Introducing Microsoft SQL Server 2016

SQL Server 2016
Mission-Critical Applications, Deeper Insights,
Hyperscale Cloud
Preview 2
Stacia Varga, Denny Cherry, Joseph D’Antoni
Introducing Microsoft
SQL Server 2016
Mission-Critical Applications,
Deeper Insights, Hyperscale
Preview 2
Stacia Varga, Denny Cherry, and
Joseph D’Antoni
Microsoft Press
A division of Microsoft Corporation
One Microsoft Way
Redmond, Washington 98052-6399
Copyright © 2016 by Microsoft Corporation
All rights reserved. No part of the contents of this book may be reproduced or transmitted in any
form or by any means without the written permission of the publisher.
ISBN: 978-1-5093-0193-5
Printed and bound in the United States of America.
First Printing
Microsoft Press books are available through booksellers and distributors worldwide. If you need
support related to this book, email Microsoft Press Support at Please tell us
what you think of this book at
This book is provided “as-is” and expresses the author’s views and opinions. The views, opinions and
information expressed in this book, including URL and other Internet website references, may change
without notice.
Some examples depicted herein are provided for illustration only and are fictitious. No real association
or connection is intended or should be inferred.
Microsoft and the trademarks listed at on the “Trademarks” webpage are
trademarks of the Microsoft group of companies. All other marks are property of their respective
Acquisitions and Developmental Editor: Devon Musgrave
Project Editor: John Pierce
Editorial Production: Flyingspress
Cover: Twist Creative  Seattle
Contents at a
Chapter 2
Better security
Chapter 3
Higher availability
Chapter 4
Improved database engine
Chapter 6
More analytics
Chapter 7
Better reporting
Chapter 2
Better security ........................................................................................................................ 1
Always Encrypted ..................................................................................................................................................................... 1
Getting started with Always Encrypted ...................................................................................................................... 1
Creating a table with encrypted values ...................................................................................................................... 7
CREATE TABLE statement for encrypted columns ................................................................................................. 7
Migrating existing tables to Always Encrypted ....................................................................................................... 9
Row-Level Security ............................................................................................................................................................... 11
Creating inline table functions .................................................................................................................................... 11
Creating security policies .............................................................................................................................................. 14
Using block predicates ................................................................................................................................................... 15
Dynamic data masking ....................................................................................................................................................... 15
Dynamic data masking of a new table .................................................................................................................... 16
Dynamic data masking of an existing table .......................................................................................................... 16
Understanding dynamic data masking and permissions ................................................................................. 17
Masking encrypted values ............................................................................................................................................ 18
Using dynamic data masking in SQL Database.................................................................................................... 18
Chapter 3
Higher availability ............................................................................................................... 20
AlwaysOn Availability Groups .......................................................................................................................................... 20
Supporting disaster recovery with basic availability groups........................................................................... 21
Using group Managed Service Accounts ............................................................................................................... 23
Triggering failover at the database level ................................................................................................................ 23
Supporting distributed transactions ......................................................................................................................... 24
Scaling out read workloads .......................................................................................................................................... 25
Defining automatic failover targets .......................................................................................................................... 26
Reviewing the improved log transport performance ........................................................................................ 27
Windows Server 2016 Technical Preview high-availability enhancements .................................................... 28
Creating workgroup clusters ....................................................................................................................................... 28
Configuring a cloud witness ........................................................................................................................................ 29
Using Storage Spaces Direct ....................................................................................................................................... 32
Introducing site-aware failover clusters .................................................................................................................. 32
Windows Server Failover Cluster logging .............................................................................................................. 33
Performing rolling cluster operating system upgrades .................................................................................... 33
Chapter 4
Improved database engine ................................................................................................ 35
TempDB enhancements ..................................................................................................................................................... 35
Configuring data files for TempDB ........................................................................................................................... 36
Eliminating specific trace flags.................................................................................................................................... 37
Query Store ............................................................................................................................................................................. 38
Enabling Query Store...................................................................................................................................................... 38
Understanding Query Store components .............................................................................................................. 39
Reviewing information in the query store .............................................................................................................. 40
Using Force Plan ............................................................................................................................................................... 42
Managing the query store ............................................................................................................................................ 43
Tuning with the query store ........................................................................................................................................ 44
Stretch Database ................................................................................................................................................................... 44
Understanding Stretch Database architecture ..................................................................................................... 45
Security and Stretch Database .................................................................................................................................... 45
Identifying tables for Stretch Database ................................................................................................................... 46
Configuring Stretch Database ..................................................................................................................................... 47
Monitoring Stretch Database ...................................................................................................................................... 48
Backup and recovery with Stretch Database ......................................................................................................... 49
Chapter 6
More analytics ...................................................................................................................... 50
Tabular enhancements ....................................................................................................................................................... 50
Accessing more data sources with DirectQuery .................................................................................................. 51
Modeling with a DirectQuery source ....................................................................................................................... 51
Working with calculated tables .................................................................................................................................. 54
Bidirectional cross-filtering .......................................................................................................................................... 56
Writing formulas ............................................................................................................................................................... 60
Introducing new DAX functions ................................................................................................................................. 60
Using variables in DAX ................................................................................................................................................... 63
R integration ........................................................................................................................................................................... 64
Installing and configuring R Services ....................................................................................................................... 64
Getting started with R Services ................................................................................................................................... 65
Using an R Model in SQL Server ................................................................................................................................ 74
Chapter 7
Better reporting ................................................................................................................... 77
Report content types ........................................................................................................................................................... 77
Paginated report development enhancements ........................................................................................................ 77
Introducing changes to paginated report authoring tools ............................................................................. 78
Exploring new data visualizations .............................................................................................................................. 79
Managing parameter layout in paginated reports ............................................................................................. 84
Mobile report development ............................................................................................................................................. 85
KPI development ................................................................................................................................................................... 85
Report access enhancements .......................................................................................................................................... 86
Accessing reports with modern browsers .............................................................................................................. 86
Viewing reports on mobile devices........................................................................................................................... 88
Printing without ActiveX ................................................................................................................................................ 88
Exporting to PowerPoint ............................................................................................................................................... 90
Pinning reports to Power BI ......................................................................................................................................... 92
Managing subscriptions ................................................................................................................................................ 93
About the authors .................................................................................................................................... 96
Better security
SQL Server 2016 introduces three new principal security features—Always
Encrypted, Row-Level Security, and dynamic data masking. While all these
features are security related, each provides a different level of data
protection within this latest version of the database platform. Throughout
this chapter, we explore the uses of these features, how they work, and
when they should be used to protect data in your SQL Server database.
Always Encrypted
Always Encrypted is a client-side encryption technology in which data is automatically encrypted not
only when it is written but also when it is read by an approved application. Unlike Transparent Data
Encryption, which encrypts the data on disk but allows the data to be read by any application that
queries the data, Always Encrypted requires your client application to use an Always Encrypted–
enabled driver to communicate with the database. By using this driver, the application securely
transfers encrypted data to the database that can then be decrypted later only by an application that
has access to the encryption key. Any other application querying the data can also retrieve the
encrypted values, but that application cannot use the data without the encryption key, thereby
rendering the data useless. Because of this encryption architecture, the SQL Server instance never sees
the unencrypted version of the data.
Note At this time, the only Always Encrypted–enabled drivers are the .NET Framework Data
Provider for SqlServer, which requires installation of .NET Framework version 4.6 on the client
computer, and the JDBC 6.0 driver. In this chapter, we refer to both of these drivers as the ADO.NET
driver for simplicity.
Getting started with Always Encrypted
Using Always Encrypted requires a small amount of preparation within the database storing the
encrypted tables. While this can be done by using a wizard in SQL Server Management Studio, using
T-SQL is a more repeatable process for production deployments, so this chapter will focus on the TSQL configuration process. The preparation is a two-step process:
Create the column master key definition
Create the column encryption key
CH A P TER 2 | Better security
Column master key definition
The column master key is a certificate that is stored within a Windows certificate store, a third-party
Hardware Security Module (HSM), or the Azure Key Vault. The application that is encrypting the data
uses the column master key to protect the various column encryption keys that handle the encryption
of the data within the columns of a database table.
Note Using an HSM, also known as an Enterprise Key Manager (EKM), requires the use of SQL Server
Enterprise Edition. In this chapter, we describe the use of a self-signed certificate that you store in the
Microsoft Certificate Store of the Windows operating system. While this is approach is not the optimal
configuration, it demonstrates the concepts of Always Encrypted and is applicable to any edition of
SQL Server.
You can create a column master key definition by using the graphical interface within SQL Server
Management Studio (SSMS) or by using T-SQL. In SSMS, connect to the SQL Server 2016 database
instance in which you want to use Always Encrypted to protect a database table. In Object Explorer,
navigate first to the database, then to Security, and then expand the Always Encrypted Keys folder to
display its two subfolders, as shown in Figure 2-1.
Figure 2-1: Always Encrypted Keys folder in SQL Server 2016 Object Explorer.
To create the column master key, right-click the Column Master Keys folder and select New Column
Master Key. In the New Column Master Key dialog box, type a name for the column master key,
specify whether to store the key in the current user’s or local machine’s certificate store or the Azure
Key Vault, and then select a certificate in the list, as shown in Figure 2-2. If there are no certificates, or
if you want to use a new self-signed certificate, click the Generate Certificate button, and then click
OK. This step creates a self-signed certificate and loads it into the certificate store of the current user
account running SSMS.
CH A P TER 2 | Better security
Figure 2-2: New Column Master Key dialog box.
Note You should perform these steps on a trusted machine, but not on the computer hosting your
SQL Server instance. That way, the data remains protected in SQL Server even if the host computer
is compromised.
After creating the certificate and configuring it as a column master key, you must then export and
distribute it to all computers hosting clients requiring access to the data. If a client application is webbased, you must load the certificate on the web server. If it is an application installed on users’
computers, then you must deploy the certificate to each user’s computer individually.
You can find applicable instructions for exporting and importing certificates for your operating system
at the following URLs:
Exporting certificates
Windows 7 and Windows Server 2008 R2:
Windows 8 and Windows Server 2012:
Windows 8.1 and Windows Server 2012 R2:
Windows 10 and Windows Server 2016:
Importing certificates
Windows 7 and Windows Server 2008 R2:
Windows 8 and Windows Server 2012:
CH A P TER 2 | Better security
Windows 8.1 and Windows Server 2012 R2:
Windows 10 and Windows Server 2016:
Certificate stores and special service accounts
When you import certificates into the certificate store on the computers with the application that
encrypts and decrypts the data, you must import the certificates into either the machine certificate
store or the certificate store of the domain account running the application.
As an alternative, you can create a column master key by using T-SQL. Although you might find that
creating the key is easier using SSMS, T-SQL scripts provide you with a repeatable process that you
can check into a source control system and keep safe in case you need to rebuild the server.
Furthermore, because best practices for SQL Server 2016 discourage installation of SSMS on the
server’s console and Windows security best practices discourage certificate installation on unsecured
systems such as users’ desktops, the use of T-SQL scripts to create column master keys is
To create a column master key, use the CREATE COLUMN MASTER KEY statement, as shown in
Example 2-1. This statement requires you to supply a name for the definition, such as MyKey, as
shown in the example. You must also set the value for KEY_STORE_PROVIDER_NAME as
MSSQL_CERTIFICATE_STORE. Last, you specify the path for the certificate in the certificate store as the
KEY_PATH value. This value begins with CurrentUser when you use a certificate stored in the user
account’s certificate store or LocalMachine when using a certificate stored in the computer’s certificate
store. The rest of the value is a random-looking string of characters that represents the thumbprint of
the selected certificate. This thumbprint is unique to each certificate.
Example 2-1: Creating a column master key
USE [Samples]
KEY_PATH = N'CurrentUser/My/DE3A770F25EBD6071305B77FB198D1AE434E6014'
Other key store providers?
You may be asking yourself what key-store providers are available besides the Microsoft SQL Server
certificate store. You can choose from several other key-store providers. One option is
MSSQL_CSP_PROVIDER, which allows you to use any HSM supporting Microsoft CryptoAPI. Another
option is MSSQL_CNG_STORE, which allows you to use any HSM supporting Cryptography API:
Next Generation. A third option is to specify AZURE_KEY_VAULT as the key-store provider, which
requires you to download and install the Azure Key Vault key store provider on the machines
accessing the protected data, which will be protected as described at http://blogs.msdn
.com/b/sqlsecurity/archive/2015/11/10/using-the-azure-key-vault-key-store-provider.aspx. Last,
you can use a custom provider, as described at
CH A P TER 2 | Better security
sqlsecurity/archive”/2015/09/25/creating-an-ad-hoc-always-encrypted-provider-using-azure-keyvault.aspx. Although this article provides an example using Azure Key Vault, you can apply the
principles to the development of a custom provider.
Finding the certificate thumbprint
You can easily locate the thumbprint of the certificate in the certificate store by using the Certificate
snap-in within the Microsoft Management Console (MMC). In MMC, on the File menu, select
Add/Remove Snap-In. In the Add Or Remove Snap-ins dialog box, select Certificates in the
Available Snap-ins list on the left, and click the Add button to move your selection to the right. The
Certificates Snap-in dialog box prompts you to select a certificate store. Choose either My User
Account or Computer Account, depending on which certificate store you are using. Click the Finish
button, and then click OK. Expand the Certificates folder to locate your certificate in the
Personal/Certificates subfolder, double-click the certificate, select the Details tab, and scroll to the
bottom, where you can see the thumbprint that you use as the value for the CREATE COLUMN
Column encryption keys
After creating a column master key, you are ready to create the encryption keys for specific columns.
The SQL Server 2016 ADO.NET driver uses column encryption keys to encrypt the data before sending
it to the SQL Server and to decrypt the data after retrieving it from the SQL Server 2016 instance. As
with the column master key, you can create column encryption keys by using T-SQL or SSMS. While
the column master keys are easier to create by using T-SQL, column encryption keys are easier to
create by using SSMS.
To create a column encryption key, use Object Explorer to connect to the database instance, navigate
to the database, then to Security, and expand the Always Encrypted Keys folder. Right-click Column
Encryption Keys, and then select New Column Encryption Key. In the New Column Encryption Key
dialog box, type a name for the new encryption key, select a Column Master Key Definition in the
drop-down list, as shown in Figure 2-3, and then click OK. You can now use the column encryption
key in the definition of a new table.
CH A P TER 2 | Better security
Figure 2-3: New Column Encryption Key dialog box.
To create a new column encryption key by using T-SQL, you use the CREATE COLUMN ENCRYPTION
KEY statement as shown in Example 2-2.
USE [Samples]
The CREATE COLUMN ENCRYPTION KEY statement accepts three parameters. The first parameter is
COLUMN MASTER KEY DEFINITION, which corresponds to the column master key definition that you
created in a previous step. The second parameter defines the encryption algorithm used to encrypt
the value of the encryption key. In SQL Server 2016, the only supported parameter value at this time is
RAS_OAEP. The third parameter is the value of the column encryption key after it has been encrypted
by the column master key definition.
Note When creating column encryption keys, you should not use an unencrypted value as the
ENCRYPTED_VALUE parameter of the CREATE COLUMN ENCRYPTION KEY statement. Otherwise,
you compromise the benefits of Always Encrypted by making data vulnerable to attack.
CH A P TER 2 | Better security
The CREATE COLUMN ENCRYPTION KEY command accepts a minimum of one VALUE block, and a
maximum of two VALUE blocks. Two VALUE blocks should be used when rotating encryption keys,
either because a key has expired or because it has become compromised. Two keys should exist
within the database long enough for all connected applications to download the new encryption keys
from the database. Depending on the application design and client connectivity, this process may take
minutes or months.
Generating new encrypted values
Given that the value is encrypted, how can new encrypted values be generated? The easiest way is
to use SSMS to open the New Column Encryption Key dialog box shown in Figure 2-3, select the
correct column master key definition, provide a name for the new encryption key, and then click the
Script button at the top of the dialog box. This selection gives you the full CREATE COLUMN
ENCRYPTION KEY statement, including a new random encrypted value. You can then add this new
value as a second encryption key and thereby easily rotate the encryption keys.
Creating a table with encrypted values
After creating the column master key definition and column encryption keys, you can create the table
to hold the encrypted values. Before you do this, you must decide what type of encryption to use,
which columns to encrypt, and whether you can index these columns. With the Always Encrypted
feature, you define column sizes normally, and SQL Server adjusts the storage size of the column
based on the encryption settings. After you create your table, you might need to change your
application to execute commands on this table using Always Encrypted. In this section, we describe
the choices you have when creating your table and adapting your application.
Encryption types
Before creating a table to contain encrypted values, you must first make a choice about each column
to be encrypted. First, will this column be used for looking up values or just returning those values? If
the column is going to be used for lookups, the column must use a deterministic encryption type,
which allows for equality operations. However, there are limitations on searching for data that has
been encrypted by using the Always Encrypted feature. SQL Server 2016 supports only equality
operations, which include equal to, not equal to, joins (which use equality), and using the value in the
GROUP BY clause. Any search using LIKE is not supported. Additionally, sorting data that is encrypted
using Always Encrypted must be done at the application level, as SQL Server will sort based on the
encrypted value rather than the decrypted value.
If the column is not going to be used for locating records, then the column should use the
randomized encryption type. This type of encryption is more secure, but it does not support searches,
joins, or grouping operations.
CREATE TABLE statement for encrypted columns
When creating tables, you use the normal CREATE TABLE syntax with some additional parameters
within the column definition, as shown in Example 2-3. Three parameters are used within the
ENCRYPTED WITH syntax for the CREATE TABLE statement. The first of these is the ENCRYPTION_TYPE
parameter, which accepts a value of RANDOMIZED or DETERMINISTIC. The second is the ALGORITHM
parameter, which only accepts a value of AEAD_AES_256_CBC_HMAC_SHA_256. The third parameter is
the COLUMN_ENCRYPTION_KEY, which is the encryption key you use to encrypt the value.
CH A P TER 2 | Better security
Example 2-3: Creating a table using Always Encrypted
CREATE TABLE [dbo].[Customers](
[CustomerId] [int] IDENTITY(1,1),
[TaxId] [varchar](11) COLLATE Latin1_General_BIN2
[FirstName] [nvarchar](50) NULL,
[LastName] [nvarchar](50) NULL,
[MiddleName] [nvarchar](50) NULL,
[Address1] [nvarchar](50) NULL,
[Address2] [nvarchar](50) NULL,
[Address3] [nvarchar](50) NULL,
[City] [nvarchar](50) NULL,
[PostalCode] [nvarchar](10) NULL,
[State] [char](2) NULL,
[BirthDate] [date]
The sample code shown in Example 2-3 creates two encrypted columns. The first encrypted column is
the TaxId column, which is encrypted as a deterministic value because our application allows a search
of customers based on their government-issued tax identification number. The second encrypted
column is the BirthDate column, which is a randomized column because our application does not
require the ability to search, join, or group by this column.
Indexing and Always Encrypted
Columns containing encrypted data can be used as key columns within indexes—provided that those
columns are encrypted by using the DETERMINISTIC encryption type. Columns encrypted by using the
RANDOMIZED encryption type return an error message when you try to create an index on those
columns. Columns encrypted by using either encryption type can be used as INCLUDE columns within
nonclustered indexes.
Because encrypted values can be indexes, no additional performance-tuning measures are required
for values encrypted with Always Encrypted beyond the indexing and tuning that you normally
perform. Additional network bandwidth and greater I/O are the only side effects that result from the
increased size of the values being returned.
