Akka Scala Documentation

Akka Scala Documentation
Release 2.5.0-RC1
Lightbend Inc
Mar 17, 2017
CONTENTS
1
2
3
4
5
Security Announcements
1.1 Receiving Security Advisories .
1.2 Reporting Vulnerabilities . . . .
1.3 Security Related Documentation
1.4 Fixed Security Vulnerabilities .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
1
1
1
1
Introduction
2.1 What is Akka? . . . . . . . . . . . .
2.2 Why Akka? . . . . . . . . . . . . . .
2.3 Getting Started . . . . . . . . . . . .
2.4 The Obligatory Hello World . . . . .
2.5 Use-case and Deployment Scenarios .
2.6 Examples of use-cases for Akka . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3
3
4
5
9
9
10
General
3.1 Terminology, Concepts . . . . . . . . .
3.2 Actor Systems . . . . . . . . . . . . .
3.3 What is an Actor? . . . . . . . . . . .
3.4 Supervision and Monitoring . . . . . .
3.5 Actor References, Paths and Addresses
3.6 Location Transparency . . . . . . . . .
3.7 Akka and the Java Memory Model . . .
3.8 Message Delivery Reliability . . . . .
3.9 Configuration . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
12
12
14
16
18
23
29
30
33
38
Actors
4.1 Actors . . . . . . . . . . . . .
4.2 Akka Typed . . . . . . . . . . .
4.3 Fault Tolerance . . . . . . . . .
4.4 Dispatchers . . . . . . . . . . .
4.5 Mailboxes . . . . . . . . . . .
4.6 Routing . . . . . . . . . . . . .
4.7 FSM . . . . . . . . . . . . . .
4.8 Persistence . . . . . . . . . . .
4.9 Persistence - Schema Evolution
4.10 Persistence Query . . . . . . .
4.11 Persistence Query for LevelDB
4.12 Testing Actor Systems . . . . .
4.13 Actor DSL . . . . . . . . . . .
4.14 Typed Actors . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
109
109
132
137
148
152
159
177
186
215
228
237
240
257
259
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Futures and Agents
266
5.1 Futures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
5.2 Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
i
6
7
8
9
Networking
6.1 Cluster Specification . . . . . . . . . .
6.2 Cluster Usage . . . . . . . . . . . . . .
6.3 Cluster Singleton . . . . . . . . . . . .
6.4 Distributed Publish Subscribe in Cluster
6.5 Cluster Client . . . . . . . . . . . . . .
6.6 Cluster Sharding . . . . . . . . . . . .
6.7 Cluster Metrics Extension . . . . . . .
6.8 Distributed Data . . . . . . . . . . . .
6.9 Remoting . . . . . . . . . . . . . . . .
6.10 Remoting (codename Artery) . . . . .
6.11 Serialization . . . . . . . . . . . . . .
6.12 I/O . . . . . . . . . . . . . . . . . . .
6.13 Using TCP . . . . . . . . . . . . . . .
6.14 Using UDP . . . . . . . . . . . . . . .
6.15 Camel . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
276
276
281
300
303
308
313
323
329
347
359
373
379
381
390
393
Utilities
7.1 Event Bus . . . .
7.2 Logging . . . . .
7.3 Scheduler . . . .
7.4 Duration . . . .
7.5 Circuit Breaker .
7.6 Akka Extensions
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
403
403
409
418
422
423
428
Streams
8.1 Introduction . . . . . . . . . . . . . . . . . .
8.2 Quick Start Guide . . . . . . . . . . . . . . .
8.3 Reactive Tweets . . . . . . . . . . . . . . . .
8.4 Design Principles behind Akka Streams . . . .
8.5 Basics and working with Flows . . . . . . . .
8.6 Working with Graphs . . . . . . . . . . . . . .
8.7 Modularity, Composition and Hierarchy . . . .
8.8 Buffers and working with rate . . . . . . . . .
8.9 Dynamic stream handling . . . . . . . . . . .
8.10 Custom stream processing . . . . . . . . . . .
8.11 Integration . . . . . . . . . . . . . . . . . . .
8.12 Error Handling . . . . . . . . . . . . . . . . .
8.13 Working with streaming IO . . . . . . . . . .
8.14 Pipelining and Parallelism . . . . . . . . . . .
8.15 Testing streams . . . . . . . . . . . . . . . . .
8.16 Overview of built-in stages and their semantics
8.17 Streams Cookbook . . . . . . . . . . . . . . .
8.18 Configuration . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
432
432
433
435
439
442
