ideally, few or no resources are left idle.
Cloudera Administration | 283
Performance Management
An application is a YARN client program consisting of one
or more tasks. Typically, a task uses all of the available
resources in the container. A task cannot consume more
than its designated allocation, ensuring that it cannot use
all of the host CPU cycles or exceed its memory allotment.
Tune your YARN hosts to optimize the use of vcores and
memory by configuring your containers to use all available
resources, beyond those required for overhead and other
services.
There are three phases to YARN tuning. The phases correspond to the tabs in the YARN tuning spreadsheet.
1. Cluster configuration, where you configure your hosts.
2. YARN configuration, where you quantify memory and vcores.
3. MapReduce configuration, where you allocate minimum and maximum resources for specific map and reduce
tasks.
YARN and MapReduce have many configurable properties. For a complete list, see Cloudera Manager Configuration
Properties. The YARN tuning spreadsheet lists the essential subset of these properties that are most likely to improve
performance for common MapReduce applications.
284 | Cloudera Administration
Performance Management
Cluster Configuration
In the Cluster Configuration tab, you define the worker host configuration and cluster size for your YARN implementation.
Step 1: Worker Host Configuration
Step 1 is to define the configuration for a single worker host computer in your cluster.
As with any system, the more memory and CPU resources available, the faster the cluster can process large amounts
of data. A machine with 8 CPUs, each with 6 cores, provides 48 vcores per host.
3 TB hard drives in a 2-unit server installation with 12 available slots in JBOD (Just a Bunch Of Disks) configuration is a
reasonable balance of performance and pricing at the time the spreadsheet was created. The cost of storage decreases
over time, so you might consider 4 TB disks. Larger disks are expensive and not required for all use cases.
Two 1-Gigabit Ethernet ports provide sufficient throughput at the time the spreadsheet was published, but 10-Gigabit
Ethernet ports are an option where price is of less concern than speed.
Step 2: Worker Host Planning
Step 2 is to allocate resources on each worker machine.
Start with at least 8 GB for your operating system, and 1 GB for Cloudera Manager. If services outside of CDH require
additional resources, add those numbers under Other Services.
The HDFS DataNode uses a minimum of 1 core and about 1 GB of memory. The same requirements apply to the YARN
NodeManager.
The spreadsheet lists three optional services. For Impala, allocate at least 16 GB for the daemon. HBase RegionServer
requires 12-16 GB of memory. Solr Server requires a minimum of 1 GB of memory.
Any remaining resources are available for YARN applications (Spark and MapReduce). In this example, 44 CPU cores
are available. Set the multiplier for vcores you want on each physical core to calculate the total available vcores.
Cloudera Administration | 285
Performance Management
Step 3: Cluster Size
Having defined the specifications for each host in your cluster, enter the number of worker hosts needed to support
your business case. To see the benefits of parallel computing, set the number of hosts to a minimum of 10.
YARN Configuration
On the YARN Configuration tab, you verify your available resources and set minimum and maximum limits for each
container.
Steps 4 and 5: Verify Settings
Step 4 pulls forward the memory and vcore numbers from step 2. Step 5 shows the total memory and vcores for the
cluster.
Step 6: Verify Container Settings on Cluster
In step 6, you can change the four values that impact the size of your containers.
The minimum number of vcores should be 1. When additional vcores are required, adding 1 at a time should result in
the most efficient allocation. Set the maximum number of vcore reservations for a container to ensure that no single
task consumes all available resources.
Set the minimum and maximum reservations for memory. The increment should be the smallest amount that can
impact performance. Here, the minimum is approximately 1 GB, the maximum is approximately 8 GB, and the increment
is 512 MB.
286 | Cloudera Administration
Performance Management
Step 6A: Cluster Container Capacity
Step 6A lets you validate the minimum and maximum number of containers in your cluster, based on the numbers you
entered.
Step 6B: Container Sanity Checking
Step 6B lets you see at a glance whether you have over-allocated resources.
MapReduce Configuration
On the MapReduce Configuration tab, you can plan for increased task-specific memory capacity.
Step 7: MapReduce Configuration
You can increase the memory allocation for the ApplicationMaster, map tasks, and reduce tasks. The minimum vcore
allocation for any task is always 1. The Spill/Sort memory allocation of 256 should be sufficient, and should be (rarely)
increased if you determine that frequent spills to disk are hurting job performance.
Cloudera Administration | 287
Performance Management
Step 7A: MapReduce Sanity Checking
Step 7A lets you verify at a glance that all of your minimum and maximum resource allocations are within the parameters
you set.
Configuring Your Cluster In Cloudera Manager
When you are satisfied with the cluster configuration estimates, use the values in the spreadsheet to set the
corresponding properties in Cloudera Manager. For more information, see Modifying Configuration Properties Using
Cloudera Manager on page 10
Table 12: Cloudera Manager Property Correspondence
Step
YARN/MapReduce Property
4
yarn.nodemanager.resource.cpu-vcores Container Virtual CPU Cores
4
yarn.nodemanager.resource.memory-mb Container Memory
6
yarn.scheduler.minimum-allocation-vcores Container Virtual CPU Cores Minimum
6
yarn-scheduler.maximum-allocation-vcores Container Virtual CPU Cores Maximum
6
yarn.scheduler.increment-allocation-vcores Container Virtual CPU Cores Increment
6
yarn.scheduler.minimum-allocation-mb Container Memory Minimum
6
yarn.scheduler.maximum-allocation-mb Container Memory Maximum
6
yarn.scheduler.increment-allocation-mb Container Memory Increment
7
yarn.app.mapreduce.am.resource.cpu-vcores ApplicationMaster Virtual CPU Cores
288 | Cloudera Administration
Cloudera Manager Equivalent
Performance Management
Step
YARN/MapReduce Property
Cloudera Manager Equivalent
7
yarn.app.mapreduce.am.resource.mb ApplicationMaster Memory
7
mapreduce.map.cpu.vcores
Map Task CPU Virtual Cores
7
mapreduce.map.memory.mb
Map Task Memory
7
mapreduce.reduce.cpu.vcores
Reduce Task CPU Virtual Cores
7
mapreduce.reduce.memory.mb
Reduce Task Memory
7
mapreduce.task.io.sort.mb
I/O Sort Memory
Cloudera Administration | 289
High Availability
High Availability
This guide is for Apache Hadoop system administrators who want to enable continuous availability by configuring
clusters without single points of failure.
HDFS High Availability
This section provides an overview of the HDFS high availability (HA) feature and how to configure and manage an HA
HDFS cluster.
Introduction to HDFS High Availability
This section assumes that the reader has a general understanding of components in an HDFS cluster. For details, see
the Apache HDFS Architecture Guide.
Background
In a standard configuration, the NameNode is a single point of failure (SPOF) in an HDFS cluster. Each cluster has a
single NameNode, and if that host or process became unavailable, the cluster as a whole is unavailable until the
NameNode is either restarted or brought up on a new host. The Secondary NameNode does not provide failover
capability.
The standard configuration reduces the total availability of an HDFS cluster in two major ways:
• In the case of an unplanned event such as a host crash, the cluster is unavailable until an operator restarts the
NameNode.
• Planned maintenance events such as software or hardware upgrades on the NameNode machine result in periods
of cluster downtime.
HDFS HA addresses the above problems by providing the option of running two NameNodes in the same cluster, in an
active/passive configuration. These are referred to as the active NameNode and the standby NameNode. Unlike the
Secondary NameNode, the standby NameNode is hot standby, allowing a fast automatic failover to a new NameNode
in the case that a host crashes, or a graceful administrator-initiated failover for the purpose of planned maintenance.
You cannot have more than two NameNodes.
Implementation
Cloudera Manager 5 and CDH 5 support Quorum-based Storage on page 290 as the only HA implementation. In contrast,
CDH 4 supports both Quorum-based Storage and shared storage using NFS. For instructions on switching to
Quorum-based storage, see Converting From an NFS-mounted Shared Edits Directory to Quorum-based Storage on
page 314.
Important: In Cloudera Manager 5, when you attempt to upgrade a CDH 4 cluster configured for HA
using an NFS-mounted shared edits directory:
• If you do not disable your HA configuration before upgrading, your HA configuration will continue
to work; but you will see a warning recommending that you switch to Quorum-based storage.
• If you do disable your HA configuration before upgrading, you will not be able to re-enable HA
with NFS-mounted shared directories. Instead, you must configure HA to use Quorum-based
storage.
Quorum-based Storage
Quorum-based Storage refers to the HA implementation that uses a Quorum Journal Manager (QJM).
290 | Cloudera Administration
High Availability
In order for the standby NameNode to keep its state synchronized with the active NameNode in this implementation,
both nodes communicate with a group of separate daemons called JournalNodes. When any namespace modification
is performed by the active NameNode, it durably logs a record of the modification to a majority of the JournalNodes.
The standby NameNode is capable of reading the edits from the JournalNodes, and is constantly watching them for
changes to the edit log. As the standby Node sees the edits, it applies them to its own namespace. In the event of a
failover, the standby ensures that it has read all of the edits from the JournalNodes before promoting itself to the
active state. This ensures that the namespace state is fully synchronized before a failover occurs.
In order to provide a fast failover, it is also necessary that the standby NameNode has up-to-date information regarding
the location of blocks in the cluster. In order to achieve this, DataNodes are configured with the location of both
NameNodes, and they send block location information and heartbeats to both.
It is vital for the correct operation of an HA cluster that only one of the NameNodes be active at a time. Otherwise,
the namespace state would quickly diverge between the two, risking data loss or other incorrect results. In order to
ensure this property and prevent the so-called "split-brain scenario," JournalNodes only ever allow a single NameNode
to be a writer at a time. During a failover, the NameNode which is to become active simply takes over the role of writing
to the JournalNodes, which effectively prevents the other NameNode from continuing in the active state, allowing the
new active NameNode to safely proceed with failover.
Note: Because of this, fencing is not required, but it is still useful; see Enabling HDFS HA on page 292.
Shared Storage Using NFS
In order for the standby NameNode to keep its state synchronized with the active NameNode, this implementation
requires that the two nodes both have access to a directory on a shared storage device (for example, an NFS mount
from a NAS).
When any namespace modification is performed by the active NameNode, it durably logs a record of the modification
to an edit log file stored in the shared directory. The standby NameNode constantly watches this directory for edits,
and when edits occur, the standby NameNode applies them to its own namespace. In the event of a failover, the
standby will ensure that it has read all of the edits from the shared storage before promoting itself to the active state.
This ensures that the namespace state is fully synchronized before a failover occurs.
In order to prevent the "split-brain scenario" in which the namespace state is diverged between the two NameNodes,
an administrator must configure at least one fencing method for the shared storage. If, during a failover, it cannot be
verified that the previous active NameNode has relinquished its active state, the fencing process is responsible for
cutting off the previous active NameNode's access to the shared edits storage. This prevents it from making any further
edits to the namespace, allowing the new active NameNode to safely proceed with failover.
Automatic Failover Implementation
Automatic failover relies on two additional components in an HDFS: a ZooKeeper quorum, and the
ZKFailoverController process (abbreviated as ZKFC). In Cloudera Manager, the ZKFC process maps to the HDFS
Failover Controller role.
Apache ZooKeeper is a highly available service for maintaining small amounts of coordination data, notifying clients
of changes in that data, and monitoring clients for failures. The implementation of HDFS automatic failover relies on
ZooKeeper for the following functions:
• Failure detection - each of the NameNode machines in the cluster maintains a persistent session in ZooKeeper.
If the machine crashes, the ZooKeeper session will expire, notifying the other NameNode that a failover should
be triggered.
• Active NameNode election - ZooKeeper provides a simple mechanism to exclusively elect a node as active. If the
current active NameNode crashes, another node can take a special exclusive lock in ZooKeeper indicating that it
should become the next active NameNode.
The ZKFailoverController (ZKFC) is a ZooKeeper client that also monitors and manages the state of the NameNode.
Each of the hosts that run a NameNode also run a ZKFC. The ZKFC is responsible for:
Cloudera Administration | 291
High Availability
• Health monitoring - the ZKFC contacts its local NameNode on a periodic basis with a health-check command. So
long as the NameNode responds promptly with a healthy status, the ZKFC considers the NameNode healthy. If
the NameNode has crashed, frozen, or otherwise entered an unhealthy state, the health monitor marks it as
unhealthy.
• ZooKeeper session management - when the local NameNode is healthy, the ZKFC holds a session open in ZooKeeper.
If the local NameNode is active, it also holds a special lock znode. This lock uses ZooKeeper's support for
"ephemeral" nodes; if the session expires, the lock node is automatically deleted.
• ZooKeeper-based election - if the local NameNode is healthy, and the ZKFC sees that no other NameNode currently
holds the lock znode, it will itself try to acquire the lock. If it succeeds, then it has "won the election", and is
responsible for running a failover to make its local NameNode active. The failover process is similar to the manual
failover described above: first, the previous active is fenced if necessary, and then the local NameNode transitions
to active state.
Configuring Hardware for HDFS HA
In order to deploy an HA cluster using Quorum-based Storage, you should prepare the following:
• NameNode hosts - These are the hosts on which you run the active and standby NameNodes. They should have
equivalent hardware to each other, and equivalent hardware to what would be used in a non-HA cluster.
• JournalNode hosts - These are the hosts on which you run the JournalNodes. Cloudera recommends that you
deploy the JournalNode daemons on the "master" host or hosts (NameNode, Standby NameNode, JobTracker,
and so on) so the JournalNodes' local directories can use the reliable local storage on those machines.
• If co-located on the same host, each JournalNode process and each NameNode process should have its own
dedicated disk. You should not use SAN or NAS storage for these directories.
• There must be at least three JournalNode daemons, since edit log modifications must be written to a majority of
JournalNodes. This will allow the system to tolerate the failure of a single host. You can also run more than three
JournalNodes, but in order to actually increase the number of failures the system can tolerate, you should run an
odd number of JournalNodes, (three, five, seven, and so on). Note that when running with N JournalNodes, the
system can tolerate at most (N - 1) / 2 failures and continue to function normally. If the requisite quorum is not
available, the NameNode will not format or start, and you will see an error similar to this:
12/10/01 17:34:18 WARN namenode.FSEditLog: Unable to determine input streams from QJM
to [10.0.1.10:8485, 10.0.1.10:8486, 10.0.1.10:8487]. Skipping.
java.io.IOException: Timed out waiting 20000ms for a quorum of nodes to respond.
Note: In an HA cluster, the standby NameNode also performs checkpoints of the namespace state,
and thus it is not necessary to run a Secondary NameNode, CheckpointNode, or BackupNode in an
HA cluster. In fact, to do so would be an error. If you are reconfiguring a non-HA-enabled HDFS cluster
to be HA-enabled, you can reuse the hardware which you had previously dedicated to the Secondary
NameNode.
Enabling HDFS HA
An HDFS high availability (HA) cluster uses two NameNodes—an active NameNode and a standby NameNode. Only
one NameNode can be active at any point in time. HDFS HA depends on maintaining a log of all namespace modifications
in a location available to both NameNodes, so that in the event of a failure, the standby NameNode has up-to-date
information about the edits and location of blocks in the cluster. For CDH 4 HA features, see the CDH 4 High Availability
Guide.
Important: Enabling and disabling HA causes a service outage for the HDFS service and all services
that depend on HDFS. Before enabling or disabling HA, ensure that there are no jobs running on your
cluster.
Enabling HDFS HA Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
292 | Cloudera Administration
High Availability
You can use Cloudera Manager to configure your CDH 4 or CDH 5 cluster for HDFS HA and automatic failover. In Cloudera
Manager 5, HA is implemented using Quorum-based storage. Quorum-based storage relies upon a set of JournalNodes,
each of which maintains a local edits directory that logs the modifications to the namespace metadata. Enabling HA
enables automatic failover as part of the same command.
Important:
• Enabling or disabling HA causes the previous monitoring history to become unavailable.
• Some parameters will be automatically set as follows once you have enabled JobTracker HA. If
you want to change the value from the default for these parameters, use an advanced
configuration snippet.
–
–
–
–
mapred.jobtracker.restart.recover: true
mapred.job.tracker.persist.jobstatus.active: true
mapred.ha.automatic-failover.enabled: true
mapred.ha.fencing.methods: shell(/bin/true)
Enabling High Availability and Automatic Failover
The Enable High Availability workflow leads you through adding a second (standby) NameNode and configuring
JournalNodes. During the workflow, Cloudera Manager creates a federated namespace.
1.
2.
3.
4.
Perform all the configuration and setup tasks described under Configuring Hardware for HDFS HA on page 292.
Ensure that you have a ZooKeeper service.
Go to the HDFS service.
Select Actions > Enable High Availability. A screen showing the hosts that are eligible to run a standby NameNode
and the JournalNodes displays.
a. Specify a name for the nameservice or accept the default name nameservice1 and click Continue.
b. In the NameNode Hosts field, click Select a host. The host selection dialog box displays.
c. Check the checkbox next to the hosts where you want the standby NameNode to be set up and click OK. The
standby NameNode cannot be on the same host as the active NameNode, and the host that is chosen should
have the same hardware configuration (RAM, disk space, number of cores, and so on) as the active NameNode.
d. In the JournalNode Hosts field, click Select hosts. The host selection dialog box displays.
e. Check the checkboxes next to an odd number of hosts (a minimum of three) to act as JournalNodes and click
OK. JournalNodes should be hosted on hosts with similar hardware specification as the NameNodes. Cloudera
recommends that you put a JournalNode each on the same hosts as the active and standby NameNodes, and
the third JournalNode on similar hardware, such as the JobTracker.
f. Click Continue.
g. In the JournalNode Edits Directory property, enter a directory location for the JournalNode edits directory
into the fields for each JournalNode host.
• You may enter only one directory for each JournalNode. The paths do not need to be the same on every
JournalNode.
• The directories you specify should be empty, and must have the appropriate permissions.
h. Extra Options: Decide whether Cloudera Manager should clear existing data in ZooKeeper, standby NameNode,
and JournalNodes. If the directories are not empty (for example, you are re-enabling a previous HA
configuration), Cloudera Manager will not automatically delete the contents—you can select to delete the
contents by keeping the default checkbox selection. The recommended default is to clear the directories. If
you choose not to do so, the data should be in sync across the edits directories of the JournalNodes and
should have the same version data as the NameNodes.
i. Click Continue.
Cloudera Manager executes a set of commands that will stop the dependent services, delete, create, and configure
roles and directories as appropriate, create a nameservice and failover controller, and restart the dependent
services and deploy the new client configuration.
Cloudera Administration | 293
High Availability
5. If you want to use other services in a cluster with HA configured, follow the procedures in Configuring Other CDH
Components to Use HDFS HA on page 309.
6. If you are running CDH 4.0 or 4.1, the standby NameNode may fail at the bootstrapStandby command with
the error Unable to read transaction ids 1-7 from the configured shared edits storage.
Use rsync or a similar tool to copy the contents of the dfs.name.dir directory from the active NameNode to
the standby NameNode and start the standby NameNode.
Important: If you change the NameNode Service RPC Port (dfs.namenode.servicerpc-address)
while automatic failover is enabled, this will cause a mismatch between the NameNode address saved
in the ZooKeeper /hadoop-ha znode and the NameNode address that the Failover Controller is
configured with. This will prevent the Failover Controllers from restarting. If you need to change the
NameNode Service RPC Port after Auto Failover has been enabled, you must do the following to
re-initialize the znode:
1. Stop the HDFS service.
2. Configure the service RPC port:
a.
b.
c.
d.
e.
Go to the HDFS service.
Click the Configuration tab.
Select Scope > NameNode.
Select Category > Ports and Addresses.
Locate the NameNode Service RPC Port property or search for it by typing its name in the
Search box.
f. Change the port value as needed.
If more than one role group applies to this configuration, edit the value for the appropriate
role group. See Modifying Configuration Properties Using Cloudera Manager on page 10.
3. On a ZooKeeper server host, run zookeeper-client.
a. Execute the following to remove the configured nameservice. This example assumes the
name of the nameservice is nameservice1. You can identify the nameservice from the
Federation and High Availability section on the HDFS Instances tab:
rmr /hadoop-ha/nameservice1
4. Click the Instances tab.
5. Select Actions > Initialize High Availability State in ZooKeeper.
6. Start the HDFS service.
Fencing Methods
To ensure that only one NameNode is active at a time, a fencing method is required for the shared edits directory.
During a failover, the fencing method is responsible for ensuring that the previous active NameNode no longer has
access to the shared edits directory, so that the new active NameNode can safely proceed writing to it.
By default, Cloudera Manager configures HDFS to use a shell fencing method
(shell(./cloudera_manager_agent_fencer.py)) that takes advantage of the Cloudera Manager Agent. However,
you can configure HDFS to use the sshfence method, or you can add your own shell fencing scripts, instead of or in
addition to the one Cloudera Manager provides.
The fencing parameters are found in the Service-Wide > High Availability category under the configuration properties
for your HDFS service.
For details of the fencing methods supplied with CDH 5, and how fencing is configured, see Fencing Configuration on
page 298.
294 | Cloudera Administration
High Availability
Enabling HDFS HA Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
This section describes the software configuration required for HDFS HA in CDH 5 and explains how to set configuration
properties and use the command line to deploy HDFS HA.
Configuring Software for HDFS HA
Configuration Overview
As with HDFS federation configuration, HA configuration is backward compatible and allows existing single NameNode
configurations to work without change. The new configuration is designed such that all the nodes in the cluster can
have the same configuration without the need for deploying different configuration files to different machines based
on the type of the node.
HA clusters reuse the Nameservice ID to identify a single HDFS instance that may consist of multiple HA NameNodes.
In addition, there is a new abstraction called NameNode ID. Each distinct NameNode in the cluster has a different
NameNode ID. To support a single configuration file for all of the NameNodes, the relevant configuration parameters
include the Nameservice ID as well as the NameNode ID.
Changes to Existing Configuration Parameters
The following configuration parameter has changed for YARN implementations:
fs.defaultFS - formerly fs.default.name, the default path prefix used by the Hadoop FS client when none is
given. (fs.default.name is deprecated for YARN implementations, but will still work.)
Optionally, you can configure the default path for Hadoop clients to use the HA-enabled logical URI. For example, if
you use mycluster as the Nameservice ID as shown below, this will be the value of the authority portion of all of your
HDFS paths. You can configure the default path in your core-site.xml file:
• For YARN:
<property>
<name>fs.defaultFS</name>
<value>hdfs://mycluster</value>
</property>
• For MRv1:
<property>
<name>fs.default.name</name>
<value>hdfs://mycluster</value>
</property>
New Configuration Parameters
To configure HA NameNodes, you must add several configuration options to your hdfs-site.xml configuration file.
The order in which you set these configurations is unimportant, but the values you choose for dfs.nameservices
and dfs.ha.namenodes.[Nameservice ID] will determine the keys of those that follow. This means that you
should decide on these values before setting the rest of the configuration options.
Configure dfs.nameservices
dfs.nameservices - the logical name for this new nameservice
Cloudera Administration | 295
High Availability
Choose a logical name for this nameservice, for example mycluster, and use this logical name for the value of this
configuration option. The name you choose is arbitrary. It will be used both for configuration and as the authority
component of absolute HDFS paths in the cluster.
Note: If you are also using HDFS federation, this configuration setting should also include the list of
other Nameservices, HA or otherwise, as a comma-separated list.
<property>
<name>dfs.nameservices</name>
<value>mycluster</value>
</property>
Configure dfs.ha.namenodes.[nameservice ID]
dfs.ha.namenodes.[nameservice ID] - unique identifiers for each NameNode in the nameservice
Configure a list of comma-separated NameNode IDs. This will be used by DataNodes to determine all the NameNodes
in the cluster. For example, if you used mycluster as the NameService ID previously, and you wanted to use nn1 and
nn2 as the individual IDs of the NameNodes, you would configure this as follows:
<property>
<name>dfs.ha.namenodes.mycluster</name>
<value>nn1,nn2</value>
</property>
Note: In this release, you can configure a maximum of two NameNodes per nameservice.
Configure dfs.namenode.rpc-address.[nameservice ID]
dfs.namenode.rpc-address.[nameservice ID].[name node ID] - the fully-qualified RPC address for each
NameNode to listen on
For both of the previously-configured NameNode IDs, set the full address and RPC port of the NameNode process.
Note that this results in two separate configuration options. For example:
<property>
<name>dfs.namenode.rpc-address.mycluster.nn1</name>
<value>machine1.example.com:8020</value>
</property>
<property>
<name>dfs.namenode.rpc-address.mycluster.nn2</name>
<value>machine2.example.com:8020</value>
</property>
Note: If necessary, you can similarly configure the servicerpc-address setting.