Application changes
The beauty of the Always Encrypted feature of SQL Server 2016 is that applications already using
stored procedures, ORMs, or parameterized T-SQL commands should require no application changes
to use Always Encrypted, unless nonequality operations are currently being used. Applications that
build SQL statements as dynamic SQL within the application and execute those commands against the
database directly need to be modified to use parameterization of their queries, a recommended
security best practice for all applications, before they can take advantage of the Always Encrypted
Another change required to make Always Encrypted work is the addition of a connection string
attribute to the connection string of the application connecting to the database:
Column Encryption Setting=enabled
CH A P TER 2 | Better security
With this setting added to the connection string, the ADO.NET driver asks the SQL Server if the
executing command includes any encrypted columns, and if so, which columns are encrypted. For
high-load applications, the use of this setting may not be the best practice, especially if a large
percentage of executing commands do not include encrypted values. Consequently, the .NET
Framework provides a new method on the SqlConnection object called
SqlCommandColumnEncryptionSetting, which has three possible values as shown in the following
Method value
Effective change
There are no Always Encrypted columns or parameters to use for the queries that
are executed by using this connection object.
There are Always Encrypted columns and/or parameters in use for the queries that
are executed by using this connection object.
There are no Always Encrypted parameters. However, executing queries using this
connection object return columns encrypted by using Always Encrypted.
Note Be aware that the use of this method can potentially require a significant amount of change
to your application code. An alternative approach is to refactor your application to use different
For the best performance of SQL Server, it is wise to request only the metadata about Always
Encrypted for those queries that use Always Encrypted. This means that in applications for which a
large percentage of queries use Always Encrypted, the connection string should be enabled and the
specific queries within the application should specify SqlCommandColumnEncryptionSetting as
Disabled. For applications for which most queries are not using Always Encrypted values, the
connection string should not be enabled, and SqlCommandColumnEncryptionSetting should be set for
Enabled or ResultSet as needed for those queries that are using Always Encrypted columns. In most
cases, applications are able to simply enable the connection string attribute, and application
performance will remain unchanged while using the encrypted data.
Note While enabling the Always Encrypted setting has been designed to be an easy-to-implement
solution for application data encryption, it is a very major change to application functionality. Like
all major changes to application functionality, there should be rigorous testing of this feature in a
testing environment, including load testing, before making this change in a production
Migrating existing tables to Always Encrypted
In a production environment, there is no direct path to migrate an unencrypted table to a table that is
protected by Always Encrypted. A multiphased approach to data migration is required to move data
from the current table into the new table. The basic approach to move data from an existing table into
an Always Encrypted table includes the following steps:
Build a new staging table.
Write a .NET application using ADO.NET to process the encryption of both existing and updated
Run the .NET application built in the prior step.
4. Drop the existing table and rename the new table to use the old table name.
CH A P TER 2 | Better security
Change the application’s connection string to include Column Encryption Setting=enabled.
Note For nonproduction environments, you can use the Always Encrypted wizard or the
Import/Export wizard in SSMS, which follow a process similar to the one we outline in this section.
Step 1: Build a new staging table
Because Always Encrypted does not support the conversion of an existing table into an Always
Encrypted table, you must build a new table. The new table should have the same schema as the
existing table. When you build the new table, the only changes you need to make are enabling the
columns to be encrypted and specifying the collation as described in Example 2-3.
A large application is likely to require a large amount of time to encrypt and move the data, and it
might not complete this process during a single maintenance window. In that case, it is helpful to
make two additional schema changes. The first change is to add a column on the production table to
track when a row is updated (if the table does not already have such a column). The second change is
to add a trigger to the production table that fires on delete and removes any rows from the new table
when the row is deleted from the production table. To reduce downtime when you move the table
with the encrypted data into production, you should create any indexes existing on the production
table on the new table before loading it with data.
Steps 2 and 3: Write a .NET application to encrypt the data and move it to the new table
Because of the design of Always Encrypted, data is encrypted only by applications using the ADO.NET
driver with parameterized queries. This design prevents you from using SSMS to move data into the
new table. Similarly, you cannot use an application to perform a simple query such as this:
The rows must be brought from the database into a .NET application and then written back to the
database using a parameterized query, one row at a time, for the data to be properly inserted as
encrypted values in the database.
For small applications, this process can be completed quickly, within a single maintenance window.
For larger applications, this processes may take several nights, which requires the application to be
aware of data changes during the business day. After the application has processed the initial push of
data from the source table to the new table, the application must run periodically to move over any
changed rows to the new table until the cutover has been completed.
Step 4: Rename the table
Once all the data has been migrated, the existing table can be dropped or renamed so that it can be
saved until testing has been completed. Then the new table can be renamed so that it now has the
production table’s name. Any indexes existing on the production table that do not exist on the new
table should be created at this time, as well as any foreign keys that exist on the old table. Once
testing is completed, if the old table is not deleted, any foreign keys using that table as a parent
should be removed to prevent issues when rows are deleted.
Step 5: Update the application’s connection string
Once the tables are changed, the application needs to know to use Always Encrypted. To do this,
change the application’s connection string to use the new Column Encryption Setting=enabled
attribute or release a new version of the application that uses the
SqlCommandColumnEncryptionSetting method on the connection object within the .NET code.
CH A P TER 2 | Better security
Using Always Encrypted in Microsoft Azure SQL Database
Always Encrypted is fully supported by the SQL Database platform. You configure Always Encrypted
for a SQL Database just as you do for an on-premises SQL Server 2016 deployment by using T-SQL
commands. At the time of this writing, there are no enhancements in the Microsoft Azure portal for
configuring Always Encrypted in SQL Database.
Row-Level Security
Row-Level Security (RLS) allows you to configure tables such that users see only the rows within the
table to which you grant them access. This feature limits which rows are returned to the user,
regardless of which application they are using, by automatically applying a predicate to the query. You
can use a filter predicate to silently filter the rows that are accessible by the user when using INSERT,
UPDATE, or DELETE statements. In addition, you can use the following block predicates to block the
block predicates return an error to the application indicating that the user is attempting to modify
rows to which the user does not have access.
You implement RLS by creating an inline table function that identifies the rows accessible to users.
The function you create can be as simple or complex as you need. Then you create a security policy to
bind the inline table function to one or more tables.
Note Although you can create a complex RLS inline table function, bear in mind that complex
queries are typically slow to execute. Besides ensuring that your function properly limits access to
specific rows in a table, you should take care that it does so with minimal impact to application
RLS is designed to simplify your application code by centralizing access logic within the database. It
should be noted that, as with any RLS solution and workarounds, it is possible for users with the ability
to execute arbitrary T-SQL commands to infer the existence of data that should be filtered, via sidechannel attacks. Therefore, RLS is intended for scenarios where the queries that users can execute are
controlled, such as through a middle-tier application.
Be aware that RLS impacts all users of a database, including members of the db_owner fixed database
role. Members of this role have the ability to remove the RLS configuration from tables in the
database. However, by doing so, all other users again have access to all rows in the table.
Note You can use branching logic in the inline table function for RLS when you need to allow
members of the db_owner fixed database role to access all rows in the table.
Creating inline table functions
The method by which users connect to a database determines how you need to write the inline table
function. In an application that connects users to the database with their individual Windows or SQL
login, the function must directly match each user’s login to a value within the table. On the other
hand, in an application that uses a single SQL login for authentication, you must modify the
application to set the session context to use a database value that sets the row-level filtering as we
explain in more detail later in this section. Either way, when you create a row-level filtering inline table
function, you must enable SCHEMABINDING and the function must return a column that contains a
value of 1 (or any other valid value) when the user can view the row.
CH A P TER 2 | Better security
Note You can implement RLS on existing tables without rebuilding the tables because the inline
table function that handles the filtering is a separate object in the database, which you then bind to
the table after you create the function. Consequently, you can quickly and easily implement RLS in
existing applications without requiring significant downtime.
Application using one login per user
When your application logs into the database engine by using each user’s Windows or SQL login, your
inline table function needs only to compare the user’s login against a table in the database to
determine whether the user has access to the requested rows. As an example, let’s say you have an
Orders application for which you want to use RLS to restrict access to order information to the person
entering the order. First, your application requires an Order table, such as the one shown in Example
2-4. When your application writes a row into this table, it must store the user’s login in the SalesRep
Example 2-4: Creating an Orders table
OrderId int,
SalesRep sysname
Your next step is to create an inline table function like the one shown in Example 2-5. In this example,
when a user queries the Orders table, the value of the SalesRep column passes into the @SalesRep
parameter of the fn_Orders function. Then, row by row, the function compares the @SalesRep
parameter value to the value returned by the USER_NAME() system function and returns a table
containing only the rows for which it finds a match between the two values.
Example 2-5: Creating an inline table function to restrict access by user login
CREATE FUNCTION dbo.fn_Orders(@SalesRep AS sysname)
SELECT 1 AS fn_Orders_result
WHERE @SalesRep = USER_NAME();
Note The data type of the parameter in your inline table function must match the corresponding
column data type in the table that you plan to secure with RLS, although it is not necessary for the
parameter name to match the column name. However, managing your code is easier if you keep
the names consistent.
Now let’s consider what happens if your database contains related information in another table, such
as the OrderDetails table shown in Example 2-6.
CH A P TER 2 | Better security
Example 2-6: Creating an OrderDetails table
OrderId int,
ProductId int,
Qty int,
Price numeric(8,2)
To apply the same security policy to this related table, you must implement additional filtering by
creating another inline table-valued function, such as the one shown in Example 2-7. Notice that you
continue to use the USER_NAME() system function to secure the table by a user-specific login.
However, this time the inline table-valued function’s parameter is @OrderId, which is used in
conjunction with the SalesRep column.
Example 2-7: Creating an inline table function to restrict access by user login in a
related table
CREATE FUNCTION dbo.fn_OrderDetails(@OrderId AS int)
SELECT 1 AS fn_Orders_result
FROM Orders
WHERE OrderId = @OrderId
AND SalesRep = USER_NAME();
Application using one login for all users
When your application uses a single login for all users of the application, also known as an application
account, you use similar logic as you do when the application passes user logins to the database. Let’s
continue with a similar example as the one in the previous section, but let’s add some additional
columns to the Orders table, as shown in Example 2-8. In this version of the Orders table, the SalesRep
column has an int data type instead of the sysname data type in the earlier example.
Example 2-8: Creating a variation of the Orders table
OrderId int,
SalesRep int,
ProductId int,
Qty int,
Price numeric(8,2)
Additionally, the inline table function changes to reflect the single login, as shown in Example 2-9.
Notice the parameter’s data type is now int instead of sysname to match the column in the table
shown in Example 2-8. In addition, the predicate in the function now uses the SESSION_CONTEXT
system function and outputs the result as an int data type to match the input parameter’s data type.
CH A P TER 2 | Better security
Example 2-9: Creating an inline table function for an application using a single login
CREATE FUNCTION dbo.fn_Orders(@SalesRep AS int)
SELECT 1 AS fn_Orders_result
You must also modify your application code to use the sp_set_session_context system stored
procedure, which sets the value returned by the SESSION_CONTEXT system function, as shown in
Example 2-10. This system stored procedure supports two parameters—the key name of the value to
add and the value to store for this key. In this example, the key name is UserID and its value is set to
the UserId of the application user, which the application passes into the stored procedure by using the
@UserId input parameter. Applications can call sp_set_session_context in line within the stored
procedures or directly at application startup when the connection is created.
Example 2-10: Using the sp_set_session_context system stored procedure
@OrderId int,
@UserId int
EXEC sp_set_session_context @key=N'UserId', @value=@UserId;
FROM Orders
WHERE OrderId = @OrderId;
Creating security policies
After creating inline table-valued functions, you next bind them to the table that you want to secure.
To do this, use the CREATE SECURITY POLICY command, as shown in Example 2-11. In the security
policy, you can define a filter predicate by specifying the inline table-valued function name, the
column name to pass to the function, and the table to which the policy applies.
Example 2-11: Creating a security policy
ADD FILTER PREDICATE dbo.fn_Orders(SalesRep) ON dbo.Orders
You can specify multiple filter predicates in the security policy when you want to filter rows in different
tables, as shown in Example 2-12.
Example 2-12: Creating one security policy for multiple tables
ADD FILTER PREDICATE dbo.fn_Orders(SalesRep) ON dbo.Orders,
ADD FILTER PREDICATE dbo.fn_OrderHistory(OrderId) ON dbo.OrderHistory
CH A P TER 2 | Better security
Using block predicates
When you use the filter predicate as shown in the examples in the preceding section, the security
policy affects “get” operations only. Users are still able to insert rows that they cannot subsequently
query. They can also update rows they can currently access and even change the rows to store values
that block further access. You must decide whether your application should allow this behavior or
should prevent users from inserting rows to which they do not have access. To do this, use a block
predicate in addition to a filter predicate.
As shown in Example 2-13, you can use both filter and block predicates in a security policy. In this
example, the security policy allows users to query for rows using the SELECT statement and returns
only rows to which the user has access. A user can insert new rows into the table as long as the
SalesRep value matches the user’s login. Otherwise, the insert fails and returns an error to the user.
Similarly, an update to the table succeeds as long as the user doesn’t attempt to change the value of
the SalesRep column. In that case, the update fails and returns an error to the user.
Example 2-13: Using block and filter predicates in a single security policy
ADD FILTER PREDICATE dbo.fn_Orders(SalesRep) ON dbo.Orders,
ADD BLOCK PREDICATE dbo.fn_Orders(SalesRep) ON dbo.Orders AFTER INSERT,
ADD BLOCK PREDICATE dbo.fn_Orders(SalesRep) ON dbo.Orders AFTER UPDATE
Note You can use a filter predicate to prevent users from updating or deleting records they cannot
read, but the filter is silent. By contrast, the block predicate always returns an error when
performing these operations.
Using RLS in SQL Database
You can use RLS in SQL database by using the same T-SQL commands described in this chapter. At
the time of this writing, you cannot use the Azure portal to implement RLS.
Dynamic data masking
When you have a database that contains sensitive data, you can use dynamic data masking to
obfuscate a portion of the data unless you specifically authorize a user to view the unmasked data. To
mask data, you can use one of the following four masking functions to control how users see the data
returned by a query:
Default Use this function to fully mask values by returning a value of XXXX (or fewer Xs if a
column length is less than 4 characters) for string data types, 0 for numeric and binary data types,
and 01.01.2000 00:00:00.0000000 for date and time data types.
Email Use this function to partially mask email addresses like this: This pattern
masks not only the email address but also the length of the email address.
CH A P TER 2 | Better security
Partial Use this function to partially mask values by using a custom definition requiring three
parameters as described in the following table:
Number of starting characters to display, starting from the first character
in the value.
Value to be displayed between the prefix and suffix characters.
Number of ending characters to display, starting from the last character in
the value.
Random Use this function to fully mask numeric values by using a random value between a lower
and upper boundary that you specify.
Random function may display unmasked data
The Random() data-masking function may on occasion display the actual value that is stored in the
table. This behavior is the result of using a random value that could match the value to mask if it is
within the specified range. You should consider whether the business rules of your application allow
for this behavior before using this masking function. Whenever possible, use a range of values
outside the possible range of values to mask to ensure that there is no possibility of an accidental
data leak. While it is possible that the random value will return the actual value, there is no way of
knowing that the displayed random value is in fact the actual value without knowing the actual
Dynamic data masking of a new table
To configure dynamic data masking for a new table, use the CREATE TABLE statement with the
MASKED WITH argument, as shown in Example 2-14. In this example, the default() function masks the
TaxId column for complete masking, and the partial() function masks the FirstName column by
displaying its first three characters and its final character and replacing the remaining characters with
Example 2-14: Creating a table with two masked columns
CREATE TABLE [dbo].[Customer](
[CustomerId] [int] IDENTITY(1,1) NOT NULL,
[TaxId] [varchar](11) MASKED WITH (FUNCTION = 'default()'),
[FirstName] [nvarchar](50) MASKED WITH (FUNCTION = 'partial(3, "xyz", 1)') NULL,
[LastName] [nvarchar](50) NULL,
[CustomerId] ASC)
Dynamic data masking of an existing table
Because dynamic data masking changes only the presentation of data returned by a query, there is no
change to the underlying table structure. That means you can easily add dynamic data masking to a
column in an existing table without rebuilding the table. To this, use the ALTER TABLE statement with
the ALTER COLUMN and ADD MASKED arguments, as shown in Example 2-15.
CH A P TER 2 | Better security
Example 2-15: Adding dynamic data masking to an existing table
ALTER TABLE [dbo].[Customers]
Likewise, you can remove dynamic data masking quickly and easily without rebuilding a table or
moving data because only metadata changes rather than the schema. You remove dynamic data
masking from a column by using the ALTER TABLE statement with the ALTER COLUMN and DROP
MASKED arguments, as shown in Example 2-16.
Example 2-16: Removing dynamic data masking from a table
ALTER TABLE [dbo].[Customers]
Understanding dynamic data masking and permissions
When you use dynamic data masking, the permissions that you assign to users affect whether users
see plain text values or masked values. Specifically, members of the db_owner fixed database role
always see plain text values, whereas users who are not members of this role see masked data by
If you need to grant a user permission to see plain text data in a table, you must grant the new
UNMASK permission at the database level. To do this, use the GRANT UNMASK statement in the
database containing the masked values, as shown in Example 2-17.
Example 2-17: Granting the UNMASK permission
Note It is not possible to grant table-level access to masked data. You can grant this privilege only
at the database level. Consequently, you can mask either all masked data within the database for a
user or none of the data.
To remove this permission, you use the REVOKE statement as shown in Example 2-18.
Example 2-18: Revoking the UNMASK permission
Figure 2-4 shows examples of query results when you apply dynamic data masking to a table. The first
query shows default and email masking. The second result set shows the same queries executed after
giving the user permissions to view masked data.
CH A P TER 2 | Better security
Figure 2-4: Query results for masked and unmasked values.
Data-masking permissions and configuration survive when you copy data from one object to another.
For example, if you copy data from a user table to a temporary table, the data remains masked in the
temporary table.
Masking encrypted values
Dynamic data masking does not work with encrypted values if you encrypt data in the application tier
or by using the Always Encrypted feature. If you encrypt data before storing it in the SQL Server
database engine, the engine cannot mask a value that it cannot decrypt. In this case, because data is
already encrypted, there is no benefit or extra protection from applying dynamic data masking.
Using dynamic data masking in SQL Database
Dynamic data masking is also available for use in SQL Database. You can configure it by using T-SQL
or by using the Microsoft Azure portal. In the Azure portal, navigate to the list of SQL Databases
within SQL DB, and then select the database to view its properties. Next, in the Settings panel, select
Dynamic Data Masking, as shown in Figure 2-5. In the Dynamic Data Masking window, a list of
masking rules is displayed in addition to a list of columns for which data masking is recommended.
You can enable data masking on those columns by clicking the Add Mask button to the right of the
column name.
CH A P TER 2 | Better security
Figure 2.5: Configuring dynamic data masking for a SQL Database in the Azure portal.
After specifying the mask function to apply to selected columns, click the Save button at the top of
the window to save the configuration changes to your SQL Database. After saving these changes,
users can no longer see the unmasked data in the SQL Database tables unless they have the unmask
privilege within the database.
CH A P TER 2 | Better security
Higher availability
In a world that is always online, maintaining uptime and streamlining
maintenance operations for your mission-critical applications are more
important than ever. In SQL Server 2016, the capabilities of the AlwaysOn
Availability Group feature continue to evolve from previous versions,
enabling you to protect data more easily and flexibly and with greater
throughput to support modern storage systems and CPUs. Furthermore,
AlwaysOn Availability Groups and AlwaysOn Failover Cluster Instances now
have higher security, reliability, and scalability. By running SQL Server 2016
on Windows Server 2016, you have more options for better managing
clusters and storage. In this chapter, we introduce the new features that
you can use to deploy more robust high-availability solutions.
AlwaysOn Availability Groups
First introduced in SQL Server 2012 Enterprise Edition, the AlwaysOn Availability Groups feature
provides data protection by sending transactions from the transaction log on the primary replica to
one or more secondary replicas, a process that is conceptually similar to database mirroring. In SQL
Server 2014, the significant enhancement to availability groups was the increase in the number of
supported secondary replicas from three to eight. SQL Server 2016 includes a number of new
enhancements that we explain in this section:
AlwaysOn Basic Availability Groups
Support for group Managed Service Accounts (gMSAs)
Database-level failover
Distributed Transaction Coordinator (DTC) support
Load balancing for readable secondary replicas
Up to three automatic failover targets
Improved log transport performance
CH A P TER 3 | Higher availability
New to availability groups?