449
462
472
476
480
498
510
512
516
519
522
547
559
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Akka HTTP Documentation (Scala) moved!
10 HowTo: Common Patterns
10.1 Throttling Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.2 Balancing Workload Across Nodes . . . . . . . . . . . . . . . . . . . . . . . . .
10.3 Work Pulling Pattern to throttle and distribute work, and prevent mailbox overflow
10.4 Ordered Termination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.5 Akka AMQP Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.6 Shutdown Patterns in Akka 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.7 Distributed (in-memory) graph processing with Akka . . . . . . . . . . . . . . . .
10.8 Case Study: An Auto-Updating Cache Using Actors . . . . . . . . . . . . . . . .
10.9 Discovering message flows in actor systems with the Spider Pattern . . . . . . . .
10.10 Scheduling Periodic Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . .
562
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
563
563
563
563
563
564
564
564
564
564
565
ii
11 Information for Akka Developers
11.1 Building Akka . . . . . . . .
11.2 Multi JVM Testing . . . . . .
11.3 I/O Layer Design . . . . . . .
11.4 Developer Guidelines . . . .
11.5 Documentation Guidelines . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
566
566
568
571
572
573
12 Project Information
12.1 Migration Guides
12.2 Issue Tracking .
12.3 Licenses . . . .
12.4 Sponsors . . . .
12.5 Project . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
577
577
590
590
591
591
13 Additional Information
13.1 Binary Compatibility Rules . .
13.2 Modules marked “May Change”
13.3 Frequently Asked Questions . .
13.4 Books . . . . . . . . . . . . . .
13.5 Videos . . . . . . . . . . . . .
13.6 Akka in OSGi . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
593
593
595
624
627
627
627
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
iii
CHAPTER
ONE
SECURITY ANNOUNCEMENTS
1.1 Receiving Security Advisories
The best way to receive any and all security announcements is to subscribe to the Akka security list.
The mailing list is very low traffic, and receives notifications only after security reports have been managed by the
core team and fixes are publicly available.
1.2 Reporting Vulnerabilities
We strongly encourage people to report such problems to our private security mailing list first, before disclosing
them in a public forum.
Following best practice, we strongly encourage anyone to report potential security vulnerabilities to security@akka.io before disclosing them in a public forum like the mailing list or as a Github issue.
Reports to this email address will be handled by our security team, who will work together with you to ensure that
a fix can be provided without delay.
1.3 Security Related Documentation
• Disabling the Java Serializer
• Remote deployment whitelist
• Remote Security
1.4 Fixed Security Vulnerabilities
1.4.1 Java Serialization, Fixed in Akka 2.4.17
Date
10 Feburary 2017
Description of Vulnerability
An attacker that can connect to an ActorSystem exposed via Akka Remote over TCP can gain remote code
execution capabilities in the context of the JVM process that runs the ActorSystem if:
• JavaSerializer is enabled (default in Akka 2.4.x)
1
Akka Scala Documentation, Release 2.5.0-RC1
• and TLS is disabled or TLS is enabled with akka.remote.netty.ssl.security.
require-mutual-authentication = false (which is still the default in Akka 2.4.x)
• or if TLS is enabled with mutual authentication and the authentication keys of a host that is allowed to
connect have been compromised, an attacker gained access to a valid certificate (e.g. by compromising a
node with certificates issued by the same internal PKI tree to get access of the certificate)
• regardless of whether untrusted mode is enabled or not
Java deserialization is known to be vulnerable to attacks when attacker can provide arbitrary types.