Configure dfs.namenode.http-address.[nameservice ID]
dfs.namenode.http-address.[nameservice ID].[name node ID] - the fully-qualified HTTP address for each
NameNode to listen on
Similarly to rpc-address above, set the addresses for both NameNodes' HTTP servers to listen on. For example:
<property>
<name>dfs.namenode.http-address.mycluster.nn1</name>
<value>machine1.example.com:50070</value>
</property>
<property>
296 | Cloudera Administration
High Availability
<name>dfs.namenode.http-address.mycluster.nn2</name>
<value>machine2.example.com:50070</value>
</property>
Note: If you have Hadoop Kerberos security features enabled, and you intend to use HSFTP, you
should also set the https-address similarly for each NameNode.
Configure dfs.namenode.shared.edits.dir
dfs.namenode.shared.edits.dir - the location of the shared storage directory
Configure the addresses of the JournalNodes which provide the shared edits storage, written to by the Active NameNode
and read by the Standby NameNode to stay up-to-date with all the file system changes the Active NameNode makes.
Though you must specify several JournalNode addresses, you should only configure one of these URIs. The URI should
be in the form:
qjournal://<host1:port1>;<host2:port2>;<host3:port3>/<journalId>
The Journal ID is a unique identifier for this nameservice, which allows a single set of JournalNodes to provide storage
for multiple federated namesystems. Though it is not a requirement, it's a good idea to reuse the Nameservice ID for
the journal identifier.
For example, if the JournalNodes for this cluster were running on the machines node1.example.com,
node2.example.com, and node3.example.com, and the nameservice ID were mycluster, you would use the
following as the value for this setting (the default port for the JournalNode is 8485):
<property>
<name>dfs.namenode.shared.edits.dir</name>
<value>qjournal://node1.example.com:8485;node2.example.com:8485;node3.example.com:8485/mycluster</value>
</property>
Configure dfs.journalnode.edits.dir
dfs.journalnode.edits.dir - the path where the JournalNode daemon will store its local state
On each JournalNode machine, configure the absolute path where the edits and other local state information used by
the JournalNodes will be stored; use only a single path per JournalNode. (The other JournalNodes provide redundancy;
you can also configure this directory on a locally-attached RAID-1 or RAID-10 array.)
For example:
<property>
<name>dfs.journalnode.edits.dir</name>
<value>/data/1/dfs/jn</value>
</property>
Now create the directory (if it doesn't already exist) and make sure its owner is hdfs, for example:
$ sudo mkdir -p /data/1/dfs/jn
$ sudo chown -R hdfs:hdfs /data/1/dfs/jn
Client Failover Configuration
dfs.client.failover.proxy.provider.[nameservice ID] - the Java class that HDFS clients use to contact
the Active NameNode
Configure the name of the Java class which the DFS client will use to determine which NameNode is the current active,
and therefore which NameNode is currently serving client requests. The only implementation which currently ships
Cloudera Administration | 297
High Availability
with Hadoop is the ConfiguredFailoverProxyProvider, so use this unless you are using a custom one. For
example:
<property>
<name>dfs.client.failover.proxy.provider.mycluster</name>
<value>org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider</value>
</property>
Fencing Configuration
dfs.ha.fencing.methods - a list of scripts or Java classes which will be used to fence the active NameNode during
a failover
It is desirable for correctness of the system that only one NameNode be in the active state at any given time.
Important: When you use Quorum-based Storage, only one NameNode will ever be allowed to write
to the JournalNodes, so there is no potential for corrupting the file system metadata in a "split-brain"
scenario. But when a failover occurs, it is still possible that the previously active NameNode could
serve read requests to clients - and these requests may be out of date - until that NameNode shuts
down when it tries to write to the JournalNodes. For this reason, it is still desirable to configure some
fencing methods even when using Quorum-based Storage.
To improve the availability of the system in the event the fencing mechanisms fail, it is advisable to configure a fencing
method which is guaranteed to return success as the last fencing method in the list.
Note: If you choose to use no actual fencing methods, you still must configure something for this
setting, for example shell(/bin/true).
The fencing methods used during a failover are configured as a carriage-return-separated list, and these will be
attempted in order until one of them indicates that fencing has succeeded.
There are two fencing methods which ship with Hadoop:
• sshfence
• shell
For information on implementing your own custom fencing method, see the org.apache.hadoop.ha.NodeFencer
class.
Configuring the sshfence fencing method
sshfence - SSH to the active NameNode and kill the process
The sshfence option uses SSH to connect to the target node and uses fuser to kill the process listening on the
service's TCP port. In order for this fencing option to work, it must be able to SSH to the target node without providing
a passphrase. Thus, you must also configure the dfs.ha.fencing.ssh.private-key-files option, which is a
comma-separated list of SSH private key files.
Important: The files must be accessible to the user running the NameNode processes (typically the
hdfs user on the NameNode hosts).
For example:
<property>
<name>dfs.ha.fencing.methods</name>
<value>sshfence</value>
</property>
<property>
298 | Cloudera Administration
High Availability
<name>dfs.ha.fencing.ssh.private-key-files</name>
<value>/home/exampleuser/.ssh/id_rsa</value>
</property>
Optionally, you can configure a non-standard username or port to perform the SSH as shown below. You can also
configure a timeout, in milliseconds, for the SSH, after which this fencing method will be considered to have failed:
<property>
<name>dfs.ha.fencing.methods</name>
<value>sshfence([[username][:port]])</value>
</property>
<property>
<name>dfs.ha.fencing.ssh.connect-timeout</name>
<value>30000</value>
<description>
SSH connection timeout, in milliseconds, to use with the builtin
sshfence fencer.
</description>
</property>
Configuring the shell fencing method
shell - run an arbitrary shell command to fence the active NameNode
The shell fencing method runs an arbitrary shell command, which you can configure as shown below:
<property>
<name>dfs.ha.fencing.methods</name>
<value>shell(/path/to/my/script.sh arg1 arg2 ...)</value>
</property>
The string between '(' and ')' is passed directly to a bash shell and cannot include any closing parentheses.
When executed, the first argument to the configured script will be the address of the NameNode to be fenced, followed
by all arguments specified in the configuration.
The shell command will be run with an environment set up to contain all of the current Hadoop configuration variables,
with the '_' character replacing any '.' characters in the configuration keys. The configuration used has already had any
NameNode-specific configurations promoted to their generic forms - for example dfs_namenode_rpc-address will
contain the RPC address of the target node, even though the configuration may specify that variable as
dfs.namenode.rpc-address.ns1.nn1.
The following variables referring to the target node to be fenced are also available:
Variable
Description
$target_host
Hostname of the node to be fenced
$target_port
IPC port of the node to be fenced
$target_address
The two variables above, combined as host:port
$target_nameserviceid
The nameservice ID of the NameNode to be fenced
$target_namenodeid
The NameNode ID of the NameNode to be fenced
You can also use these environment variables as substitutions in the shell command itself. For example:
<property>
<name>dfs.ha.fencing.methods</name>
<value>shell(/path/to/my/script.sh --nameservice=$target_nameserviceid
$target_host:$target_port)</value>
</property>
If the shell command returns an exit code of 0, the fencing is determined to be successful. If it returns any other exit
code, the fencing was not successful and the next fencing method in the list will be attempted.
Cloudera Administration | 299
High Availability
Note: This fencing method does not implement any timeout. If timeouts are necessary, they should
be implemented in the shell script itself (for example, by forking a subshell to kill its parent in some
number of seconds).
Automatic Failover Configuration
The previous sections describe how to configure manual failover. In that mode, the system will not automatically trigger
a failover from the active to the standby NameNode, even if the active node has failed. This section describes how to
configure and deploy automatic failover. For an overview of how automatic failover is implemented, see Automatic
Failover Implementation on page 291.
Deploying ZooKeeper
In a typical deployment, ZooKeeper daemons are configured to run on three or five nodes. Since ZooKeeper itself has
light resource requirements, it is acceptable to collocate the ZooKeeper nodes on the same hardware as the HDFS
NameNode and Standby Node. Operators using MapReduce v2 (MRv2) often choose to deploy the third ZooKeeper
process on the same node as the YARN ResourceManager. It is advisable to configure the ZooKeeper nodes to store
their data on separate disk drives from the HDFS metadata for best performance and isolation.
See the ZooKeeper documentation for instructions on how to set up a ZooKeeper ensemble. In the following sections
we assume that you have set up a ZooKeeper cluster running on three or more nodes, and have verified its correct
operation by connecting using the ZooKeeper command-line interface (CLI).
Configuring Automatic Failover
Note: Before you begin configuring automatic failover, you must shut down your cluster. It is not
currently possible to transition from a manual failover setup to an automatic failover setup while the
cluster is running.
Configuring automatic failover requires two additional configuration parameters. In your hdfs-site.xml file, add:
<property>
<name>dfs.ha.automatic-failover.enabled</name>
<value>true</value>
</property>
This specifies that the cluster should be set up for automatic failover. In your core-site.xml file, add:
<property>
<name>ha.zookeeper.quorum</name>
<value>zk1.example.com:2181,zk2.example.com:2181,zk3.example.com:2181</value>
</property>
This lists the host-port pairs running the ZooKeeper service.
As with the parameters described earlier in this document, these settings may be configured on a per-nameservice
basis by suffixing the configuration key with the nameservice ID. For example, in a cluster with federation enabled,
you can explicitly enable automatic failover for only one of the nameservices by setting
dfs.ha.automatic-failover.enabled.my-nameservice-id.
There are several other configuration parameters which you can set to control the behavior of automatic failover, but
they are not necessary for most installations. See the configuration section of the Hadoop documentation for details.
Initializing the HA state in ZooKeeper
After you have added the configuration keys, the next step is to initialize the required state in ZooKeeper. You can do
so by running the following command from one of the NameNode hosts.
300 | Cloudera Administration
High Availability
Note: The ZooKeeper ensemble must be running when you use this command; otherwise it will not
work properly.
$ hdfs zkfc -formatZK
This will create a znode in ZooKeeper in which the automatic failover system stores its data.
Securing access to ZooKeeper
If you are running a secure cluster, you will probably want to ensure that the information stored in ZooKeeper is also
secured. This prevents malicious clients from modifying the metadata in ZooKeeper or potentially triggering a false
failover.
In order to secure the information in ZooKeeper, first add the following to your core-site.xml file:
<property>
<name>ha.zookeeper.auth</name>
<value>@/path/to/zk-auth.txt</value>
</property>
<property>
<name>ha.zookeeper.acl</name>
<value>@/path/to/zk-acl.txt</value>
</property>
Note the '@' character in these values – this specifies that the configurations are not inline, but rather point to a file
on disk.
The first configured file specifies a list of ZooKeeper authentications, in the same format as used by the ZooKeeper
CLI. For example, you may specify something like digest:hdfs-zkfcs:mypassword where hdfs-zkfcs is a unique
username for ZooKeeper, and mypassword is some unique string used as a password.
Next, generate a ZooKeeper Access Control List (ACL) that corresponds to this authentication, using a command such
as the following:
$ java -cp $ZK_HOME/lib/*:$ZK_HOME/zookeeper-3.4.2.jar
org.apache.zookeeper.server.auth.DigestAuthenticationProvider hdfs-zkfcs:mypassword
output: hdfs-zkfcs:mypassword->hdfs-zkfcs:P/OQvnYyU/nF/mGYvB/xurX8dYs=
Copy and paste the section of this output after the '->' string into the file zk-acls.txt, prefixed by the string "digest:".
For example:
digest:hdfs-zkfcs:vlUvLnd8MlacsE80rDuu6ONESbM=:rwcda
To put these ACLs into effect, rerun the zkfc -formatZK command as described above.
After doing so, you can verify the ACLs from the ZooKeeper CLI as follows:
[zk: localhost:2181(CONNECTED) 1] getAcl /hadoop-ha
'digest,'hdfs-zkfcs:vlUvLnd8MlacsE80rDuu6ONESbM=
: cdrwa
Automatic Failover FAQ
Is it important that I start the ZKFC and NameNode daemons in any particular order?
No. On any given node you may start the ZKFC before or after its corresponding NameNode.
What additional monitoring should I put in place?
You should add monitoring on each host that runs a NameNode to ensure that the ZKFC remains running. In some
types of ZooKeeper failures, for example, the ZKFC may unexpectedly exit, and should be restarted to ensure that
the system is ready for automatic failover. Additionally, you should monitor each of the servers in the ZooKeeper
quorum. If ZooKeeper crashes, automatic failover will not function.
Cloudera Administration | 301
High Availability
What happens if ZooKeeper goes down?
If the ZooKeeper cluster crashes, no automatic failovers will be triggered. However, HDFS will continue to run
without any impact. When ZooKeeper is restarted, HDFS will reconnect with no issues.
Can I designate one of my NameNodes as primary/preferred?
No. Currently, this is not supported. Whichever NameNode is started first will become active. You may choose to
start the cluster in a specific order such that your preferred node starts first.
How can I initiate a manual failover when automatic failover is configured?
Even if automatic failover is configured, you can initiate a manual failover. It will perform a coordinated failover.
Deploying HDFS High Availability
After you have set all of the necessary configuration options, you are ready to start the JournalNodes and the two HA
NameNodes.
Important: Before you start:Make sure you have performed all the configuration and setup tasks
described under Configuring Hardware for HDFS HA on page 292 and Configuring Software for HDFS
HA on page 295, including initializing the HA state in ZooKeeper if you are deploying automatic failover.
Install and Start the JournalNodes
1. Install the JournalNode daemons on each of the machines where they will run.
To install JournalNode on Red Hat-compatible systems:
$ sudo yum install hadoop-hdfs-journalnode
To install JournalNode on Ubuntu and Debian systems:
$ sudo apt-get install hadoop-hdfs-journalnode
To install JournalNode on SLES systems:
$ sudo zypper install hadoop-hdfs-journalnode
2. Start the JournalNode daemons on each of the machines where they will run:
sudo service hadoop-hdfs-journalnode start
Wait for the daemons to start before formatting the primary NameNode (in a new cluster) and before starting the
NameNodes (in all cases).
Format the NameNode (if new cluster)
If you are setting up a new HDFS cluster, format the NameNode you will use as your primary NameNode; see Formatting
the NameNode.
Important: Make sure the JournalNodes have started. Formatting will fail if you have configured the
NameNode to communicate with the JournalNodes, but have not started the JournalNodes.
Initialize the Shared Edits directory (if converting existing non-HA cluster)
If you are converting a non-HA NameNode to HA, initialize the shared edits directory with the edits data from the local
NameNode edits directories:
hdfs namenode -initializeSharedEdits
302 | Cloudera Administration
High Availability
Start the NameNodes
1. Start the primary (formatted) NameNode:
$ sudo service hadoop-hdfs-namenode start
2. Start the standby NameNode:
$ sudo -u hdfs hdfs namenode -bootstrapStandby
$ sudo service hadoop-hdfs-namenode start
Note: If Kerberos is enabled, do not use commands in the form sudo -u <user> <command>;
they will fail with a security error. Instead, use the following commands: $ kinit <user> (if
you are using a password) or $ kinit -kt <keytab> <principal> (if you are using a keytab)
and then, for each command executed by this user, $ <command>
Starting the standby NameNode with the -bootstrapStandby option copies over the contents of the primary
NameNode's metadata directories (including the namespace information and most recent checkpoint) to the standby
NameNode. (The location of the directories containing the NameNode metadata is configured using the configuration
options dfs.namenode.name.dir and dfs.namenode.edits.dir.)
You can visit each NameNode's web page by browsing to its configured HTTP address. Notice that next to the configured
address is the HA state of the NameNode (either "Standby" or "Active".) Whenever an HA NameNode starts and
automatic failover is not enabled, it is initially in the Standby state. If automatic failover is enabled the first NameNode
that is started will become active.
Restart Services (if converting existing non-HA cluster)
If you are converting from a non-HA to an HA configuration, you need to restart the JobTracker and TaskTracker (for
MRv1, if used), or ResourceManager, NodeManager, and JobHistory Server (for YARN), and the DataNodes:
On each DataNode:
$ sudo service hadoop-hdfs-datanode start
On each TaskTracker system (MRv1):
$ sudo service hadoop-0.20-mapreduce-tasktracker start
On the JobTracker system (MRv1):
$ sudo service hadoop-0.20-mapreduce-jobtracker start
Verify that the JobTracker and TaskTracker started properly:
sudo jps | grep Tracker
On the ResourceManager system (YARN):
$ sudo service hadoop-yarn-resourcemanager start
On each NodeManager system (YARN; typically the same ones where DataNode service runs):
$ sudo service hadoop-yarn-nodemanager start
On the MapReduce JobHistory Server system (YARN):
$ sudo service hadoop-mapreduce-historyserver start
Cloudera Administration | 303
High Availability
Deploy Automatic Failover (if it is configured)
If you have configured automatic failover using the ZooKeeper FailoverController (ZKFC), you must install and start the
zkfc daemon on each of the machines that runs a NameNode. Proceed as follows.
To install ZKFC on Red Hat-compatible systems:
$ sudo yum install hadoop-hdfs-zkfc
To install ZKFC on Ubuntu and Debian systems:
$ sudo apt-get install hadoop-hdfs-zkfc
To install ZKFC on SLES systems:
$ sudo zypper install hadoop-hdfs-zkfc
To start the zkfc daemon:
$ sudo service hadoop-hdfs-zkfc start
It is not important that you start the ZKFC and NameNode daemons in a particular order. On any given node you can
start the ZKFC before or after its corresponding NameNode.
You should add monitoring on each host that runs a NameNode to ensure that the ZKFC remains running. In some
types of ZooKeeper failures, for example, the ZKFC may unexpectedly exit, and should be restarted to ensure that the
system is ready for automatic failover.
Additionally, you should monitor each of the servers in the ZooKeeper quorum. If ZooKeeper crashes, then automatic
failover will not function. If the ZooKeeper cluster crashes, no automatic failovers will be triggered. However, HDFS
will continue to run without any impact. When ZooKeeper is restarted, HDFS will reconnect with no issues.
Verifying Automatic Failover
After the initial deployment of a cluster with automatic failover enabled, you should test its operation. To do so, first
locate the active NameNode. As mentioned above, you can tell which node is active by visiting the NameNode web
interfaces.
Once you have located your active NameNode, you can cause a failure on that node. For example, you can use kill
-9 <pid of NN> to simulate a JVM crash. Or you can power-cycle the machine or its network interface to simulate
different kinds of outages. After you trigger the outage you want to test, the other NameNode should automatically
become active within several seconds. The amount of time required to detect a failure and trigger a failover depends
on the configuration of ha.zookeeper.session-timeout.ms, but defaults to 5 seconds.
If the test does not succeed, you may have a misconfiguration. Check the logs for the zkfc daemons as well as the
NameNode daemons in order to further diagnose the issue.
Upgrading an HDFS HA Configuration to the Latest Release
Upgrading to CDH 5
Important: NFS shared storage is not supported in CDH 5. If you are using an HDFS HA configuration
using NFS shared storage, disable the configuration before you begin the upgrade. You can redeploy
HA using Quorum-based storage either before or after the upgrade.
To upgrade an HDFS HA configuration using Quorum-base storage from CDH 4 to the latest release, follow the directions
for upgrading a cluster under Upgrading from CDH 4 to CDH 5.
304 | Cloudera Administration
High Availability
Disabling and Redeploying HDFS HA
Disabling and Redeploying HDFS HA Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
1.
2.
3.
4.
5.
6.
Go to the HDFS service.
Select Actions > Disable High Availability.
Select the hosts for the NameNode and the SecondaryNameNode and click Continue.
Select the HDFS checkpoint directory and click Continue.
Confirm that you want to take this action.
Update the Hive Metastore NameNode.
Cloudera Manager ensures that one NameNode is active, and saves the namespace. Then it stops the standby NameNode,
creates a SecondaryNameNode, removes the standby NameNode role, and restarts all the HDFS services.
Disabling and Redeploying HDFS HA Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
If you need to unconfigure HA and revert to using a single NameNode, either permanently or for upgrade or testing
purposes, proceed as follows.
Important: If you have been using NFS shared storage in CDH 4, you must unconfigure it before
upgrading to CDH 5. Only Quorum-based storage is supported in CDH 5. If you already using
Quorum-based storage, you do not need to unconfigure it in order to upgrade.
Step 1: Shut Down the Cluster
1. Shut down Hadoop services across your entire cluster. Do this from Cloudera Manager; or, if you are not using
Cloudera Manager, run the following command on every host in your cluster:
$ for x in `cd /etc/init.d ; ls hadoop-*` ; do sudo service $x stop ; done
2. Check each host to make sure that there are no processes running as the hdfs, yarn, mapred or httpfs users
from root:
# ps -aef | grep java
Step 2: Unconfigure HA
1. Disable the software configuration.
• If you are using Quorum-based storage and want to unconfigure it, unconfigure the HA properties described
under Enabling HDFS HA Using the Command Line on page 295.
If you intend to redeploy HDFS HA later, comment out the HA properties rather than deleting them.
• If you were using NFS shared storage in CDH 4, you must unconfigure the properties described below before
upgrading to CDH 5.
2. Move the NameNode metadata directories on the standby NameNode. The location of these directories is
configured by dfs.namenode.name.dir and dfs.namenode.edits.dir. Move them to a backup location.
Cloudera Administration | 305
High Availability
Step 3: Restart the Cluster
for x in `cd /etc/init.d ; ls hadoop-*` ; do sudo service $x start ; done
Properties to unconfigure to disable an HDFS HA configuration using NFS shared storage
Important: HDFS HA with NFS shared storage is not supported in CDH 5. Comment out or delete
these properties before attempting to upgrade your cluster to CDH 5. (If you intend to configure HA
with Quorum-based storage under CDH 5, you should comment them out rather than deleting them,
as they are also used in that configuration.)
Unconfigure the following properties:
• In your core-site.xml file:
fs.defaultFS (formerly fs.default.name)
Optionally, you may have configured the default path for Hadoop clients to use the HA-enabled logical URI. For
example, if you used mycluster as the nameservice ID as shown below, this will be the value of the authority
portion of all of your HDFS paths.
<property>
<name>fs.default.name/name>
<value>hdfs://mycluster</value>
</property>
• In your hdfs-site.xml configuration file:
dfs.nameservices
<property>
<name>dfs.nameservices</name>
<value>mycluster</value>
</property>
Note: If you are also using HDFS federation, this configuration setting will include the list of other
nameservices, HA or otherwise, as a comma-separated list.
dfs.ha.namenodes.[nameservice ID]
A list of comma-separated NameNode IDs used by DataNodes to determine all the NameNodes in the cluster. For
example, if you used mycluster as the nameservice ID, and you used nn1 and nn2 as the individual IDs of the
NameNodes, you would have configured this as follows:
<property>
<name>dfs.ha.namenodes.mycluster</name>
<value>nn1,nn2</value>
</property>
dfs.namenode.rpc-address.[nameservice ID]
For both of the previously-configured NameNode IDs, the full address and RPC port of the NameNode process.
For example:
<property>
<name>dfs.namenode.rpc-address.mycluster.nn1</name>
<value>machine1.example.com:8020</value>
</property>
<property>
<name>dfs.namenode.rpc-address.mycluster.nn2</name>
306 | Cloudera Administration
High Availability
<value>machine2.example.com:8020</value>
</property>
Note: You may have similarly configured the servicerpc-address setting.
dfs.namenode.http-address.[nameservice ID]
The addresses for both NameNodes' HTTP servers to listen on. For example:
<property>
<name>dfs.namenode.http-address.mycluster.nn1</name>
<value>machine1.example.com:50070</value>
</property>
<property>
<name>dfs.namenode.http-address.mycluster.nn2</name>
<value>machine2.example.com:50070</value>
</property>
Note: If you have Hadoop's Kerberos security features enabled, and you use HSFTP, you will have
set the https-address similarly for each NameNode.
dfs.namenode.shared.edits.dir
The path to the remote shared edits directory which the standby NameNode uses to stay up-to-date with all the
file system changes the Active NameNode makes. You should have configured only one of these directories,
mounted read/write on both NameNode machines. The value of this setting should be the absolute path to this
directory on the NameNode machines. For example:
<property>
<name>dfs.namenode.shared.edits.dir</name>
<value>file:///mnt/filer1/dfs/ha-name-dir-shared</value>
</property>
dfs.client.failover.proxy.provider.[nameservice ID]
The name of the Java class that the DFS client uses to determine which NameNode is the current active, and
therefore which NameNode is currently serving client requests. The only implementation which shipped with
Hadoop is the ConfiguredFailoverProxyProvider. For example:
<property>
<name>dfs.client.failover.proxy.provider.mycluster</name>
<value>org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider</value>
</property>
dfs.ha.fencing.methods - a list of scripts or Java classes which will be used to fence the active NameNode during
a failover.
Note: If you implemented your own custom fencing method, see the
org.apache.hadoop.ha.NodeFencer class.