If you are still using database mirroring, there are several reasons to transition your high-availability
strategy to availability groups. Database mirroring is deprecated as of SQL Server 2012, for
example, and basic availability groups are now included in SQL Server 2016 Standard Edition as a
replacement. Also, if you are exploring options for high-availability/disaster-recovery (HA/DR)
solutions but have never implemented availability groups, SQL Server 2016 provides several
benefits to consider.
Whereas database mirroring occurs at the database level, using a single thread to perform the data
replication, data is moved within availability groups by using a worker pool, which provides better
throughput and reduces CPU overhead. When your application requires multiple databases, you
can assign the databases to a single availability group to ensure that they all fail over at the same
time. By contrast, the unit of failover with database mirroring is a single database. With database
mirroring, you use a SQL Server witness instance to manage automatic failover, but with availability
groups you rely on Windows Server Failover Clustering (WSFC) to arbitrate uptime and connections.
Furthermore, clustering is a more robust solution than database mirroring because it provides
additional levels of protection.
A key benefit of availability groups is the ability to scale out replicas that you can configure to
support both high-availability and disaster-recovery requirements. For high-availability scenarios,
you should locate two or three servers in the same geographic location, configured to use
synchronous-commit mode and automatic failover. That said, automatic failover should be used
only in low-latency scenarios because writes to the primary replica are not considered complete
until they reach the transaction log on the secondary replica. For disaster-recovery scenarios in
which the servers are more than 100 kilometers apart, asynchronous-commit mode is a better
choice to minimize the performance impact on the primary replica.
Another benefit of availability groups is the ability for databases on a secondary replica to support
online reads as well as database backups. This capability allows you to implement a scale-out
architecture for reporting solutions by having multiple copies of secondary replicas in multiple
geographies. You provide connectivity to the availability group by using a virtual IP address called
the listener, which you configure to connect transparently to the primary replica or to a secondary
replica for reading. Figure 3-1 is a diagram of an availability group with replicas in New York, Los
Angeles, and Seattle and a listener to which clients connect.
Figure 3-1: An AlwaysOn Availability Group with a primary replica and two secondary replicas.
Supporting disaster recovery with basic availability groups
You can now use basic availability groups in the Standard Edition to automatically fail over a single
database. The use of basic availability groups is subject to the following limitations:
Two replicas (one primary and one secondary)
CH A P TER 3 | Higher availability
One availability database
No read access on secondary replica
No backups on secondary replica
No availability group listener
No support in an existing availability group to add or remove a replica
No support for upgrading a basic availability group to an advanced availability group
Despite these limitations, with a basic availability group you get benefits similar to database mirroring
in addition to other features. For each replica, you can choose either synchronous-commit or
asynchronous-commit mode, which determines whether the primary replica waits for the secondary
replica to confirm that it has hardened the log. Moreover, performance is better because basic
availability groups use the same improved log transport operations that we describe later in this
The steps to configure basic availability groups are similar to those for regular availability groups, with
some exceptions. You start by using the New Availability Group Wizard, which you launch in SQL
Server Management Studio (SSMS) by right-clicking the AlwaysOn High Availability folder in Object
Explorer. When you reach the Specify Replicas page, you click the Add Replica button to add the
primary and secondary replicas, but then the button becomes unavailable, as shown in Figure 3-2. In
addition, you cannot change the value for the Readable Secondary drop-down list, nor can you access
the Backup Preferences or Listener tabs.
Figure 3-2: Configuring replicas for a basic availability group.
CH A P TER 3 | Higher availability
Note Although including an Azure replica in your disaster-recovery architecture is fully supported
for basic availability groups, the New Availability Group Wizard does not allow you the option to
add it. However, you can perform this step separately by using the Add Azure Replica Wizard, which
is described at
Using group Managed Service Accounts
To comply with regulatory auditing requirements, DBAs or system administrators in a large enterprise
must frequently reset service account passwords across SQL Server instances. However, managing
individual service account passwords involves a high degree of risk because downtime is likely to
occur if anything goes wrong. To address this problem, Microsoft enhanced Windows Server 2012 so
that you can more easily manage passwords for a service account in Active Directory by creating a
single service account for your SQL Server instances and then delegating permissions to each of those
servers. By default, Active Directory changes the password for a group Managed Service Account
(gMSA) every thirty days, although you can adjust the password-change interval to satisfy your audit
In SQL Server 2012 and SQL Server 2014, you can implement this feature only in standalone
configurations. In SQL Server 2016, you can now use gMSAs with both availability groups and failover
clusters. If you are using Windows Server 2012 R2 as your operating system, you must install
KB298082 to ensure that services can seamlessly log on after a password change. However, no
patches are required if you install SQL Server 2016 on Windows Server 2016.
Triggering failover at the database level
Beginning in SQL Server 2012, AlwaysOn Availability Groups and AlwaysOn Failover Cluster Instances
(FCIs) use the sp_server_diagnostics stored procedure to periodically monitor the health of a server.
The default behavior is to fail over an availability group or an FCI when the health monitoring reveals
any of the following conditions:
The stored procedure returns an error condition.
The SQL Service service is not running.
The SQL Server instance is not responding.
However, in versions earlier than SQL Server 2016, this check does not account for database-level
failures. Beginning in SQL Server 2016, you can enable Database Level Health Detection when you
create an availability group, as shown in Figure 3-3. This way, any error that causes a database to be
suspect or go offline also triggers a failover of the availability group.
Note The FailureConditionLevel property determines the conditions that trigger a failover. For
normal operations, the default value is suitable. However, you can reduce or increase this property’s
value if necessary. To learn more, see “Configure FailureConditionLevel Property Settings” at
CH A P TER 3 | Higher availability
Figure 3-3: Creating a new availability group with database-level health detection.
Note Enabling Database Level Health Detection needs to be weighed carefully with the needs of
your application and its intended behavior in response to a failover. If your application can support
a database failover, Database Level Health Detection can enhance your total availability and uptime.
Automatic page repair
An important capability of availability groups is automatic page repair. If the primary replica cannot
read a page, it requests a fresh copy of the page from a secondary. However, in the event that the
primary replica cannot be repaired, such as when storage fails completely, you might be able to
bring your secondary replica online as a primary replica.
Supporting distributed transactions
One of the features long supported in FCIs, but not in availability groups, is the use of the Distributed
Transaction Coordinator (DTC). This feature is required if your application performs transactions that
must be consistent across multiple instances. When running SQL Server 2016 on Windows Server
2016, you can now implement support for distributed transactions when you create a new availability
group. To do this, you select the Per Database DTC check box (shown earlier in Figure 3-3) or by using
the T-SQL command shown in Example 3-1. Note that you cannot add DTC support to an existing
CH A P TER 3 | Higher availability
availability group. By enabling DTC support, your application can distribute transactions between
separate SQL Server instances or between SQL Server and another DTC-compliant server, such as
Oracle or WebSphere.
Example 3-1: Creating an availability group with DTC support
Note Because each database in an availability group is synchronized independently while the
cross-database transaction manager operates at the SQL Server instance level, an active crossdatabase transaction might be lost during an availability group failover. Consequently, crossdatabase transactions are not supported for databases hosted by one SQL Server or within the
same availability group.
Scaling out read workloads
You can use availability groups for scale-out reads across multiple secondary copies of the availability
group. In SQL Server 2016, as in the previous version, you can scale up to as many as eight secondary
replicas, but the scale-out reads feature is not enabled by default. To support scale-out reads, you
must configure read-only routing by using the T-SQL commands shown in Example 3-2. You must
also create an availability group listener and direct connections to this listener. In addition, the
connection string must include the ApplicationIntent=ReadOnly keyword.
Example 3-2: Configuring read-only routing
CH A P TER 3 | Higher availability
Note You can also use Windows PowerShell to configure a read-only routing list as described at
In SQL Server 2012 and SQL Server 2014, the read traffic is directed to the first available replica,
without consideration for load balancing. An alternative solution requires the use of third-party
hardware or software load balancers to route traffic equally across the secondary copies of the
databases. In SQL Server 2016, you can now balance loads across replicas by using nested
parentheses in the read-only routing list, as shown in Example 3-3. In this example, connection
requests first try the load-balanced set containing Server1 and Server2. If neither replica in that set is
available, the request continues by sequentially trying other replicas defined in the list, Server3 and
Server4 in this example. Only one level of nested parentheses is supported at this time.
Example 3-3: Defining a load-balanced replica list
READ_ONLY_ROUTING_LIST = (('Server1','Server2'), 'Server3', 'Server4')
Defining automatic failover targets
In SQL Server 2012 and SQL Server 2014, you can define a maximum of two replicas running in an
automatic failover set, but now SQL Server 2016 allows for a third replica to support a topology such
as is shown in Figure 3-4. In this example, the replicas on Node01, Node02, and Node03 are
configured as an automatic failover set. As long as data is synchronized between the primary replica
and one of the secondary replicas, failover can take place in an automatic fashion with no data loss.
Figure 3-4: Availability Group topology with three automatic failover targets.
When configuring availability group failover, you can choose from among the following failover
Automatic Failover A failover that occurs automatically on the failure of the primary replica,
which is supported only when both the primary replica and at least one secondary replica are
configured with AUTOMATIC failover mode and the secondary replica is currently synchronized.
Planned Manual Failover (without data loss) A failover that is typically initiated by an
administrator for maintenance purposes. This requires synchronous-commit mode, and the
databases must currently be synchronized.
Forced Failover (with possible data loss) A failover that occurs when the availability group is
configured with asynchronous-commit mode, or the databases in the availability group are not
currently synchronized.
CH A P TER 3 | Higher availability
For automatic failover to work, you must configure all members of the availability group for
synchronous-commit mode and for automatic failover. You typically configure automatic failover for
high-availability scenarios, such as rolling updates to SQL Server. In this configuration, any
uncommitted transactions that have not reached the secondary replica are rolled back in the event of
failover, thereby providing near zero data loss.
Reviewing the improved log transport performance
When AlwaysOn Availability Groups were first introduced in SQL Server 2012, solid-state storage
devices (SSDs) were far less prevalent in enterprise IT architectures than they are now. SSDs enable
more throughput, which can be problematic on a standalone system and can overwhelm the ability to
write to the secondary database. In prior versions of SQL Server, the underlying processes responsible
for synchronizing data between replicas are shared among availability groups, database mirroring,
Service Broker, and replication. In SQL Server 2016, these processes are completely rewritten, resulting
in a streamlined protocol with better throughput and lower CPU utilization.
Although the process has been refactored, the sequence of internal operations for the log transport,
shown in Figure 3-5, continues to include the following steps:
Log flush Log data is generated and flushed to disk on the primary replica in preparation
for replication to the secondary replica. It then enters the send queue.
Log capture Logs for each database are captured on the primary replica, compressed, and sent
to the corresponding queue on the secondary replica. This process runs continuously as long as
database replicas are connecting. If this process is not able to scan and enqueue the messages
quickly enough, the log send queue continues to grow.
Send The messages are removed from the queue and sent to each secondary replica across the
4. Log receive/Log cache Each secondary replica gets messages from the primary replica and
then caches the messages.
Log hardened The log is flushed on the secondary replica, and then a notification is sent to the
primary replica to acknowledge completion of the transaction.
Redo pages
The flushed pages are retrieved from the redo queue and applied to the secondary
Figure 3-5: Log transport operations for AlwaysOn Availability Groups.
CH A P TER 3 | Higher availability
Bottlenecks can occur in this process during the log-capture step on the primary replica and the redo
step on the secondary replica. In previous versions of SQL Server, both steps were single-threaded.
Consequently, bottlenecks might occur during large index rebuilds on availability groups with highspeed storage and on local networks, because these single-threaded steps had trouble keeping up
with the stream of log records. However, in SQL Server 2016 these steps can use multiple threads that
run in parallel, resulting in significant performance improvements. Furthermore, the compression
functions in the log-capture step have been replaced by a newer Windows compression function that
delivers up to five times better performance. During testing with high-throughput storage devices,
speeds up to 500 MB/s have been observed. Considering that this throughput is a compressed
stream, the redo step is receiving 1 GB/s, which should support the busiest applications on the fastest
Microsoft Azure high-availability/disaster-recovery licensing changes
Hybrid disaster-recovery scenarios are becoming increasingly popular. If you choose to implement
hybrid disaster recovery, be sure to maintain symmetry between on-premises and cloud solutions.
The license mobility benefit included with software assurance (SA) allows you to use a secondary
copy of SQL Server in any high-availability or disaster-recovery scenario without purchasing another
license for it. In the past, you could not use this benefit with the SQL Server images on Azure Virtual
Machines. Now you can deploy a new SQL Server image on an Azure Virtual Machine without
incurring charges as long as the secondary replica is not active. This means you can automate the
scale-out of your high-availability/disaster-recovery solutions with a minimum of effort and cost.
Windows Server 2016 Technical Preview highavailability enhancements
Nearly every version of Windows Server since Windows Server 2008 R2 has had major enhancements
to the operating system’s failover clustering stack as a result of development investments in related
technologies. First, Hyper-V, the virtualization platform in the operating system, uses the clustering
stack for its high-availability and disaster-recovery scenarios. Microsoft Azure also uses this same
functionality. Because SQL Server has failover clustering at the center of its high-availability/disasterrecovery technologies, it also takes advantage of the clustering features in the operating system.
Sometimes these features are visible from the database tier, allowing you to make configuration
changes, but other features from the operating system, such as dynamic quorum, enhance SQL
Server’s uptime without requiring configuration. Windows Server 2016 Server Technical Preview
includes the following features that enhance SQL Server’s uptime:
Workgroup clusters
Cloud witness
Storage Spaces Direct
Troubleshooting enhancements to Windows Server Failover Clusters (WSFC)
Cluster operating system rolling upgrade
Creating workgroup clusters
Earlier in this chapter, we explained how basic availability groups replace nearly all the functionality in
database mirroring. The one advantage that database mirroring has over availability groups in prior
CH A P TER 3 | Higher availability
versions of SQL Server is the ability to provide data protection across Active Directory (AD) domains or
without an AD domain. Starting with Windows Server 2016 Technical Preview and SQL Server 2016,
you can now create a workgroup cluster with nondomain servers as well as servers attached to
different AD domains. However, there is no mechanism for using a file share witness. Instead, you
must create a cloud witness, as we describe in the next section, or a shared disk.
Each server that you want to add to a workgroup cluster requires a primary DNS suffix in the full
computer name, as shown in Figure 3-6. You can add this suffix by clicking the More button in the
Computer Name/Domain Changes dialog box, which you access from System Properties for the
Figure 3-6: A server with a DNS suffix assigned to a workgroup.
Note In the current release of Windows Server 2016 Technical Preview, this feature is enabled by
default. However, you must use PowerShell to create a workgroup cluster because the Failover
Cluster Manager snap-in does not support this functionality. Furthermore, configuring a crossdomain cluster is more complex than configuring a workgroup cluster because Kerberos is required
to make cross-domain authentication work correctly. You can learn more about configuration
requirements for both scenarios by referring to “Workgroup and Multi-domain clusters in Windows
Server 2016” at
Configuring a cloud witness
Maintaining quorum is the most important function of any clustering software. In SQL Server, the
most frequently deployed model is Node And File Share Majority. One of the challenges when you are
designing a disaster-recovery architecture is to decide where to place the file share witness.
Microsoft’s official recommendation is to place it in a third data center, assuming you have a
primary/secondary availability group configuration. Many organizations already have two data
centers, but fewer have a third data center. Windows Server 2016 Technical Preview introduces a new
feature, the cloud witness, that address this problem.
CH A P TER 3 | Higher availability
You create a cloud witness by using the Cluster Quorum Wizard. Before you launch this wizard, you
must have an active Azure subscription and a storage account. To launch the wizard, right-click the
server in the Failover Cluster Manager, point to More Actions, select Configure Cluster Quorum
Settings, select the Select The Quorum Witness option, and then select the Configure A Cloud Witness
option, as shown in Figure 3-7.
Figure 3-7: Creating a cloud witness for a cluster quorum.
On the next page of the wizard, provide the name of your Azure storage account, copy the storage
account key from the Azure portal to the clipboard, and type the Azure service endpoint, as shown in
Figure 3-8.
CH A P TER 3 | Higher availability
Figure 3-8: Addition of Azure credentials to cloud witness configuration.
When you successfully complete the wizard, the cloud witness is displayed in the Cluster Core
Resources pane in the Failover Configuration Manager snap-in, as shown in Figure 3-9.
Figure 3-9: Successful addition of a cloud witness to cluster core resources.
Because you select the Azure region when you create a storage account, you have control over the
placement of your cloud witness. All communications to and from Azure storage are encrypted by
default. This new feature is suitable for most disaster-recovery architectures and is easy to configure
at minimal costs.
Note More information on this topic is available in “Understanding Quorum Configurations in a
Failover Cluster” at
CH A P TER 3 | Higher availability
Using Storage Spaces Direct
Windows Server 2016 Technical Preview introduces the Storage Spaces Direct feature, which
seamlessly integrates several existing Windows Server features to build a software-defined storage
stack, as shown in Figure 3-10. These features include Scale-Out File Server, Clustered Shared Volume
File Systems (CSVFS), and Failover Clustering.
Figure 3-10: Storage options in Storage Spaces Direct.
The main use case for this feature is high-performance primary storage for Hyper-V virtual files.
Additionally, storage tiers are built into the solution. If you require a higher tier of performance for
TempDB volumes, you can configure Storage Spaces Direct accordingly. SQL Server can take full
advantage of this feature set because the infrastructure supports AlwaysOn Availability Groups and
AlwaysOn Failover Cluster Instances, thereby providing a much lower total cost of ownership
compared with traditional enterprise storage solutions.
Note See “Storage Spaces Direct in Windows Server 2016 Technical Preview” at to learn more.
Introducing site-aware failover clusters
Windows Server 2016 Technical Preview also introduces site-aware clusters. As a consequence, you
can now group nodes in stretched clusters based on their physical location. This capability enhances
key clustering operations such as failover behavior, placement policies, heartbeat between nodes, and
quorum behavior.
CH A P TER 3 | Higher availability
One of the key features of interest to SQL Server professionals is failover affinity, which allows
availability groups to fail over within the same site before failing to a node in a different site.
Additionally, you can now configure the threshold and site delay for heartbeating, which is the
network ping that ensures the cluster can talk to all its nodes.
You can not only specify a site for a cluster node, you can also define a primary location, known as a
preferred site, for your cluster. Dynamic quorum ensures that the preferred site stays online in the
event of a failure by lowering the weights of the disaster-recovery site.
Note Currently (in Windows Server 2016 TP4), the site-awareness functionality is only enabled
through PowerShell and not through Failover Cluster Manager. More information is available at
“Site-aware Failover Clusters in Windows Server 2016” at
Windows Server Failover Cluster logging
Troubleshooting complex cluster problems has always been challenging. One of the goals of WSFC
logging in Windows Server 2016 is to simplify some of these challenges. First, the top of the cluster
log now shows the UTC offset of the server and notes whether the cluster is using UTC or local time.
The cluster log also dumps all cluster objects, such as networks, storage, or roles, into a commaseparated list with headers for easy review in tools such as Excel. In addition, there is a new logging
model call DiagnosticVerbose that offers the ability to keep recent logs in verbose logging while
maintaining a history in normal diagnostic mode. This compromise saves space but also provides
verbose logging as needed.
Note Additional information is available at “Windows Server 2016 Failover Cluster Troubleshooting
Enhancements – Cluster Log” at
Performing rolling cluster operating system upgrades
In prior versions of SQL Server, if your SQL Server instance was running in any type of clustered
environment and an operating system upgrade was required, you built a new cluster on the new
operating system and then migrated the storage to the new cluster. Some DBAs use log shipping to
bring the downtime to an absolute minimum, but this approach is complex and, more importantly,
requires a second set of hardware. With rolling cluster operating system upgrades in Windows Server
2016, the process is more straightforward.