Akka Remoting uses Java serialiser as default configuration which makes it vulnerable in its default form. The
documentation of how to disable Java serializer was not complete. The documentation of how to enable mutual
authentication was missing (only described in reference.conf).
To protect against such attacks the system should be updated to Akka 2.4.17 or later and be configured with
disabled Java serializer. Additional protection can be achieved when running in an untrusted network by enabling
TLS with mutual authentication.
Please subscribe to the akka-security mailing list to be notified promptly about future security issues.
Severity
The
CVSS
score
of
this
vulnerability
AV:A/AC:M/Au:N/C:C/I:C/A:C/E:F/RL:TF/RC:C.
is
6.8
(Medium),
based
on
vector
Rationale for the score:
• AV:A - Best practice is that Akka remoting nodes should only be accessible from the adjacent network, so
in good setups, this will be adjacent.
• AC:M - Any one in the adjacent network can launch the attack with non-special access privileges.
• C:C, I:C, A:C - Remote Code Execution vulnerabilities are by definition CIA:C.
Affected Versions
• Akka 2.4.16 and prior
• Akka 2.5-M1 (milestone not intended for production)
Fixed Versions
We have prepared patches for the affected versions, and have released the following versions which resolve the
issue:
• Akka 2.4.17 (Scala 2.11, 2.12)
Binary and source compatibility has been maintained for the patched releases so the upgrade procedure is as
simple as changing the library dependency.
It will also be fixed in 2.5-M2 or 2.5.0-RC1.
Acknowledgements
We would like to thank Alvaro Munoz at Hewlett Packard Enterprise Security & Adrian Bravo at Workday for
their thorough investigation and bringing this issue to our attention.
1.4. Fixed Security Vulnerabilities
2
CHAPTER
TWO
INTRODUCTION
2.1 What is Akka?
«resilient elastic distributed real-time transaction processing»
We believe that writing correct distributed, concurrent, fault-tolerant and scalable applications is too hard. Most
of the time it’s because we are using the wrong tools and the wrong level of abstraction. Akka is here to change
that. Using the Actor Model we raise the abstraction level and provide a better platform to build scalable, resilient
and responsive applications—see the Reactive Manifesto for more details. For fault-tolerance we adopt the “let
it crash” model which the telecom industry has used with great success to build applications that self-heal and
systems that never stop. Actors also provide the abstraction for transparent distribution and the basis for truly
scalable and fault-tolerant applications.
Akka is Open Source and available under the Apache 2 License.
Download from http://akka.io/downloads.
Please note that all code samples compile, so if you want direct access to the sources, have a look over at the Akka
Docs subproject on github: for Java and Scala.
2.1.1 Akka implements a unique hybrid
Actors
Actors give you:
• Simple and high-level abstractions for distribution, concurrency and parallelism.
• Asynchronous, non-blocking and highly performant message-driven programming model.
• Very lightweight event-driven processes (several million actors per GB of heap memory).
See the chapter for Scala or Java.
Fault Tolerance
• Supervisor hierarchies with “let-it-crash” semantics.
• Actor systems can span over multiple JVMs to provide truly fault-tolerant systems.
• Excellent for writing highly fault-tolerant systems that self-heal and never stop.
See Fault Tolerance (Scala) and Fault Tolerance (Java).
3
Akka Scala Documentation, Release 2.5.0-RC1
Location Transparency
Everything in Akka is designed to work in a distributed environment: all interactions of actors use pure message
passing and everything is asynchronous.
For an overview of the cluster support see the Java and Scala documentation chapters.
Persistence
State changes experienced by an actor can optionally be persisted and replayed when the actor is started or
restarted. This allows actors to recover their state, even after JVM crashes or when being migrated to another
node.
You can find more details in the respective chapter for Java or Scala.