• The sshfence fencing method
sshfence - SSH to the active NameNode and kill the process
For example:
<property>
<name>dfs.ha.fencing.methods</name>
Cloudera Administration | 307
High Availability
<value>sshfence</value>
</property>
<property>
<name>dfs.ha.fencing.ssh.private-key-files</name>
<value>/home/exampleuser/.ssh/id_rsa</value>
</property>
Optionally, you may have configured a non-standard username or port to perform the SSH, as shown below,
and also a timeout, in milliseconds, for the SSH:
<property>
<name>dfs.ha.fencing.methods</name>
<value>sshfence([[username][:port]])</value>
</property>
<property>
<name>dfs.ha.fencing.ssh.connect-timeout</name>
<value>30000</value>
<description>
SSH connection timeout, in milliseconds, to use with the builtin
sshfence fencer.
</description>
</property>
• The shell fencing method
shell - run an arbitrary shell command to fence the active NameNode
The shell fencing method runs an arbitrary shell command, which you may have configured as shown below:
<property>
<name>dfs.ha.fencing.methods</name>
<value>shell(/path/to/my/script.sh arg1 arg2 ...)</value>
</property>
Automatic failover: If you configured automatic failover, you configured two additional configuration parameters.
• In your hdfs-site.xml:
<property>
<name>dfs.ha.automatic-failover.enabled</name>
<value>true</value>
</property>
• In your core-site.xml file, add:
<property>
<name>ha.zookeeper.quorum</name>
<value>zk1.example.com:2181,zk2.example.com:2181,zk3.example.com:2181</value>
</property>
Other properties: There are several other configuration parameters which you may have set to control the behavior
of automatic failover, though they were not necessary for most installations. See the configuration section of the
Hadoop documentation for details.
Redeploying HDFS High Availability
If you need to redeploy HA using Quorum-based storage after temporarily disabling it, proceed as follows:
1. Shut down the cluster as described in Step 1: Shut Down the Cluster on page 305.
2. Uncomment the properties you commented out in Step 2: Unconfigure HA on page 305.
3. Deploy HDFS HA, following the instructions under Deploying HDFS High Availability on page 302.
308 | Cloudera Administration
High Availability
Configuring Other CDH Components to Use HDFS HA
You can use the HDFS high availability NameNodes with other components of CDH.
Configuring HBase to Use HDFS HA
Configuring HBase to Use HDFS HA Using the Command Line
To configure HBase to use HDFS HA, proceed as follows.
Shut Down the HBase Cluster
1. Stop the Thrift server and clients:
sudo service hbase-thrift stop
2. Stop the cluster by shutting down the Master and the RegionServers:
• Use the following command on the Master host:
sudo service hbase-master stop
• Use the following command on each host hosting a RegionServer:
sudo service hbase-regionserver stop
Configure hbase.rootdir
Change the distributed file system URI in hbase-site.xml to the name specified in the dfs.nameservices property
in hdfs-site.xml. The clients must also have access to hdfs-site.xml's dfs.client.* settings to properly use
HA.
For example, suppose the HDFS HA property dfs.nameservices is set to ha-nn in hdfs-site.xml. To configure
HBase to use the HA NameNodes, specify that same value as part of your hbase-site.xml's hbase.rootdir value:
<!-- Configure HBase to use the HA NameNode nameservice -->
<property>
<name>hbase.rootdir</name>
<value>hdfs://ha-nn/hbase</value>
</property>
Restart HBase
1. Start the HBase Master.
2. Start each of the HBase RegionServers.
HBase-HDFS HA Troubleshooting
Problem: HMasters fail to start.
Solution: Check for this error in the HMaster log:
2012-05-17 12:21:28,929 FATAL master.HMaster (HMaster.java:abort(1317)) - Unhandled
exception. Starting shutdown.
java.lang.IllegalArgumentException: java.net.UnknownHostException: ha-nn
at
org.apache.hadoop.security.SecurityUtil.buildTokenService(SecurityUtil.java:431)
at
org.apache.hadoop.hdfs.NameNodeProxies.createNonHAProxy(NameNodeProxies.java:161)
at org.apache.hadoop.hdfs.NameNodeProxies.createProxy(NameNodeProxies.java:126)
...
Cloudera Administration | 309
High Availability
If so, verify that Hadoop's hdfs-site.xml and core-site.xml files are in your hbase/conf directory. This may
be necessary if you put your configurations in non-standard places.
Upgrading the Hive Metastore to Use HDFS HA
The Hive metastore can be configured to use HDFS high availability.
Upgrading the Hive Metastore to Use HDFS HA Using Cloudera Manager
1. Go the Hive service.
2. Select Actions > Stop.
Note: You may want to stop the Hue and Impala services first, if present, because they depend
on the Hive service.
3.
4.
5.
6.
Click Stop to confirm the command.
Back up the Hive metastore database.
Select Actions > Update Hive Metastore NameNodes and confirm the command.
Select Actions > Start.
Restart the Hue and Impala services if you stopped them prior to updating the metastore.
Upgrading the Hive Metastore to Use HDFS HA Using the Command Line
To configure the Hive metastore to use HDFS HA, change the records to reflect the location specified in the
dfs.nameservices property, using the Hive metatool to obtain and change the locations.
Note: Before attempting to upgrade the Hive metastore to use HDFS HA, shut down the metastore
and back it up to a persistent store.
If you are unsure which version of Avro SerDe is used, use both the serdePropKey and tablePropKey arguments.
For example:
$ hive --service metatool -listFSRoot
hdfs://oldnamenode.com/user/hive/warehouse
$ metatool -updateLocation hdfs://nameservice1 hdfs://oldnamenode.com -tablePropKey
avro.schema.url
-serdePropKey schema.url
$ hive --service metatool -listFSRoot
hdfs://nameservice1/user/hive/warehouse
where:
• hdfs://oldnamenode.com/user/hive/warehouse identifies the NameNode location.
• hdfs://nameservice1 specifies the new location and should match the value of the dfs.nameservices
property.
• tablePropKey is a table property key whose value field may reference the HDFS NameNode location and hence
may require an update. To update the Avro SerDe schema URL, specify avro.schema.url for this argument.
• serdePropKey is a SerDe property key whose value field may reference the HDFS NameNode location and hence
may require an update. To update the Haivvero schema URL, specify schema.url for this argument.
Note: The Hive metatool is a best effort service that tries to update as many Hive metastore records
as possible. If it encounters an error during the update of a record, it skips to the next record.
Configuring Hue to Work with HDFS HA
1. Add the HttpFS role.
2. After the command has completed, go to the Hue service.
3. Click the Configuration tab.
310 | Cloudera Administration
High Availability
4. Locate the HDFS Web Interface Role property or search for it by typing its name in the Search box.
5. Select the HttpFS role you just created instead of the NameNode role, and save your changes.
6. Restart the Hue service.
Configuring Impala to Work with HDFS HA
1. Complete the steps to reconfigure the Hive metastore database, as described in the preceding section. Impala
shares the same underlying database with Hive, to manage metadata for databases, tables, and so on.
2. Issue the INVALIDATE METADATA statement from an Impala shell. This one-time operation makes all Impala
daemons across the cluster aware of the latest settings for the Hive metastore database. Alternatively, restart
the Impala service.
Configuring Oozie to Use HDFS HA
To configure an Oozie workflow to use HDFS HA, use the HDFS nameservice instead of the NameNode URI in the
<name-node> element of the workflow.
Example:
<action name="mr-node">
<map-reduce>
<job-tracker>${jobTracker}</job-tracker>
<name-node>hdfs://ha-nn
where ha-nn is the value of dfs.nameservices in hdfs-site.xml.
Administering an HDFS High Availability Cluster
Manually Failing Over to the Standby NameNode
Manually Failing Over to the Standby NameNode Using Cloudera Manager
If you are running an HDFS service with HA enabled, you can manually cause the active NameNode to failover to the
standby NameNode. This is useful for planned downtime—for hardware changes, configuration changes, or software
upgrades of your primary host.
1.
2.
3.
4.
Go to the HDFS service.
Click the Instances tab.
Select Actions > Manual Failover. (This option does not appear if HA is not enabled for the cluster.)
From the pop-up, select the NameNode that should be made active, then click Manual Failover.
Note: For advanced use only: You can set the Force Failover checkbox to force the selected
NameNode to be active, irrespective of its state or the other NameNode's state. Forcing a failover
will first attempt to failover the selected NameNode to active mode and the other NameNode
to standby mode. It will do so even if the selected NameNode is in safe mode. If this fails, it will
proceed to transition the selected NameNode to active mode. To avoid having two NameNodes
be active, use this only if the other NameNode is either definitely stopped, or can be transitioned
to standby mode by the first failover step.
5. When all the steps have been completed, click Finish.
Cloudera Manager transitions the NameNode you selected to be the active NameNode, and the other NameNode to
be the standby NameNode. HDFS should never have two active NameNodes.
Manually Failing Over to the Standby NameNode Using the Command Line
To initiate a failover between two NameNodes, run the command hdfs haadmin -failover.
This command causes a failover from the first provided NameNode to the second. If the first NameNode is in the
Standby state, this command simply transitions the second to the Active state without error. If the first NameNode is
in the Active state, an attempt will be made to gracefully transition it to the Standby state. If this fails, the fencing
methods (as configured by dfs.ha.fencing.methods) will be attempted in order until one of the methods succeeds.
Cloudera Administration | 311
High Availability
Only after this process will the second NameNode be transitioned to the Active state. If no fencing method succeeds,
the second NameNode will not be transitioned to the Active state, and an error will be returned.
Note: Running hdfs haadmin -failover from the command line works whether you have
configured HA from the command line or using Cloudera Manager. This means you can initiate a
failover manually even if Cloudera Manager is unavailable.
Moving an HA NameNode to a New Host
Moving an HA NameNode to a New Host Using Cloudera Manager
See Moving Highly Available NameNode, Failover Controller, and JournalNode Roles Using the Migrate Roles Wizard
on page 137.
Moving an HA NameNode to a New Host Using the Command Line
Use the following steps to move one of the NameNodes to a new host.
In this example, the current NameNodes are called nn1 and nn2, and the new NameNode is nn2-alt. The example
assumes the following:
• nn2-alt is already a member of this CDH 5 HA cluster.
• Automatic failover is configured.
• A JournalNode on nn2, in addition to the NameNode service, is to be moved to nn2-alt.
The procedure moves the NameNode and JournalNode services from nn2 to nn2-alt, reconfigures nn1 to recognize
the new location of the JournalNode, and restarts nn1 and nn2-alt in the new HA configuration.
Step 1: Make sure that nn1 is the active NameNode
Make sure that the NameNode that is not going to be moved is active; in this example, nn1 must be active. You can
use the NameNodes' web UIs to see which is active; see Start the NameNodes on page 303.
If nn1 is not the active NameNode, use the hdfs haadmin -failover command to initiate a failover from nn2 to
nn1:
hdfs haadmin -failover nn2 nn1
Step 2: Stop services on nn2
Once you have made sure that the node to be moved is inactive, stop services on that node; in this example, stop
services on nn2. Stop the NameNode, the ZKFC daemon if this an automatic-failover deployment, and the JournalNode
if you are moving it:
1. Stop the NameNode daemon:
$ sudo service hadoop-hdfs-namenode stop
2. Stop the ZKFC daemon if it is running:
$ sudo service hadoop-hdfs-zkfc stop
3. Stop the JournalNode daemon if it is running:
$ sudo service hadoop-hdfs-journalnode stop
4. Make sure these services are not set to restart on boot. If you are not planning to use nn2 as a NameNode again,
you may want to remove the services.
312 | Cloudera Administration
High Availability
Step 3: Install the NameNode daemon on nn2-alt
See the instructions for installing hadoop-hdfs-namenode under Step 3: Install CDH 5 with YARN or Step 4: Install
CDH 5 with MRv1.
Step 4: Configure HA on nn2-alt
See Enabling HDFS HA on page 292 for the properties to configure on nn2-alt in core-site.xml and hdfs-site.xml,
and explanations and instructions. You should copy the values that are already set in the corresponding files on nn2.
• If you are relocating a JournalNode to nn2-alt, follow these directions to install it, but do not start it yet.
• If you are using automatic failover, make sure you follow the instructions for configuring the required properties
on nn2-alt and initializing the HA state in ZooKeeper.
Note: You do not need to shut down the cluster to do this if automatic failover is already
configured as your failover method; shutdown is required only if you are switching from manual
to automatic failover.
Step 5: Copy the contents of the dfs.name.dir and dfs.journalnode.edits.dir directories to nn2-alt
Use rsync or a similar tool to copy the contents of the dfs.name.dir directory, and the
dfs.journalnode.edits.dir directory if you are moving the JournalNode, from nn2 to nn2-alt.
Step 6: If you are moving a JournalNode, update dfs.namenode.shared.edits.dir on nn1
If you are relocating a JournalNode from nn2 to nn2-alt, update dfs.namenode.shared.edits.dir in
hdfs-site.xml on nn1 to reflect the new hostname. See this section for more information about
dfs.namenode.shared.edits.dir.
Step 7: If you are using automatic failover, install the zkfc daemon on nn2-alt
For instructions, see Deploy Automatic Failover (if it is configured) on page 304, but do not start the daemon yet.
Step 8: Start services on nn2-alt
Start the NameNode; start the ZKFC for automatic failover; and install and start a JournalNode if you want one to run
on nn2-alt:
1. Start the JournalNode daemon:
$ sudo service hadoop-hdfs-journalnode start
2. Start the NameNode daemon:
$ sudo service hadoop-hdfs-namenode start
3. Start the ZKFC daemon:
$ sudo service hadoop-hdfs-zkfc start
4. Set these services to restart on boot; for example on a RHEL-compatible system:
$ sudo chkconfig hadoop-hdfs-namenode on
$ sudo chkconfig hadoop-hdfs-zkfc on
$ sudo chkconfig hadoop-hdfs-journalnode on
Cloudera Administration | 313
High Availability
Step 9: If you are relocating a JournalNode, fail over to nn2-alt
hdfs haadmin -failover nn1 nn2-alt
Step 10: If you are relocating a JournalNode, restart nn1
Restart the NameNode daemon on nn1 to force it to re-read the configuration:
$ sudo service hadoop-hdfs-namenode stop
$ sudo service hadoop-hdfs-namenode start
Other HDFS haadmin Commands
After your HA NameNodes are configured and started, you have access to additional commands to administer your
HA HDFS cluster. Specifically, you should familiarize yourself with the subcommands of the hdfs haadmin command.
This page describes high-level uses of some important subcommands. For specific usage information of each
subcommand, run hdfs haadmin -help <command>.
getServiceState
getServiceState - Determine whether the given NameNode is active or standby.
Connect to the provided NameNode to determine its current state, printing either "standby" or "active" to STDOUT as
appropriate. This subcommand might be used by cron jobs or monitoring scripts, which need to behave differently
based on whether the NameNode is currently active or standby.
checkHealth
checkHealth - Check the health of the given NameNode.
Connect to the provided NameNode to check its health. The NameNode can perform some diagnostics on itself,
including checking if internal services are running as expected. This command returns 0 if the NameNode is healthy,
non-zero otherwise. You can use this command for monitoring purposes.
Using the dfsadmin Command When HA Is Enabled
By default, applicable dfsadmin command options are run against both active and standby NameNodes. To limit an
option to a specific NameNode, use the -fs option. For example,
To turn safe mode on for both NameNodes, run:
hdfs dfsadmin -safemode enter
To turn safe mode on for a single NameNode, run:
hdfs dfsadmin -fs hdfs://<host>:<port> -safemode enter
For a full list of dfsadmin command options, run: hdfs dfsadmin -help.
Converting From an NFS-mounted Shared Edits Directory to Quorum-based Storage
Converting From an NFS-mounted Shared Edits Directory to Quorum-based Storage Using Cloudera Manager
Converting a HA configuration from using an NFS-mounted shared edits directory to Quorum-based storage involves
disabling the current HA configuration then enabling HA using Quorum-based storage.
1. Disable HA.
2. Although the standby NameNode role is removed, its name directories are not deleted. Empty these directories.
3. Enable HA with Quorum-based storage.
Converting From an NFS-mounted Shared Edits Directory to Quorum-based Storage Using the Command Line
To switch from shared storage using NFS to Quorum-based storage, proceed as follows:
314 | Cloudera Administration
High Availability
1. Disable HA.
2. Redeploy HA using Quorum-based storage.
Changing a Nameservice Name for Highly Available HDFS Using Cloudera Manager
For background on HDFS namespaces and HDFS high availability, see Managing Federated Nameservices on page 134
and Enabling HDFS HA Using Cloudera Manager on page 292.
1. Stop all services except ZooKeeper.
2. On a ZooKeeper server host, run zookeeper-client.
a. Execute the following to remove the configured nameservice. This example assumes the name of the
nameservice is nameservice1. You can identify the nameservice from the Federation and High Availability
section on the HDFS Instances tab:
rmr /hadoop-ha/nameservice1
3. In the Cloudera Manager Admin Console, update the NameNode nameservice name.
a.
b.
c.
d.
Go to the HDFS service.
Click the Configuration tab.
Type nameservice in the Search field.
For the NameNode Nameservice property, type the nameservice name in the NameNode (instance_name)
field. The name must be unique and can contain only alphanumeric characters.
e. Type quorum in the Search field.
f. For the Quorum-based Storage Journal name property, type the nameservice name in the NameNode
(instance_name) field.
g. Click Save Changes to commit the changes.
4. Click the Instances tab.
5. In the Federation and High Availability pane, select Actions > Initialize High Availability State in ZooKeeper.
6. Go to the Hive service.
7. Select Actions > Update Hive Metastore NameNodes.
8. Go to the HDFS service.
9. Click the Instances tab.
10. Select the checkboxes next to the JournalNode role instances.
11. Select Actions for Selected > Start.
12. Click a NameNode role instance.
13. Select Actions > Initialize Shared Edits Directory.
14. Click the Cloudera Manager logo to return to the Home page.
15. Redeploy client configuration files.
16. Start all services except ZooKeeper.
MapReduce (MRv1) and YARN (MRv2) High Availability
This section covers:
YARN (MRv2) ResourceManager High Availability
The YARN ResourceManager is responsible for tracking the resources in a cluster and scheduling applications (for
example, MapReduce jobs). Before CDH 5, the ResourceManager was a single point of failure in a YARN cluster. The
ResourceManager high availability (HA) feature adds redundancy in the form of an active-standby ResourceManager
pair to remove this single point of failure. Furthermore, upon failover from the active ResourceManager to the standby,
the applications can resume from the last state saved to the state store; for example, map tasks in a MapReduce job
Cloudera Administration | 315
High Availability
are not executed again if a failover to a new active ResourceManager occurs after the completion of the map phase.
This allows events such the following to be handled without any significant performance effect on running applications:
• Unplanned events such as machine crashes
• Planned maintenance events such as software or hardware upgrades on the machine running the ResourceManager
ResourceManager HA requires ZooKeeper and HDFS services to be running.
Architecture
ResourceManager HA is implemented by means of an active-standby pair of ResourceManagers. On start-up, each
ResourceManager is in the standby state; the process is started, but the state is not loaded. When one of the
ResourceManagers is transitioning to the active state, the ResourceManager loads the internal state from the designated
state store and starts all the internal services. The stimulus to transition to active comes from either the administrator
(through the CLI) or through the integrated failover controller when automatic failover is enabled. The subsections
that follow provide more details about the components of ResourceManager HA.
ResourceManager Restart
Restarting the ResourceManager allows for the recovery of in-flight applications if recovery is enabled. To achieve this,
the ResourceManager stores its internal state, primarily application-related data and tokens, to the
ResourceManagerStateStore; the cluster resources are re-constructed when the NodeManagers connect. The
available alternatives for the state store are MemoryResourceManagerStateStore (a memory-based implementation),
FileSystemResourceManagerStateStore (file system-based implementation; HDFS can be used for the file
system), and ZKResourceManagerStateStore (ZooKeeper-based implementation).
Fencing
When running two ResourceManagers, a split-brain situation can arise where both ResourceManagers assume they
are active. To avoid this, only a single ResourceManager should be able to perform active operations and the other
ResourceManager should be "fenced". The ZooKeeper-based state store (ZKResourceManagerStateStore) allows
only a single ResourceManager to make changes to the stored state, implicitly fencing the other ResourceManager.
This is accomplished by the ResourceManager claiming exclusive create-delete permissions on the root znode. The
ACLs on the root znode are automatically created based on the ACLs configured for the store; in case of secure clusters,
Cloudera recommends that you set ACLs for the root node such that both ResourceManagers share read-write-admin
access, but have exclusive create-delete access. The fencing is implicit and doesn't require explicit configuration (as
fencing in HDFS and MRv1 does). You can plug in a custom "Fencer" if you choose to – for example, to use a different
implementation of the state store.
Configuration and FailoverProxy
In an HA setting, you should configure two ResourceManagers to use different ports (for example, ports on different
hosts). To facilitate this, YARN uses the notion of an ResourceManager Identifier (rm-id). Each ResourceManager has
a unique rm-id, and all the RPC configurations (<rpc-address>; for example yarn.resourcemanager.address) for
that ResourceManager can be configured via <rpc-address>.<rm-id>. Clients, ApplicationMasters, and
NodeManagers use these RPC addresses to talk to the active ResourceManager automatically, even after a failover.
To achieve this, they cycle through the list of ResourceManagers in the configuration. This is done automatically and
doesn't require any configuration (as it does in HDFS and MapReduce (MRv1)).
Automatic Failover
By default, ResourceManager HA uses ZKFC (ZooKeeper-based failover controller) for automatic failover in case the
active ResourceManager is unreachable or goes down. Internally, the StandbyElector is used to elect the active
ResourceManager. The failover controller runs as part of the ResourceManager (not as a separate process as in HDFS
and MapReduce v1) and requires no further setup after the appropriate properties are configured in yarn-site.xml.
You can plug in a custom failover controller if you prefer.
Manual Transitions and Failover
You can use the command-line tool yarn rmadmin to transition a particular ResourceManager to active or standby
state, to fail over from one ResourceManager to the other, to get the HA state of an ResourceManager, and to monitor
an ResourceManager's health.
316 | Cloudera Administration
High Availability
Configuring YARN (MRv2) ResourceManager High Availability Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
You can use Cloudera Manager to configure CDH 5 or higher for ResourceManager high availability (HA). Cloudera
Manager supports automatic failover of the ResourceManager. It does not provide a mechanism to manually force a
failover through the Cloudera Manager user interface.
Important: Enabling or disabling HA will cause the previous monitoring history to become unavailable.
Enabling High Availability
1. Go to the YARN service.
2. Select Actions > Enable High Availability. A screen showing the hosts that are eligible to run a standby
ResourceManager displays. The host where the current ResourceManager is running is not available as a choice.
3. Select the host where you want the standby ResourceManager to be installed, and click Continue. Cloudera
Manager proceeds to execute a set of commands that stop the YARN service, add a standby ResourceManager,
initialize the ResourceManager high availability state in ZooKeeper, restart YARN, and redeploy the relevant client
configurations.
4. Work preserving recovery is enabled for the ResourceManager by default when you enable ResourceManager HA
in Cloudera Manager. For more information, including instructions on disabling work preserving recovery, see
Work Preserving Recovery for YARN Components on page 323.
Note: ResourceManager HA doesn't affect the JobHistory Server (JHS). JHS doesn't maintain any
state, so if the host fails you can simply assign it to a new host. You can also enable process auto-restart
by doing the following:
1.
2.
3.
4.
5.
Go to the YARN service.
Click the Configuration tab.
Select Scope > JobHistory Server.
Select Category > Advanced.
Locate the Automatically Restart Process property or search for it by typing its name in the Search
box.
6. Click Edit Individual Values
7. Select the JobHistory Server Default Group.
8. Restart the JobHistory Server role.
Disabling High Availability
1. Go to the YARN service.
2. Select Actions > Disable High Availability. A screen showing the hosts running the ResourceManagers displays.
3. Select which ResourceManager (host) you want to remain as the single ResourceManager, and click Continue.
Cloudera Manager executes a set of commands that stop the YARN service, remove the standby ResourceManager
and the Failover Controller, restart the YARN service, and redeploy client configurations.
Configuring YARN (MRv2) ResourceManager High Availability Using the Command Line
To configure and start ResourceManager HA, proceed as follows.
Cloudera Administration | 317
High Availability
Stop the YARN daemons
Stop the MapReduce JobHistory service, ResourceManager service, and NodeManager on all nodes where they are
running, as follows:
$ sudo service hadoop-mapreduce-historyserver stop
$ sudo service hadoop-yarn-resourcemanager stop
$ sudo service hadoop-yarn-nodemanager stop
Configure Manual Failover, and Optionally Automatic Failover
To configure failover:
Note:
Configure the following properties in yarn-site.xml as shown, whether you are configuring manual
or automatic failover. They are sufficient to configure manual failover. You need to configure additional
properties for automatic failover.
Name
Used On
Default Value
Recommended Value Description
yarn.resourcemanager. ResourceManager, false
ha.enabled
NodeManager,
Client
true
yarn.resourcemanager. ResourceManager, (None)
ha.rm-ids
NodeManager,
Client
Cluster-specific, Comma-separated list
yarn.resourcemanager. ResourceManager
ha.id
ResourceManager-specific, Id of the current
(None)
e.g., rm1,rm2
e.g., rm1
yarn.resourcemanager. ResourceManager, (None)
address.<rm-id>
Client
Cluster-specific
Enable HA
of ResourceManager
ids in this cluster.