Specifically, SQL Server requires approximately five minutes of downtime in the rolling upgrade
scenario illustrated in Figure 3-11. In general, the process drains one node at a time from the cluster,
performs a clean install of Windows Server 2016, and then adds the node back into the cluster. Until
the cluster functional level is raised in the final step of the upgrade process, you can continue to add
new cluster nodes with Windows Server 2012 R2 and roll back the entire cluster to Windows Server
2012 R2.
CH A P TER 3 | Higher availability
Figure 3-11: State transitions during a rolling operating system upgrade.
Note You use Failover Cluster Manager and PowerShell to manage the cluster upgrade. See
“Cluster Operating System Rolling Upgrade” at
/dn850430.aspx to learn more.
CH A P TER 3 | Higher availability
database engine
In past releases of SQL Server, Microsoft has targeted specific areas for
improvement. In SQL Server 2005, the storage engine was new. In SQL
Server 2008, the emphasis was on server consolidation. Now, in SQL Server
2016, you can find enhanced functionality across the entire database
engine. With Microsoft now managing more than one million SQL Server
databases through its Database as a Service (DBaaS) offering—Microsoft
Azure SQL Database—it is able to respond more quickly to opportunities
to enhance the product and validate those enhancements
comprehensively before adding features to the on-premises version of
SQL Server. SQL Server 2016 is a beneficiary of this new development
paradigm and includes many features that are already available in SQL
Database. In this chapter, we explore a few of the key new features, which
enable you to better manage growing data volumes and changing data
systems, manage query performance, and reduce barriers to entry for
hybrid cloud architectures.
TempDB enhancements
TempDB is one of the components for which performance is critical in SQL Server because the
database engine uses it for temporary tables, query memory spills, index rebuilds, Service Broker, and
a multitude of other internal functions. TempDB file behavior has been enhanced and automated in
SQL Server 2016 to eliminate many performance problems related to the basic configuration of the
server. These changes allow administrators to focus their efforts on more pressing performance and
data issues in their environments.
CH A P TER 4 |
Improved database engine
Configuring data files for TempDB
In earlier versions of SQL Server, the default configuration uses one data file for TempDB. This
limitation sometimes results in page-latch contention, which has frequently been misdiagnosed by
administrators as a storage input/output (I/O) problem for SQL Server. However, the pages for
TempDB are typically in memory and therefore not contributing to I/O contention issues. Instead,
three special types of pages are the cause of the page-latch contention issue: Global Allocation Map
(GAM), Shared Global Allocation Map (SGAM), and Page Free Space (PFS). Each database file can
contain many of these page types, which are responsible for identifying where to write incoming data
in a physical data file. Whenever a process in SQL Server needs to use any of these files, a latch is
taken. A latch is similar to a lock but is more lightweight. Latches are designed to be quickly turned on
and just as quickly turned off when not needed. The problem with TempDB is that each data file has
only one GAM, SGAM, and PFS page per four gigabytes of space, and a lot of processes are trying to
access those pages, as shown in Figure 4-1. Subsequent requests begin to queue, and wait times for
processes at the end of the queue increase from milliseconds to seconds.
Figure 4-1: Contention in TempDB.
An easy way to remedy TempDB page-latch contention in SQL Server is to add more data files. In turn,
SQL Server creates more of the three special types of pages and gives SQL Server more throughput to
TempDB. Importantly, the files should all be the same size. SQL Server uses a proportional fill
algorithm that tries to fill the largest files first, leading to hotspots and more latch contention.
However, because the default setting creates only one file, many database administrators have not
been aware of the solution. Even after learning about the need to create multiple files, there was often
confusion about the correct number of files to configure, especially when factoring in virtual
machines, hyperthreading, and cores versus CPU sockets.
In 2011, Microsoft released the following guidance for TempDB configuration:
As a general rule, if the number of logical processors is less than or equal to 8, use the same number of
data files as logical processors. If the number of logical processors is greater than 8, use 8 data files and
then if contention continues, increase the number of data files by multiples of 4 (up to the number of
logical processors) until the contention is reduced to acceptable levels or make changes to the
CH A P TER 4 | Improved database engine
Note For more detail, see “Recommendations to reduce allocation contention in SQL Server
tempdb database,” at
Accordingly, in SQL Server 2016, this recommendation is built into the product setup. When you
install SQL Server, the default configuration for TempDB now adapts to your environment, as shown in
Figure 4-2. The setup wizard no longer creates a single file by default; instead, it assigns a default
number of files based on the number of logical processors that it detects on the server, up to a
maximum of 8. You can adjust the size of the files and the autogrowth rate if you like. Always monitor
the growth of these files carefully, as performance is affected by file growth even when instant file
initialization is enabled.
Figure 4-2: Configuring TempDB in SQL Server 2016.
Note SQL Server defaults to a conservative setting of 8 megabytes (MB) for Initial Size and 64 MB
for Autogrowth. A best practice is to start with an initial file size of 4,092 MB, with an autogrowth
setting of 512 MB, as the initial file size is still small by most standards. Many DBAs dedicate a
standard-size file system (typically 100–200 GB) to TempDB and allocate 90 percent of it to the data
files. This sizing can reduce contention and also prevents any uncontrolled TempDB growth from
impacting user databases.
Eliminating specific trace flags
Trace flags are commonly used by administrators to perform diagnostics or to change the behavior of
SQL Server. With TempDB in earlier releases of SQL Server, administrators use trace flags 1117 and
1118 to improve performance. In SQL Server 2016, the effect achieved by enabling these two trace
flags has been built into the database engine, rendering them unnecessary.
Trace flag 1117
Trace flag (TF) 1117 is related strictly to file groups and how data files grow within them. A file group
is a logical container for one or more data files within a database. TF 1117 forces all data files in the
same file group to grow at the same rate, which prevents one file from growing more than others,
leading to the hotspot issue described earlier in this chapter. Enabling this trace flag in earlier versions
of SQL Server is a minor tradeoff in performance. For example, if you were using multiple data files in
user databases, this trace flag affects them as well as TempDB’s data files. Depending on your
CH A P TER 4 | Improved database engine
scenario, that could be problematic—an example would be if you had a file group that you did not
want to grow as a single unit. Starting with SQL Server 2016, the behavior to grow all data files at the
same rate is built into TempDB by default, which means you no longer need this trace flag.
Trace flag 1118
Administrators use trace flag 1118 to change page allocation from a GAM page. When you enable TF
1118, SQL Server allocates eight pages, or one extent, at a time to create a dedicated (or uniform)
extent, in contrast to the default behavior to allocate a single page from a mixed extent. Unlike with
TF 1117, there was no potential downside to enabling TF 1118—it is generally recommended for all
SQL Server implementations in earlier releases. Starting with SQL Server 2016, all allocations of
TempDB pages use uniform extent allocation, thus eliminating the need to use TF 1118.
Query Store
One of the most common scenarios you likely encounter is a user reporting that a query is suddenly
running more slowly than in the past or that a long-running job that once took 3 hours is now taking
10. These performance degradations could be the result of changes in data causing out-of-date
statistics or changes in execution parameters or be caused simply by reaching a tipping point in
hardware capabilities. In previous versions of SQL Server, troubleshooting these issues requires you to
gather data from the plan cache and parse it by using XML Query (xQuery), which can take
considerable effort. Even then, you might not have all the information you need, unless you are
actively running traces to baseline the user’s environment.
The new Query Store feature in SQL Server 2016 simplifies identification of performance outliers,
manages execution plan regression, and allows for easier upgrades between versions of SQL Server. It
has two main goals—to simplify identification of performance issues and to simplify performance
troubleshooting for queries caused by changes in execution plans. The query store also acts as a flight
data recorder for the database, capturing query run-time statistics and providing a dashboard to sort
queries by resource consumption. This vast collection of data serves not only as a resource for the
automated functions of the query store, but also as a troubleshooting resource for the DBA.
This feature is one of the biggest enhancements to the SQL Server database engine since the
introduction of dynamic management views (DMVs) into the database engine in SQL Server 2005. The
query store gives unprecedented insight into the operations of a database. Whether you want to find
the workloads in an instance, perform an in-depth analysis across executions of the same code, or fix
a pesky parameter-sniffing problem, the query store offers a vast metastore of data, allowing you to
quickly find performance issues.
Enabling Query Store
Query Store manages its metadata in the local database, but it is disabled by default. To enable it in
SQL Server Management Studio (SSMS), open Object Explorer, connect to the database engine,
navigate to the database for which you want to enable Query Store, right-click the database, select
Properties, and then click Query Store in the Database Properties dialog box. You can change the
Operation Mode (Requested) value from Off to Read Only or Read Write. By selecting Read Write, as
shown in Figure 4-3, you enable Query Store to record the run-time information necessary to make
better decisions about queries.
CH A P TER 4 | Improved database engine
Figure 4-3: Enabling Query Store.
You can also use the T-SQL ALTER DATABASE command to enable Query Store, as shown in Example
Example 4-1: Enabling Query Store
ALTER DATABASE AdventureWorks2014
Understanding Query Store components
The query store contains two stores: a plan store that persists the execution plans, and a run-time
stats store that persists the statistics surrounding query execution, such as CPU, I/O, memory, and
other metrics. SQL Server retains this data until the space allocated to Query Store is full. To reduce
the impact on performance, SQL Server writes information to each of these stores asynchronously.
Note The default space allocation for Query Store is 100 MB.
You can use the following five catalog views, as shown in Figure 4-4, to return metadata and query
execution history from the query store:
execution statistics are collected.
Run-time execution statistics for queries.
Start and end times for the intervals over which run-time
Execution plan information for queries.
CH A P TER 4 | Improved database engine
Query information and its overall aggregated run-time execution statistics.
Query text as entered by the user, including white space, hints, and
Figure 4-4: Query Store catalog views.
Reviewing information in the query store
The change in query execution plans over time can be a troubleshooting challenge unless you
periodically mine the procedure cache to capture query plans. However, plans might be evicted from
the cache as a server comes under memory pressure. If you use real-time querying, you have access
only to the most recently cached plan. By using Query Store, as long as it is properly configured, you
always have access to the information you need. One way to review this information is by using the
dashboard views available in SSMS when you expand the Query Store folder for the database node, as
shown in Figure 4-5. By taking advantage of this data, you can quickly isolate problems and be more
productive in your tuning efforts.
CH A P TER 4 | Improved database engine
Figure 4-5: Query Store dashboards available in SSMS.
After enabling Query Store for a database, you have access to the following four dashboards:
Regressed Queries Use this dashboard to review queries that might have regressed because of
execution plan changes. The dashboard allows you to view the queries and their plans as well as
to select queries based on statistics (total, average, minimum, maximum, and standard deviation)
by query metric (duration, CPU time, memory consumption, logical reads, logical writes, and
physical reads) for the top 25 regressed queries over the last hour.
Overall Resource Consumption Use this dashboard to visualize overall resource consumption
during the last month in four charts: duration, execution count, CPU time, and logical reads. You
have the option to toggle between a chart view and a grid view of the query store data.
Top Resource Consuming Queries Use this dashboard to review queries in the set of top 25
resource consumers during the last hour. You can filter the queries by using the same criteria
available in the Regressed Queries dashboard.
Tracked Queries
Use this dashboard to monitor a specify query.
All the dashboards except Overall Resource Consumption allow you to view the execution plan for a
query. In addition, you have the option to force an execution plan at the click of a button in the
dashboard, which is one of the most powerful features of the query store. However, the plan must still
exist in the query plan cache to use this feature.
You can customize Query Store dashboards to show more data or to use a different time interval. To
do this, double-click a dashboard to open it, and then click the Configure button at the top of the
dashboard to display and edit the configuration dialog box, as shown in Figure 4-6.
CH A P TER 4 | Improved database engine
Figure 4-6: Configuring a Query Store dashboard.
Alternatively, you can query a DMV directly, which is a powerful approach for quickly isolating poorly
performing queries. Example 4-2 shows a T-SQL statement to return the poorest performing queries
over the last hour.
Example 4-2: Finding the poorest performing queries over the last hour
SELECT TOP 10 rs.avg_duration, qt.query_sql_text, q.query_id,
qt.query_text_id, p.plan_id, GETUTCDATE() AS CurrentUTCTime,
FROM sys.query_store_query_text AS qt
JOIN sys.query_store_query AS q
ON qt.query_text_id = q.query_text_id
JOIN sys.query_store_plan AS p
ON q.query_id = p.query_id
JOIN sys.query_store_runtime_stats AS rs
ON p.plan_id = rs.plan_id
WHERE rs.last_execution_time > DATEADD(hour, -1, GETUTCDATE())
ORDER BY rs.avg_duration DESC;
Using Force Plan
The generation of an execution plan is CPU intensive. To reduce the workload on the database engine,
SQL Server generates a plan once and stores it in a cache. Generally, caching the plan is good for
database performance, but it can also lead to a situation known as parameter sniffing. The query
optimizer uses parameter sniffing to minimize the number of recompiled queries. This situation occurs
when a stored procedure is initially run with a given parameter against a table having a skewed
number of values. You can use the query store’s Force Plan option to address this problem.
To better understand parameter sniffing, consider an example in which you create a stored procedure
like the one shown in Example 4-3.
CH A P TER 4 | Improved database engine
Example 4-3: Understanding parameter sniffing
SET value=2
Now let’s assume that you have a table such as the one shown here:
In this simple example of skewed values in a table, seven values have an ID of 1, and one value has an
ID of 2. If you first run this procedure with a parameter value of 2, the execution plan generated by the
database optimizer is likely to be less than optimal. Then, when you later execute the procedure with a
parameter value of 1, SQL Server reuses the suboptimal plan.
Because skewed data might force your procedures into plans that are less than optimal for many
queries, you have the opportunity to force the plan that is best optimized for all executions of a given
stored procedure. While this approach might not offer the best performance for all values of a
procedure’s parameter, forcing a plan can give you more consistent overall performance and better
performance on average. SQL Server honors plan forcing during recompilation for in-memory, natively
compiled procedures, but the same is not true for disk-based modules.
You can also unforce a plan by using either the Query Store interface in SSMS or the
sp_query_store_unforce_plan stored procedure. You might unforce a plan after your data changes
significantly or when the underlying code changes enough to render the existing plan invalid.
Managing the query store
The query store is extremely helpful, but it does require some management. As we explained earlier in
this chapter, the query store is not enabled by default. You must enable it on each user database
individually. In addition, a best practice is to enable it on the model database.
Note At the time of this writing, Query Store is not currently included in the Database Properties
dialog box in SSMS for the model database. To add it, you must enable Query Store by using the
following code:
After enabling the query store, you might need to change the space allocated to the query store from
the default of 100 MB per database. If you have a busy database, this allocation might not be large
enough to manage execution plans and their related metadata. When this space fills up, the query
store reverts to a read-only mode and no longer provides up-to-date execution statistics.
CH A P TER 4 | Improved database engine
The size of your query store is also directly related to the statistics collection interval. The default for
this value is 60 minutes, but you can adjust it to a higher frequency if you need more finely grained
data. However, capturing data at a higher frequency requires more space for the query store.
Another setting to consider is size-based cleanup mode. By default, the query store converts to readonly mode when full. When you enable size-based cleanup, SQL Server flushes older queries and
plans as new data comes in, thereby continually providing the latest data. Another option for space
conservation is adjusting the capture mode of the query store from ALL to AUTO, which eliminates the
capture of queries having insignificant compile and execution detail.
Tuning with the query store
After enabling the query store and collecting data over a baseline period, you now have a wealth of
data and options to start troubleshooting performance issues. The query store allows you to spend
more time troubleshooting problem queries and improving them, rather than on trying to find the
proverbial needle in a haystack. A simple approach is to start troubleshooting queries on the basis of
highest resource consumption. For example, you can look at queries consuming the most CPU and
logical I/Os. After identifying poorly performing queries, you can then consider the following options:
If multiple plans are associated with a query, identify the best-performing plan and use the
Force Plan option to request it for future executions.
If you observe a large gap between the estimated rows and the actual rows in a query,
updating statistics might help performance.
If query logic is problematic overall, work with your development team to optimize the query
Stretch Database
One of the more common refrains in IT infrastructure organizations in recent years has been the high
costs of storage. A combination of regulatory and business requirements for long-term data retention,
as well as the presence of more data sources, means enterprises are managing ever-increasing
volumes of data. While the price of storage has dropped, as anyone who owns enterprise storage
knows, the total cost of ownership (TCO) for enterprise storage commonly used for databases is still
very high. Redundant arrays of independent disks (RAID), support contracts, management software,
geographical redundancy, and storage administrators all add to the high total cost of enterprise
Another factor in the cost of storage is the lack of support for online data archiving in many thirdparty applications. To address this problem, a common approach is to use file groups and partitioning
to move older data to slower disks. Although this approach can be effective, it also comes with high
managerial overhead because it involves storage administrators in provisioning the storage and
requires active management of partitions.
Perhaps more important than the TCO of enterprise storage is the impact of large databases and
tables on overall administration and availability of the systems. As tables grow to millions and even
billions of rows, index maintenance and performance tuning become significantly more complex.
These large databases also affect availability service-level agreements as restore times can often
exceed service-level agreements required by the business.
SQL Server 2016 introduces a new hybrid feature called Stretch Database that combines the power of
Azure SQL Database with an on-premises SQL Server instance to provide nearly bottomless storage at
a significantly lower cost, plus enterprise-class security and near-zero management overhead. With
Stretch Database, you can store cold, infrequently accessed data in Azure, usually with no changes to
CH A P TER 4 | Improved database engine
application code. All administration and security policies are still managed from the same local SQL
Server database as before.
Understanding Stretch Database architecture
Enabling Stretch Database for a SQL Server 2016 table creates a new Stretch Database in Azure, an
external data source in SQL Server, and a remote endpoint for the database, as shown in Figure 4-7.
User logins query the stretch table in the local SQL Server database, and Stretch Database rewrites the
query to run local and remote queries according to the locality of the data. Because only system
processes can access the external data source and the remote endpoint, user queries cannot be issued
directly against the remote database.
Figure 4-7: Stretch Database architecture.
Security and Stretch Database
One of the biggest concerns about cloud computing is the security of data leaving an organization’s
data center. In addition to the world-class physical security provided at Azure data centers, Stretch
Database includes several additional security measures. If required, you have the option to enable
Transparent Data Encryption to provide encryption at rest. All traffic into and out of the remote
database is encrypted and certificate validation is mandatory. This ensures that data never leaves SQL
Server in plain text and the target in Azure is always verified.
The external resource that references the Azure SQL Stretch Database can only be used by system
processes and is not accessible by users. (See Figure 4-8.) Furthermore, it has no impact on the
underlying security model of a stretch table.
CH A P TER 4 | Improved database engine
Figure 4-8: External resource for Stretch Database.
The security model in your on-premises database has a couple of components. The first requirement
is to enable “remote data archive” for the instance. You will need to have either sysadmin or
serveradmin permission. Once you have enabled this feature, you can configure databases for stretch,
move data to your stretch database, and query data in your stretch database. To enable Stretch
Database at the individual database level, you must have the CONTROL DATABASE permission. You
will also need ALTER privileges on the tables you are looking to stretch.
As you would for a SQL Database that you provision manually, you must also create a firewall rule for
the remote SQL Stretch Database database. That way, only safe IP addresses can connect to it. The
creation of this firewall rule is part of the automation in the Stretch Database wizard if you enable
your SQL Server database for stretch via SQL Server Management Studio.
Identifying tables for Stretch Database
Not all tables are ideal candidates for Stretch Database. In the current release, you cannot enable
stretch for a table if it has any of the following characteristics:
More than 1,023 columns
Memory-optimized tables
Common language runtime (CLR) data types
Check constraints
Default constraints
Computed columns
After eliminating tables with these characteristics from consideration, you have two options for
identifying which of the remaining eligible tables in your environment are good candidates for
stretching. First, you can use T-SQL to find large tables and then work with your application teams to
determine the typical rate of change. A table with a high number of rows that are infrequently read is
a good candidate. The other, more automated option is to use the Stretch Database Advisor, which is
part of the SQL Server 2016 Upgrade Advisor. This advisor checks the current limitations for Stretch
Database and then shows the best candidates for stretching based on benefits and costs, as shown in
Figure 4-9.