2.1.2 Scala and Java APIs
Akka has both a Scala Documentation and a java-api.
2.1.3 Akka can be used in different ways
Akka is a toolkit, not a framework: you integrate it into your build like any other library without having to follow
a particular source code layout. When expressing your systems as collaborating Actors you may feel pushed more
towards proper encapsulation of internal state, you may find that there is a natural separation between business
logic and inter-component communication.
Akka applications are typically deployed as follows:
• as a library: used as a regular JAR on the classpath or in a web app.
• packaged with sbt-native-packager.
• packaged and deployed using Lightbend ConductR.
2.1.4 Commercial Support
Akka is available from Lightbend Inc. under a commercial license which includes development or production
support, read more here.
2.2 Why Akka?
2.2.1 What features can the Akka platform offer, over the competition?
Akka provides scalable real-time transaction processing.
Akka is a unified runtime and programming model for:
• Scale up (Concurrency)
• Scale out (Remoting)
• Fault tolerance
One thing to learn and admin, with high cohesion and coherent semantics.
Akka is a very scalable piece of software, not only in the context of performance but also in the size of applications
it is useful for. The core of Akka, akka-actor, is very small and easily dropped into an existing project where you
need asynchronicity and lockless concurrency without hassle.
2.2. Why Akka?
4
Akka Scala Documentation, Release 2.5.0-RC1
You can choose to include only the parts of Akka you need in your application. With CPUs growing more and
more cores every cycle, Akka is the alternative that provides outstanding performance even if you’re only running
it on one machine. Akka also supplies a wide array of concurrency-paradigms, allowing users to choose the right
tool for the job.
2.2.2 What’s a good use-case for Akka?
We see Akka being adopted by many large organizations in a big range of industries:
• Investment and Merchant Banking
• Retail
• Social Media
• Simulation
• Gaming and Betting
• Automobile and Traffic Systems
• Health Care
• Data Analytics
and much more. Any system with the need for high-throughput and low latency is a good candidate for using
Akka.
Actors let you manage service failures (Supervisors), load management (back-off strategies, timeouts and
processing-isolation), as well as both horizontal and vertical scalability (add more cores and/or add more machines).
Here’s what some of the Akka users have to say about how they are using Akka: http://stackoverflow.com/
questions/4493001/good-use-case-for-akka
All this in the ApacheV2-licensed open source project.
2.3 Getting Started
2.3.1 Prerequisites
Akka requires that you have Java 8 or later installed on your machine.
Lightbend Inc. provides a commercial build of Akka and related projects such as Scala or Play as part of the
Lightbend Reactive Platform which is made available for Java 6 in case your project can not upgrade to Java 8 just
yet. It also includes additional commercial features or libraries.
2.3.2 Getting Started Guides and Template Projects
The best way to start learning Akka is to download Lightbend Activator and try out one of Akka Template Projects.
2.3.3 Download
There are several ways to download Akka. You can download it as part of the Lightbend Platform (as described
above). You can download the full distribution, which includes all modules. Or you can use a build tool like
Maven or SBT to download dependencies from the Akka Maven repository.
2.3. Getting Started
5
Akka Scala Documentation, Release 2.5.0-RC1
2.3.4 Modules
Akka is very modular and consists of several JARs containing different features.
• akka-actor – Classic Actors, Typed Actors, IO Actor etc.
• akka-agent – Agents, integrated with Scala STM
• akka-camel – Apache Camel integration
• akka-cluster – Cluster membership management, elastic routers.
• akka-cluster-sharding – Cluster Sharding of actors.
• akka-cluster-tools – Additoinal Cluster utilities, such as Singleton, Pub/Sub and Client.
• akka-distributed-data – Cluster data with CRDTs.