ResourceManager.
Must be set explicitly
on each
ResourceManager to
the appropriate value.
The value of
yarn.resourcemanager.
address
(Client-ResourceManager
RPC) for this
ResourceManager.
Must be set for all
ResourceManagers.
yarn.resourcemanager. ResourceManager, (None)
scheduler.address.<rm-id>
Client
Cluster-specific
The value of
yarn.resourcemanager.
scheduler.address
(AM-ResourceManager
RPC) for this
ResourceManager.
Must beset for all
ResourceManagers.
yarn.resourcemanager. ResourceManager, (None)
admin.address.<rm-id>
Client/Admin
Cluster-specific
The value of
yarn.resourcemanager.
admin.address
(ResourceManager
administration) for
318 | Cloudera Administration
High Availability
Name
Used On
Default Value
Recommended Value Description
this
ResourceManager.
Must be set for all
ResourceManagers.
yarn.resourcemanager. ResourceManager, (None)
resource-tracker.address.
<rm-id>
NodeManager
Cluster-specific
The value of
yarn.resourcemanager.
resource-tracker.address
(NM-ResourceManager
RPC) for this
ResourceManager.
Must be set for all
ResourceManagers.
yarn.resourcemanager. ResourceManager, (None)
webapp.address.<rm-id> Client
Cluster-specific
The value of
yarn.resourcemanager.
webapp.address
(ResourceManager
webapp) for this
ResourceManager.Must
be set for all
ResourceManagers.
yarn.resourcemanager. ResourceManager
recovery.enabled
false
true
Enable job recovery
on ResourceManager
restart or failover.
yarn.resourcemanager. ResourceManager
store.class
org.apache.hadoop.
yarn.server.
resourcemanager.
recovery.
FileSystemResourceManagerStateStore
org.apache.
hadoop.yarn.
server.
resourcemanager.
recovery.
ZKResourceManagerStateStore
The
ResourceManagerStateStore
implementation to
use to store the
ResourceManager's
internal state. The
ZooKeeper- based
store supports fencing
implicitly. That it, it
allows a single
ResourceManager to
make multiple
changes at a time, and
hence is
recommended.
yarn.resourcemanager. ResourceManager
zk-address
(None)
Clusterspecific
The ZooKeeper
quorum to use to
store the
ResourceManager's
internal state.
yarn.resourcemanager. ResourceManager
zk-acl
world:anyone:rwcda Clusterspecific
The ACLs the
ResourceManager
uses for the znode
structure to store the
internal state.
yarn.resourcemanager.zk- ResourceManager
state-store.root-node.acl
(None)
The ACLs used for the
root node of the
ZooKeeper state
Clusterspecific
Cloudera Administration | 319
High Availability
Name
Used On
Default Value
Recommended Value Description
store. The ACLs set
here should allow
both
ResourceManagers to
read, write, and
administer, with
exclusive access to
create and delete. If
nothing is specified,
the root node ACLs
are automatically
generated on the
basis of the ACLs
specified through
yarn.resourcemanager.zk-acl.
But that leaves a
security hole in a
secure setup.
To configure automatic failover:
Configure the following additional properties in yarn-site.xml to configure automatic failover.
Configure work preserving recovery:
Optionally, you can configure work preserving recovery for the Resource Manager and Node Managers. See Work
Preserving Recovery for YARN Components on page 323.
Name
Used On
Default Value
Recommended Value Description
yarn.resourcemanager. ResourceManager
ha.automatic-failover.enabled
true
true
Enable automatic
failover
yarn.resourcemanager. ResourceManager
ha.automatic-failover.embedded
true
true
Use the
EmbeddedElectorService
to pick an active
ResourceManager
from the ensemble
yarn.resourcemanager. ResourceManager
cluster-id
No default value.
Clusterspecific
Cluster name used by
the
ActiveStandbyElector
to elect one of the
ResourceManagers as
leader.
The following is a sample yarn-site.xml showing these properties configured, including work preserving recovery
for both ResourceManager and NM:
<configuration>
<!-- Resource Manager Configs -->
<property>
<name>yarn.resourcemanager.connect.retry-interval.ms</name>
<value>2000</value>
</property>
<property>
<name>yarn.resourcemanager.ha.enabled</name>
<value>true</value>
320 | Cloudera Administration
High Availability
</property>
<property>
<name>yarn.resourcemanager.ha.automatic-failover.enabled</name>
<value>true</value>
</property>
<property>
<name>yarn.resourcemanager.ha.automatic-failover.embedded</name>
<value>true</value>
</property>
<property>
<name>yarn.resourcemanager.cluster-id</name>
<value>pseudo-yarn-rm-cluster</value>
</property>
<property>
<name>yarn.resourcemanager.ha.rm-ids</name>
<value>rm1,rm2</value>
</property>
<property>
<name>yarn.resourcemanager.ha.id</name>
<value>rm1</value>
</property>
<property>
<name>yarn.resourcemanager.scheduler.class</name>
<value>org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FairScheduler</value>
</property>
<property>
<name>yarn.resourcemanager.recovery.enabled</name>
<value>true</value>
</property>
<property>
<name>yarn.resourcemanager.store.class</name>
<value>org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKResourceManagerStateStore</value>
</property>
<property>
<name>yarn.resourcemanager.zk-address</name>
<value>localhost:2181</value>
</property>
<property>
<name>yarn.app.mapreduce.am.scheduler.connection.wait.interval-ms</name>
<value>5000</value>
</property>
<property>
<name>yarn.resourcemanager.work-preserving-recovery.enabled</name>
<value>true</value>
</property>
<!-- ResourceManager1 configs -->
<property>
<name>yarn.resourcemanager.address.rm1</name>
<value>host1:23140</value>
</property>
<property>
<name>yarn.resourcemanager.scheduler.address.rm1</name>
<value>host1:23130</value>
</property>
<property>
<name>yarn.resourcemanager.webapp.https.address.rm1</name>
<value>host1:23189</value>
</property>
<property>
<name>yarn.resourcemanager.webapp.address.rm1</name>
<value>host1:23188</value>
</property>
<property>
<name>yarn.resourcemanager.resource-tracker.address.rm1</name>
<value>host1:23125</value>
</property>
<property>
<name>yarn.resourcemanager.admin.address.rm1</name>
Cloudera Administration | 321
High Availability
<value>host1:23141</value>
</property>
<!-- ResourceManager2 configs -->
<property>
<name>yarn.resourcemanager.address.rm2</name>
<value>host2:23140</value>
</property>
<property>
<name>yarn.resourcemanager.scheduler.address.rm2</name>
<value>host2:23130</value>
</property>
<property>
<name>yarn.resourcemanager.webapp.https.address.rm2</name>
<value>host2:23189</value>
</property>
<property>
<name>yarn.resourcemanager.webapp.address.rm2</name>
<value>host2:23188</value>
</property>
<property>
<name>yarn.resourcemanager.resource-tracker.address.rm2</name>
<value>host2:23125</value>
</property>
<property>
<name>yarn.resourcemanager.admin.address.rm2</name>
<value>host2:23141</value>
</property>
<!-- Node Manager Configs -->
<property>
<description>Address where the localizer IPC is.</description>
<name>yarn.nodemanager.localizer.address</name>
<value>0.0.0.0:23344</value>
</property>
<property>
<description>NM Webapp address.</description>
<name>yarn.nodemanager.webapp.address</name>
<value>0.0.0.0:23999</value>
</property>
<property>
<name>yarn.nodemanager.aux-services</name>
<value>mapreduce_shuffle</value>
</property>
<property>
<name>yarn.nodemanager.local-dirs</name>
<value>/tmp/pseudo-dist/yarn/local</value>
</property>
<property>
<name>yarn.nodemanager.log-dirs</name>
<value>/tmp/pseudo-dist/yarn/log</value>
</property>
<property>
<name>mapreduce.shuffle.port</name>
<value>23080</value>
</property>
<property>
<name>yarn.resourcemanager.work-preserving-recovery.enabled</name>
<value>true</value>
</property>
</configuration>
Restart the YARN daemons
Start the MapReduce JobHistory server, ResourceManager, and NodeManager on all nodes where they were previously
running, as follows:
$ sudo service hadoop-mapreduce-historyserver start
$ sudo service hadoop-yarn-resourcemanager start
$ sudo service hadoop-yarn-nodemanager start
322 | Cloudera Administration
High Availability
Using yarn rmadmin to Administer ResourceManager HA
You can use yarn rmadmin on the command line to manage your ResourceManager HA deployment. yarn rmadmin
has the following options related to ResourceManager HA:
[-transitionToActive serviceId]
[-transitionToStandby serviceId]
[-getServiceState serviceId]
[-checkHealth <serviceId]
[-help <command>]
where serviceId is the rm-id.
Note: Even though -help lists the -failover option, it is not supported by yarn rmadmin.
Work Preserving Recovery for YARN Components
CDH 5.2 introduces work preserving recovery for the YARN ResourceManager and NodeManager. With work preserving
recovery enabled, if a ResourceManager or NodeManager restarts, no in-flight work is lost. You can configure work
preserving recovery separately for a ResourceManager or NodeManager.
Note: YARN does not support high availability for the Job History Server (JHS). If the JHS goes down,
Cloudera Manager will restart it automatically.
Prerequisites
To use work preserving recovery for the ResourceManager, you need to first enable High Availability for the
ResourceManager. See YARN (MRv2) ResourceManager High Availability on page 315 for more information.
Configuring Work Preserving Recovery Using Cloudera Manager
Enabling Work Preserving Recovery on ResourceManager with Cloudera Manager
If you use Cloudera Manager and you enable YARN (MRv2) ResourceManager High Availability on page 315, work
preserving recovery is enabled by default for the ResourceManager.
Disabling Work Preserving Recovery on ResourceManager with Cloudera Manager
To disable the feature for the ResourceManager, change the value of
yarn.resourcemanager.work-preserving-recovery.enabled to false in the yarn-site.xml using an
advanced configuration snippet.
1.
2.
3.
4.
In Cloudera Manager>Home>Status, click the yarn link.
Click the Configuration tab.
Search for ResourceManager yarn-site.xml.
In the ResourceManager Base Group field, enter the following XML element, setting the feature to false.
<property>
<name>yarn.resourcemanager.work-preserving-recovery.enabled</name>
<value>false</value>
</property>
5. Click Save Changes.
Cloudera Administration | 323
High Availability
Enabling Work Preserving Recovery on NodeManager with Cloudera Manager
The default value for the recovery directory is ${hadoop.tmp.dir}/yarn-nm-recovery. This location usually
points to the /tmp directory on the local filesystem. Because many operating systems do not preserve the contents
of the /tmp directory across a reboot, Cloudera strongly recommends that you change the location of
yarn.nodemanager.recover.dir to a different directory on the local filesystem. The example below uses
/home/cloudera/recovery.
To enable work preserving recovery for a given NodeManager:
1. Edit the advanced configuration snippet for yarn-site.xml on that NodeManager, and set the value of
yarn.nodemanager.recovery.enabled to true.
2. Configure the directory on the local filesystem where state information is stored when work preserving recovery
is enabled.
a.
b.
c.
d.
e.
In Cloudera Manager Home, click the yarn link.
Click the Configuration tab.
Search for yarn.nodemanager.recovery.dir.
Enter the directory path in the NodeManager Base Group field (for example, /home/cloudera/recovery).
Click Save Changes.
Configuring Work Preserving Recovery Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
After enabling YARN (MRv2) ResourceManager High Availability on page 315, add the following property elements to
yarn-site.xml on the ResourceManager and all NodeManagers.
1. Set the value of yarn.resourcemanager.work-preserving-recovery.enabled to true to enable work
preserving recovery for the ResourceManager, and set the value of yarn.nodemanager.recovery.enabled
to true for the NodeManager.
2. For each NodeManager, configure the directory on the local filesystem where state information is stored when
work preserving recovery is enabled, Set yarn.nodemanager.recovery.dir to a local filesystem directory.
The default value is ${hadoop.tmp.dir}/yarn-nm-recovery. This location usually points to the /tmp directory
on the local filesystem. Because many operating systems do not preserve the contents of the /tmp directory
across a reboot, Cloudera strongly recommends changing the location of yarn.nodemanager.recover.dir
to a different directory on the local filesystem. The example below uses /home/cloudera/recovery.
3. Configure a valid RPC address for the NodeManager by setting yarn.nodemanager.address to an address with
a specific port number (such as 0.0.0.0:45454). Ephemeral ports (default is port 0) cannot be used for the
NodeManager's RPC server; this could cause the NodeManager to use different ports before and after a restart,
preventing clients from connecting to the NodeManager. The NodeManager RPC address is also important for
auxiliary services that run in a YARN cluster.
Auxiliary services should be designed to support recoverability by reloading the previous state after a NodeManager
restarts. An example auxiliary service, the ShuffleHandler service for MapReduce, follows the correct pattern for an
auxiliary service that supports work preserving recovery of the NodeManager.
Example Configuration for Work Preserving Recovery
The following example configuration can be used with a Cloudera Manager advanced configuration snippet or added
to yarn-site.xml directly if you do not use Cloudera Manager. Adjust the configuration to suit your environment.
<property>
<name>yarn.resourcemanager.work-preserving-recovery.enabled</name>
324 | Cloudera Administration
High Availability
<value>true</value>
<description>Whether to enable work preserving recovery for the Resource
Manager</description>
</property>
<property>
<name>yarn.nodemanager.recovery.enabled</name>
<value>true</value>
<description>Whether to enable work preserving recovery for the Node
Manager</description>
</property>
<property>
<name>yarn.nodemanager.recovery.dir</name>
<value>/home/cloudera/recovery</value>
<description>The location for stored state on the Node Manager, if work preserving
recovery
is enabled.</description>
</property>
<property>
<name>yarn.nodemanager.address</name>
<value>0.0.0.0:45454</value>
</property>
MapReduce (MRv1) JobTracker High Availability
Follow the instructions in this section to configure high availability (HA) for JobTracker.
Configuring MapReduce (MRv1) JobTracker High Availability Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
You can use Cloudera Manager to configure CDH 4.3 or higher for JobTracker high availability (HA). Although it is
possible to configure JobTracker HA with CDH 4.2, it is not recommended. Rolling restart, decommissioning of
TaskTrackers, and rolling upgrade of MapReduce from CDH 4.2 to CDH 4.3 are not supported when JobTracker HA is
enabled.
Cloudera Manager supports automatic failover of the JobTracker. It does not provide a mechanism to manually force
a failover through the Cloudera Manager user interface.
Important: Enabling or disabling JobTracker HA will cause the previous monitoring history to become
unavailable.
Enabling JobTracker High Availability
The Enable High Availability workflow leads you through adding a second (standby) JobTracker:
1. Go to the MapReduce service.
2. Select Actions > Enable High Availability. A screen showing the hosts that are eligible to run a standby JobTracker
displays. The host where the current JobTracker is running is not available as a choice.
3. Select the host where you want the Standby JobTracker to be installed, and click Continue.
4. Enter a directory location on the local filesystem for each JobTracker host. These directories will be used to store
job configuration data.
• You may enter more than one directory, though it is not required. The paths do not need to be the same on
both JobTracker hosts.
• If the directories you specify do not exist, they will be created with the appropriate permissions. If they already
exist, they must be empty and have the appropriate permissions.
• If the directories are not empty, Cloudera Manager will not delete the contents.
5. Optionally use the checkbox under Advanced Options to force initialize the ZooKeeper znode for auto-failover.
6. Click Continue. Cloudera Manager executes a set of commands that stop the MapReduce service, add a standby
JobTracker and Failover controller, initialize the JobTracker high availability state in ZooKeeper, create the job
status directory, restart MapReduce, and redeploy the relevant client configurations.
Cloudera Administration | 325
High Availability
Disabling JobTracker High Availability
1. Go to the MapReduce service.
2. Select Actions > Disable High Availability. A screen showing the hosts running the JobTrackers displays.
3. Select which JobTracker (host) you want to remain as the single JobTracker, and click Continue. Cloudera Manager
executes a set of commands that stop the MapReduce service, remove the standby JobTracker and the Failover
Controller, restart the MapReduce service, and redeploy client configurations.
Configuring MapReduce (MRv1) JobTracker High Availability Using the Command Line
If you are running MRv1, you can configure the JobTracker to be highly available. You can configure either manual or
automatic failover to a warm-standby JobTracker.
Note:
• As with HDFS High Availability on page 290, the JobTracker high availability feature is backward
compatible; that is, if you do not want to enable JobTracker high availability, you can simply keep
your existing configuration after updating your hadoop-0.20-mapreduce,
hadoop-0.20-mapreduce-jobtracker, and hadoop-0.20-mapreduce-tasktracker
packages, and start your services as before. You do not need to perform any of the actions
described on this page.
To use the high availability feature, you must create a new configuration. This new configuration is designed such that
all the hosts in the cluster can have the same configuration; you do not need to deploy different configuration files to
different hosts depending on each host's role in the cluster.
In an HA setup, the mapred.job.tracker property is no longer a host:port string, but instead specifies a logical
name to identify JobTracker instances in the cluster (active and standby). Each distinct JobTracker in the cluster has a
different JobTracker ID. To support a single configuration file for all of the JobTrackers, the relevant configuration
parameters are suffixed with the JobTracker logical name as well as the JobTracker ID.
The HA JobTracker is packaged separately from the original (non-HA) JobTracker.
Important: You cannot run both HA and non-HA JobTrackers in the same cluster. Do not install the
HA JobTracker unless you need a highly available JobTracker. If you install the HA JobTracker and later
decide to revert to the non-HA JobTracker, you will need to uninstall the HA JobTracker and re-install
the non-HA JobTracker.
JobTracker HA reuses the mapred.job.tracker parameter in mapred-site.xml to identify a JobTracker
active-standby pair. In addition, you must enable the existing mapred.jobtracker.restart.recover,
mapred.job.tracker.persist.jobstatus.active, and mapred.job.tracker.persist.jobstatus.hours
parameters, as well as a number of new parameters, as discussed below.
Use the sections that follow to install, configure and test JobTracker HA.
Replacing the non-HA JobTracker with the HA JobTracker
This section provides instructions for removing the non-HA JobTracker and installing the HA JobTracker.
Important: The HA JobTracker cannot be installed on a node on which the non-HA JobTracker is
installed, and vice versa. If the JobTracker is installed, uninstall it following the instructions below
before installing the HA JobTracker. Uninstall the non-HA JobTracker whether or not you intend to
install the HA JobTracker on the same node.
Removing the non-HA JobTracker
You must remove the original (non-HA) JobTracker before you install and run the HA JobTracker. First, you need to
stop the JobTracker and TaskTrackers.
326 | Cloudera Administration
High Availability
To stop the JobTracker and TaskTrackers:
1. Stop the TaskTrackers: On each TaskTracker system:
$ sudo service hadoop-0.20-mapreduce-tasktracker stop
2. Stop the JobTracker: On the JobTracker system:
$ sudo service hadoop-0.20-mapreduce-jobtracker stop
3. Verify that the JobTracker and TaskTrackers have stopped:
$ ps -eaf | grep -i job
$ ps -eaf | grep -i task
To remove the JobTracker:
• On Red Hat-compatible systems:
$ sudo yum remove hadoop-0.20-mapreduce-jobtracker
• On SLES systems:
$ sudo zypper remove hadoop-0.20-mapreduce-jobtracker
• On Ubuntu systems:
sudo apt-get remove hadoop-0.20-mapreduce-jobtracker
Installing the HA JobTracker
Use the following steps to install the HA JobTracker package, and optionally the ZooKeeper failover controller package
(needed for automatic failover).
Step 1: Install the HA JobTracker package on two separate nodes
On each JobTracker node:
• On Red Hat-compatible systems:
$ sudo yum install hadoop-0.20-mapreduce-jobtrackerha
• On SLES systems:
$ sudo zypper install hadoop-0.20-mapreduce-jobtrackerha
• On Ubuntu systems:
sudo apt-get install hadoop-0.20-mapreduce-jobtrackerha
Step 2: (Optionally) install the failover controller package
If you intend to enable automatic failover, you need to install the failover controller package.
Note: The instructions for automatic failover assume that you have set up a ZooKeeper cluster running
on three or more nodes, and have verified its correct operation by connecting using the ZooKeeper
command-line interface (CLI). See the ZooKeeper documentation for instructions on how to set up a
ZooKeeper ensemble.
Cloudera Administration | 327
High Availability
Install the failover controller package as follows:
On each JobTracker node:
• On Red Hat-compatible systems:
$ sudo yum install hadoop-0.20-mapreduce-zkfc
• On SLES systems:
$ sudo zypper install hadoop-0.20-mapreduce-zkfc
• On Ubuntu systems:
sudo apt-get install hadoop-0.20-mapreduce-zkfc
Configuring and Deploying Manual Failover
Proceed as follows to configure manual failover:
1.
2.
3.
4.
Configure the JobTrackers, TaskTrackers, and Clients
Start the JobTrackers
Activate a JobTracker
Verify that failover is working
Step 1: Configure the JobTrackers, TaskTrackers, and Clients
Changes to existing configuration parameters
Property name
Default
Used on
Description
mapred.job.tracker
local
JobTracker, TaskTracker,
client
In an HA setup, the logical
name of the JobTracker
active-standby pair. In a
non-HA setup
mapred.job.tracker is a
host:port string specifying
the JobTracker's RPC
address, but in an HA
configuration the logical
name must not include a
port number.
mapred.jobtracker.restart. false
recover
JobTracker
Whether to recover jobs
that were running in the
most recent active
JobTracker. Must be set to
true for JobTracker HA.
mapred.job.tracker.persist. false
jobstatus.active
JobTracker
Whether to make job status
persistent in HDFS. Must be
set to true for JobTracker
HA.
mapred.job.tracker.persist. 0
jobstatus.hours
JobTracker
The number of hours job
status information is
retained in HDFS. Must be
greater than zero for
JobTracker HA.
328 | Cloudera Administration
High Availability
Property name
Default
mapred.job.tracker.persist. /jobtracker/jobsInfo
jobstatus.dir
Used on
Description
JobTracker
The HDFS directory in which
job status information is
kept persistently. The
directory must exist and be
owned by the mapred user.
New configuration parameters
Property name
Default
Used on
Description
mapred.jobtrackers.<name> None
JobTracker, TaskTracker, client
A comma-separated pair of IDs
for the active and standby
JobTrackers. The <name> is the
value of
mapred.job.tracker.
mapred.jobtracker.rpc- None
address.<name>.<id>
JobTracker, TaskTracker, client
The RPC address of an
individual JobTracker. <name>
refers to the value of
mapred.job.tracker; <id>
refers to one or other of the
values in
mapred.jobtrackers.<name>.
mapred.job.tracker.http. None
address.<name>.<id>
JobTracker, TaskTracker
The HTTP address of an
individual JobTracker. (In a
non-HA setup
mapred.job.tracker.http.address
(with no suffix) is the
JobTracker's HTTP address.)
mapred.ha.jobtracker. None
rpc-address.<name>.<id>
JobTracker, failover controller
The RPC address of the HA
service protocol for the
JobTracker. The JobTracker
listens on a separate port for
HA operations which is why
this property exists in addition
to
mapred.jobtracker.rpc-address.<name>.<id>.
mapred.ha.jobtracker. None
http-redirect-address.<name>.<id>
JobTracker
The HTTP address of an
individual JobTracker that
should be used for HTTP
redirects. The standby
JobTracker will redirect all web
traffic to the active, and will
use this property to discover
the URL to use for redirects. A
property separate from
mapred.job.tracker.http.
address.<name>.<id> is
needed since the latter may be
a wildcard bind address, such
as 0.0.0.0:50030, which is
Cloudera Administration | 329
High Availability
Property name
Default
Used on
Description
not suitable for making
requests. Note also that
mapred.ha.jobtracker.http-redirect-address.<name>.<id>
is the HTTP redirect address for
the JobTracker with ID <id>
for the pair with the logical
name <name> - that is, the
address that should be used
when that JobTracker is active,
and not the address that
should be redirected to when
that JobTracker is the standby.
mapred.ha.jobtracker.id None
JobTracker
The identity of this JobTracker
instance. Note that this is
optional since each JobTracker
can infer its ID from the
matching address in one of the
mapred.jobtracker.rpc-address.
<name>.<id> properties. It is
provided for testing purposes.
mapred.client.failover. None
proxy.provider.<name>
TaskTracker, client
The failover provider class. The
only class available is
org.apache.hadoop.mapred.