CH A P TER 4 | Improved database engine
Figure 4-9: Analyzing candidates for Stretch Database in SQL Server 2016 Upgrade Advisor.
The best applications for Stretch Database are systems for which you are required to keep cold data
for extended periods. By working with your application teams to understand which of your systems fit
these scenarios, you can implement Stretch Database strategically to meet business requirements
while reducing overall storage TCO and meeting business SLAs.
Configuring Stretch Database
Before you can configure Stretch Database in SQL Server, you must have an Azure account in place
and change the REMOTE DATA ARCHIVE configuration option at the SQL Server instance level. To
make this change, execute the command shown in Example 4-4.
Example 4-4: Changing the REMOTE DATA ARCHIVE configuration option
EXEC sp_configure 'remote data archive', '1';
You can then configure stretch, using the wizard that you launch by right-clicking the database in
Object Explorer, pointing to Stretch, and clicking Enable. The wizard prompts you to supply a
password for a database master key and select the table to stretch and then validates whether the
table is eligible for stretch. Next, you sign in with your Azure credentials, select a subscription, and
then select an Azure region. For performance reasons, choose the Azure region closest to your onpremises location.
Next, you have the option to create a new server or use an existing server. There is no impact on your
existing SQL Databases if you choose to use an existing server. Your next step is to provide
administrator credentials for the new SQL Database and to create a firewall rule allowing your onpremises databases to connect to SQL Database. When you click Finish on the last page of the wizard,
the wizard provisions Stretch Database and begins migrating data to the new SQL Database.
Note As an alternative to using the wizard, you can perform the steps necessary to configure a
database and a table for stretch by using T-SQL commands. For more information, see “Enable
Stretch Database for a database” at
CH A P TER 4 | Improved database engine
Monitoring Stretch Database
SQL Server 2016 includes a dashboard in SSMS to monitor Stretch Database. To view it, right-click the
database name in Object Explorer, select Stretch Database, and then select Monitor to display the
dashboard shown in Figure 4-10.
Figure 4-10: Monitoring Stretch Database in SSMS.
In this dashboard, you can see which tables are configured for stretch in addition to the number of
rows eligible for stretch and the number of local rows. In Figure 4-10, all rows are stretched. You can
also change the migration state of a table. The default state is Outbound, which means data is moving
into Azure. However, you can pause the migration of the data.
Enabling Stretch Database also creates an Extended Events session called StretchDatabase_Health.
You can view the extended events associated with this session by clicking the View Stretch Database
Health Events link above the Stretch Configured Tables section of the dashboard. Additionally, you
can explore two DMVs associated with Stretch Database: sys.dm_db_rda_migration_status and
Note Most common problems you encounter with Stretch Database are likely to be network or
firewall related. As your first troubleshooting step, work with a network administrator to ensure that
you can reach your SQL Database over port 1433, which is a commonly blocked outbound port on
many networks.
CH A P TER 4 | Improved database engine
Another monitoring tool at your disposal is the new Remote Query operator in the execution plan for
a stretch table, as shown in Figure 4-11. SQL Server 2016 also includes the Concatenation operator to
merge the results of the on-premises data with the remote query results.
Figure 4-11: Reviewing the execution plan for a stretch table.
An important design pattern with Stretch Database is to ensure that your queries do not regularly
retrieve unnecessary rows. Running poorly written queries against a stretch table can apply adverse
performance. When troubleshooting performance issues on stretched tables, start your tuning effort
as you would on a regular on-premises database. After eliminating issues related to your on-premises
instance, examine the Azure portal to understand how the workload affects the stretch database.
If your remote query performance is still not sufficient, you have several options for tuning. First,
ensure that your remote database is in the Azure data center nearest your on-premises data center to
reduce latency. Next, monitor the Azure portal to observe the performance characteristics of the
underlying Azure database. You might need to increase the service tier of the SQL Stretch Database.
Last, work with your network administrator to guarantee quality of service between your site and your
remote database.
Backup and recovery with Stretch Database
Backup and recovery of a stretch-enabled database does not include the SQL Stretch Database
containing your remote tables. Nonetheless, your data remains protected because SQL Stretch
Database leverages the built-in backup features of SQL Database. Accordingly, SQL Database is
constantly making full and transaction log backups. The retention period for these backups is
determined by the service tier of the database. However, when you back up your on-premises
database, you are taking a shallow backup. In other words, your backup contains only the data that
remains on-premises and does not include the migrated data.
To restore a database, follow these steps:
Restore your on-premises SQL Server database.
Create a master key for the stretch-enabled database.
Create a database-scoped credential for your SQL Database.
4. Run the restore procedure.
CH A P TER 4 | Improved database engine
More analytics
Better and faster analytics capabilities have been built into SQL Server
2016. Enhancements to tabular models provide greater flexibility for the
design of models, and an array of new tools helps you develop solutions
more quickly and easily. As an option in SQL Server 2016, you can now use
SQL Server R Services to build secure, advanced-analytics solutions at
enterprise scale. By using R Services, you can explore data and build
predictive models by using R functions in-database. You can then deploy
these models for production use in applications and reporting tools.
Tabular enhancements
In general, tabular models are relatively easy to develop in SQL Server Analysis Services. You can build
such a solution directly from a wide array of sources in their native state without having to create a set
of tables as a star schema in a relational database. You can then see the results of your modeling
within the design environment. However, there are some inherent limitations in the scalability and
complexity of the solutions you can build. In the latest release of SQL Server, some of these limitations
have been removed to better support enterprise requirements. In addition, enhancements to the
modeling process make controlling the behavior and content of your model easier. In this section, we
review the following enhancements that help you build better analytics solutions in SQL Server 2016:
More data sources accessible in DirectQuery mode
Choice of using all, some, or no data during modeling in DirectQuery mode
Calculated tables
Bidirectional cross-filtering
Formula bar enhancements
New Data Analysis Expressions (DAX) functions
Using DAX variables
CH A P TER 6 |
More analytics
Accessing more data sources with DirectQuery
One of the benefits of using tabular models in Analysis Services is the ability to use data from a variety
of data sources, both relational and nonrelational. Although prior versions of SQL Server support a
quite extensive list of data sources, not all of those sources are available to use with DirectQuery, the
feature in tabular models that retrieves data from the data source when a query is run instead of
importing data into memory in advance of querying. Having live access to more data sources means
that users can get answers to questions more quickly, and you have less administrative overhead to
maintain in your analytic infrastructure.
In previous versions of SQL Server, you are limited to using SQL Server 2005 or later for a model in
DirectQuery mode. In SQL Server 2016, the list of data sources supported for DirectQuery now
includes the following:
SQL Server 2008 or later
Azure SQL Database
Analytics Platform System (formerly Parallel Data Warehouse)
Oracle 9i, 10g, 11g, and 12g
Teradata V2R6, V2
When should you use DirectQuery?
Tabular models can compress and cache large volumes of data in memory for high-performance
queries. DirectQuery might be a better option in some cases, but only if you are using a single data
source. In general, you should use DirectQuery if any of the following situations apply: your users
require real-time access to data, the volume of data is larger than the memory available to Analysis
Services, or you prefer to rely on row-level security in the database engine.
Using DirectQuery can potentially have an adverse impact on query performance. If your source is
SQL Server 2012 or later, you should consider implementing columnstore indexes so that
DirectQuery can take advantage of query optimization provided by the database engine.
Even if you create a tabular model in in-memory mode, you can always switch to DirectQuery mode
at any time. If you do this, any data previously stored in the cache is flushed, but the metadata is
There are some drawbacks to using DirectQuery mode that you should consider before choosing it
for your model. First, you cannot create calculated columns or calculated tables in the model, nor
can you add a pasted table. An alternative is to use corresponding logic to create a derived column
or a view in the underlying source. Second, because Analysis Services translates the DAX formulas
and measures of your model into SQL statements, you might encounter errors or inconsistent
behavior for some DAX functions that do not have a counterpart in SQL, such as time-intelligence
functions or some statistical functions. In that case, you might be able to create a derived column in
the source. You can see a list of functions that are not supported in DirectQuery mode at
To learn more about DirectQuery mode in general, see
Modeling with a DirectQuery source
During the tabular modeling process, you import data from data sources into the design environment,
unless your model is configured in DirectQuery mode. A new element in this process in SQL Server
2016 is that you can specify whether to create a model by using all the data (which is the only option
in earlier versions of SQL Server), no data (which is the new default), or a subset of data based on a
query you supply.
CH A P TER 6 | More analytics
To use the later two options, double-click the Model.bim file in Solution Explorer (if it is not already
open in SQL Server Data Tools), and then click the file again to display its properties in the Properties
window. If necessary, select SQL Server 2016 RTM (1200) in the Compatibility Level drop-down list.
Then select On in the DirectQuery Mode drop-down list.
Important If you upgrade the model from in-memory to DirectQuery mode, you cannot revert to a
lower compatibility level.
To connect your model to the data source, you still use the Table Import Wizard, which you launch by
selecting Import From Data Source from the Model menu. You select a relational database and then
the tables for your model, but the data is no longer imported for DirectQuery-mode models. Instead,
you work with an empty model that contains only metadata, such as column names and relationships,
as shown in Figure 6-1. You can continue by configuring properties for columns, defining
relationships, or working with the model in diagram view, just as you normally would if you import
data instead. In DirectQuery mode, the data stays in the source until you generate a query in Excel or
another presentation-layer application.
Figure 6-1: Working with a model in DirectQuery mode.
When you work without data in the model designer, the modeling process is likely to be faster
because you no longer have to wait for calculations to be performed against the entire data set as you
add or change measures. However, you might prefer to view sample data to help you better review
the model during its design. To do this, select a table, and then select Partitions on the Table menu.
Select the existing partition, which has the prefix DirectQuery, click the Copy button, and then select
the copy of the partition (which has the prefix Sample). Click the Edit SQL Query button to add a
WHERE clause to the partition query, as shown in Figure 6-2, which returns a subset of the original
CH A P TER 6 | More analytics
partition query. When you close the Partition Manager and process the partition, the subset of rows
defined by your query is displayed in the model designer.
Figure 6-2: Configuring sample data for a partition in DirectQuery mode.
Note If you create multiple sample partitions, the model designer displays the combined results
from all the queries.
Notice the Set As DirectQuery button below the list of partitions in Figure 6-2. This button appears
only when you select a sample partition in the list. When you click this button, you reset the partition
that Analysis Services uses to retrieve data for user queries. For this reason, there can be only one
DirectQuery partition defined for a table at any time. If you select the DirectQuery button, the button’s
caption displays Set As Sample instead.
You can also use a sample data partition with the Analyze Table In Excel feature to test the model
from the presentation layer. It’s helpful to test the results of calculations in measures and to check
relationships between tables even when you are using only a subset of data. To do this, select Analyze
In Excel from the Model menu. When the model is in DirectQuery mode, the Analyze In Excel dialog
box requires you to choose one of two options, Sample Data View or Full Data View, as shown in
Figure 6-3.
CH A P TER 6 | More analytics
Figure 6-3: Using the sample data view in DirectQuery mode to analyze data in Excel.
If you create a sample partition for one table, you should create sample partitions for all tables.
Otherwise, when you use the Analyze Table In Excel option, you see null values for columns in tables
without a sample partition. For example, in Figure 6-4, CalendarYear is placed in rows but displays null
values, and Total Sales is a measure added to the Internet Sales table for which sample data is
Figure 6-4: Viewing sample data for multiple tables in Excel in which a sample partition is defined for Internet Sales
Your sample partition can be a copy of the partition’s data, if you prefer that. Simply make a copy of
the DirectQuery partition and omit the addition of a WHERE clause. This approach is useful if your
table is relatively small, and it also allows you to better confirm that the sample partitions defined for
other tables are correct, as shown in Figure 6-5.
Figure 6-5: Viewing sample data for multiple tables in Excel after adding a sample partition for the Date table.
Working with calculated tables
A new feature for tabular models in SQL Server 2016 is the calculated table—as long as you are
working with a model that is not in DirectQuery mode. A calculated table is built by using a DAX
expression. A model that includes calculated tables might require more memory and more processing
time than a model without calculated tables, but it can be useful in situations for which you do not
have an existing data warehouse with cleansed and transformed data. This technique is useful in the
following situations:
CH A P TER 6 | More analytics
Creating small data sets to satisfy simple requirements without adding a lot of overhead to your
technical infrastructure.
Prototyping a solution before building a complete solution.
Creating a simple date table by using the new CALENDAR() or CALENDARAUTO() functions.
Separating a role-playing dimension into multiple tables for simpler modeling.
When you create a calculated table, you can choose to use only a few columns from a source table,
combine columns from multiple tables, or apply complex expressions to filter and transform existing
data into a new table. As a simple example, the FactInternetSales table in the AdventureWorksDW
sample database contains the following three columns: OrderDateKey, ShipDateKey, and
OrderDateKey. These three columns have a foreign-key relationship to a single role-playing
dimension, DimDate. Instead of activating a relationship to change the context of a query from Order
Date to Ship Date or Due Date, as required in earlier versions, you can now create one calculated table
for ShipDate and another for DueDate.
To create a calculated table, you must set your model to compatibility level 1200. Select New
Calculated Table from the Table menu or click the Create A New Table Calculated From A DAX
Formula tab at the bottom of the model designer, as shown in Figure 6-6. The designer displays a new
empty table in the model. In the formula bar above the table, enter a DAX expression or query that
returns a table.
Figure 6-6: Adding a new calculated table.
To continue with our example, you can use a simple expression such as =Date to copy an existing
role-playing dimension table, Date (renamed from DimDate). The model evaluates the expression and
displays the results. You can then rename the calculated table, add relationships, add calculated
columns, and perform any other activity that you normally would with a regular table. The tab at the
bottom of the model designer includes an icon that identifies the table as a calculated table, as shown
in Figure 6-7.
CH A P TER 6 | More analytics
Figure 6-7: Viewing a calculated table in the model designer.
Note Some interesting uses for calculated tables are described at and Although these articles describe uses for calculated tables in Power BI, you can now use
the same approach in tabular models.
Bidirectional cross-filtering
Another feature new to tabular models is bidirectional cross-filtering. This rather complex-sounding
name allows cross-filtering across a table relationship in two directions rather than one direction,
which has always been a feature in tabular models. This means that you no longer need to create
complex DAX expressions to produce specific results, as long as you set your model to compatibility
level 1200.
To better understand the implications of this new feature, we’ll start by reviewing one-directional
cross-filtering. A one-directional cross-filter applies in a one-to-many relationship such as that
between a dimension table and a fact table. Typically, the fact table contains a column with a
relationship to a corresponding column in the dimension table. In a tabular model based on
AdventureWorksDW, FactInternetSales (which you can rename as Internet Sales) has the ProductKey
column, and you can use this column to define a relationship in a single direction from the
DimProduct dimension table (renamed as Product) to the fact table. If you import these tables into
the model at the same time, the foreign-key relationship defined in the tables is automatically
detected and configured for you. Otherwise, you can manually create the relationship.
CH A P TER 6 | More analytics
When you use a PivotTable to query the model, you see how the dimension labels in the rows become
the filter context to the value from the fact table, which appears on the same row, as shown in Figure
6-8. In this example, each row is a product line from the Product table. To derive the aggregate value
in the Sales Count column in each row, the Analysis Services engine filters the table for the current
product line value and computes the count aggregate for sales having that value. Because of the onedirectional filter in the relationship, each entry for Sales Count is not only an aggregate value but also
a filtered value based on the current row’s dimension value. If no relationship existed between the two
tables, the Sales Count value would display the total aggregated value of 60,398 in each row because
no filter would be applicable.
Figure 6-8: Viewing the effect of a one-directional filter between Product and Internet Sales.
Although you can create measures in any table in a tabular model, the behavior you see in a
PivotTable might not produce the results you want if you add a measure to a dimension table. Let’s
say that you add a distinct count of products to the Product table and then add Calendar Year to your
query. In this case, a one-directional relationship exists between Date and Internet Sales, which can be
combined with the one-directional relationship between Product and Internet Sales to compute Sales
Count by year and by product line, as shown in Figure 6-9. However, because the relationship
between Product and Internet Sales is one-directional from Product to Internet Sales, terminating at
Internet Sales, there is no relationship chain that goes from Date to Internet Sales to Product that
provides the filter context necessary to compute the distinct count measure. Consequently, the
distinct count by product line, which is in the same table and thereby provides filter context, repeats
across all years for each product line.
Figure 6-9: Viewing the effect of a one-directional filter between Product and Internet Sales and Date and Internet
Sales on measures in the Product table.
You can override this behavior by changing the relationship between Product and Internet Sales to a
bidirectional relationship. To do this, select Manage Relationships from the Table menu, double-click
CH A P TER 6 | More analytics
the relationship between these two tables, and select To Both Tables in the Filter Direction drop-down
list, as shown in Figure 6-10.
Figure 6-10: Setting a bidirectional cross-filter in a table relationship.
With this change, the filter context of the current year applies to both the Internet Sales table (as it did
previously) to correctly aggregate Sales Count, and to the Product table to correctly aggregate
Distinct Product Count, as shown in Figure 6-11.
Figure 6-11: Viewing the effect of a bidirectional filter between Product and Internet Sales and a one-directional filter
between Date and Internet Sales on measures in the Product table.
Another problem that bidirectional cross-filtering can solve is the modeling of a many-to-many
relationship, which in earlier versions of SQL Server required you to create complex DAX expressions.
An example of a many-to-many relationship in the AdventureWorksDW database is the one in which
the Internet Sales table stores individual sales orders by line number. The DimSalesReason table stores
the reasons a customer indicated for making the purchase, such as price or quality. Because a
customer could choose zero, one, or more reasons for any item sold, a factless fact table called
FactInternetSalesReason is in the database to relate the sales reasons by sales order and line number.
You can add this table to a tabular model and easily aggregate values in the Internet Sales table by
sales reason after making a few adjustments to the model.
CH A P TER 6 | More analytics
Because the structure of the two fact tables in this example does not allow you to define a
relationship between them, you must add a calculated column called Sales Key (or another unique
name that you prefer) to each of them to uniquely identify the combination of sales order and line
number. To do this, you use a DAX expression similar to this: =[SalesOrderNumber]&""&[SalesOrderLineNumber] in each fact table. You can then create a relationship between the two
tables using this common column and set the direction to To Both Tables, as shown in Figure 6-12.
Figure 6-12: Defining a many-to-many relationship.
Then, when you create a PivotTable to review sales counts by sales reason, the many-to-many
relationship is correctly evaluated, as shown in Figure 6-13, even though there is no direct relationship
between the Sales Reason table and Internet Sales. The grand total continues to correctly reflect the
count of sales, which is less than the sum of the individual rows in the PivotTable. This is expected
behavior for a many-to-many relationship because of the inclusion of the same sale with multiple
sales reasons.
Figure 6-13: Viewing the effect of a many-to-many relationship in a PivotTable.
Important Although you might be tempted to configure bidirectional filtering on all relationships
to address all possible situations, it is possible for this configuration to overfilter results
unexpectedly. Therefore, you should test the behavior of each filter direction change to ensure that
you get the results you want.
CH A P TER 6 | More analytics
By default, a relationship between two tables is one-directional unless you change this behavior at the
model or environment level. At the model level, open the model properties, and then choose Both
Directions in the Default Filter Direction drop-down list. This change applies only to your current
model; any new models you create will default to single direction. If you prefer to change the default
for all new models, select Options from the Tools menu, expand Analysis Services Tabular Designers in
the navigation pane on the left, select New Project Settings, and then select Both Directions in the
Default Filter Direction drop-down list, as shown in Figure 6-14.
Figure 6-14: Setting the default filter direction for new projects.
Writing formulas
The user interface for the formula bar in the model designer has been improved by the addition of the
following changes, which help you write and review DAX formulas more easily:
Syntax coloring Functions are now displayed in a blue font, variables in a cyan font, and string
constants in a red font to distinguish these expression elements more easily from fields and other
IntelliSense Errors are now identified by a wavy red underscore, and typing a few characters
displays a function, table, or column name that begins with matching characters.
Formatting You can persist tabs and multiple lines by pressing Alt+Enter in your expression to
improve legibility. You can also include a comment line by typing // as a prefix to your comment.
Formula fixup In a model set to compatibility level 1200, the model designer automatically
updates measures that reference a renamed column or table.
Incomplete formula preservation In a model set to compatibility level 1200, you can enter an
incomplete formula, save and close the model, and then return to your work at a later time.
Introducing new DAX functions
The current release of SQL Server 2016 includes many new functions, which are described in the
following table:
CH A P TER 6 | More analytics
Function type
Date and time
Returns a table with one column named Date,
containing a date range based on start and end
dates that you provide as arguments.
Returns a table with one column named Date,
containing a date range based on minimum and
maximum dates present in the model’s tables.
Returns the count of intervals (such as DAY or
WEEK) between the start and end dates provided
as arguments.
Includes rows with missing values in the result set.
Returns a table containing the results of a left
semi-join between two tables, replacing common
columns in these tables with a single index
Returns a Boolean value for each row to indicate
whether two values (such as a column value and a
constant value) are the same.
Returns the arccosine, or inverse cosine, of a
Returns the inverse hyperbolic cosine of a
Returns the arcsine, or inverse sine, of a number.
Returns the inverse hyperbolic sine of a number.
Returns the arctangent, or inverse tangent, of a
Returns the inverse hyperbolic tangent of a
Returns the total number of possible groups of a
number for a specified number of items.
Returns the number of combinations with
repetitions of a number for a specified number of
Returns the cosine of a number.
Returns the hyperbolic cosine of a number.
Converts radians into degrees.
Rounds a number up to the nearest even integer.
Returns a decimal number for e raised to the
power of a given number.
Returns the greatest common divisor between
two specified numbers as an integer without a
Rounds a number up to the nearest integer or to
the nearest multiple of significance.
Returns the least common multiple between two
specified integers.
Math and Trig
CH A P TER 6 | More analytics
Function type
Returns a number rounded to the nearest
multiple of a specified number.
Rounds a number up to the nearest odd integer.
Returns the value of pi with 15-digit accuracy.
Returns the product of values in a column.
Returns the product of an expression evaluated
for each row in a table.
Performs division on the numerator and
denominator provided as arguments and returns
the integer portion of the result.
Converts degrees to radians.
Returns the sine of a number.
Returns the hyperbolic sine of a number.
Returns the square root of pi multiplied by a
specified number.
Returns the tangent of a number.
Returns the hyperbolic tangent of a number.
Returns the internal rate of return.
Returns the net present value.
Returns the beta distribution of a sample.
Returns the inverse of the beta cumulative
probability density function.
Returns the inverse of the left-tailed probability of
the chi-squared distribution.
Returns the inverse of the right-tailed probability
of the chi-squared distribution.
Returns the confidence interval as a range of
Returns the confidence interval for a population
mean using a student’s t distribution.
Returns the exponential distribution.
Returns the geometric mean of numbers in a
Returns the geometric mean of an expression
evaluated for each row in a column.
Returns the median of numbers in a column.
Returns the median of an expression evaluated
for each row in a column.
Returns the kth percentile of values in a range,
where k is in the range 0 to 1, exclusive.
Returns the kth percentile of values in a range,
where k is in the range 0 to 1, inclusive.
Returns the kth percentile of an expression
evaluated for each row in a column, where k is in
the range 0 to 1, exclusive.
CH A P TER 6 | More analytics
Function type
Returns the kth percentile of an expression
evaluated for each row in a column, where k is in
the range 0 to 1, inclusive.
Concatenates the results of an expression
evaluated for each row in a table.
Returns a table with selected columns with the
results of evaluating an expression for each seat
of GroupBy values.
Returns a table containing values in one table
that are also in a second table.
Returns a Boolean value indicating whether a
table is empty.
Returns a table after performing an inner join of
two tables.
Returns a table after performing a left outer join
of two tables.
Returns a table containing combinations of values
from two tables for which the combination is
Returns a table containing all rows from two
Using variables in DAX
You can now include variables in a DAX expression to break up a complex expression into a series of
easier-to-read expressions. Another benefit of using variables is the reusability of logic within the
same expression, which might possibly improve query performance.
Here’s an example that focuses on sales of products in categories other than bikes and finds the ratio
of the sales of these products with a unit price less than $50 to all sales of these products. First, to
create this measure without variables and without using intermediate measures, you would use the
expression shown in Example 6-1.
Example 6-1: Creating a complex DAX expression without variables
Non Bikes Sales Under $50 % of Total:=
filter(values(Category[CategoryName]), Category[CategoryName]<> "Bikes"),
calculate(sum([SalesAmount]),'Internet Sales'[UnitPrice]<50)
filter(values(Category[CategoryName]), Category[CategoryName]<> "Bikes"),
To reproduce the same results by using an expression with variables, you can use the expression
shown in Example 6-2. You can use as many variables as you like in the expression. Use the VAR
keyword to introduce each variable, then use the RETURN keyword for the final expression to resolve
CH A P TER 6 | More analytics
the expression and return to the Analysis Services engine. Notice that a variable can be a scalar
variable or a table.
Example 6-2: Creating a complex DAX expression with variables
Non Bikes Sales Under $50 % of Total:=
// create a table for all categories except Bikes
tNonBikes = filter(values(Category[CategoryName]), Category[CategoryName]<> "Bikes")
// get the total of sales for tNonBikes table where UnitPrice is less than 50
NonBikeSalesUnder50 = sumx(tNonBikes,
calculate(sum([SalesAmount]),'Internet Sales'[UnitPrice]<50))
// get the total of all sales for tNonBikes table
NonBikeAllSales = sumx(tNonBikes,
// divide the first total by the second total
NonBikeSalesUnder50 / NonBikeAllSales
As an alternative, you could create intermediate measures for NonBikeSalesUnder50 and
NonBikeAllSales and then divide the former by the latter to obtain the final result. That approach
would be preferable if you were to require the results of the intermediate measures in other
expressions because variables are limited in scope to a single expression. If these results are not
required elsewhere, consolidating the logic into one expression and using variables helps you to more
easily see the flow of the expression evaluation.
R integration
R is a popular open-source programming language used by data scientists, statisticians, and data
analysts for advanced analytics, data exploration, and machine learning. Despite its popularity, the use
of R in an enterprise environment can be challenging. Many tools for R operate in a single-threaded,
memory-bound desktop environment, which puts constraints on the volume of data that you can
analyze. In addition, moving sensitive data from a server environment to the desktop removes it from
the security controls built into the database.
SQL Server R Services, the result of Microsoft’s acquisition in 2015 of Revolution Analytics, resolves
these challenges by integrating a unique R distribution into the SQL Server platform. You can execute
R code directly in a SQL Server database and reuse the code in another platform, such as Hadoop. The
workload shifts from the desktop to the server, where R Services performs multithreaded, multicore,
and parallelized multiprocessor computations at high speed while maintaining the necessary levels of
security for your data. Using R Services, you can build intelligent, predictive applications that you can
easily deploy to production.
Installing and configuring R Services
To use SQL Server R Services, you must install a collection of components to prepare a SQL Server
instance to support the R distribution. In addition, each client workstation requires an installation of
the R distribution and libraries specific to R Services. In this section, we provide the links to download
the requisite files and additional information necessary to prepare your environment to use R Services.
Server configuration
In the server environment, you start by installing the following components on the computer hosting
SQL Server 2016:
CH A P TER 6 | More analytics
Advanced Analytics Extensions
a SQL Server instance.
Microsoft R Open (MRO) 3.2.2 for Revolution R Enterprise 7.5.0 An enhanced open-source
R distribution that you download from This is a prerequisite for Revolution R Enterprise.
Revolution R Enterprise 7.5.0 (RRE-7.5.0) A collection of libraries that you download from This provides connectivity to
SQL Server and executes R packages in a high-performance, parallel architecture.
A database engine feature that you select during installation of
At the time of writing, you must perform several postinstallation tasks. Because this information is
subject to change, refer to “Post-Installation Server Configuration (SQL Server R Services)” at for the most up-to-date instructions.
Note The default memory-allocation settings might allow SQL Server to consume most of the
available physical memory without leaving adequate memory for R. Consider changing the
maximum memory for SQL Server to 80 percent of available physical memory. Refer to “Server
Memory Server Configuration Options” at
for more information about configuring server memory.
Client workstation
To prepare a client workstation to work with R Services, install the following components:
Microsoft .NET Framework 3.5 If this version of the .NET Framework is not yet installed on the
client workstation, you can manually install it by following the instructions at
MRO 3.2.2 for RRE-7.5.0 Download this component from, and install it to load the open-source R run-time libraries
onto the workstation. This is a prerequisite for RRE.
RRE-7.5.0 Download this component from
/details.aspx?id=49505, and install it to add connectivity to SQL Server and enhanced R packages
for high-performance.
An R integrated development environment (IDE)
You can use any R IDE that you prefer.
Note At the time of writing, the integration of Revolution R into SQL Server is not complete. A list
of current limitations is available at
Getting started with R Services
Although you execute your R code and run computations on SQL Server, you develop and test by
using an R IDE of your choice. In this section, we describe how to prepare your data for exploration
with R functions, how to build and use a predictive model, and how to test the accuracy of your
Note The examples in this chapter are derived from “Data Science End-to-End Walkthrough,”
available at, which includes additional
topics about working with R and provides a PowerShell script you can download and use to prepare
the data set used in the examples. This data set contains public data about New York City taxi fares,
passenger counts, pickup and drop-off locations, and whether a tip was given.
CH A P TER 6 | More analytics
You can also learn more about working with R Services in “Data Science Deep Dive: Using the
RevoScaleR Packages” at
User permissions
Before users can begin executing R on a database, you must ensure that each user has read
permissions to the data. In addition, you must add each user to the db_rrerole in SQL Server
Management Studio (SSMS) by running the code shown in Example 6-3.
Example 6-3: Adding a user to the db_rrerole
USE [master]
CREATE USER [<user name>] FOR LOGIN [<login name>] WITH
ALTER ROLE [db_rrerole] ADD MEMBER [<user name>];
Compute context
Before you can execute R on your data, you must use the RxSetComputeContext function in the R IDE
to set the compute context for functions in the RevoScaleR package in RRE to run on SQL Server.
Although you can use a single line of code to run this command, you can assign values to variables in
separate lines of code and then use the variables as arguments to the function, as shown in Example
Example 6-4: Setting compute context to SQL Server
connStr <- "Driver=SQL Server; Server=<srv>; Database=NYCTaxi_Sample; Uid=<login>;
sqlShareDir <- paste("C:\\AllShare\\",Sys.getenv("USERNAME"),sep="")
sqlWait <- TRUE
sqlConsoleOutput <- FALSE
cc <- RxInSqlServer(connectionString = connStr, shareDir = sqlShareDir,
wait = sqlWait, consoleOutput = sqlConsoleOutput)
Creating variables and assigning values is simple to do in R. As shown in Example 6-4, you define a
name for the variable and then use the assignment operator (<- ) followed by the value to assign. The
value can be a string, a Boolean value, or an array, to name only a few object types.
In Example 6-4, several variables store values for use as arguments in the RxInSqlServer function. This
function is responsible for creating the connection to a SQL Server database and sharing objects
between the server context and your local computer context. In this example, it takes the following
connectionString An ODBC connection string for SQL Server. At the time of this writing, you
must use a SQL login in the connection string.
shareDir A temporary directory in which to store R objects shared between the local compute
context and the server compute context.
wait A Boolean value to control whether the job will be blocking or nonblocking. Use TRUE for
blocking, which prevents you from running other R code until the job completes. Use FALSE for
nonblocking, which allows you to run other R code while the job continues to execute.
CH A P TER 6 | More analytics
consoleOutput A Boolean value that controls whether the output of R execution on the SQL
Server displays locally in the R console.
Another variable stores the result of the RxInSQLServer function and is passed as an argument to the
rxSetComputeContext function. Now your subsequent RevoScaleR functions run on the server instance.
Note The RevoScaleR package enables the scalable, high-performance, multicore analytic
functions. In this chapter, we explore several functions in this package. Setting the compute context
affects only the RevoScaleR functions. Open-source R functions continue to execute locally.
Important At the time of this writing, the RevoScaleR package requires a SQL login with the
necessary permissions to create tables and read data in a database.
Data source
To execute R commands against data, you define a data source. A data source is a subset of data from
your database and can be a table, a view, or a SQL query. By creating a data source, you create only a
reference to a result set. Data never leaves the database. Example 6-5 shows how to create a data
source object in the R IDE by first assigning a T-SQL query string to a variable, passing the variable to
the RxSqlServerData function, and storing the data source reference in another variable.
Example 6-5: Creating a data source
sampleDataQuery <- "select top 1000 tipped, fare_amount, passenger_count, trip_time_in_secs,
trip_distance, pickup_datetime, dropoff_datetime, pickup_longitude, pickup_latitude,
dropoff_longitude, dropoff_latitude from nyctaxi_sample"
inDataSource <- RxSqlServerData(sqlQuery = sampleDataQuery, connectionString = connStr,
colClasses = c(pickup_longitude = "numeric", pickup_latitude = "numeric",
dropoff_longitude = "numeric", dropoff_latitude = "numeric"),
stringsAsFactors=TRUE, rowsPerRead=500)
In this example, the RxSqlServerData function takes the following arguments:
connectionString An ODBC connection string for SQL Server. At the time of this writing, you
must use a SQL login in the connection string.
colClasses A character vector that maps the column types between SQL Server and R. For the
purposes of this section, a character vector is a string. In this case, the string must contain the
names of columns in the query paired with one of the following allowable column types: logical,
integer, float32, numeric, character, factor, int16, uint16, or date.
rowsPerRead Number of rows read into a chunk. R Services processes chunks of data and
aggregates the results. Use this argument to control the chunk size to manage memory usage. If
this value is too high, processing can slow as the result of inadequate memory resources,
although a value that is too low might also adversely affect processing.
A string representing a valid SQL query.
Note You can replace the sqlQuery argument with the table argument if you prefer to reference
an entire table. You cannot use both arguments together.
CH A P TER 6 | More analytics
Tip The sample data in this section does not contain categorical data. When your data contains
categories, such as age groups or geographic regions, you should consider including the
stringsAsFactors argument with the RxSqlServerData function. This argument is a Boolean value that
controls whether to convert strings to factors. A factor is an R object type that is used in statistical
functions. Functions such as rxSummary return more complete results for factors as compared to
Data exploration
After creating a data source, you can use statistical functions or create plots and graphic objects with
which to explore your data in the R IDE. A good starting point is the rxGetVarInfo function to display
basic information about the structure of your data source. To do this, use the following code in your R
rxGetVarInfo(data = inDataSource)
Executing this code returns the results shown in Figure 6-14. The rxGetVarInfo function returns
metadata about the columns of your data, which are called variables when working in R. The metadata
includes the name and data type for each variable.
Figure 6-14: Executing the rxGetVarInfo function in the R console.
Another common function to use for becoming familiar with your data is rxSummary. For a basic
statistical summary of your data, as shown in Figure 6-15, use the following code:
rxSummary(~., data = inDataSource)
Figure 6-15: Executing the rxSummary function in the R console.
You can also use graphical objects such as a histogram to explore your data. Figure 6-16 shows the
results of the rxHistogram function, which plots the count of observations in your data for each
distinct value of fare_amount under 50 (to ignore outliers) by using the following code:
rxHistogram(~(fare_amount), data = inDataSource, title = "Fare Amounts Under $50", endVal=50)
CH A P TER 6 | More analytics
Figure 6-16: Viewing the plot created by executing the rxHistogram function.
Note To achieve the results shown in Figure 6-16, the code shown in Example 6-5 was modified to
select the top 100,000 rows from the table and then executed again prior to executing the
rxHistogram function. You can further fine-tune the appearance of the histogram by adding
arguments to apply formatting to the axes and configure other style settings. For more information,
refer to the function documentation at
Spatial data can be plotted on a map as another option for exploring data. Because a common
security practice is to prevent SQL Server from accessing the Internet, you cannot perform the
complete operation on the server. Instead, you use the local context to make a geocoding call to
Google Maps to obtain a graphical layer for the map and send it to the server context to get the plot
points for individual locations. To start the process, create a custom function to get the plot points, as
shown in Example 6-6.
Example 6-6: Creating a custom function to plot spatial data
mapPlot <- function(inDataSource, googMap){
ds <- rxImport(inDataSource)
p <- ggmap(googMap)+
geom_point(aes(x = pickup_longitude, y =pickup_latitude ), data=ds, alpha =.5,
color="darkred", size = 1.5)
In this example, you use the function function to define a custom function and supply an argument list
in parentheses. The arguments in this case are inDataSource, which is the RxSqlServerData object
created in an earlier example, and googMap, which you create in a later step. R allows you to reference
this argument in the definition without testing for its existence because you are not yet attempting to
execute the function.
CH A P TER 6 | More analytics
The body of the function appears between the two braces. The first two lines use the library function
to load the ggmap1 and mapproj packages to ensure that the functions they provide are available and
ready to use on your server. When you install MRO, a core set of packages and libraries is available for
immediate use. A package is a collection of objects that can include code, data, or documentation that
you use to extend base R, whereas a library is a directory containing packages. There are thousands of
add-on packages contributed by the open-source community that you can download and use to build
your analytical application.
Next, the rxImport function loads your data into server memory as a data frame called ds. This step
prepares the data for use in open-source R functions because these functions cannot run in-database.
The variable p is a plot object consisting of two layers. The ggmap function produces a map from the
object passed in as an argument, which we explain later in this section. The + operator adds another
layer to the plot object. You can add as many layers as necessary by using this technique. The
geom_point function creates a scatterplot of pickup_longitude on the x-axis and pickup_latitude on the
y-axis from the in-memory data frame. The alpha, color, and size arguments set point transparency,
point color, and point size, respectively. The final line of code assigns a tag, myplot, to the p variable
and converts the object to a list data type, which is the return value of the custom function.
Next you execute the geocoding call to get the map, call the custom function to send the map and
combine it with your plot points, and render the results in the local context, as shown in Example 6-7.
Example 6-7: Creating a custom function to plot spatial data
gc <- geocode("Manhattan", source = "google")
googMap <- get_googlemap(center = as.numeric(gc), zoom = 12, maptype = 'roadmap',
color = 'color')
myplots <- rxExec(mapPlot, inDataSource, googMap, timesToRun = 1)
The first two lines are calls to load ggmap and mapproj again, but this time in the local context. Then
the geocode function takes a street address or a place name as its first argument and sets the source
to google as the second argument. The other possible source is dsk, which is the Data Science Toolkit
(, but this source tends to return results more slowly or timeout.
Next, the get_googlemap function uses the latitude and longitude coordinates stored in the gc variable
to set the center of the map that it downloads from Google. The zoom argument takes an integer
value ranging from 3 (for continent) to 21 (for building) to indicate the level of detail to display in the
map. You can set the color argument to either color or black and white. The resulting map is stored in
the googMap variable that you pass to the mapPlot function.
In the next line, the rxExec function executes the function specified as the first argument (the custom
function mapPlot, in this case) on SQL Server, using the arguments passed as a list as subsequent
arguments, until it encounters rxExec arguments such as timesToRun. The variable myplots stores the
results of execution, which is a list containing one item called myplot.
Last, the plot function takes the first item in the myplots list and renders the object on the local
computer. The result is a static map of Manhattan with multiple points representing pickup locations
overlayed on the map, as shown in Figure 6-17.
D. Kahle and H. Wickham, “ggmap: Spatial Visualization with ggplot2,” The R Journal no. 5(1): 144–
CH A P TER 6 | More analytics
Figure 6-17: Viewing the map plot created by executing the custom mapPlot function.
Note Before executing this code, the data source query was adjusted to return only 1,000 rows. It
is important to note that the map is created locally and passed by a function that runs in the server
context. The data is serialized back to the local computer where you view it in the Plot window in
the R IDE.