ConfiguredFailoverProxyProvider.
mapred.client.failover. 15
max.attempts
TaskTracker, client
The maximum number of times
to try to fail over.
mapred.client.failover. 500
sleep.base.millis
TaskTracker, client
The time to wait before the
first failover.
mapred.client.failover. 1500
sleep.max.millis
TaskTracker, client
The maximum amount of time
to wait between failovers (for
exponential backoff).
mapred.client.failover. 0
connection.retries
TaskTracker, client
The maximum number of times
to retry between failovers.
mapred.client.failover. 0
connection.retries.on.
timeouts
TaskTracker, client
The maximum number of times
to retry on timeouts between
failovers.
mapred.ha.fencing.methods None
failover controller
A list of scripts or Java classes
that will be used to fence the
active JobTracker during
failover.
Only one JobTracker should be
active at any given time, but
you can simply configure
mapred.ha.fencing.methods
as shell(/bin/true) since
the JobTrackers fence
themselves, and split-brain is
330 | Cloudera Administration
High Availability
Property name
Default
Used on
Description
avoided by the old active
JobTracker shutting itself down
if another JobTracker takes
over.
Make changes and additions similar to the following to mapred-site.xml on each node.
Note: It is simplest to configure all the parameters on all nodes, even though not all of the parameters
will be used on any given node. This also makes for robustness if you later change the roles of the
nodes in your cluster.
<?xml version="1.0"?>
<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
<!-- Put site-specific property overrides in this file. -->
<configuration>
<property>
<name>mapred.job.tracker</name>
<value>logicaljt</value>
<!-- host:port string is replaced with a logical name -->
</property>
<property>
<name>mapred.jobtrackers.logicaljt</name>
<value>jt1,jt2</value>
<description>Comma-separated list of JobTracker IDs.</description>
</property>
<property>
<name>mapred.jobtracker.rpc-address.logicaljt.jt1</name>
<!-- RPC address for jt1 -->
<value>myjt1.myco.com:8021</value>
</property>
<property>
<name>mapred.jobtracker.rpc-address.logicaljt.jt2</name>
<!-- RPC address for jt2 -->
<value>myjt2.myco.com:8022</value>
</property>
<property>
<name>mapred.job.tracker.http.address.logicaljt.jt1</name>
<!-- HTTP bind address for jt1 -->
<value>0.0.0.0:50030</value>
</property>
<property>
<name>mapred.job.tracker.http.address.logicaljt.jt2</name>
<!-- HTTP bind address for jt2 -->
<value>0.0.0.0:50031</value>
</property>
<property>
<name>mapred.ha.jobtracker.rpc-address.logicaljt.jt1</name>
<!-- RPC address for jt1 HA daemon -->
<value>myjt1.myco.com:8023</value>
</property>
<property>
<name>mapred.ha.jobtracker.rpc-address.logicaljt.jt2</name>
<!-- RPC address for jt2 HA daemon -->
<value>myjt2.myco.com:8024</value>
</property>
Cloudera Administration | 331
High Availability
<property>
<name>mapred.ha.jobtracker.http-redirect-address.logicaljt.jt1</name>
<!-- HTTP redirect address for jt1 -->
<value>myjt1.myco.com:50030</value>
</property>
<property>
<name>mapred.ha.jobtracker.http-redirect-address.logicaljt.jt2</name>
<!-- HTTP redirect address for jt2 -->
<value>myjt2.myco.com:50031</value>
</property>
<property>
<name>mapred.jobtracker.restart.recover</name>
<value>true</value>
</property>
<property>
<name>mapred.job.tracker.persist.jobstatus.active</name>
<value>true</value>
</property>
<property>
<name>mapred.job.tracker.persist.jobstatus.hours</name>
<value>1</value>
</property>
<property>
<name>mapred.job.tracker.persist.jobstatus.dir</name>
<value>/jobtracker/jobsInfo</value>
</property>
<property>
<name>mapred.client.failover.proxy.provider.logicaljt</name>
<value>org.apache.hadoop.mapred.ConfiguredFailoverProxyProvider</value>
</property>
<property>
<name>mapred.client.failover.max.attempts</name>
<value>15</value>
</property>
<property>
<name>mapred.client.failover.sleep.base.millis</name>
<value>500</value>
</property>
<property>
<name>mapred.client.failover.sleep.max.millis</name>
<value>1500</value>
</property>
<property>
<name>mapred.client.failover.connection.retries</name>
<value>0</value>
</property>
<property>
<name>mapred.client.failover.connection.retries.on.timeouts</name>
<value>0</value>
</property>
<property>
<name>mapred.ha.fencing.methods</name>
<value>shell(/bin/true)</value>
</property>
</configuration>
332 | Cloudera Administration
High Availability
Note:
In pseudo-distributed mode you need to specify mapred.ha.jobtracker.id for each JobTracker,
so that the JobTracker knows its identity.
But in a fully-distributed setup, where the JobTrackers run on different nodes, there is no need to set
mapred.ha.jobtracker.id, since the JobTracker can infer the ID from the matching address in
one of the mapred.jobtracker.rpc-address.<name>.<id> properties.
Step 2: Start the JobTracker daemons
To start the daemons, run the following command on each JobTracker node:
$ sudo service hadoop-0.20-mapreduce-jobtrackerha start
Step 3: Activate a JobTracker
Note:
• You must be the mapred user to use mrhaadmin commands.
• If Kerberos is enabled, do not use sudo -u mapred when using the hadoop mrhaadmin
command. Instead, you must log in with the mapred Kerberos credentials (the short name must
be mapred). See Configuring Hadoop Security in CDH 5 for more information.
Unless automatic failover is configured, both JobTrackers will be in a standby state after the jobtrackerha daemons
start up.
If Kerberos is not enabled, use the following commands:
To find out what state each JobTracker is in:
$ sudo -u mapred hadoop mrhaadmin -getServiceState <id>
where <id> is one of the values you configured in the mapred.jobtrackers.<name> property – jt1 or jt2 in our
sample mapred-site.xml files.
To transition one of the JobTrackers to active and then verify that it is active:
$ sudo -u mapred hadoop mrhaadmin -transitionToActive <id>
$ sudo -u mapred hadoop mrhaadmin -getServiceState <id>
where <id> is one of the values you configured in the mapred.jobtrackers.<name> property – jt1 or jt2 in our
sample mapred-site.xml files.
With Kerberos enabled, log in as the mapred user and use the following commands:
To log in as the mapred user and kinit:
$ sudo su - mapred
$ kinit -kt mapred.keytab mapred/<fully.qualified.domain.name>
To find out what state each JobTracker is in:
$ hadoop mrhaadmin -getServiceState <id>
where <id> is one of the values you configured in the mapred.jobtrackers.<name> property – jt1 or jt2 in our
sample mapred-site.xml files.
Cloudera Administration | 333
High Availability
To transition one of the JobTrackers to active and then verify that it is active:
$ hadoop mrhaadmin -transitionToActive <id>
$ hadoop mrhaadmin -getServiceState <id>
where <id> is one of the values you configured in the mapred.jobtrackers.<name> property – jt1 or jt2 in our
sample mapred-site.xml files.
Step 4: Verify that failover is working
Use the following commands, depending whether or not Kerberos is enabled.
If Kerberos is not enabled, use the following commands:
To cause a failover from the currently active to the currently inactive JobTracker:
$ sudo -u mapred hadoop mrhaadmin -failover <id_of_active_JobTracker>
<id_of_inactive_JobTracker>
For example, if jt1 is currently active:
$ sudo -u mapred hadoop mrhaadmin -failover jt1 jt2
To verify the failover:
$ sudo -u mapred hadoop mrhaadmin -getServiceState <id>
For example, if jt2 should now be active:
$ sudo -u mapred hadoop mrhaadmin -getServiceState jt2
With Kerberos enabled, use the following commands:
To log in as the mapred user and kinit:
$ sudo su - mapred
$ kinit -kt mapred.keytab mapred/<fully.qualified.domain.name>
To cause a failover from the currently active to the currently inactive JobTracker:
$ hadoop mrhaadmin -failover <id_of_active_JobTracker> <id_of_inactive_JobTracker>
For example, if jt1 is currently active:
$ hadoop mrhaadmin -failover jt1 jt2
To verify the failover:
$ hadoop mrhaadmin -getServiceState <id>
For example, if jt2 should now be active:
$ hadoop mrhaadmin -getServiceState jt2
Configuring and Deploying Automatic Failover
To configure automatic failover, proceed as follows:
1. Configure a ZooKeeper ensemble (if necessary)
2. Configure parameters for manual failover
3. Configure failover controller parameters
334 | Cloudera Administration
High Availability
4. Initialize the HA state in ZooKeeper
5. Enable automatic failover
6. Verify automatic failover
Step 1: Configure a ZooKeeper ensemble (if necessary)
To support automatic failover you need to set up a ZooKeeper ensemble running on three or more nodes, and verify
its correct operation by connecting using the ZooKeeper command-line interface (CLI). See the ZooKeeper documentation
for instructions on how to set up a ZooKeeper ensemble.
Note: If you are already using a ZooKeeper ensemble for automatic failover, use the same ensemble
for automatic JobTracker failover.
Step 2: Configure the parameters for manual failover
See the instructions for configuring the TaskTrackers and JobTrackers under Configuring and Deploying Manual Failover.
Step 3: Configure failover controller parameters
Use the following additional parameters to configure a failover controller for each JobTracker. The failover controller
daemons run on the JobTracker nodes.
New configuration parameters
Property name
Default
mapred.ha.automatic-failover.enabled false
Configure on
Description
failover controller
Set to true to enable
automatic failover.
mapred.ha.zkfc.port
8019
failover controller
The ZooKeeper failover
controller port.
ha.zookeeper.quorum
None
failover controller
The ZooKeeper quorum
(ensemble) to use for
MRZKFailoverController.
Add the following configuration information to mapred-site.xml:
<property>
<name>mapred.ha.automatic-failover.enabled</name>
<value>true</value>
</property>
<property>
<name>mapred.ha.zkfc.port</name>
<value>8018</value>
<!-- Pick a different port for each failover controller when running one machine
-->
</property>
Add an entry similar to the following to core-site.xml:
<property>
<name>ha.zookeeper.quorum</name>
<value>zk1.example.com:2181,zk2.example.com:2181,zk3.example.com:2181 </value>
<!-- ZK ensemble addresses -->
</property>
Cloudera Administration | 335
High Availability
Note: If you have already configured automatic failover for HDFS, this property is already properly
configured; you use the same ZooKeeper ensemble for HDFS and JobTracker HA.
Step 4: Initialize the HA State in ZooKeeper
After you have configured the failover controllers, the next step is to initialize the required state in ZooKeeper. You
can do so by running one of the following commands from one of the JobTracker nodes.
Note: The ZooKeeper ensemble must be running when you use this command; otherwise it will not
work properly.
$ sudo service hadoop-0.20-mapreduce-zkfc init
or
$ sudo -u mapred hadoop mrzkfc -formatZK
This will create a znode in ZooKeeper in which the automatic failover system stores its data.
Note: If you are running a secure cluster, see also Securing access to ZooKeeper.
Step 5: Enable automatic failover
To enable automatic failover once you have completed the configuration steps, you need only start the jobtrackerha
and zkfc daemons.
To start the daemons, run the following commands on each JobTracker node:
$ sudo service hadoop-0.20-mapreduce-zkfc start
$ sudo service hadoop-0.20-mapreduce-jobtrackerha start
One of the JobTrackers will automatically transition to active.
Step 6: Verify automatic failover
After enabling automatic failover, you should test its operation. To do so, first locate the active JobTracker. To find out
what state each JobTracker is in, use the following command:
$ sudo -u mapred hadoop mrhaadmin -getServiceState <id>
where <id> is one of the values you configured in the mapred.jobtrackers.<name> property – jt1 or jt2 in our
sample mapred-site.xml files.
Note: You must be the mapred user to use mrhaadmin commands.
Once you have located your active JobTracker, you can cause a failure on that node. For example, you can use kill
-9 <pid of JobTracker> to simulate a JVM crash. Or you can power-cycle the machine or its network interface
to simulate different kinds of outages. After you trigger the outage you want to test, the other JobTracker should
automatically become active within several seconds. The amount of time required to detect a failure and trigger a
failover depends on the configuration of ha.zookeeper.session-timeout.ms, but defaults to 5 seconds.
336 | Cloudera Administration
High Availability
If the test does not succeed, you may have a misconfiguration. Check the logs for the zkfc and jobtrackerha
daemons to diagnose the problem.
Usage Notes
Using the JobTracker Web UI
To use the JobTracker Web UI, use the HTTP address of either JobTracker (that is, the value of
mapred.job.tracker.http.address.<name>.<id> for either the active or the standby JobTracker). Note the
following:
• If you use the URL of the standby JobTracker, you will be redirected to the active JobTracker.
• If you use the URL of a JobTracker that is down, you will not be redirected - you will simply get a "Not Found" error
from your browser.
Turning off Job Recovery
After a failover, the newly active JobTracker by default restarts all jobs that were running when the failover occurred.
For Sqoop 1 and HBase jobs, this is undesirable because they are not idempotent (that is, they do not behave the same
way on repeated execution). For these jobs you should consider setting mapred.job.restart.recover to false
in the job configuration (JobConf).
Cloudera Navigator Key Trustee Server High Availability
Key Trustee Server high availability applies to read operations only. If either Key Trustee Server fails, the KeyProvider
automatically retries fetching keys from the functioning server. New write operations (for example, creating new
encryption keys) are not allowed unless both Key Trustee Servers are operational.
If a Key Trustee Server fails, the following operations are impacted:
• HDFS Encryption
– You cannot create new encryption keys for encryption zones.
– You can write to and read from existing encryption zones, but you cannot create new zones.
• Cloudera Navigator Encrypt
– You cannot register new Cloudera Navigator Encrypt clients.
– You can continue reading and writing encrypted data, including creating new mount points, using existing
clients.
Cloudera recommends monitoring both Key Trustee Servers. If a Key Trustee Server fails catastrophically, restore it
from backup to a new host with the same hostname and IP address as the failed host. See Backing Up and Restoring
Key Trustee Server for more information. Cloudera does not support PostgreSQL promotion to convert a passive Key
Trustee Server to an active Key Trustee Server.
Configuring Key Trustee Server High Availability Using Cloudera Manager
For new installations, use the Set up HDFS Data At Rest Encryption wizard and follow the instructions in Enabling HDFS
Encryption Using the Wizard. When prompted, make sure that the Enable High Availability option is selected.
If you already have a Key Trustee Server service, and want to enable high availability, use the Add Role Instances wizard
for the Key Trustee Server service instead to add the Passive Key Trustee Server and Passive Database roles.
Important: You must assign the Key Trustee Server and Database roles to the same host. Assign the
Active Key Trustee Server and Active Database roles to one host, and the Passive Key Trustee Server
and Passive Database roles to a separate host.
After completing the Add Role Instances wizard, the Passive Key Trustee Server and Passive Database roles fail to start.
Complete the following manual actions to start these roles:
Cloudera Administration | 337
High Availability
1. Stop the Key Trustee Server service (Key Trustee Server service > Actions > Stop).
2. Run the Set Up Key Trustee Server Database command (Key Trustee Server service > Actions > Set Up Key Trustee
Server Database).
3. Run the following command on the Active Key Trustee Server:
$ sudo rsync -zav --exclude .ssl /var/lib/keytrustee/.keytrustee
root@keytrustee02.example.com:/var/lib/keytrustee/.
Replace keytrustee02.example.com with the hostname of the Passive Key Trustee Server.
4. Run the following command on the Passive Key Trustee Server:
$ sudo ktadmin init
5. Start the Key Trustee Server service (Key Trustee Server service > Actions > Start).
Important: Starting or restarting the Key Trustee Server service attempts to start the Active
Database and Passive Database roles. If the Active Database is not running when the Passive
Database attempts to start, the Passive Database fails to start. If this occurs, manually restart the
Passive Database role after confirming that the Active Database role is running.
6. Enable synchronous replication (Key Trustee Server service > Actions > Setup Enable Synchronous Replication
in HA mode).
7. Restart the Key Trustee Server service (Key Trustee Server service > Actions > Restart).
Configuring Key Trustee Server High Availability Using the Command Line
Install and configure a second Key Trustee Server following the instructions in Installing Cloudera Navigator Key Trustee
Server.
Once you have installed and configured the second Key Trustee Server, initialize the active Key Trustee Server by
running the following commands on the active Key Trustee Server host:
Important: For Key Trustee Server 5.4.0 and higher, the ktadmin init-master command is
deprecated, and should not be used. Use the ktadmin init command instead. If you are using SSH
software other than OpenSSH, pre-create the SSH key on the active Key Trustee Server before
continuing:
$ sudo -u keytrustee ssh-keygen -t rsa /var/lib/keytrustee/.ssh/id_rsa
$ sudo ktadmin init --external-address keytrustee01.example.com
$ sudo rsync -zav --exclude .ssl /var/lib/keytrustee/.keytrustee
root@keytrustee02.example.com:/var/lib/keytrustee/.
$ sudo ktadmin db --bootstrap --port 11381 --pg-rootdir /var/lib/keytrustee/db --slave
keytrustee02.example.com
## For RHEL/CentOS 7, use 'sudo systemctl [stop|start] <service_name>' instead of 'sudo
service <service_name> [stop|start]' ##
$ sudo service keytrustee-db stop
$ sudo service keytrustee-db start
$ sudo service keytrusteed start
Replace keytrustee01.example.com with the fully qualified domain name (FQDN) of the active Key Trustee Server.
Replace keytrustee02.example.com with the FQDN of the passive Key Trustee Server. Cloudera recommends using
the default /var/lib/keytrustee/db directory for the PostgreSQL database.
338 | Cloudera Administration
High Availability
To use a different port for the database, modify the ktadmin init and ktadmin db commands as follows:
$ sudo ktadmin init --external-address keytrustee01.example.com --db-connect
postgresql://localhost:<port>/keytrustee?host=/tmp
$ sudo ktadmin db --bootstrap --port <port> --pg-rootdir /var/lib/keytrustee/db --slave
keytrustee02.example.com
If you use a database directory other than /var/lib/keytrustee/db, create or edit the
/etc/sysconfig/keytrustee-db file and add the following line:
ARGS="--pg-rootdir /path/to/db"
The ktadmin init command generates a self-signed certificate that the Key Trustee Server uses for HTTPS
communication.
Initialize the passive Key Trustee Server by running the following commands on the passive host:
$ sudo ktadmin init-slave --master keytrustee01.example.com --pg-rootdir
/var/lib/keytrustee/db --no-import-key --no-start
## For RHEL/CentOS 7, use 'sudo systemctl [stop|start] <service_name>' instead of 'sudo
service <service_name> [stop|start]' ##
$ sudo service keytrustee-db start
$ sudo ktadmin init --external-address keytrustee02.example.com
$ sudo service keytrusteed start
Replace keytrustee02.example.com with the fully qualified domain name (FQDN) of the passive Key Trustee Server.
Replace keytrustee01.example.com with the FQDN of the active Key Trustee Server. Cloudera recommends using
the default /var/lib/keytrustee/db directory for the PostgreSQL database.
To use a different port for the database, modify the ktadmin init-slave command as follows:
$ sudo ktadmin init-slave --master keytrustee01.example.com --pg-port <port> --pg-rootdir
/var/lib/keytrustee/db --no-import-key --no-start
If you use a database directory other than /var/lib/keytrustee/db, create or edit the
/etc/sysconfig/keytrustee-db file and add the following line:
ARGS="--pg-rootdir /path/to/db"
The ktadmin init-slave command performs an initial database sync by running the pg_basebackup command.
The database directory must be empty for this step to work. For information on performing an incremental backup,
see the PostgreSQL documentation.
Note: The /etc/init.d/postgresql script does not work when the PostgreSQL database is started
by Key Trustee Server, and cannot be used to monitor the status of the database. Use
/etc/init.d/keytrustee-db instead.
The ktadmin init command generates a self-signed certificate that the Key Trustee Server uses for HTTPS
communication. Instructions for using alternate certificates (for example, if you have obtained certificates from a
trusted Certificate Authority) are provided later.
Enable Synchronous Replication
Key Trustee Server high availability requires synchronous replication to ensure that all rows in the database are inserted
in at least two hosts, which protects against key loss.
To enable synchronous replication, run the following command on the active Key Trustee Server:
$ sudo ktadmin enable-synchronous-replication --pg-rootdir /var/lib/keytrustee/db
If you modified the default database location, replace /var/lib/keytrustee/db with the modified path.
Cloudera Administration | 339
High Availability
(Optional) Replace Self-Signed Certificates with CA-Signed Certificates
Important: Because clients connect to Key Trustee Server using its fully qualified domain name
(FQDN), certificates must be issued to the FQDN of the Key Trustee Server host. If you are using
CA-signed certificates, ensure that the generated certificates use the FQDN, and not the short name.
If you have a CA-signed certificate for Key Trustee Server, see Managing Key Trustee Server Certificates for instructions
on how to replace the self-signed certificates.
Recovering a Key Trustee Server
If a Key Trustee Server fails, restore it from backup as soon as possible. If the Key Trustee Server hosts fails completely,
make sure that you restore the Key Trustee Server to a new host with the same hostname and IP address as the failed
host.
For more information, see Backing Up and Restoring Key Trustee Server.
Key Trustee KMS High Availability
CDH 5.4.0 and higher supports Key Trustee KMS high availability. For new installations, you can use the Set up HDFS
Data At Rest Encryption wizard to install and configure Key Trustee KMS high availability. If you have an existing
standalone Key Trustee KMS service, use the following procedure to enable Key Trustee KMS high availability:
1. Back up the Key Trustee KMS private key and configuration directory. See Backing Up and Restoring Key Trustee
Server for more information.
2. If you do not have a ZooKeeper service in your cluster, add one using the instructions in Adding a Service on page
36.
3. Run the Add Role Instances wizard for the Key Trustee KMS service (Key Trustee KMS service > Actions > Add
Role Instances).
4. Click Select hosts and check the box for the host where you want to add the additional Key Management Server
Proxy role. See Resource Planning for Data at Rest Encryption for considerations when selecting a host. Click OK
and then Continue.
5. On the Review Changes page of the wizard, confirm the authorization code, organization name, and Key Trustee
Server settings, and then click Finish.
6. Go to Key Trustee KMS service > Configuration and make sure that the ZooKeeper Service dependency is set to
the ZooKeeper service for your cluster.
7. Synchronize the Key Trustee KMS private key.
Warning: It is very important that you perform this step. Failure to do so leaves Key Trustee KMS
in a state where keys are intermittently inaccessible, depending on which Key Trustee KMS host
a client interacts with, because cryptographic key material encrypted by one Key Trustee KMS
host cannot be decrypted by another. If you are already running multiple Key Trustee KMS hosts
with different private keys, immediately back up all Key Trustee KMS hosts, and contact Cloudera
Support for assistance correcting the issue.
To determine whether the Key Trustee KMS private keys are different, compare the MD5 hash
of the private keys. On each Key Trustee KMS host, run the following command:
$ md5sum /var/lib/kms-keytrustee/keytrustee/.keytrustee/secring.gpg
If the outputs are different, contact Cloudera Support for assistance. Do not attempt to synchronize
existing keys. If you overwrite the private key and do not have a backup, any keys encrypted by
that private key are permanently inaccessible, and any data encrypted by those keys is permanently
irretrievable. If you are configuring Key Trustee KMS high availability for the first time, continue
synchronizing the private keys.
340 | Cloudera Administration
High Availability
Cloudera recommends following security best practices and transferring the private key using offline media, such
as a removable USB drive. For convenience (for example, in a development or testing environment where maximum
security is not required), you can copy the private key over the network by running the following rsync command
on the original Key Trustee KMS host:
rsync -zav /var/lib/kms-keytrustee/keytrustee/.keytrustee
root@ktkms02.example.com:/var/lib/kms-keytrustee/keytrustee/.
Replace ktkms02.example.com with the hostname of the Key Trustee KMS host that you are adding.
8. Restart the Key Trustee KMS service (Key Trustee KMS service > Actions > Restart).
9. Restart the cluster.
10. Redeploy the client configuration (Home > Cluster-wide > Deploy Client Configuration).
11. Re-run the steps in Validating Hadoop Key Operations.
High Availability for Other CDH Components
This section provides information on high availability for CDH components independently of HDFS. See also Configuring
Other CDH Components to Use HDFS HA on page 309.
For details about HA for Impala, see Using Impala through a Proxy for High Availability.
For details about HA for Cloudera Search, see Using Search through a Proxy for High Availability.
HBase High Availability
Most aspects of HBase are highly available in a standard configuration. A cluster typically consists of one Master and
three or more RegionServers, with data stored in HDFS. To ensure that every component is highly available, configure
one or more backup Masters. The backup Masters run on other hosts than the active Master.
Enabling HBase High Availability Using Cloudera Manager
1. Go to the HBase service.
2. Follow the process for adding a role instance and add a backup Master to a different host than the one on which
the active Master is running.