Data transformation
Besides exploring data with R, you can also use R to transform data to enhance it for use in predictive
modeling. However, when working with large data volumes, R transformations might not perform as
optimally as similar transformations done by using a T-SQL function. You are not limited to these
options, though. You might prefer to use T-SQL scripts or Integration Services to preprocess the data
before using the data with R Services.
Note You use the rxDataStep function in conjunction with custom functions to perform
transformations by using the RevoScaleR package on the server. You can learn more about this
function at
The taxi data currently includes coordinates for pickup and drop-off locations, which you can use to
compute the linear distance. The database also includes a custom function, fnCalculateDistance, to use
for this computation. To set up a new data source using a random sample that includes the computed
distance, execute the code shown in Example 6-8 in your R IDE.
Example 6-8: Adding a data source with a feature computed in T-SQL
modelQuery = "SELECT tipped, fare_amount, passenger_count, trip_time_in_secs,trip_distance,
pickup_datetime, dropoff_datetime,
dbo.fnCalculateDistance(pickup_latitude, pickup_longitude, dropoff_latitude,
dropoff_longitude) as direct_distance,
pickup_latitude, pickup_longitude, dropoff_latitude, dropoff_longitude
FROM nyctaxi_sample
tablesample (1 percent) repeatable (98052)"
modelDataSource = RxSqlServerData(sqlQuery = modelQuery,
colClasses = c(pickup_longitude = "numeric", pickup_latitude = "numeric",
dropoff_longitude = "numeric", dropoff_latitude = "numeric",
CH A P TER 6 | More analytics
passenger_count = "numeric", trip_distance = "numeric",
trip_time_in_secs = "numeric", direct_distance = "numeric"),
connectionString = connStr)
Predictive model creation
After preparing your data, you can create a model by using any of the functions available in the
RevoScaleR package. The RxLogit function is a good choice for classification problems. It uses logistic
regression to estimate the probability of a variable with two possible values. In the sample code
shown in Example 6-9, the goal is to predict whether a tip was given. The summary function provides
statistical information about the resulting model, as shown in Figure 6-18.
Example 6-9: Creating a logistic regression model
logitObj <- rxLogit(tipped ~ passenger_count + trip_distance + trip_time_in_secs +
direct_distance, data = modelDataSource)
Figure 6-18: Viewing the summary of a logistic regression predictive model.
Note To learn more about the rxLogit function, see
Model usage
After you build a predictive model, you can apply it to a data source to predict the dependent variable
value, score the prediction, and store the results in a table by using the rxPredict function. You can see
the code necessary to perform these steps in Example 6-10. In this example, you define a table
without a schema in the rxSqlServerData function. The output from the rxPredict function returns the
schema and creates the table, which means the SQL login associated with the rxSqlServerData
function (as defined in the connection string) must have permissions to create a table, or the
execution of the code fails.
Example 6-10: Predicting values
scoredOutput <- RxSqlServerData(
connectionString = connStr,
table = "taxiScoreOutput")
CH A P TER 6 | More analytics
rxPredict(modelObject = logitObj, data = modelDataSource, outData = scoredOutput,
predVarNames = "Score", type = "response", writeModelVars = TRUE, overwrite = TRUE)
Figure 6-19 shows the results of the output stored in the table. A value below 0.5 in the Score column
indicates a tip is not likely.
Figure 6-19: Viewing the output of the rxPredict function in the taxiScoreOutput table in SSMS.
Note For simplicity in this example, the data used to train the model is also used to test the
model. Typically, you partition the data, using one set to train the model and one set to test the
To learn more about the rxPredict function, see
Model accuracy
After you create a model, you can use R functions to test its accuracy. ROCR is a useful package for
testing the performance of classification models. Example 6-11 shows the code to install and load this
library by using the install.packages and library functions.
Example 6-11: Testing a model’s accuracy
if (!('ROCR' %in% rownames(installed.packages()))){
scoredOutput <- rxImport(scoredOutput)
pr <- prediction(scoredOutput$Score, scoredOutput$tipped)
prf <- performance(pr, measure = "tpr", x.measure = "fpr")
To use the functions in ROCR, you must bring data from the server into your local environment by
using the rxImport function. Next, you need to load the results of your predictions into a prediction
object by using the prediction function, which takes the score from your model as its first argument
and the predicted value as the second argument. Notice that the format of these arguments uses the
name of the data source first, then a $ symbol, which is followed by the data source column.
The performance function takes the prediction object as the first argument and then you specify
measures to return. In this case, tpr and fpr represent true positive rate and false positive rate and are
only two of many different types of performance metrics that the performance function returns. You
can store the performance results in a variable that you can then plot, as shown in Figure 6-20.
CH A P TER 6 | More analytics
Figure 6-20: Viewing the plot of the prediction model performance.
Note You can see the other performance metrics accessible with the performance function at
Using an R Model in SQL Server
After creating an R model, you can deploy it to SQL Server for use in applications and other tools. You
can then invoke the model by calling the sp_execute_external_script stored procedure. You can call a
model to score data in batch mode or to score data for an individual case. The sample database
includes two stored procedures that allow you to perform each of these tasks.
Model deployment
This process requires you to serialize your model as a hexadecimal string that you send to the server
and store in a varbinary(max) column in a database, as shown in Example 6-12. The serialize function
produces the string, and the paste function ensures that the result is a single string. Then the RODBC
package is installed to use the odbcDriverConnect function to open a connection to SQL Server. Next,
the paste function concatenates the serialized string with a call to the PersistModel stored procedure
to produce a query string that is passed into the sqlQuery function and executed. The PersistModel
stored procedure is a custom stored procedure in the sample database that inserts a record into the
nyc_taxi_models table.
Example 6-12: Deploying a model to SQL Server
modelbin <- serialize(logitObj, NULL)
modelbinstr=paste(modelbin, collapse="")
if (!('RODBC' %in% rownames(installed.packages()))){
conn <- odbcDriverConnect(connStr )
q<-paste("EXEC PersistModel @m='", modelbinstr,"'", sep="")
sqlQuery (conn, q)
Batch mode invocation of a model
The stored procedure in the sample database that invokes the model in batch mode is shown in
Example 6-13. This stored procedure, PredictTipBatchMode, retrieves the stored model and stores it in
a variable that becomes a parameter for the sp_execute_external_script stored procedure. You pass the
data to score as a query string into PredictTipBatchMode. It becomes a data frame called InputDataSet
CH A P TER 6 | More analytics
used in the rxPredict function that sp_execute_external_script executes. The output of this stored
procedure is a set of rows containing a score for each row in the input.
Example 6-13: Creating a stored procedure to invoke a model in batch mode
CREATE PROCEDURE [dbo].[PredictTipBatchMode] @inquery nvarchar(max)
DECLARE @lmodel2 varbinary(max) = (SELECT TOP 1 model
FROM nyc_taxi_models);
EXEC sp_execute_external_script @language = N'R',
@script = N'
mod <- unserialize(as.raw(model));
OutputDataSet<-rxPredict(modelObject = mod, data = InputDataSet, outData = NULL,
predVarNames = "Score", type = "response", writeModelVars = FALSE, overwrite = TRUE);
@input_data_1 = @inquery,
@params = N'@model varbinary(max)',
@model = @lmodel2
WITH RESULT SETS ((Score float));
Individual scoring mode invocation of a model
Rather than score a set of rows in batch mode, you can score a single case. Example 6-14 shows the
PredictTipSingleMode stored procedure in the sample database, which illustrates this approach. It is
similar to the previous example, except the PredictTipSingleMode stored procedure defines input
parameters for each of the variables in your training data set. These parameters are then sent to a
table-valued helper function in the sample database that computes the linear distance, and the result
becomes the InputDataSet data frame. The output is a single value that represents the probability of a
Example 6-14: Creating a stored procedure to invoke a model in single mode
CREATE PROCEDURE [dbo].[PredictTipSingleMode] @passenger_count int = 0,
@trip_distance float = 0,
@trip_time_in_secs int = 0,
@pickup_latitude float = 0,
@pickup_longitude float = 0,
@dropoff_latitude float = 0,
@dropoff_longitude float = 0
DECLARE @inquery nvarchar(max) = N'
SELECT * FROM [dbo].[fnEngineerFeatures]( @passenger_count, @trip_distance, @trip_time_in_secs,
@pickup_latitude, @pickup_longitude, @dropoff_latitude, @dropoff_longitude)'
DECLARE @lmodel2 varbinary(max) = (SELECT TOP 1 model FROM nyc_taxi_models);
EXEC sp_execute_external_script @language = N'R',
@script = N'
mod <- unserialize(as.raw(model));
OutputDataSet<-rxPredict(modelObject = mod, data = InputDataSet, outData = NULL,
predVarNames = "Score", type = "response", writeModelVars = FALSE, overwrite = TRUE);
@input_data_1 = @inquery,
@params = N'@model varbinary(max), @passenger_count int,
CH A P TER 6 | More analytics
@trip_distance float, @trip_time_in_secs int, @pickup_latitude float, @pickup_longitude float,
@dropoff_latitude float, @dropoff_longitude float',
@model = @lmodel2,
@passenger_count =@passenger_count ,
WITH RESULT SETS ((Score float));
CH A P TER 6 | More analytics
Better reporting
For report developers, Reporting Services in SQL Server 2016 has a more
modern development environment, two new data visualizations, and
improved parameter layout options. Users also benefit from a new web
portal that supports modern web browsers and mobile access to reports.
In this chapter, we’ll explore these new features in detail.
Report content types
This release of Reporting Services includes both enhanced and new report content types:
Paginated reports Paginated reports are the traditional content type for which Reporting
Services is especially well suited. You use this content type when you need precise control over
the layout, appearance, and behavior of each element in your report. Users can view a paginated
report online, export it to another format, or receive it on a scheduled basis by subscribing to the
report. A paginated report can consist of a single page or hundreds of pages, based on the
dataset associated with the report. The need for this type of report continues to persist in most
organizations, as well as the other report content types that are now available in the Microsoft
reporting platform.
Mobile reports In early 2015, Microsoft acquired Datazen Software to make it easier to deploy
reports to mobile devices, regardless of operating system and form factor. This content type is
best when you need touch-responsive and easy-to-read reports that are displayed on smaller
screens, communicate key metrics effectively at a glance, and support drill-through to view
supporting details. In SQL Server 2016, users can view both paginated and mobile reports through
the web portal interface of the on-premises report server.
Key performance indicators (KPIs) A KPI is a simple type of report content that you can add to
the report server to display metrics and trends at a glance. This content type uses colors to
indicate progress toward a goal and an optional visualization to show how values trend over time.
Paginated report development enhancements
In this release of Reporting Services, the authoring tools for paginated reports work much like they
did in previous releases, but with some enhancements. The first noticeable change is the overall
appearance of the authoring tools. In addition, these tools have been augmented by the addition of
new visualizations and a new interface for working with parameters.
CH A P TER 7 | Better reporting
Introducing changes to paginated report authoring tools
As in prior versions of Reporting Services, there are two methods for authoring paginated reports:
Report Designer A full-featured report development environment available as one of the
business intelligence templates installed in the new SQL Server Data Tools for Visual Studio 2015
Report Builder
A standalone application that shares many common features with Report
Report Designer
Microsoft has released a new updated business intelligence template in SSDT that you download from This business intelligence template includes an
updated version of Report Designer that allows you to develop reports for multiple versions of
Reporting Services. By default, you can develop reports for SQL Server 2016 Reporting Services or
later, as shown in Figure 7-1, but you can change the TargetServerVersion property in the project’s
properties to target SQL Server 2008, SQL Server 2008 R2, SQL Server 2012, or SQL Server 2014.
Report authors may continue to use SQL Server Data Tools for Business Intelligence in Visual Studio
2013 to develop reports for these earlier versions, but the new features specific to SQL Server 2016
that we discuss later in this chapter are not supported.
Figure 7-1: A new default value for the TargetServerVersion property in the project’s properties.
Report Builder
Report Builder is an alternative report-development tool for power users and report developers who
need only to create or edit one report at a time. You can start the ClickOnce version of the Report
Builder by clicking the Report Builder button on the web portal toolbar on your report server at
http://<servername>/reports. You can also download and install a standalone version of Report
Builder from and then use
the Windows Start menu to open it after installation. Previous versions of Report Builder use the light
blue Office 2007 appearance, but the most recent version of Report Builder, shown in Figure 7-2, uses
the same darker theme that appears in both Office 2016 and the Power BI Desktop application and
continues to use a ribbon interface like Office applications.
CH A P TER 7 | Better reporting
Figure 7-2: New Report Builder interface.
Exploring new data visualizations
All data visualizations included in prior versions of Reporting Services continue to be available, but the
SQL Server 2016 version includes two new types of data visualizations:
Tree map A tree map represents hierarchical categories as rectangles with relative sizes.
Sunburst A sunburst chart is a hierarchical representation of data that uses circles for each level.
Tree map
A tree map is useful to show how parts contribute to a whole. Each rectangle represents the sum of a
value and is sized according to the percentage of its value relative to the total of values for all
rectangles in the tree map. The rectangles are positioned within the tree map with the largest
category in the upper-left corner of the parent rectangle and the smallest category in the lower-right
corner. Each rectangle can contain another collection of rectangles that break down its values by
another category that represents a lower level in a hierarchy.
As an example, in the tree map shown in Figure 7-3, the first level shows the United States as the
largest category, followed by Canada, with the second largest category, and then progressively smaller
rectangles are displayed for France, United Kingdom, Germany, and Australia. For each of these
country/region categories, business type is the next lower level in the hierarchy, and rectangles for
each distinct business type are displayed using the same pattern of largest to smallest from top left to
bottom right within a country’s/region’s rectangle. In this example, the largest business type in the
United States is Value Added Reseller, followed by Warehouse, and then Specialty Bike Shop.
CH A P TER 7 | Better reporting
Figure 7-3: Tree map showing sales hierarchically by country/region and by business type.
To add a tree map to your report, you use the same technique as you do for any chart. Whether using
Report Designer or Report Builder, you insert a chart into the report by choosing Chart from the
toolbox or ribbon, and then select Tree Map in the Shape collection of chart types in the Select Chart
Type dialog box, as shown in Figure 7-4.
Figure 7-4: Selection of a tree map in the Select Chart Type dialog box.
To configure the chart, click anywhere on its surface to open the Chart Data pane. Then click the
button with the plus symbol to add fields to the Values, Category Groups, or Series Groups areas, as
shown in Figure 7-5. The value field determines the size of a rectangle for category groups and series
groups. Each series group field is associated with a different color and becomes the outermost
collection of rectangles. For example, with SalesTerritoryCountry as a series group, each
country/region is identifiable by color in the tree map. Within each country’s/region’s rectangle, each
distinct value within a category group is represented by a separate rectangle. In this case, each
country’s/region’s rectangle contains three rectangles—Specialty Bike Shop, Value Added Reseller,
and Warehouse. The proportion of an individual business type’s sales amount value relative to a
country’s/region’s total sales determines the size of its rectangle.
CH A P TER 7 | Better reporting
Figure 7-5: Configuring the Chart Data pane for a tree map.
To improve the legibility of a tree map, you should consider making the following changes to specific
chart properties:
Size You should increase the size of the chart because the default size, 3 inches wide by 2 inches
high, is too small to view the data labels that are enabled by default. Click the chart object, but
take care to click an element such as the Chart Title or a series in the chart, and then adjust the
Size properties, Width and Height, in the Properties pane.
Legend To maximize the space of the chart area allocated to the tree map, consider moving the
legend above or below the chart. To do this, right-click the legend, select Legend Properties, and
then select one of the Legend Position options to reposition the legend.
Data labels Even after resizing the chart, you might find that the default 10 point font size used
for the labels is too large to display labels in each rectangle or that the black font is difficult to
read when the series color is dark. To reduce the size of the font and change its color to improve
the visibility of the data labels, click the chart to display the Chart Data pane, click the field in the
Values area, and then locate the Labels section in the Properties pane. When you expand this
section, you can change font properties such as size and color as needed.
Note The size of rectangles in a tree map might continue to affect the visibility of the data labels
even if you reduce the font size to 6 points. If the smaller label text cannot fit within the width of its
rectangle, the label is not displayed.
Tooltip One way to compensate for missing data labels in small rectangles, or to add more
context to a tree map, is to add a tooltip, as shown in Figure 7-6. To do this, right-click a rectangle
in the chart, select Series Properties, click the expression button next to the Tooltip box in the
Series Properties dialog box, and type an expression such as this:
=Fields!BusinessType.Value + " : " + Format(Sum(Fields!SalesAmount.Value), "C0")
CH A P TER 7 | Better reporting
Figure 7-6: Tooltip displayed above a selected rectangle in a tree map.
You can add more than one field to the Category Groups or Series Groups areas of the Chart Data
pane. However, the meaning of the chart is easier to discern if you add the second field only to the
Series Groups area so that different colors help viewers distinguish values, as shown in Figure 7-7. If
you add a second field to the Category Groups area, more rectangles are displayed in the tree map,
but it’s more difficult to interpret the hierarchical arrangement without extensive customization of the
tree map’s elements.
Figure 7-7: Tree map displaying two series groups.
A sunburst chart is a type of visualization that is a hybrid of a pie chart, using slices of a circle to
represent the proportional value of a category to the total. However, a sunburst chart includes
multiple circles to represent levels of hierarchical data. Color is the highest level of a hierarchy if a
series group is added to the chart, but it is not required. If no series group is defined, the innermost
circle becomes the highest level of the hierarchy. Each lower level moves farther from the center of
the circle, with the outermost circle as the lowest level of detail. Within each type of grouping, color or
circle, the slices are arranged in clockwise order, with the largest value appearing first and the smallest
value appearing last in the slice.
As an example, in Figure 7-8, color is used to identify sales amount by year across all circles, with the
largest color slice starting at the twelve o’clock position in the circle. At a glance, a viewer can easily
see the relative contribution of each year to total sales and which year had the greatest number of
sales. Next, the inner circle slices each color by country/region, again sorting the countries/regions
from largest to smallest in clockwise order. The outer circle further subdivides the countries/regions
by business type. In this example, some of the slices are too small for the labels to be displayed.
CH A P TER 7 | Better reporting
Figure 7-8: Example of a sunburst chart.
To produce a sunburst, you insert a chart into the report and select Sunburst from the Shape
collection of chart types. Click the chart to open the Chart Data pane and use the button with the plus
symbol to add fields to the Values, Category Groups, or Series Groups areas, as shown in Figure 7-9.
The value field determines the size of a slice for category groups and series groups. Each series group
field is associated with a different color and becomes the first division of the total value into
proportional slices, although the inclusion of a series group is optional. Category groups then further
subdivide values into slices, with the first category group in the list as the inner circle, and each
subsequent category group added to the chart as another outer circle moving from the center.
Figure 7-9: Chart Data pane configured for a sunburst chart.
As for a tree map, a sunburst chart’s default properties are likely to produce a chart that is difficult to
read. Therefore, you should consider modifying the following chart properties:
CH A P TER 7 | Better reporting
Size The minimum recommended size for a sunburst chart is 5 inches wide. Click the chart
object (but not an element such as the Chart) and then increase the Size properties, Width and
Height, in the Properties pane.
Legend More space is allocated to the sunburst chart when you move the legend above or
below the chart. To do this, right-click the legend, select Legend Properties, and select one of the
Legend Position options to reposition the legend.
Data labels Reducing the label size and changing the font color are likely to improve legibility.
To fix these properties, click the chart to display the Chart Data pane, click the field in the Values
area, expand the Labels section in the Properties pane, and change the font size and color
Note Some sunburst slices can still be too small for some data labels even if you reduce the font
size to 6 points.
Tooltip To help users understand the values in a sunburst chart when data labels are missing
from small slices, consider adding a tooltip by right-clicking a slice in the chart, selecting Series
Properties, clicking the expression button next to the Tooltip box in the Series Properties dialog
box, and then typing an expression such as this:
=Fields!BusinessType.Value + " : " + Fields!SalesTerritoryCountry.Value + " : " +
Format(Sum(Fields!SalesAmount.Value), "C0")
Managing parameter layout in paginated reports
In previous versions of Reporting Services, there was no option for configuring the layout of
parameters unless you designed a custom interface to replace Report Manager for accessing reports.