Enabling HBase High Availability Using the Command Line
To configure backup Masters, create a new file in the conf/ directory which will be distributed across your cluster,
called backup-masters. For each backup Master you wish to start, add a new line with the hostname where the
Master should be started. Each host that will run a Master needs to have all of the configuration files available. In
general, it is a good practice to distribute the entire conf/ directory across all cluster nodes.
After saving and distributing the file, restart your cluster for the changes to take effect. When the master starts the
backup Masters, messages are logged. In addition, you can verify that an HMaster process is listed in the output of
the jps command on the nodes where the backup Master should be running.
$ jps
15930
16194
15838
16010
HRegionServer
Jps
HQuorumPeer
HMaster
To stop a backup Master without stopping the entire cluster, first find its process ID using the jps command, then
issue the kill command against its process ID.
$ kill 16010
Cloudera Administration | 341
High Availability
HBase Read Replicas
CDH 5.4 introduces read replicas. Without read replicas, only one RegionServer services a read request from a client,
regardless of whether RegionServers are colocated with other DataNodes that have local access to the same block.
This ensures consistency of the data being read. However, a RegionServer can become a bottleneck due to an
underperforming RegionServer, network problems, or other reasons that could cause slow reads.
With read replicas enabled, the HMaster distributes read-only copies of regions (replicas) to different RegionServers
in the cluster. One RegionServer services the default or primary replica, which is the only replica which can service
write requests. If the RegionServer servicing the primary replica is down, writes will fail.
Other RegionServers serve the secondary replicas, follow the primary RegionServer and only see committed updates.
The secondary replicas are read-only, and are unable to service write requests. The secondary replicas can be kept up
to date by reading the primary replica's HFiles at a set interval or by replication. If they use the first approach, the
secondary replicas may not reflect the most recent updates to the data when updates are made and the RegionServer
has not yet flushed the memstore to HDFS. If the client receives the read response from a secondary replica, this is
indicated by marking the read as "stale". Clients can detect whether or not the read result is stale and react accordingly.
Replicas are placed on different RegionServers, and on different racks when possible. This provides a measure of high
availability (HA), as far as reads are concerned. If a RegionServer becomes unavailable, the regions it was serving can
still be accessed by clients even before the region is taken over by a different RegionServer, using one of the secondary
replicas. The reads may be stale until the entire WAL is processed by the new RegionServer for a given region.
For any given read request, a client can request a faster result even if it comes from a secondary replica, or if consistency
is more important than speed, it can ensure that its request is serviced by the primary RegionServer. This allows you
to decide the relative importance of consistency and availability, in terms of the CAP Theorem, in the context of your
application, or individual aspects of your application, using Timeline Consistency semantics.
Timeline Consistency
Timeline Consistency is a consistency model which allows for a more flexible standard of consistency than the default
HBase model of strong consistency. A client can indicate the level of consistency it requires for a given read (Get or
Scan) operation. The default consistency level is STRONG, meaning that the read request is only sent to the RegionServer
servicing the region. This is the same behavior as when read replicas are not used. The other possibility, TIMELINE,
sends the request to all RegionServers with replicas, including the primary. The client accepts the first response, which
includes whether it came from the primary or a secondary RegionServer. If it came from a secondary, the client can
choose to verify the read later or not to treat it as definitive.
Enabling Read Replica Support
Note:
Before you enable read-replica support, make sure to account for their increased heap memory
requirements. Although no additional copies of HFile data are created, read-only replicas regions have
the same memory footprint as normal regions and need to be considered when calculating the amount
of increased heap memory required. For example, if your table requires 8 GB of heap memory, when
you enable three replicas, you need about 24 GB of heap memory.
To enable support for read replicas in HBase, several properties must be set.
Table 13: HBase Read Replica Properties
Property Name
Default Value
hbase.region.replica.replication.enabled false
342 | Cloudera Administration
Description
The mechanism for refreshing the
secondary replicas. If set to false,
secondary replicas are not guaranteed to
be consistent at the row level. Secondary
replicas are refreshed at intervals controlled
by a timer
High Availability
Property Name
Default Value
Description
(hbase.regionserver.storefile.refresh.period),
and so are guaranteed to be at most that
interval of milliseconds behind the primary
RegionServer. Secondary replicas read from
the HFile in HDFS, and have no access to
writes that have not been flushed to the
HFile by the primary RegionServer.
If true, replicas are kept up to date using
replication. and the column family has the
attribute
REGION_MEMSTORE_REPLICATION set to
false, Using replication for read
replication of hbase:meta is not
supported, and
REGION_MEMSTORE_REPLICATION must
always be set to false on the column
family.
hbase.regionserver.storefile.refresh.period 0 (disabled)
The period, in milliseconds, for refreshing
the store files for the secondary replicas.
The default value of 0 indicates that the
feature is disabled. Secondary replicas
update their store files from the primary
RegionServer at this interval.
If refreshes occur too often, this can create
a burden for the NameNode. If refreshes
occur too infrequently, secondary replicas
will be less consistent with the primary
RegionServer.
hbase.master.loadbalancer.class
The Java class used for balancing the load
of all HBase clients. The default
implementation is the
balancer.StochasticLoadBalancer StochasticLoadBalancer, which is the
only load balancer that supports reading
(the class name is split for data from secondary RegionServers.
formatting purposes)
org.apache.hadoop.hbase.master.
hbase.ipc.client.allowsInterrupt true
Whether or not to enable interruption of
RPC threads at the client. The default value
of true enables primary RegionServers to
access data from other regions' secondary
replicas.
hbase.client.primaryCallTimeout.get 10 ms
The timeout period, in milliseconds, an
HBase client's will wait for a response
before the read is submitted to a secondary
replica if the read request allows timeline
consistency. The default value is 10. Lower
values increase the number of remote
procedure calls while lowering latency.
hbase.client.primaryCallTimeout.multiget 10 ms
The timeout period, in milliseconds, before
an HBase client's multi-get request, such as
HTable.get(List<GET>)), is submitted
to a secondary replica if the multi-get
Cloudera Administration | 343
High Availability
Property Name
Default Value
Description
request allows timeline consistency. Lower
values increase the number of remote
procedure calls while lowering latency.
Configure Read Replicas Using Cloudera Manager
1.
2.
3.
4.
5.
Select Clusters > HBase.
Click the Configuration tab.
Select Scope > HBase or HBase Service-Wide.
Select Category > Advanced.
Locate the HBase Service Advanced Configuration Snippet (Safety Valve) for hbase-site.xml property or search
for it by typing its name in the Search box.
6. Using the same XML syntax as Configure Read Replicas Using the Command Line on page 344 and the chart above,
create a configuration and paste it into the text field.
7. Click Save Changes to commit the changes.
Configure Read Replicas Using the Command Line
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
1. Add the properties from Table 13: HBase Read Replica Properties on page 342 to hbase-site.xml on each
RegionServer in your cluster, and configure each of them to a value appropriate to your needs. The following
example configuration shows the syntax.
<property>
<name>hbase.regionserver.storefile.refresh.period</name>
<value>0</value>
</property>
<property>
<name>hbase.ipc.client.allowsInterrupt</name>
<value>true</value>
<description>Whether to enable interruption of RPC threads at the client. The default
value of true is
required to enable Primary RegionServers to access other RegionServers in secondary
mode. </description>
</property>
<property>
<name>hbase.client.primaryCallTimeout.get</name>
<value>10</value>
</property>
<property>
<name>hbase.client.primaryCallTimeout.multiget</name>
<value>10</value>
</property>
2. Restart each RegionServer for the changes to take effect.
Activating Read Replicas On a Table
After enabling read replica support on your RegionServers, configure the tables for which you want read replicas to
be created. Keep in mind that each replica increases the amount of storage used by HBase in HDFS.
344 | Cloudera Administration
High Availability
At Table Creation
To create a new table with read replication capabilities enabled, set the REGION_REPLICATION property on the table.
Use a command like the following, in HBase Shell:
hbase> create 'myTable', 'myCF', {REGION_REPLICATION => '3'}
By Altering an Existing Table
You can also alter an existing column family to enable or change the number of read replicas it propagates, using a
command similar to the following. The change will take effect at the next major compaction.
hbase> disable 'myTable'
hbase> alter 'myTable', 'myCF', {REGION_REPLICATION => '3'}
hbase> enable 'myTable'
Requesting a Timeline-Consistent Read
To request a timeline-consistent read in your application, use the get.setConsistency(Consistency.TIMELINE)
method before performing the Get or Scan operation.
To check whether the result is stale (comes from a secondary replica), use the isStale() method of the result object.
Use the following examples for reference.
Get Request
Get get = new Get(key);
get.setConsistency(Consistency.TIMELINE);
Result result = table.get(get);
Scan Request
Scan scan = new Scan();
scan.setConsistency(CONSISTENCY.TIMELINE);
ResultScanner scanner = table.getScanner(scan);
Result result = scanner.next();
Scan Request to a Specific Replica
This example overrides the normal behavior of sending the read request to all known replicas, and only sends it to the
replica specified by ID.
Scan scan = new Scan();
scan.setConsistency(CONSISTENCY.TIMELINE);
scan.setReplicaId(2);
ResultScanner scanner = table.getScanner(scan);
Result result = scanner.next();
Detecting a Stale Result
Result result = table.get(get);
if (result.isStale()) {
...
}
Cloudera Administration | 345
High Availability
Getting and Scanning Using HBase Shell
You can also request timeline consistency using HBase Shell, allowing the result to come from a secondary replica.
hbase> get 'myTable', 'myRow', {CONSISTENCY => "TIMELINE"}
hbase> scan 'myTable', {CONSISTENCY => 'TIMELINE'}
Hive Metastore High Availability
You can enable Hive metastore high availability (HA), so that your cluster is resilient to failures due to a metastore that
becomes unavailable. Each metastore is independent; they do not use a quorum.
Prerequisites
• Cloudera recommends that each instance of the metastore runs on a separate cluster host, to maximize high
availability.
• Hive metastore HA requires a database that is also highly available, such as MySQL with replication in active-active
mode. Refer to the documentation for your database of choice to configure it correctly.
Limitations
Sentry HDFS synchronization does not support Hive metastore HA.
Enabling Hive Metastore High Availability Using Cloudera Manager
Minimum Required Role: Configurator (also provided by Cluster Administrator, Full Administrator)
1. Go to the Hive service.
2. If you have a secure cluster, enable the Hive token store. Non-secure clusters can skip this step.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
a.
b.
c.
d.
Click the Configuration tab.
Select Scope > Hive Metastore Server.
Select Category > Advanced.
Locate the Hive Metastore Delegation Token Store property or search for it by typing its name In the Search
box.
e. Select org.apache.hadoop.hive.thrift.DBTokenStore.
f. Click Save Changes to commit the changes.
3. Click the Instances tab.
4. Click Add Role Instances.
5. Click the text field under Hive Metastore Server.
6. Check the box by the host on which to run the additional metastore and click OK.
7. Click Continue and click Finish.
8. Check the box by the new Hive Metastore Server role.
9. Select Actions for Selected > Start, and click Start to confirm.
10. Click Close and click to display the stale configurations page.
11. Click Restart Stale Services and click Restart Now.
12. Click Finish after the cluster finishes restarting.
Enabling Hive Metastore High Availability Using the Command Line
To configure the Hive metastore for high availability, you configure each metastore to store its state in a replicated
database, then provide the metastore clients with a list of URIs where metastores are available. The client starts with
the first URI in the list. If it does not get a response, it randomly picks another URI in the list and attempts to connect.
This continues until the client receives a response.
346 | Cloudera Administration
High Availability
Important:
• If you use Cloudera Manager, do not use these command-line instructions.
• This information applies specifically to CDH 5.5.x. If you use a lower version of CDH, see the
documentation for that version located at Cloudera Documentation.
1. Configure Hive on each of the cluster hosts where you want to run a metastore, following the instructions at
Configuring the Hive Metastore.
2. On the server where the master metastore instance runs, edit the /etc/hive/conf.server/hive-site.xml
file, setting the hive.metastore.uris property's value to a list of URIs where a Hive metastore is available for
failover.
<property>
<name>hive.metastore.uris</name>
<value>thrift://metastore1.example.com,thrift://metastore2.example.com,thrift://metastore3.example.com</value>
<description> URI for client to contact metastore server </description>
</property>
3. If you use a secure cluster, enable the Hive token store by configuring the value of the
hive.cluster.delegation.token.store.class property to
org.apache.hadoop.hive.thrift.DBTokenStore.
<property>
<name>hive.cluster.delegation.token.store.class</name>
<value>org.apache.hadoop.hive.thrift.DBTokenStore</value>
</property>
4. Save your changes and restart each Hive instance.
5. Connect to each metastore and update it to use a nameservice instead of a NameNode, as a requirement for high
availability.
a. From the command-line, as the Hive user, retrieve the list of URIs representing the filesystem roots:
hive --service metatool -listFSRoot
b. Run the following command with the --dry-run option, to be sure that the nameservice is available and
configured correctly. This will not change your configuration.
hive --service metatool -updateLocation nameservice-uri namenode-uri -dryRun
c. Run the same command again without the --dry-run option to direct the metastore to use the nameservice
instead of a NameNode.
hive --service metatool -updateLocation nameservice-uri namenode-uri
6. Test your configuration by stopping your main metastore instance, and then attempting to connect to one of the
other metastores from a client. The following is an example of doing this on a RHEL or Fedora system. The example
first stops the local metastore, then connects to the metastore on the host metastore2.example.com and runs
the SHOW TABLES command.
$ sudo service hive-metastore stop
$ /usr/lib/hive/bin/beeline
beeline> !connect jdbc:hive2://metastore2.example.com:10000 username password
org.apache.hive.jdbc.HiveDriver
0: jdbc:hive2://localhost:10000> SHOW TABLES;
show tables;
+-----------+
| tab_name |
+-----------+
+-----------+
Cloudera Administration | 347
High Availability
No rows selected (0.238 seconds)
0: jdbc:hive2://localhost:10000>
7. Restart the local metastore when you have finished testing.
$ sudo service hive-metastore start
Hue High Availability
This page explains how to configure Hue for high availability with Cloudera Manager and at the command line. It
assumes that you have the Hue service installed and one or more Hue server roles defined. If not, see Adding a Hue
Service and Role Instance on page 176.
Important: Cloudera strongly recommends an external database for clusters with multiple Hue
servers (or "hue" users). With the default embedded database (one per server), in a multi-server
environment, the data on server "A" appears lost when working on server "B" and vice versa. Use an
external database, and configure each server to point to it to ensure that no matter which server is
being used by Hue, your data is always accessible.
Configuring a Cluster for Hue High Availability Using Cloudera Manager
Minimum Required Role: Cluster Administrator (also provided by Full Administrator)
In Cloudera Manager, you can configure your Hue cluster for high availability by adding a Hue load balancer and multiple
Hue servers.
Prerequisite
• An external database was added and each Hue server is configured to use it. See Using an External Database for
Hue Using Cloudera Manager on page 179.
Configuring Hue for High Availability in Cloudera Manager
1. Go to the Hue service.
2. Select the Hue Instances tab.
3. Add one or more Hue servers to an existing Hue server role. At least two Hue server roles are required for high
availability:
a.
b.
c.
d.
Click Add Role Instances.
Click Select hosts under Hue Server (HS).
Check the box for each host on which you want a Hue server (which adds HS icons).
Click OK.
4. Add one load balancer:
a. Click Select hosts under Load Balancer (LB).
b. Check the box for the host on which you want the load balancer (which adds an LB icon).
c. Click OK.
5. Click Continue.
6. Select the Configuration tab and review the options to ensure they meet your needs. Some items to consider are:
• Hue Load Balancer Port - The Apache Load Balancer listens on this port. The default is 8889.
• Path to TLS/SSL Certificate File - The TLS/SSL certificate file.
• Path to TLS/SSL Private Key File - The TLS/SSL private key file.
7. Save any configuration changes.
8. Start the load balancer and new Hue servers:
348 | Cloudera Administration
High Availability
a. Check the box by each new role type.
b. Select Actions for Selected > Start.
c. Click Start and Close.
For more information, see Automatic High Availability and Load Balancing of Hue.
Configuring a Cluster for Hue High Availability Using the Command Line
This section applies to unmanaged deployments without Cloudera Manager. It explains how to install and configure
nginx from the command line. To make the Hue service highly available, configure at least two instances of the Hue
service, each on a different host. Also configure the nginx load balancer to direct traffic to an alternate host if the
primary host becomes unavailable. For advanced configurations, see the nginx documentation.
Prerequisites
• The Hue service is installed and two or more Hue server roles are configured.
• You have network access through SSH to the host machines with the Hue server role(s).
• An external database was added each Hue server is configured to use it. See Using an External Database for Hue
Using Cloudera Manager on page 179.
Installing and Configuring the nginx Load Balancer
To enable high availability for Hue, install the nginx load balancer on one of the Hue instances. Clients access this
instance through a designated port number, and the nginx load balancer directs the request to one of the Hue server
instances.
1. With SSH, log in as the root user to the host machine of one of the Hue instances.
2. Install nginx:
Red Hat/Centos:
yum install nginx
Debian/Ubuntu:
apt-get install nginx
3. Create the following Hue cluster configuration file:
/etc/nginx/conf.d/hue.conf
4. Configure hue.conf with the following template:
server {
server_name NGINX_HOSTNAME;
charset utf-8;
listen 8001;
# Or if running hue on https://
## listen 8001 ssl;
## ssl_certificate /path/to/ssl/cert;
## ssl_certificate_key /path/to/ssl/key;
client_max_body_size 0;
location / {
proxy_pass http://hue;
proxy_set_header Host $http_host;
proxy_set_header X-Forwarded-For $remote_addr;
# Or if the upstream Hue instances are running behind https://
## proxy_pass https://hue;
}
Cloudera Administration | 349
High Availability
location /static/ {
# Uncomment to expose the static file directories.
## autoindex on;
# If Hue was installed with packaging install:
## alias /usr/lib/hue/build/static/;
# Or if on a parcel install:
alias /opt/cloudera/parcels/CDH/lib/hue/build/static/;
expires 30d;
add_header Cache-Control public;
}
}
upstream hue {
ip_hash;
# List all the Hue instances here for high availability.
server HUE_HOST1:8888 max_fails=3;
server HUE_HOST2:8888 max_fails=3;
...
}
5. Update the following in the hue.conf file:
• Replace NGINX_HOSTNAME with the URL of the host where you installed nginx. For example:
server_name myhost-2.myco.com;
• In the location/static block, comment or uncomment the alias lines, depending on whether your
cluster was installed using parcels or packages. (The comment indicator is #.)
• In the upstream hue block, list all the hostnames, including port number, of the Hue instances in your
cluster. For example:
server myhost-1.myco.com:8888 max_fails=3;
server myhost-2.myco.com:8888 max_fails=3;
server myhost-4.myco.com:8888 max_fails=3;
• If the Hue service in your cluster is configured to use TLS/SSL:
– Uncomment these lines, and replace the path names with the paths for your cluster:
## listen 8001 ssl;
## ssl_certificate /path/to/ssl/cert;
## ssl_certificate_key /path/to/ssl/key;
– Uncomment the following line in the location / block:
## proxy_pass https://hue;
and comment out the following line:
proxy_pass http://hue;
– Comment out the following line:
listen 8001
6. Start nginx:
sudo service nginx start
350 | Cloudera Administration
High Availability
7. Test your installation by opening the Hue application in a web browser, using the following URL:
• Without TLS/SSL: http://NGINX_HOSTNAME:8001
• With TLS/SSL: https://NGINX_HOSTNAME:8001
NGINX_HOSTNAME is the name of the host machine where you installed nginx.
Note: To check if Hue is still running, use http://<HUE_HOST>/desktop/debug/is_alive.
A status code of 200 indicates that Hue is still running.
8. Test high availability:
a. Stop the Hue service instance on the host where you installed nginx.
b. Access the Hue application as described in the previous step. If you can connect to the Hue application, you
have successfully enabled high availability.
9. If necessary, configure your backend database for high availability. Consult the vendor documentation for the
database that you configured for Hue.
Llama High Availability
Note: Though Impala can be used together with YARN via simple configuration of Static Service Pools
in Cloudera Manager, the use of the general-purpose component Llama for integrated resource
management within YARN is no longer supported with CDH 5.5 / Impala 2.3 and higher.
Llama high availability (HA) uses an active-standby architecture, in which the active Llama is automatically elected
using the ZooKeeper-based ActiveStandbyElector. The active Llama accepts RPC Thrift connections and communicates
with YARN. The standby Llama monitors the leader information in ZooKeeper, but doesn't accept RPC Thrift connections.
Fencing
Only one of the Llamas should be active to ensure the resources are not partitioned. Llama uses ZooKeeper access
control lists (ACLs) to claim exclusive ownership of the cluster when transitioning to active, and monitors this ownership
periodically. If another Llama takes over, the first one realizes it within this period.
Reclaiming Cluster Resources
To claim resources from YARN, Llama spawns YARN applications and runs unmanaged ApplicationMasters. When a
Llama goes down, the resources allocated to all the YARN applications spawned by it are not reclaimed until YARN
times out those applications (the default timeout is 10 minutes). On Llama failure, these resources are reclaimed by
means of a Llama that kills any YARN applications spawned by this pair of Llamas.
Enabling Llama High Availability Using Cloudera Manager
You can enable Llama high availability when you enable integrated resource management. If you chose to create a
single Llama instance at that time, follow these steps to enable Llama high availability:
1. Go to the Impala service.
2. Select Actions > Enable High Availability.
3. Click the Impala Llama ApplicationMaster Hosts field to display a dialog for choosing Llama hosts.
The following shortcuts for specifying hostname patterns are supported:
• Range of hostnames (without the domain portion)
Range Definition
Matching Hosts
10.1.1.[1-4]
10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4
Cloudera Administration | 351
High Availability
Range Definition
Matching Hosts
host[1-3].company.com
host1.company.com, host2.company.com, host3.company.com
host[07-10].company.com
host07.company.com, host08.company.com, host09.company.com,
host10.company.com
• IP addresses
• Rack name
4.
5.
6.
7.
Specify or select one or more hosts and click OK.
Click Continue.
Click Continue. A progress screen displays with a summary of the wizard actions.
Click Finish.
Disabling Llama High Availability Using Cloudera Manager
You can disable Llama high availability when you disable integrated resource management. If you choose not to disable
integrated resource management, follow these steps to disable Llama high availability:
1.
2.
3.
4.
5.
Go to the Impala service.
Select Actions > Disable High Availability.
Choose the host on which Llama runs after high availability is disabled.
Click Continue. A progress screen displays with a summary of the wizard actions.
Click Finish.
Configuring Oozie for High Availability
In CDH 5, you can configure multiple active Oozie servers against the same database. Oozie high availability is
“active-active” or “hot-hot” so that both Oozie servers are active at the same time, with no failover. High availability
for Oozie is supported in both MRv1 and MRv2 (YARN).
Requirements
The requirements for Oozie high availability are:
• Multiple active Oozie servers, preferably identically configured.
• JDBC JAR in the same location across all Oozie hosts (for example, /var/lib/oozie/).
• External database that supports multiple concurrent connections, preferably with HA support. The default Derby
database does not support multiple concurrent connections.
• ZooKeeper ensemble with distributed locks to control database access, and service discovery for log aggregation.
• Load balancer (preferably with HA support, for example HAProxy), virtual IP, or round-robin DNS to provide a
single entry point (of the multiple active servers), and for callbacks from the Application Master or JobTracker.
For information on setting up TLS/SSL communication with Oozie HA enabled, see Additional Considerations when
Configuring TLS/SSL for Oozie HA.
Configuring Oozie High Availability Using Cloudera Manager
Minimum Required Role: Full Administrator
Important: Enabling or disabling high availability makes the previous monitoring history unavailable.
Enabling Oozie High Availability
1. Ensure that the requirements are satisfied.
2. In the Cloudera Manager Admin Console, go to the Oozie service.
352 | Cloudera Administration
High Availability
3. Select Actions > Enable High Availability to see eligible Oozie server hosts. The host running the current Oozie
server is not eligible.
4. Select the host on which to install an additional Oozie server and click Continue.
5. Specify the host and port of the Oozie load balancer, and click Continue. Cloudera Manager stops the Oozie servers,
adds another Oozie server, initializes the Oozie server High Availability state in ZooKeeper, configures Hue to
reference the Oozie load balancer, and restarts the Oozie servers and dependent services.
Disabling Oozie High Availability
1. In the Cloudera Manager Admin Console, go to the Oozie service.
2. Select Actions > Disable High Availability to see all hosts currently running Oozie servers.
3. Select the one host to run the Oozie server and click Continue. Cloudera Manager stops the Oozie service, removes
the additional Oozie servers, configures Hue to reference the Oozie service, and restarts the Oozie service and
dependent services.
Configuring Oozie High Availability Using the Command Line
For installation and configuration instructions for configuring Oozie HA using the command line, see
https://archive.cloudera.com/cdh5/cdh/5/oozie.
Search High Availability
Mission critical, large-scale online production systems need to make progress without downtime despite some issues.