Now in both Report Designer and Report Builder, you can use a Parameters pane to control the
relative position of parameters and to organize parameters into groups.
Note In Report Builder, you can change the visibility of the Parameters pane by selecting or
clearing the new Parameters check box on the View tab of the ribbon.
The new Parameters pane is a 4x2 grid that displays above the report design surface. To add a report
parameter to the grid, you can continue to use the Report Data pane as you have in previous versions
of Reporting Services. As an alternative, in the Parameters pane, right-click an empty cell and select
Add Parameter, as shown in Figure 7-10, to open the Report Parameter Properties dialog box. Notice
that the context menu that appears when you right-click a cell also includes commands to add or
remove rows or columns, delete a parameter, or view a selected parameter’s properties.
Figure 7-10: Adding a new parameter to a report by using the Parameters pane in Report Builder.
CH A P TER 7 | Better reporting
Note When you add a report parameter by using the Parameters pane, the parameter is added
automatically to the Report Data pane. You can easily access a parameter’s properties by doubleclicking it in either location.
After adding a parameter, you can drag it to a new location. Consider using empty rows or columns to
create groupings of parameters, as shown in Figure 7-11.
Figure 7-11: Using separate columns to group parameters in the Parameter pane.
Note If you design a report with cascading parameters, the sequence of parameters in the Report
Data pane remains important. Cascading parameters are a set of at least two parameters in which a
child parameter’s available list of values is dependent on the user’s selection of another parameter
value, the parent parameter. The parent parameter must be displayed above the child parameter in
the Report Data pane.
You cannot control the size of an unused parameter column, but the rendered report displays each
column with enough separation to clearly distinguish groups, as shown Figure 7-12. You can create
more separation between column groups by inserting another empty column in the Parameters pane.
Figure 7-12: Parameter groups in a rendered report.
Mobile report development
Mobile reports display data concisely for use on mobile devices. The acquisition of Datazen by
Microsoft brings a suite of tools supporting the development of mobile reports into the Reporting
Services platform, but these tools are currently in various states of integration. To create mobile
reports, you use the SQL Server Mobile Report Publisher (which you can download from the Microsoft
Store for Windows 8 and Windows 10).
Note The Mobile Report Publisher is not available at the time of this writing. This section will be
updated with more details about Mobile Report Publisher in the final version of this ebook.
Mobile reports enable you to create data mash-ups from a variety of data sources. You can use the
same data sources and shared data sets published to the report server to connect data to mobile
report elements such as gauges and charts, among others.
KPI development
In the CTP 3.2 release of SQL Server 2016, you use the Reporting Services web portal to create KPIs.
From the main portal page at http://<yourserver>/reports, click the Preview The New Reporting
Services link at the top of the page, click New in the toolbar, and then click KPI. A new KPI screen is
displayed, as shown in Figure 7-13.
CH A P TER 7 | Better reporting
Figure 7-13: Creating a new KPI.
To configure a KPI, you specify up to four values: Value, the amount to monitor; Goal, the target
amount to compare with Value; Status, a value to set the color of the background; and Trend, a set of
values to visualize. For each of these values, you can set its value manually, associate it with a field in a
shared dataset on the report server, or leave its value empty. (If you choose to use a shared dataset,
remember that you can specify a cache refresh plan to update the KPI as frequently as necessary.)
Last, you can choose to optionally include one of the following visualizations: column chart, line chart,
step chart, or area chart.
Note Datasets for Value, Goal, and Status must return a single row of data. If you choose to use a
query for Status, the query must return -1 for red, 0 for amber, and 1 for green. A query for Trend
must return a sorted set of one or more values for use as data points in the visualization.
Report access enhancements
The user-facing side of Reporting Services also benefits from several enhancements in this release.
First, browser rendering and broader support has been upgraded to accommodate modern web
standards. Furthermore, the ActiveX control is no longer required to print from the web portal. Next,
users can export reports directly to PowerPoint. Last, the process of working with subscriptions in the
web portal has been improved with several new capabilities to streamline and simplify subscription
Accessing reports with modern browsers
When Reporting Services was initially added to the SQL Server platform, it was optimized for Internet
Explorer 5. Since then, web standards have changed. As a result, modern browsers that are optimized
for newer web standards such as HTML5 and CSS3 have emerged and grown in popularity. But
however popular these browsers might be for users on a day-to-day basis, earlier versions of
Reporting Services do not render reports consistently in these browsers at best or do not render them
at all at worst. In SQL Server 2016, Reporting Services is redesigned with a new renderer that supports
HTML5 and has no dependency on features specific to Internet Explorer, so users can have a
consistent experience across supported browsers. The following table shows the browsers currently
supported by the latest version of Reporting Services by operating system:
CH A P TER 7 | Better reporting
Explorer 10
and 11
Apple Safari
2012 and
2012 R2
Mac OS
X 10.710.10
iOS 6-9
for iPad
8 and 8.1
2008 R2
Regardless of which browser you use, the first time you attempt to open a report, an error message is
displayed if you have not configured the browser to run scripts. In response to the message, you can
click to continue to view the report without scripts. In that case, the report renders in HTML, but the
features supported by the report viewer are not displayed, such as the report toolbar and the
document map.
Note Enhancing the renderer to work across all browsers is a huge undertaking. Despite extensive
testing, it is possible that a particular combination of report elements that worked well in an earlier
version of Reporting Services no longer renders properly. If you find that a report does not render
correctly with the new rendering engine, you can click the Switch To Compatibility Mode link on the
right side of the report viewer toolbar to revert rendering to Reporting Services’ prior style of
rendering. You can also click the Send Feedback button next to this link if you continue to have a
problem rendering a report. Clicking this link opens the SQL Server Reporting Services Forum on
MSDN, where you can use the Ask A Question button to create a post describing the problem you
are experiencing.
Not only is the rendering engine updated, but the Report Manager web application used for report
access is no longer available. Instead, users access reports by using the new Reporting Services web
portal, shown in Figure 7-14. The web portal includes a Favorites page on which you can organize
reports by type: KPIs, mobile reports, and paginated reports. You can switch to the Browse page to
view reports by navigating through folders.
CH A P TER 7 | Better reporting
Figure 7-14: The home page of the new Reporting Services web portal displaying the Favorites tab.
Note Mobile reports are not available in the new web portal in SQL Server 2016 CTP 3.2 but will
be available in a future release of SQL Server 2016. This section will be updated in the final ebook.
Viewing reports on mobile devices
In addition to using the web portal to view mobile reports rendered as HTML5 pages in a web
browser, you can also interact with these reports through a native user interface on the following
major mobile platforms:
Windows 8 or later On your tablets and touch-enabled devices, you can use semantic zoom
while viewing reports. In addition, you can pin dashboards and KPIs to the Start screen.
iOS8 or later You can access published dashboards and KPIs while online and review KPI
summary data when offline.
Printing without ActiveX
Earlier versions of Reporting Services require users to install ActiveX to enable a control in Internet
Explorer that allows them to print a paginated report from the browser. However, for security reasons,
many enterprise users do not have the necessary permissions to install software on their computers,
including ActiveX controls. Furthermore, many modern browsers do not support ActiveX.
Consequently, in SQL Server 2016, Reporting Services provides a new solution by generating a printerfriendly PDF version of the report with the option to override the default page size.
When you click the printer icon in the report viewer toolbar, Reporting Services checks for the
existence of the Acrobat PDF browser plug-in in Internet Explorer. If it does not exist, an error
message prompts you to install the plug-in. However, if your browser does not have the plug-in, you
are still able to print if you clear the error message. After you clear the error message, or if you are
using a browser other than Internet Explorer, the Print dialog box is displayed, as shown in Figure 788
CH A P TER 7 | Better reporting
15. This dialog box allows you to adjust the paper size and page orientation by using the respective
drop-down lists before printing your report.
Figure 7-15: Print dialog box for browser without PDF control.
When you click the Print button in this dialog box in Internet Explorer, the operating system’s Print
dialog box displays more options for selecting which pages to print, the number of copies to print,
and so on. If you choose to cancel at this point, the operating system’s Print dialog box closes, and
you then see another type of Print dialog box that displays a preview of the first page of your report,
as shown in Figure 7-16. At the bottom of this dialog box is the Click Here To View The PDF Of Your
Report link, which allows you to open your report in Acrobat Reader if it is installed on your computer.
Otherwise, you can download the PDF to store it for later viewing once you have installed the
necessary software.
Note When you use Edge as your browser and click the Print button in Reporting Services’ Print
dialog box, another tab opens in the browser and displays your report because Edge has a built-in
PDF viewer.
In Chrome, when you click Print, a message appears and indicates that the report is being
converted to PDF, and then Chrome’s Print dialog box displays.
In Safari, a message indicates that your PDF file is ready and includes the link Click Here To View
The PDF Of Your Report. When you click the link, the PDF file downloads and the Preview
application opens to display your report.
CH A P TER 7 | Better reporting
Figure 7-16: Print dialog box with option to view the PDF of your report.
Just as in prior versions, report server administrators can control whether users see the print icon in
the report viewer toolbar. However, the Enable Download For the ActiveX Client Print Control check
box is no longer available for this purpose when configuring report server properties because this
control is no longer supported. Instead, you change one of the advanced properties that controls the
presence of the print icon. To do this, open SQL Server Management Studio by using Run As
Administrator, connect to the Report Server, right-click the server node, select Properties, select the
Advanced tab in the Server Properties dialog box, and change the EnableClientPrinting property from
its default setting of True to False.
Exporting to PowerPoint
One of the many benefits of Reporting Services is the ability to export a report to a variety of different
formats, such as Excel or Word. In the SQL Server 2016 release, the list of available options is
expanded to include another popular Office application, PowerPoint. When you click the Export
button in the report viewer toolbar, you now see PowerPoint listed as an option, as shown in
Figure 7-17.
CH A P TER 7 | Better reporting
Figure 7-17: Choosing PowerPoint as an option for exporting a report.
You can also now use PowerPoint as a rendering format when configuring a subscription.
When you select the PowerPoint export option from the list, the PPTX file downloads to your
computer. You then have the option to save it or, if you have PowerPoint installed on your computer,
to open the file. In general, each page of your report becomes a separate slide in PowerPoint, as
shown in Figure 7-18, although some report items might span multiple slides. Just as you must factor
in the rendered page size during report development if you know that users plan to export to PDF or
Word, you must ensure report items can fit on a single PowerPoint slide where possible. Otherwise,
the Reporting Services rendering engine will divide the report item into two or more smaller pieces
and allocate each piece to a separate slide, as shown in the third and fourth PowerPoint slides in
Figure 7-18, which collectively represents the third page of a report when the page is rendered in
HTML. Notice that objects from a report do not consume the entire vertical space within a PowerPoint
Figure 7-18: A report rendered as a PowerPoint file.
CH A P TER 7 | Better reporting
Note Although in an earlier section of this chapter we recommend placing legend items above or
below a tree map or sunburst chart to maximize chart space, this recommendation is not applicable
to reports that you plan to export to PowerPoint because the vertical space is more constrained.
If you click the Enable Editing button that appears when PowerPoint opens the file, you can interact
with the objects added to the file. For example, you can edit freestanding text boxes containing static
text such as a report title or page numbers from the page header or footer. Report items such as a
chart or a matrix are added as picture objects and cannot be edited, although they can be resized and
rearranged by moving them to a different location on the same slide or copying and pasting them to
a different slide.
Pinning reports to Power BI
One of the ways that Reporting Services is integrating hybrid and on-premises reporting is a new
feature that allows you to pin a report in the web portal to a Power BI dashboard. This capability has
several requirements, however. You must be using Azure Active Directory (Azure AD), and the Power
BI dashboard that you want to use must be part of an Azure AD managed tenant.
To enable this feature, your Windows login must be a member of the Azure AD managed tenant and
also be the system administrator for both Reporting Services and the database hosting the Reporting
Services databases. Using these administrative credentials, launch Reporting Services Configuration
Manager, click the Power BI Integration tab, click the Register With Power BI button, and provide your
Power BI login details.
Before you can pin a report to the dashboard, it must be configured to use stored credentials and SQL
Server Agent must be running because Power BI uses a Reporting Services subscription to manage
the scheduled refresh of the report. Furthermore, you can pin a report that contains only charts,
gauges, or maps that are not nested inside other report items. To pin a report meeting these
requirements, open the report in the web portal and click the Pin To Power BI Dashboard button in
the web portal toolbar. A sign-in dialog box is displayed in which you must supply your Power BI login
credentials. The first time you pin a report, another dialog box asks for permission to update your
Power BI app. Next, items in your report that are eligible for pinning are displayed in the browser.
Click the item, select a dashboard, and then choose an hourly, daily, or weekly frequency for updating
the report, as shown in Figure 7-19.
Figure 7-19: Selecting a dashboard for pinning a report.
A dialog box confirms the success or failure of the operation. If the pinning operation succeeds, you
can click a link in the dialog box to open a web browser window and view your dashboard in Power BI.
Your report shows as a tile in the dashboard, as shown in Figure 7-20, and will refresh periodically
according to the schedule you set. When you click the report tile in the dashboard, a new browser
window opens to display your report in the web portal from the report server from which it
CH A P TER 7 | Better reporting
Figure 7-20: Displaying a Reporting Services report as a report tile in a Power BI dashboard.
Managing subscriptions
Subscription functionality does not change in SQL Server 2016 in general. You still configure
subscriptions to deliver reports to named recipients or to a designated file share. However, there are a
few new subscription-management features that we explore in this section:
Subscription description You can include a subscription description when creating or changing
a subscription, which makes it easier to identify a specific subscription when many exist for a
single report.
Subscription owner change
change its owner.
Interface support for changing subscription status Whether you have one or many
subscriptions set up on the server, the web portal interface now includes Enable and Disable
buttons to quickly change the status of subscriptions.
File share credentials File share subscriptions have a new option to use administrator-defined
credentials to add files to a file share.
After adding a subscription to the report server, you can easily
Subscription description
The subscription definition page now includes a Subscription Properties section, as shown in Figure 721, that is displayed when you create or edit a subscription. You can use this description to distinguish
this subscription from others, which is helpful when you have several subscriptions associated with a
single report. For example, use this column to describe recipients, the schedule, the delivery type, and
other report delivery options so that you no longer have to edit the subscription to determine its
CH A P TER 7 | Better reporting
Figure 7-21: A portion of a subscription definition showing the new Subscription Properties section.
When you add a description to a subscription, the description is displayed in the web portal on the
Subscriptions page that you can access for a specific report or on the My Subscriptions page, where
you can see all reports for which you have created subscriptions, as shown in Figure 7-22. You can
sort subscriptions by the Description column by clicking the column header.
Figure 7-22: My Subscriptions page in the web portal with a new column for the subscription description.
Subscription owner change
By default, the user credentials are set as the owner of a subscription when a new subscription is
created and cannot be changed during subscription creation. In prior versions of Reporting Services, a
change of owner was possible only programmatically. Now you can edit a subscription in the web
portal to change its owner. This feature is particularly helpful when users change roles in an
organization. Both the current owner and the report server administrator have permissions to change
the owner when editing the subscription in the web portal.
This feature is available in both native and SharePoint-integrated modes.
Interface support for changing subscription status
In previous versions of Reporting Services, you can pause and resume a schedule to control when
related subscriptions are active. Now there are an Enable and a Disable button in the web portal
toolbar when you view subscriptions for an individual report or view the My Subscriptions page. This
capability allows you more fine-grained control over the execution of specific subscriptions. When you
disable a subscription, the icon to the left of the subscription displays a warning symbol and the
Status column value changes to Disabled, as shown in Figure 7-23.
Figure 7-23: My Subscriptions page in the web portal displaying a disabled report.
This feature is available in both native and SharePoint-integrated modes.
CH A P TER 7 | Better reporting
File share credentials
Rather than instructing users how to define credentials required to save a subscription to a file share,
report server administrators can configure the report server to use a single domain user account that
users can select when defining a file share subscription. To do this, open Reporting Services
Configuration Manager and access the new Subscription Settings page. You enable this feature by
selecting the Specify A File Share check box and adding a domain user account and password, as
shown in Figure 7-24.
Figure 7-24: Subscription Settings page in Reporting Services Configuration Manager.
This feature is available only in native mode.
When this feature is enabled, the user can choose to associate the configured file share account with a
subscription when setting the report delivery options for a file share subscription, as shown in Figure
7-25. Using this file share account is not required, however. The user can instead select Use The
Following Windows User Credentials and supply the domain user name and password.
Figure 7-25: The Use File Share Account option when configuring a file share subscription.
CH A P TER 7 | Better reporting
About the
Stacia Varga is a consultant, educator, mentor, and writer who has specialized in businessintelligence solutions since 1999. During that time she authored or coauthored several books about BI
as Stacia Misner. Her last book was Introducing Microsoft SQL Server 2014 (Microsoft Press, 2014). She
has also written articles for SQL Server Magazine and Technet and has produced multiple BI video
courses available through Pluralsight. In addition, she has been recognized for her contributions to
the technical community as a Microsoft Data Platform MVP since 2011. Stacia provides consulting and
custom education services through her company, Data Inspirations; speaks frequently at conferences
serving the SQL Server community worldwide; and serves as the chapter leader of her local PASS user
group, SQL Server Society of Las Vegas. She holds a BA in social sciences from Washington State
University. Stacia writes about her experiences with BI at and tweets as
Joseph D'Antoni is a principal consultant for Denny Cherry and Associates Consulting. He is well
versed in SQL Server performance tuning and database infrastructure design, with more than a
decade of experience working in both Fortune 500 and smaller firms. Joseph is a frequent speaker at
major technical events worldwide. In addition, he blogs about a variety of technology topics at and tweets as @jdanton. Joseph holds a BS in computer information systems from
Louisiana Tech and an MBA from North Carolina State University.
Denny Cherry is the owner, founder, and principal consultant for Denny Cherry and Associates
Consulting. His primary areas of focus are system architecture, performance tuning, and data
replication. Denny has been recognized in the technical community as a Microsoft Data Platform MVP,
VMware vExpert, and EMC Elect. He holds certifications for SQL Server from the MCDBA for SQL
Server 2000 up through Microsoft Certified Master for SQL Server 2008. He is also a Microsoft
Certified Trainer. Denny has written dozens of articles for SQL Server Magazine, Technet, and, among others. In addition, he has authored and coauthored multiple books,
including The Basics of Digital Privacy: Simple Tools to Protect Your Personal Information and Your
Identity Online (Syngress, 2013) and Securing SQL Server: Protecting Your Database from Attackers, 2nd
Edition (Syngress, 2012). Denny speaks at events worldwide, blogs at, and tweets
as @mrdenny.
Free ebooks
From technical overviews to drilldowns on special topics, get
free ebooks from Microsoft Press at:
Download your free ebooks in PDF, EPUB, and/or Mobi for
Kindle formats.
Look for other great resources at Microsoft Virtual Academy,
where you can learn new skills and help advance your career
with free Microsoft training delivered by experts.
Microsoft Press
Hear about
it first.
Get the latest news from Microsoft Press sent
to your inbox.
• New and upcoming books
• Special offers
• Free eBooks
• How-to articles
Sign up today at
Visit us today at
•Hundreds of titles available – Books, eBooks, and
online resources from industry experts
• Free U.S. shipping
•eBooks in multiple formats – Read on your computer,
tablet, mobile device, or e-reader
•Print & eBook Best Value Packs
•eBook Deal of the Week – Save
up to 60% on featured titles
•Newsletter and special offers
– Be the first to hear about new
releases, specials, and more
•Register your book – Get
additional benefits
Now that
read the
Tell us what you think!
Was it useful?
Did it teach you what you wanted to learn?
Was there room for improvement?
Let us know at
Your feedback goes directly to the staff at Microsoft Press,
and we read every one of your responses. Thanks in advance!