Cloudera Search provides two routes to configurable, highly available, and fault-tolerant data ingestion:
• Near Real Time (NRT) ingestion using the Flume Solr Sink
• MapReduce based batch ingestion using the MapReduceIndexerTool
Production versus Test Mode
Some exceptions are generally transient, in which case the corresponding task can simply be retried. For example,
network connection errors or timeouts are recoverable exceptions. Conversely, tasks associated with an unrecoverable
exception cannot simply be retried. Corrupt or malformed parser input data, parser bugs, and errors related to unknown
Solr schema fields produce unrecoverable exceptions.
Different modes determine how Cloudera Search responds to different types of exceptions.
• Configuration parameter isProductionMode=false (Non-production mode or test mode): Default configuration.
Cloudera Search throws exceptions to quickly reveal failures, providing better debugging diagnostics to the user.
• Configuration parameter isProductionMode=true (Production mode): Cloudera Search logs and ignores
unrecoverable exceptions, enabling mission-critical large-scale online production systems to make progress without
downtime, despite some issues.
Note: Categorizing exceptions as recoverable or unrecoverable addresses most cases, though it is
possible that an unrecoverable exception could be accidentally misclassified as recoverable. Cloudera
provides the isIgnoringRecoverableExceptions configuration parameter to address such a
case. In a production environment, if an unrecoverable exception is discovered that is classified as
recoverable, change isIgnoringRecoverableExceptions to true. Doing so allows systems to
make progress and avoid retrying an event forever. This configuration flag should only be enabled if
a misclassification bug has been identified. Please report such bugs to Cloudera.
If Cloudera Search throws an exception according the rules described above, the caller, meaning Flume Solr Sink and
MapReduceIndexerTool, can catch the exception and retry the task if it meets the criteria for such retries.
Near Real Time Indexing with the Flume Solr Sink
The Flume Solr Sink uses the settings established by the isProductionMode and
isIgnoringRecoverableExceptions parameters. If a SolrSink does nonetheless receive an exception, the SolrSink
Cloudera Administration | 353
High Availability
rolls the transaction back and pauses. This causes the Flume channel, which is essentially a queue, to redeliver the
transaction's events to the SolrSink approximately five seconds later. This redelivering of the transaction event retries
the ingest to Solr. This process of rolling back, backing off, and retrying continues until ingestion eventually succeeds.
Here is a corresponding example Flume configuration file flume.conf:
agent.sinks.solrSink.isProductionMode = true
agent.sinks.solrSink.isIgnoringRecoverableExceptions = true
In addition, Flume SolrSink automatically attempts to load balance and failover among the hosts of a SolrCloud before
it considers the transaction rollback and retry. Load balancing and failover is done with the help of ZooKeeper, which
itself can be configured to be highly available.
Further, Cloudera Manager can configure Flume so it automatically restarts if its process crashes.
To tolerate extended periods of Solr downtime, you can configure Flume to use a high-performance transactional
persistent queue in the form of a FileChannel. A FileChannel can use any number of local disk drives to buffer significant
amounts of data. For example, you might buffer many terabytes of events corresponding to a week of data. Further,
using the Replicating Channel Selector Flume feature, you can configure Flume to replicate the same data both into
HDFS as well as into Solr. Doing so ensures that if the Flume SolrSink channel runs out of disk space, data delivery is
still delivered to HDFS, and this data can later be ingested from HDFS into Solr using MapReduce.
Many machines with many Flume Solr Sinks and FileChannels can be used in a failover and load balancing configuration
to improve high availability and scalability. Flume SolrSink servers can be either co-located with live Solr servers serving
end user queries, or Flume SolrSink servers can be deployed on separate industry standard hardware for improved
scalability and reliability. By spreading indexing load across a large number of Flume SolrSink servers you can improve
scalability. Indexing load can be replicated across multiple Flume SolrSink servers for high availability, for example
using Flume features such as Load balancing Sink Processor.
Batch Indexing with MapReduceIndexerTool
The Mappers and Reducers of the MapReduceIndexerTool follow the settings established by the isProductionMode
and isIgnoringRecoverableExceptions parameters. However, if a Mapper or Reducer of the
MapReduceIndexerTool does receive an exception, it does not retry at all. Instead it lets the MapReduce task fail
and relies on the Hadoop Job Tracker to retry failed MapReduce task attempts several times according to standard
Hadoop semantics. Cloudera Manager can configure the Hadoop Job Tracker to be highly available. On
MapReduceIndexerTool startup, all data in the output directory is deleted if that output directory already exists. To
retry an entire job that has failed, rerun the program using the same arguments.
For example:
hadoop ... MapReduceIndexerTool ... -D isProductionMode=true -D
isIgnoringRecoverableExceptions=true ...
Configuring Cloudera Manager for High Availability With a Load Balancer
This section provides an example of configuring Cloudera Manager 5 for high availability using a TCP load balancer.
The procedures describe how to configure high availability using a specific, open-source load balancer. Depending on
the operational requirements of your CDH deployment, you can select a different load balancer. You can use either a
hardware or software load balancer, but must be capable of forwarding all Cloudera Manager ports to backing server
instances. (See Ports Used by Cloudera Manager and Cloudera Navigator for more information about the ports used
by Cloudera Manager.)
This topic discusses Cloudera Manager high availability in the context of active-passive configurations only; active-active
configurations are currently unsupported. For information about active-active configuration options, see
http://en.wikipedia.org/wiki/High-availability_cluster.
354 | Cloudera Administration
High Availability
Important: Cloudera Support supports all of the configuration and modification to Cloudera software
detailed in this document. However, Cloudera Support is unable to assist with issues or failures with
the third-party software that is used. Use of any third-party software, or software not directly covered
by Cloudera Support, is at the risk of the end user.
Introduction to Cloudera Manager Deployment Architecture
Cloudera Manager consists of the following software components:
Figure 3: Cloudera Manager Architecture
•
•
•
•
•
Cloudera Manager Server
Cloudera Management Service
Relational databases (several)
Filesystem-based runtime state storage (used by some services that are part of Cloudera Management Service)
Cloudera Manager Agent (one instance per each managed host)
You can locate the Cloudera Manager Server and Cloudera Management Service on different hosts (with each role of
the Cloudera Management Service, such as the Event Server or the Alert Server and so on, possibly located on different
hosts).
Cloudera Manager Server and some of the Cloudera Management Service roles (such as Cloudera Navigator) use a
relational database to store their operational data. Some other services (such as the Host Monitor and the Service
Monitor) use the filesystem (through LevelDB) to store their data.
High availability in the context of Cloudera Manager involves configuring secondary failover instances for each of these
services and also for the persistence components (the relational database and the file system) that support these
services. For simplicity, this document assumes that all of the Cloudera Management Service roles are located on a
single machine.
Cloudera Administration | 355
High Availability
The Cloudera Manager Agent software includes an agent and a supervisor process. The agent process handles RPC
communication with Cloudera Manager and with the roles of the Cloudera Management Service, and primarily handles
configuration changes to your roles. The supervisor process handles the local Cloudera-deployed process lifecycle and
handles auto-restarts (if configured) of failed processes.
Prerequisites for Setting up Cloudera Manager High Availability
• A multi-homed TCP load balancer, or two TCP load balancers, capable of proxying requests on specific ports to
one server from a set of backing servers.
– The load balancer does not need to support termination of TLS/SSL connections.
– This load balancer can be hardware or software based, but should be capable of proxying multiple ports.
HTTP/HTTPS-based load balancers are insufficient because Cloudera Manager uses several non-HTTP-based
protocols internally.
– This document uses HAProxy, a small, open-source, TCP-capable load balancer, to demonstrate a workable
configuration.
• A networked storage device that you can configure to be highly available. Typically this is an NFS store, a SAN
device, or a storage array that satisfies the read/write throughput requirements of the Cloudera Management
Service. This document assumes the use of NFS due to the simplicity of its configuration and because it is an easy,
vendor-neutral illustration.
• The procedures in this document require ssh access to all the hosts in the cluster where you are enabling high
availability for Cloudera Manager.
The Heartbeat Daemon and Virtual IP Addresses
You may have configured Cloudera Manager high availability by configuring virtual IP addresses and using the Heartbeat
daemon (http://linux-ha.org/wiki/Heartbeat). The original Heartbeat package is deprecated; however, support and
maintenance releases are still available through LinBit (
http://www.linbit.com/en/company/news/125-linbit-takes-over-heartbeat-maintenance).
Cloudera recommends using Corosync and Pacemaker (both currently maintained through ClusterLabs). Corosync is
an open-source high-availability tool commonly used in the open-source community.
Editions of this document released for Cloudera Manager4 and CDH 4 also used virtual IP addresses that move as a
resource from one host to another on failure. Using virtual IP addresses has several drawbacks:
• Questionable reliance on outdated Address Resolution Protocol (ARP) behavior to ensure that the IP-to-MAC
translation works correctly to resolve to the new MAC address on failure.
• Split-brain scenarios that lead to problems with routing.
• A requirement that the virtual IP address subnet be shared between the primary and the secondary hosts, which
can be onerous if you deploy your secondaries off site.
Therefore, Cloudera no longer recommend the use of virtual IP addresses, and instead recommends using a dedicated
load balancer.
Single-User Mode, TLS, and Kerberos
High availability, as described in this document, supports the following:
• Single-user mode. You must run all commands as the root user (unless specified otherwise). These procedures
do not alter or modify the behavior of how CDH services function.
• TLS and Kerberized deployments. For more information, see TLS and Kerberos Configuration for Cloudera Manager
High Availability on page 384.
High-Level Steps to Configure Cloudera Manager High Availability
To configure Cloudera Manager for high availability, follow these high-level steps. Click each step to see detailed
procedures.
356 | Cloudera Administration
High Availability
Important:
Unless stated otherwise, run all commands mentioned in this topic as the root user.
You do not need to stop the CDH cluster to configure Cloudera Manager high availability.
•
•
•
•
Step 1: Setting Up Hosts and the Load Balancer on page 357.
Step 2: Installing and Configuring Cloudera Manager Server for High Availability on page 362.
Step 3: Installing and Configuring Cloudera Management Service for High Availability on page 366.
Step 4: Automating Failover with Corosync and Pacemaker on page 372.
Step 1: Setting Up Hosts and the Load Balancer
At a high level, you set up Cloudera Manager Server and Cloudera Management Service roles (including Cloudera
Navigator) on separate hosts, and make sure that network access to those hosts from other Cloudera services and to
the Admin Console occurs through the configured load balancer.
Cloudera Manager Server, Cloudera Navigator, and all of the Cloudera Management Service roles that use a relational
database should use an external database server, located off-host. You must make sure that these databases are
configured to be highly available. See Database High Availability Configuration on page 383.
You configure other Cloudera Management Service roles (such as the Service Monitor and Host Monitor roles) that
use a file-backed storage mechanism to store their data on a shared NFS storage mechanism.
Figure 4: High-level layout of components for Cloudera Manager high availability
Cloudera Administration | 357
High Availability
Creating Hosts for Primary and Secondary Servers
For this example, Cloudera recommends using four hosts for Cloudera Manager services. All of these hosts must resolve
forward and reverse DNS lookups correctly:
•
•
•
•
Cloudera Manager Server primary host (hostname: CMS1)
Cloudera Management Service primary host (hostname: MGMT1)
Cloudera Manager Server secondary host (hostname: CMS2)
Cloudera Management Service secondary host (hostname: MGMT2)
Note: The hostnames used here are placeholders and are used throughout this document. When
configuring your cluster, substitute the actual names of the hosts you use in your environment.
In addition, Cloudera recommends the following:
• Do not host the Cloudera Manager or Cloudera Management Service roles on existing hosts in a CDH cluster,
because this complicates failover configuration, and overlapping failure domains can cause problems with fault
containment and error tracing.
• Configure both the primary and the secondary hosts using the same host configuration. This helps to ensure that
failover does not lead to decreased performance.
• Host the primary and secondary hosts on separate power and network segments within your organization to limit
overlapping failure domains.
Setting up the Load Balancer
This procedure demonstrates configuring the load balancer as two separate software load balancers using HAProxy,
on two separate hosts for demonstration clarity. (To reduce cost, you might prefer to set up a single load balancer
with two network interfaces.) You use one HAProxy host for Cloudera Manager Server and another for the Cloudera
Management Service.
Note: HAProxy is used here for demonstration purposes. Production-level performance requirements
determine the load balancer that you select for your installation.
HAProxy version 1.5.2 is used for these procedures.
1. Reserve two hostnames in your DNS system, and assign them to each of the load balancer hosts. (The names
CMSHostname, and MGMTHostname are used in this example; substitute the correct hostname for your
environment.) These hostnames will be the externally accessible hostnames for Cloudera Manager Server and
Cloudera Management Service. (Alternatively, use one load balancer with separate, resolvable IP addresses—one
each to back CMSHostname and MGMTHostname respectively).
• CMSHostname is used to access Cloudera Manager Admin Console.
• MGMTHostname is used for internal access to the Cloudera Management Service from Cloudera Manager
Server and Cloudera Manager Agents.
2. Set up two hosts using any supported Linux distribution (RHEL, CentOS, Ubuntu or SUSE; see Supported Operating
Systems) with the hostnames listed above. See the HAProxy documentation for recommendations on configuring
the hardware of these hosts.
3. Install the version of HAProxy that is recommended for the version of Linux installed on the two hosts:
RHEL/CentOS:
$ yum install haproxy
358 | Cloudera Administration
High Availability
Ubuntu (use a current Personal Package Archive (PPA) for 1.5 from http://haproxy.debian.net):
$ apt-get install haproxy
SUSE:
$ zypper install haproxy
4. Configure HAProxy to autostart on both the CMSHostname and MGMTHostname hosts:
RHEL, CentOS, and SUSE:
$ chkconfig haproxy on
Ubuntu:
$ update-rc.d haproxy defaults
5. Configure HAProxy.
• On CMSHostname, edit the /etc/haproxy/haproxy.cfg files and make sure that the ports listed at Ports
Used by Cloudera Manager and Cloudera Navigator for “Cloudera Manager Server” are proxied. For Cloudera
Manager 5, this list includes the following ports as defaults:
– 7180
– 7182
– 7183
Sample HAProxy Configuration for CMSHostname
listen cmf :7180
mode tcp
option tcplog
server cmfhttp1 CMS1:7180 check
server cmfhttp2 CMS2:7180 check
listen cmfavro :7182
mode tcp
option tcplog
server cmfavro1 CMS1:7182 check
server cmfavro2 CMS2:7182 check
#ssl pass-through, without termination
listen cmfhttps :7183
mode tcp
option tcplog
server cmfhttps1 CMS1:7183 check
server cmfhttps2 CMS2:7183 check
• On MGMTHostname, edit the /etc/haproxy/haproxy.cfg file and make sure that the ports for Cloudera
Management Service are proxied (see Ports Used by Cloudera Manager and Cloudera Navigator). For Cloudera
Manager 5, this list includes the following ports as defaults:
–
–
–
–
–
–
–
–
–
5678
7184
7185
7186
7187
8083
8084
8086
8087
Cloudera Administration | 359
High Availability
–
–
–
–
–
–
–
–
–
8091
9000
9994
9995
9996
9997
9998
9999
10101
Example HAProxy Configuration for MGMTHostname
listen mgmt1 :5678
mode tcp
option tcplog
server mgmt1a MGMT1 check
server mgmt1b MGMT2 check
listen mgmt2 :7184
mode tcp
option tcplog
server mgmt2a MGMT1 check
server mgmt2b MGMT2 check
listen mgmt3 :7185
mode tcp
option tcplog
server mgmt3a MGMT1
server mgmt3b MGMT2
listen mgmt4 :7186
mode tcp
option tcplog
server mgmt4a MGMT1
server mgmt4b MGMT2
listen mgmt5 :7187
mode tcp
option tcplog
server mgmt5a MGMT1
server mgmt5b MGMT2
check
check
check
check
check
check
listen mgmt6 :8083
mode tcp
option tcplog
server mgmt6a MGMT1 check
server mgmt6b MGMT2 check
listen mgmt7 :8084
mode tcp
option tcplog
server mgmt7a MGMT1 check
server mgmt7b MGMT2 check
listen mgmt8 :8086
mode tcp
option tcplog
server mgmt8a MGMT1 check
server mgmt8b MGMT2 check
listen mgmt9 :8087
mode tcp
option tcplog
server mgmt9a MGMT1 check
server mgmt9b MGMT2 check
listen mgmt10 :8091
mode tcp
option tcplog
server mgmt10a MGMT1 check
server mgmt10b MGMT2 check
listen mgmt-agent :9000
mode tcp
option tcplog
server mgmt-agenta MGMT1 check
server mgmt-agentb MGMT2 check
360 | Cloudera Administration
High Availability
listen mgmt11 :9994
mode tcp
option tcplog
server mgmt11a MGMT1
server mgmt11b MGMT2
listen mgmt12 :9995
mode tcp
option tcplog
server mgmt12a MGMT1
server mgmt12b MGMT2
listen mgmt13 :9996
mode tcp
option tcplog
server mgmt13a MGMT1
server mgmt13b MGMT2
listen mgmt14 :9997
mode tcp
option tcplog
server mgmt14a MGMT1
server mgmt14b MGMT2
listen mgmt15 :9998
mode tcp
option tcplog
server mgmt15a MGMT1
server mgmt15b MGMT2
listen mgmt16 :9999
mode tcp
option tcplog
server mgmt16a MGMT1
server mgmt16b MGMT2
listen mgmt17 :10101
mode tcp
option tcplog
server mgmt17a MGMT1
server mgmt17b MGMT2
check
check
check
check
check
check
check
check
check
check
check
check
check
check
After updating the configuration, restart HAProxy on both the MGMTHostname and CMSHostname hosts:
$ service haproxy restart
Setting up the Database
1. Create databases on your preferred external database server. See Cloudera Manager and Managed Service
Datastores.
Important: The embedded Postgres database cannot be configured for high availability and
should not be used in a high-availability configuration.
2. Configure your databases to be highly available. Consult the vendor documentation for specific information.
MySQL, PostgreSQL, and Oracle each have many options for configuring high availability. See Database High
Availability Configuration on page 383 for some external references on configuring high availability for your Cloudera
Manager databases.
Setting up an NFS Server
The procedures outlined for setting up the Cloudera Manager Server and Cloudera Management Service hosts presume
there is a shared store configured that can be accessed from both the primary and secondary instances of these hosts.
This usually requires that this store be accessible over the network, and can be one of a variety of remote storage
mechanisms (such as an iSCSI drive, a SAN array, or an NFS server).
Cloudera Administration | 361
High Availability
Note: Using NFS as a shared storage mechanism is used here for demonstration purposes. Refer to
your Linux distribution documentation on production NFS configuration and security. Production-level
performance requirements determine the storage that you select for your installation.
This section describes how to configure an NFS server and assumes that you understand how to configure highly
available remote storage devices. Further details are beyond the scope and intent of this guide.
There are no intrinsic limitations on where this NFS server is located, but because overlapping failure domains can
cause problems with fault containment and error tracing, Cloudera recommends that you not co-locate the NFS server
with any CDH or Cloudera Manager servers or the load-balancer hosts detailed in this document.
1. Install NFS on your designated server:
RHEL/CentOS
$ yum install nfs-utils nfs-utils-lib
Ubuntu
$ apt-get install nfs-kernel-server
SUSE
$ zypper install nfs-kernel-server
2. Start nfs and rpcbind, and configure them to autostart:
RHEL/CentOS:
$ chkconfig nfs on
$ service rpcbind start
$ service nfs start
Ubuntu:
$ update-rc.d nfs defaults
$ service rpcbind start
$ service nfs-kernel-server
SUSE:
$ chkconfig nfs on
$ service rpcbind start
$ service nfs-kernel-server start
Note: Later sections describe mounting the shared directories and sharing them between the primary
and secondary instances.
Step 2: Installing and Configuring Cloudera Manager Server for High Availability
You can use an existing Cloudera Manager installation and extend it to a high-availability configuration, as long as you
are not using the embedded PostgreSQL database.
This section describes how to install and configure a failover secondary for Cloudera Manager Server that can take
over if the primary fails.
This section does not cover installing instances of Cloudera Manager Agent on CMS1 or CMS2 and configuring them to
be highly available. See Installation.
362 | Cloudera Administration
High Availability
Setting up NFS Mounts for Cloudera Manager Server
1. Create the following directories on the NFS server you created in a previous step:
$ mkdir -p /media/cloudera-scm-server
2. Mark these mounts by adding these lines to the /etc/exports file on the NFS server:
/media/cloudera-scm-server CMS1(rw,sync,no_root_squash,no_subtree_check)
/media/cloudera-scm-server CMS2(rw,sync,no_root_squash,no_subtree_check)
3. Export the mounts by running the following command on the NFS server:
$ exportfs -a
4. Set up the filesystem mounts on CMS1 and CMS2 hosts:
a. If you are updating an existing installation for high availability, stop the Cloudera Manager Server if it is
running on either of the CMS1 or CMS2 hosts by running the following command:
$ service cloudera-scm-server stop
b. Make sure that the NFS mount helper is installed:
RHEL/CentOS:
$ yum install nfs-utils-lib
Ubuntu:
$ apt-get install nfs-common
SUSE:
$ zypper
install nfs-client
c. Make sure that rpcbind is running and has been restarted:
$ service rpcbind restart
5. Create the mount points on both CMS1 and CMS2:
a. If you are updating an existing installation for high availability, copy the /var/lib/cloudera-scm-server
file from your existing Cloudera Manager Server host to the NFS server with the following command (NFS
refers to the NFS server you created in a previous step):
$ scp -r /var/lib/cloudera-scm-server/ NFS:/media/cloudera-scm-server
b. Set up the /var/lib/cloudera-scm-server directory on the CMS1 and CMS2 hosts:
$ rm -rf /var/lib/cloudera-scm-server
$ mkdir -p /var/lib/cloudera-scm-server
c. Mount the following directory to the NFS mounts, on both CMS1 and CMS2:
$ mount -t nfs NFS:/media/cloudera-scm-server /var/lib/cloudera-scm-server
Cloudera Administration | 363
High Availability
d. Set up fstab to persist the mounts across restarts by editing the /etc/fstab file on CMS1 and CMS2 and
adding the following lines:
NFS:/media/cloudera-scm-server /var/lib/cloudera-scm-server nfs
auto,noatime,nolock,intr,tcp,actimeo=1800 0 0
Installing the Primary
Updating an Existing Installation for High Availability
You can retain your existing Cloudera Manager Server as-is, if the deployment meets the following conditions:
• The Cloudera Management Service is located on a single host that is not the host where Cloudera Manager Server
runs.
• The data directories for the roles of the Cloudera Management Service are located on a remote storage device
(such as an NFS store), and they can be accessed from both primary and secondary installations of the Cloudera
Management Service.
If your deployment does not meet these conditions, Cloudera recommends that you uninstall Cloudera Management
Services by stopping the existing service and deleting it.
Important: Deleting the Cloudera Management Service leads to loss of all existing data from the Host
Monitor and Service Monitor roles that store health and monitoring information for your cluster on
the local disk associated with the host(s) where those roles are installed.
To delete and remove the Cloudera Management Service:
1. Open the Cloudera Manager Admin Console and go to the Home page.
2. Click Cloudera Management Service > Stop.
3. Click Cloudera Management Service > Delete.
Fresh Installation
Use either of the installation paths (B or C) specified in the documentation to install Cloudera Manager Server, but do
not add “Cloudera Management Service” to your deployment until you complete Step 3: Installing and Configuring
Cloudera Management Service for High Availability on page 366, which describes how to set up the Cloudera Management
Service.
See:
• Installation Path B - Manual Installation Using Cloudera Manager Packages
• Installation Path C - Manual Installation Using Cloudera Manager Tarballs
You can now start the freshly-installed Cloudera Manager Server on CMS1:
$ service cloudera-scm-server start
Before proceeding, verify that you can access the Cloudera Manager Admin Console at http://CMS1:7180.
If you have just installed Cloudera Manager, click the Cloudera Manager logo to skip adding new hosts and to gain
access to the Administration menu, which you need for the following steps.
HTTP Referer Configuration
Cloudera recommends that you disable the HTTP Referer check because it causes problems for some proxies and load
balancers. Check the configuration manual of your proxy or load balancer to determine if this is necessary.
To disable HTTP Referer in the Cloudera Manager Admin Console:
1. Select Administration > Settings.
2. Select Category > Security.
3. Deselect the HTTP Referer Check property.
364 | Cloudera Administration
High Availability
Before proceeding, verify that you can access the Cloudera Manager Admin Console through the load balancer at
http://CMSHostname:7180.
TLS and Kerberos Configuration
To configure Cloudera Manager to use TLS encryption or authentication, or to use Kerberos authentication, see TLS
and Kerberos Configuration for Cloudera Manager High Availability on page 384.
Installing the Secondary
Setting up the Cloudera Manager Server secondary requires copying certain files from the primary to ensure that they
are consistently initialized.
1. On the CMS2 host, install the cloudera-manager-server package using Installation Path B or Installation Path
C.
See:
• Installation Path B - Manual Installation Using Cloudera Manager Packages
• Installation Path C - Manual Installation Using Cloudera Manager Tarballs
2. When setting up the database on the secondary, copy the /etc/cloudera-scm-server/db.properties file
from host CMS1 to host CMS2 at /etc/cloudera-scm-server/db.properties. For example:
$ mkdir -p /etc/cloudera-scm-server
$ scp [<ssh-user>@]CMS1:/etc/cloudera-scm-server/db.properties
/etc/cloudera-scm-server/db.properties
3. If you configured Cloudera Manager TLS encryption or authentication, or Kerberos authentication in your primary
installation, see TLS and Kerberos Configuration for Cloudera Manager High Availability on page 384 for additional
configuration steps.
4. Do not start the cloudera-scm-server service on this host yet, and disable autostart on the secondary to avoid
automatically starting the service on this host.
RHEL/CentOS/SUSEL:
$ chkconfig cloudera-scm-server off
Ubuntu:
$ update-rc.d -f cloudera-scm-server remove
(You will also disable autostart on the primary when you configure automatic failover in a later step.) Data corruption
can result if both primary and secondary Cloudera Manager Server instances are running at the same time, and
it is not supported. :
Testing Failover
Test failover manually by using the following steps:
1. Stop cloudera-scm-server on your primary host (CMS1):
$ service cloudera-scm-server stop
2. Start cloudera-scm-server on your secondary host (CMS2):
$ service cloudera-scm-server start
3. Wait a few minutes for the service to load, and then access the Cloudera Manager Admin Console through a web
browser, using the load-balanced hostname (for example: http://CMSHostname:CMS_port).
Now, fail back to the primary before configuring the Cloudera Management Service on your installation:
Cloudera Administration | 365
High Availability
1. Stop cloudera-scm-server on your secondary machine (CMS2):
$ service cloudera-scm-server stop
2. Start cloudera-scm-server on your primary machine (CMS1):
$ service cloudera-scm-server start
3. Wait a few minutes for the service to load, and then access the Cloudera Manager Admin Console through a web
browser, using the load-balanced hostname (for example: http://CMSHostname:7180).
Updating Cloudera Manager Agents to use the Load Balancer
After completing the primary and secondary installation steps listed previously, update the Cloudera Manager Agent
configuration on all of the hosts associated with this Cloudera Manager installation, except the MGMT1, MGMT2, CMS1,
and CMS2 hosts, to use the load balancer address:
1. Connect to a shell on each host where CDH processes are installed and running. (The MGMT1, MGMT2, CMS1, and
CMS2 hosts do not need to be modified as part of this step.)
2. Update the /etc/cloudera-scm-agent/config.ini file and change the server_host line:
server_host = <CMSHostname>
3. Restart the agent (this command starts the agents if they are not running):
$ service cloudera-scm-agent restart
Step 3: Installing and Configuring Cloudera Management Service for High Availability
This section demonstrates how to set up shared mounts on MGMT1 and MGMT2, and then install Cloudera Management
Service to use those mounts on the primary and secondary servers.
Important: Do not start the primary and secondary servers that are running Cloudera Management
Service at the same time. Data corruption can result.
Setting up NFS Mounts for Cloudera Management Service
1. Create directories on the NFS server:
$
$
$
$
$
$
$
mkdir
mkdir
mkdir
mkdir
mkdir
mkdir
mkdir
-p
-p
-p
-p
-p
-p
-p
/media/cloudera-host-monitor
/media/cloudera-scm-agent
/media/cloudera-scm-eventserver
/media/cloudera-scm-headlamp
/media/cloudera-service-monitor
/media/cloudera-scm-navigator
/media/etc-cloudera-scm-agent
2. Mark these mounts by adding the following lines to the /etc/exports file on the NFS server:
/media/cloudera-host-monitor MGMT1(rw,sync,no_root_squash,no_subtree_check)
/media/cloudera-scm-agent MGMT1(rw,sync,no_root_squash,no_subtree_check)
/media/cloudera-scm-eventserver MGMT1(rw,sync,no_root_squash,no_subtree_check)
/media/cloudera-scm-headlamp MGMT1(rw,sync,no_root_squash,no_subtree_check)
/media/cloudera-service-monitor MGMT1(rw,sync,no_root_squash,no_subtree_check)
/media/cloudera-scm-navigator MGMT1(rw,sync,no_root_squash,no_subtree_check)
/media/etc-cloudera-scm-agent MGMT1(rw,sync,no_root_squash,no_subtree_check)
/media/cloudera-host-monitor MGMT2(rw,sync,no_root_squash,no_subtree_check)
/media/cloudera-scm-agent MGMT2(rw,sync,no_root_squash,no_subtree_check)
/media/cloudera-scm-eventserver MGMT2(rw,sync,no_root_squash,no_subtree_check)
/media/cloudera-scm-headlamp MGMT2(rw,sync,no_root_squash,no_subtree_check)
/media/cloudera-service-monitor MGMT2(rw,sync,no_root_squash,no_subtree_check)
366 | Cloudera Administration
High Availability
/media/cloudera-scm-navigator MGMT2(rw,sync,no_root_squash,no_subtree_check)
/media/etc-cloudera-scm-agent MGMT2(rw,sync,no_root_squash,no_subtree_check)
3. Export the mounts running the following command on the NFS server:
$ exportfs -a
4. Set up the filesystem mounts on MGMT1 and MGMT2 hosts:
a. Make sure that the NFS mount helper is installed:
RHEL/CentOS:
$ yum install nfs-utils-lib
Ubuntu:
$ apt-get install nfs-common
SUSE:
$ zypper
install nfs-client
b. Create the mount points on both MGMT1 and MGMT2:
$
$
$
$
$
$
$
mkdir
mkdir
mkdir
mkdir
mkdir
mkdir
mkdir
-p
-p
-p
-p
-p
-p
-p
/var/lib/cloudera-host-monitor
/var/lib/cloudera-scm-agent
/var/lib/cloudera-scm-eventserver
/var/lib/cloudera-scm-headlamp
/var/lib/cloudera-service-monitor
/var/lib/cloudera-scm-navigator
/etc/cloudera-scm-agent
c. Mount the following directories to the NFS mounts, on both MGMT1 and MGMT2 (NFS refers to the server NFS
hostname or IP address):
$
$
$
$
$
$
$
mount
mount
mount
mount
mount
mount
mount
-t
-t
-t
-t
-t
-t
-t
nfs
nfs
nfs
nfs
nfs
nfs
nfs
NFS:/media/cloudera-host-monitor /var/lib/cloudera-host-monitor
NFS:/media/cloudera-scm-agent /var/lib/cloudera-scm-agent
NFS:/media/cloudera-scm-eventserver /var/lib/cloudera-scm-eventserver
NFS:/media/cloudera-scm-headlamp /var/lib/cloudera-scm-headlamp
NFS:/media/cloudera-service-monitor /var/lib/cloudera-service-monitor
NFS:/media/cloudera-scm-navigator /var/lib/cloudera-scm-navigator
NFS:/media/etc-cloudera-scm-agent /etc/cloudera-scm-agent
5. Set up fstab to persist the mounts across restarts. Edit the /etc/fstab file and add these lines:
NFS:/media/cloudera-host-monitor /var/lib/cloudera-host-monitor nfs
auto,noatime,nolock,intr,tcp,actimeo=1800 0 0
NFS:/media/cloudera-scm-agent /var/lib/cloudera-scm-agent nfs
auto,noatime,nolock,intr,tcp,actimeo=1800 0 0
NFS:/media/cloudera-scm-eventserver /var/lib/cloudera-scm-eventserver nfs
auto,noatime,nolock,intr,tcp,actimeo=1800 0 0
NFS:/media/cloudera-scm-headlamp /var/lib/cloudera-scm-headlamp nfs
auto,noatime,nolock,intr,tcp,actimeo=1800 0 0
NFS:/media/cloudera-service-monitor /var/lib/cloudera-service-monitor nfs
auto,noatime,nolock,intr,tcp,actimeo=1800 0 0
NFS:/media/cloudera-scm-navigator /var/lib/cloudera-scm-navigator nfs
auto,noatime,nolock,intr,tcp,actimeo=1800 0 0
NFS:/media/etc-cloudera-scm-agent /etc/cloudera-scm-agent nfs
auto,noatime,nolock,intr,tcp,actimeo=1800 0 0
Cloudera Administration | 367
High Availability
Installing the Primary
1. Connect to a shell on MGMT1, and then install the cloudera-manager-daemons and cloudera-manager-agent
packages:
a. Install packages cloudera-manager-daemons and cloudera-manager-agent packages using instructions
from Installation Path B (See Installation Path B - Manual Installation Using Cloudera Manager Packages).
b. Install the Oracle Java JDK version that is required for your deployment, if it is not already installed on the
host. See Supported JDK Versions.
2. Configure the agent to report its hostname as <MGMTHostname> to Cloudera Manager. This ensures that the
connections from the Cloudera Manager Agents on the CDH cluster hosts report to the correct Cloudera
Management Service host in the event of a failover.
a. Edit the/etc/cloudera-scm-agent/config.ini file to update the following lines:
server_host=CMSHostname
listening_hostname=MGMTHostname
b. Edit the /etc/hosts file and add MGMTHostname as an alias for your public IP address for MGMT1 by adding
a line like this at the end of your /etc/hosts file:
MGMT1 IP MGMTHostname
c. Confirm that the alias has taken effect by running the ping command. For example:
[root@MGMT1 ~]# ping MGMTHostname
PING MGMTHostname (MGMT1 IP) 56(84) bytes of data.
64 bytes from MGMTHostname (MGMT1 IP): icmp_seq=1 ttl=64 time=0.034 ms
64 bytes from MGMTHostname (MGMT1 IP): icmp_seq=2 ttl=64 time=0.018 ms
...
d. Make sure that the cloudera-scm user and the cloudera-scm group have access to the mounted directories
under /var/lib, by using the chown command on cloudera-scm. For example, run the following on MGMT1:
$
$
$
$
$
$
chown
chown
chown
chown
chown
chown
-R
-R
-R
-R
-R
-R
cloudera-scm:cloudera-scm
cloudera-scm:cloudera-scm
cloudera-scm:cloudera-scm
cloudera-scm:cloudera-scm
cloudera-scm:cloudera-scm
cloudera-scm:cloudera-scm
/var/lib/cloudera-scm-eventserver
/var/lib/cloudera-scm-navigator
/var/lib/cloudera-service-monitor
/var/lib/cloudera-host-monitor
/var/lib/cloudera-scm-agent
/var/lib/cloudera-scm-headlamp
Note: The cloudera-scm user and the cloudera-scm group are the default owners as
specified in Cloudera Management Service advanced configuration. If you alter these settings,
or are using single-user mode, modify the above chown instructions to use the altered user
or group name.
e. Restart the agent on MGMT1 (this also starts the agent if it is not running):
$ service cloudera-scm-agent restart
f. Connect to the Cloudera Manager Admin Console running on <CMSHostname> and:
a. Go to the Hosts tab and make sure that a host with name <MGMTHostname> is reported. (If it is not
available yet, wait for it to show up before you proceed.)
b. Click Add Cloudera Management Service.
368 | Cloudera Administration
High Availability
g. Make sure you install all of the roles of the Cloudera Management Service on the host named MGMTHostname.
h. Proceed through the steps to configure the roles of the service to use your database server, and use defaults
for the storage directory for Host Monitor or Service Monitor.
i. After you have completed the steps, wait for the Cloudera Management Service to finish starting, and verify
the health status of your clusters as well as the health of the Cloudera Management Service as reported in
the Cloudera Manager Admin Console. The health status indicators should be green, as shown:
The service health for Cloudera Management Service might, however, show as red:
Cloudera Administration | 369
High Availability
In this case, you need to identify whether the health test failure is caused by the Hostname and Canonical
Name Health Check for the MGMTHostname host, which might look like this:
This test can fail in this way because of the way you modified /etc/hosts on MGMT1 and MGMT2 to allow
the resolution of MGMTHostname locally. This test can be safely disabled on the MGMTHostname host from
the Cloudera Manager Admin Console.
j. If you are configuring Kerberos and TLS/SSL, see TLS and Kerberos Configuration for Cloudera Manager High
Availability on page 384 for configuration changes as part of this step.
Installing the Secondary
1. Stop all Cloudera Management Service roles using the Cloudera Manager Admin Console:
a. On the Home > Status tab, click
to the right of Cloudera Management Service and select Stop.
b. Click Stop to confirm. The Command Details window shows the progress of stopping the roles.
c. When Command completed with n/n successful subcommands appears, the task is complete. Click Close.
2. Stop the cloudera-scm-agent service on MGMT1:
$ service cloudera-scm-agent stop
3. Install cloudera-manager-daemons and cloudera-manager-agent:
a. Install the cloudera-manager-daemons and cloudera-manager-agent packages using instructions
from Installation Path B. See Installation Path B - Manual Installation Using Cloudera Manager Packages.
370 | Cloudera Administration
High Availability
b. Install the Oracle Java JDK version that is required for your deployment, if it is not already installed on the
host. See Supported JDK Versions.
4. Configure the agent to report its hostname as MGMTHostname to Cloudera Manager, as described previously in
Installing the Primary on page 368.
a. Make sure that /etc/cloudera-scm-agent/config.ini has the following lines (because this is a shared
mount with the primary, it should be the same as in the primary installation):
server_host=<CMHostname>
listening_hostname=<MGMTHostname>
b. Edit the /etc/hosts file and add MGMTHostname as an alias for your public IP address for MGMT1, by adding
a line like this at the end of your /etc/hosts file:
<MGMT2-IP> <MGMTHostname>
c. Confirm that the alias is working by running the ping command. For example:
[root@MGMT2 ~]# ping MGMTHostname
PING MGMTHostname (MGMT2 IP) 56(84) bytes of data.
64 bytes from MGMTHostname (MGMT2 IP): icmp_seq=1 ttl=64 time=0.034 ms
64 bytes from MGMTHostname (MGMT2 IP): icmp_seq=2 ttl=64 time=0.018 ms
5. Start the agent on MGMT2 by running the following command:
$ service cloudera-scm-agent start
6. Log into the Cloudera Manager Admin Console in a web browser and start all Cloudera Management Service roles.
This starts the Cloudera Management Service on MGMT2.
a. Wait for the Cloudera Manager Admin Console to report that the services have started.
b. Confirm that the services have started on this host by running the following command on MGMT2:
$ ps -elf | grep “scm”
You should see ten total processes running on that host, including the eight Cloudera Management Service
processes, a Cloudera Manager Agent process, and a Supervisor process.
c. Test the secondary installation through the Cloudera Management Admin Console, and inspect the health
of the Cloudera Management Service roles, before proceeding.
Note:
Make sure that the UID and GID for the cloudera-scm user on the primary and secondary Cloudera
Management Service hosts are same; this ensures that the correct permissions are available on the
shared directories after failover.
Failing Back to the Primary
Before finishing the installation, fail back to the primary host (MGMT1):
1. Stop the cloudera-scm-agent service on MGMT2:
$ service cloudera-scm-agent hard_stop_confirmed
Cloudera Administration | 371
High Availability
2. Start the cloudera-scm-agent service on MGMT1:
$ service cloudera-scm-agent start
Step 4: Automating Failover with Corosync and Pacemaker
Corosync and Pacemaker are popular high-availability utilities that allow you to configure Cloudera Manager to fail
over automatically.
This document describes one way to set up clustering using these tools. Actual setup can be done in several ways,
depending on the network configuration of your environment.
Prerequisites:
1. Install Pacemaker and Corosync on CMS1, MGMT1, CMS2, and MGMT2, using the correct versions for your Linux
distribution:
Note: The versions referred to for setting up automatic failover in this document are Pacemaker
1.1.11 and Corosync 1.4.7. See http://clusterlabs.org/wiki/Install to determine what works best
for your Linux distribution.
RHEL/CentOS:
$ yum install pacemaker corosync
Ubuntu:
$ apt-get install pacemaker corosync
SUSE:
$ zypper install pacemaker corosync
2. Make sure that the crm tool exists on all of the hosts. This procedure uses the crm tool, which works with Pacemaker
configuration. If this tool is not installed when you installed Pacemaker (verify this by running which crm), you
can download and install the tool for your distribution using the instructions at http://crmsh.github.io/installation.
About Corosync and Pacemaker
• By default, Corosync and Pacemaker are not autostarted as part of the boot sequence. Cloudera recommends
leaving this as is. If the machine crashes and restarts, manually make sure that failover was successful and determine
the cause of the restart before manually starting these processes to achieve higher availability.
– If the /etc/defaults/corosync file exists, make sure that START is set to yes in that file:
START=yes
– Make sure that Corosync is not set to start automatically, by running the following command:
RHEL/CentOS/SUSE:
$ chkconfig corosync off
Ubuntu:
$ update-rc.d -f corosync remove
• Note which version of Corosync is installed. The contents of the configuration file for Corosync (corosync.conf)
that you edit varies based on the version suitable for your distribution. Sample configurations are supplied in this
document and are labeled with the Corosync version.
372 | Cloudera Administration
High Availability
• This document does not demonstrate configuring Corosync with authentication (with secauth set to on). The
Corosync website demonstrates a mechanism to encrypt traffic using symmetric keys. A simple example is available
at http://docs.openstack.org/high-availability-guide/content/_setting_up_corosync.html.
• Firewall configuration:
Corosync uses UDP transport on ports 5404 and 5405, and these ports must be open for both inbound and
outbound traffic on all hosts. If you are using IP tables, run a command similar to the following:
$ sudo iptables -I INPUT -m state --state NEW -p udp -m multiport --dports 5404,5405 -j
ACCEPT
$ sudo iptables -I OUTPUT -m state --state NEW -p udp -m multiport --sports 5404,5405
-j ACCEPT
Setting up Cloudera Manager Server
Set up a Corosync cluster over unicast, between CMS1 and CMS2, and make sure that the hosts can “cluster” together.
Then, set up Pacemaker to register Cloudera Manager Server as a resource that it monitors and to fail over to the
secondary when needed.
Setting up Corosync
1. Edit the /etc/corosync/corosync.conf file on CMS1 and replace the entire contents with the following text
(use the correct version for your environment):
Corosync version 1.x:
compatibility: whitetank
totem {
version: 2
secauth: off
interface {
member {
memberaddr: CMS1
}
member {
memberaddr: CMS2
}
ringnumber: 0
bindnetaddr: CMS1
mcastport: 5405
}
transport: udpu
}
logging {
fileline: off
to_logfile: yes
to_syslog: yes
logfile: /var/log/cluster/corosync.log
debug: off
timestamp: on
logger_subsys {
subsys: AMF
debug: off
}
}
service {
# Load the Pacemaker Cluster Resource Manager
name: pacemaker
ver: 1
#
}
Corosync version 2.x:
totem {
version: 2
secauth: off
Cloudera Administration | 373
High Availability
cluster_name: cmf
transport: udpu
}
nodelist {
node {
ring0_addr: CMS1
nodeid: 1
}
node {
ring0_addr: CMS2
nodeid: 2
}
}
quorum {
provider: corosync_votequorum
two_node: 1
}
2. Edit the /etc/corosync/corosync.conf file on CMS2, and replace the entire contents with the following text
(use the correct version for your environment):
Corosync version 1.x:
compatibility: whitetank
totem {
version: 2
secauth: off
interface {
member {
memberaddr: CMS1
}
member {
memberaddr: CMS2
}
ringnumber: 0
bindnetaddr: CMS2
mcastport: 5405
}
transport: udpu
}
logging {
fileline: off
to_logfile: yes
to_syslog: yes
logfile: /var/log/cluster/corosync.log
debug: off
timestamp: on
logger_subsys {
subsys: AMF
debug: off
}
}
service {
# Load the Pacemaker Cluster Resource Manager
name: pacemaker
ver: 1
#
}
Corosync version 2.x:
totem {
version: 2
secauth: off
cluster_name: cmf
transport: udpu
}
374 | Cloudera Administration
High Availability
nodelist {
node {
ring0_addr: CMS1
nodeid: 1
}
node {
ring0_addr: CMS2
nodeid: 2
}
}
quorum {
provider: corosync_votequorum
two_node: 1
}
3. Restart Corosync on CMS1 and CMS2 so that the new configuration takes effect:
$ service corosync restart
Setting up Pacemaker
You use Pacemaker to set up Cloudera Manager Server as a cluster resource.
See the Pacemaker configuration reference at
http://clusterlabs.org/doc/en-US/Pacemaker/1.1-plugin/html/Clusters_from_Scratch/ for more details about Pacemaker
options.
The following steps demonstrate one way, recommended by Cloudera, to configure Pacemaker for simple use:
1. Disable autostart for Cloudera Manager Server (because you manage its lifecycle through Pacemaker) on both
CMS1 and CMS2:
RHEL/CentOS/SUSE:
$ chkconfig cloudera-scm-server off
Ubuntu:
$ update-rc.d -f cloudera-scm-server remove
2. Make sure that Pacemaker has been started on both CMS1 and CMS2:
$ /etc/init.d/pacemaker start
3. Make sure that crm reports two nodes in the cluster:
# crm status
Last updated: Wed Mar 4 18:55:27 2015
Last change: Wed Mar 4 18:38:40 2015 via crmd on CMS1
Stack: corosync
Current DC: CMS1 (1) - partition with quorum
Version: 1.1.10-42f2063
2 Nodes configured
0 Resources configured
4. Change the Pacemaker cluster configuration (on either CMS1 or CMS2):
$ crm configure property no-quorum-policy=ignore
$ crm configure property stonith-enabled=false
$ crm configure rsc_defaults resource-stickiness=100
These commands do the following:
• Disable quorum checks. (Because there are only two nodes in this cluster, quorum cannot be established.)
Cloudera Administration | 375
High Availability
• Disable STONITH explicitly (see Enabling STONITH (Shoot the other node in the head) on page 376).
• Reduce the likelihood of the resource being moved among hosts on restarts.
5. Add Cloudera Manager Server as an LSB-managed resource (either on CMS1 or CMS2):
$ crm configure primitive cloudera-scm-server lsb:cloudera-scm-server
6. Verify that the primitive has been picked up by Pacemaker:
$ crm_mon
For example:
$ crm_mon
Last updated: Tue Jan 27 15:01:35 2015
Last change: Mon Jan 27 14:10:11 2015
Stack: classic openais (with plugin)
Current DC: CMS1 - partition with quorum
Version: 1.1.11-97629de
2 Nodes configured, 2 expected votes
1 Resources configured
Online: [ CMS1 CMS2 ]
cloudera-scm-server (lsb:cloudera-scm-server): Started CMS1
At this point, Pacemaker manages the status of the cloudera-scm-server service on hosts CMS1 and CMS2, ensuring
that only one instance is running at a time.
Note: Pacemaker expects all lifecycle actions, such as start and stop, to go through Pacemaker;
therefore, running direct service start or service stop commands breaks that assumption.
Testing Failover with Pacemaker
Test Pacemaker failover by running the following command to move the cloudera-scm-server resource to CMS2:
$ crm resource move cloudera-scm-server <CMS2>
Test the resource move by connecting to a shell on CMS2 and verifying that the cloudera-scm-server process is
now active on that host. It takes usually a few minutes for the new services to come up on the new host.
Enabling STONITH (Shoot the other node in the head)
The following link provides an explanation of the problem of fencing and ensuring (within reasonable limits) that only
one host is running a shared resource at a time:
http://clusterlabs.org/doc/en-US/Pacemaker/1.1-plugin/html-single/Clusters_from_Scratch/index.html#idm140457872046640
As noted in that link, you can use several methods (such as IPMI) to achieve reasonable guarantees on remote host
shutdown. Cloudera recommends enabling STONITH, based on the hardware configuration in your environment.
Setting up the Cloudera Manager Service
Setting Up Corosync
1. Edit the /etc/corosync/corosync.conf file on MGMT1 and replace the entire contents with the contents
below; make sure to use the correct section for your version of Corosync:
Corosync version 1.x:
compatibility: whitetank
totem {
version: 2
secauth: off
interface {
member {
memberaddr: MGMT1
376 | Cloudera Administration
High Availability
}
member {
memberaddr: MGMT2
}
ringnumber: 0
bindnetaddr: MGMT1
mcastport: 5405
}
transport: udpu
}
logging {
fileline: off
to_logfile: yes
to_syslog: yes
logfile: /var/log/cluster/corosync.log
debug: off
timestamp: on
logger_subsys {
subsys: AMF
debug: off
}
}
service {
# Load the Pacemaker Cluster Resource Manager
name: pacemaker
ver: 1
#
}
Corosync version 2.x:
totem {
version: 2
secauth: off
cluster_name: mgmt
transport: udpu
}
nodelist {
node {
ring0_addr: MGMT1
nodeid: 1
}
node {
ring0_addr: MGMT2
nodei