• akka-osgi – Utilities for using Akka in OSGi containers
• akka-osgi-aries – Aries blueprint for provisioning actor systems
• akka-remote – Remote Actors
• akka-slf4j – SLF4J Logger (event bus listener)
• akka-stream – Reactive stream processing
• akka-testkit – Toolkit for testing Actor systems
In addition to these stable modules there are several which are on their way into the stable core but are still marked
may change at this point. This does not mean that they do not function as intended, it primarily means that their
API has not yet solidified enough in order to be considered frozen. You can help accelerating this process by
giving feedback on these modules on our mailing list.
• akka-contrib – an assortment of contributions which may or may not be moved into core modules, see
External Contributions for more details.
The filename of the actual JAR is for example akka-actor_2.11-2.5.0-RC1.jar (and analog for the
other modules).
How to see the JARs dependencies of each Akka module is described in the Dependencies section.
2.3.5 Using a release distribution
Download the release you need from http://akka.io/downloads and unzip it.
2.3.6 Using a snapshot version
The Akka nightly snapshots are published to http://repo.akka.io/snapshots/ and are versioned with both
SNAPSHOT and timestamps. You can choose a timestamped version to work with and can decide when to update
to a newer version.
Warning: The use of Akka SNAPSHOTs, nightlies and milestone releases is discouraged unless you know
what you are doing.
2.3.7 Using a build tool
Akka can be used with build tools that support Maven repositories.
2.3. Getting Started
6
Akka Scala Documentation, Release 2.5.0-RC1
2.3.8 Maven repositories
For Akka version 2.1-M2 and onwards:
Maven Central
For previous Akka versions:
Akka Repo
2.3.9 Using Akka with Maven
The simplest way to get started with Akka and Maven is to check out the Lightbend Activator tutorial named Akka
Main in Java.
Since Akka is published to Maven Central (for versions since 2.1-M2), it is enough to add the Akka dependencies
to the POM. For example, here is the dependency for akka-actor:
<dependency>
<groupId>com.typesafe.akka</groupId>
<artifactId>akka-actor_2.11</artifactId>
<version>2.5.0-RC1</version>
</dependency>
For snapshot versions, the snapshot repository needs to be added as well:
<repositories>
<repository>
<id>akka-snapshots</id>
<snapshots>
<enabled>true</enabled>
</snapshots>
<url>http://repo.akka.io/snapshots/</url>
</repository>
</repositories>
Note: for snapshot versions both SNAPSHOT and timestamped versions are published.
2.3.10 Using Akka with SBT
The simplest way to get started with Akka and SBT is to use Lightbend Activator with one of the SBT templates.
Summary of the essential parts for using Akka with SBT:
SBT installation instructions on http://www.scala-sbt.org/release/tutorial/Setup.html
build.sbt file:
name := "My Project"
version := "1.0"
scalaVersion := "2.11.8"
libraryDependencies +=
"com.typesafe.akka" %% "akka-actor" % "2.5.0-RC1"
Note: the libraryDependencies setting above is specific to SBT v0.12.x and higher. If you are using an older
version of SBT, the libraryDependencies should look like this:
libraryDependencies +=
"com.typesafe.akka" % "akka-actor_2.11" % "2.5.0-RC1"
2.3. Getting Started
7
Akka Scala Documentation, Release 2.5.0-RC1
For snapshot versions, the snapshot repository needs to be added as well:
resolvers += "Akka Snapshot Repository" at "http://repo.akka.io/snapshots/"
2.3.11 Using Akka with Gradle
Requires at least Gradle 1.4 Uses the Scala plugin
apply plugin: 'scala'
repositories {
mavenCentral()
}
dependencies {
compile 'org.scala-lang:scala-library:2.11.8'
}
tasks.withType(ScalaCompile) {
scalaCompileOptions.useAnt = false
}
dependencies {
compile group: 'com.typesafe.akka', name: 'akka-actor_2.11', version: '2.5.0-RC1'
compile group: 'org.scala-lang', name: 'scala-library', version: '2.11.8'
}
For snapshot versions, the snapshot repository needs to be added as well:
repositories {
mavenCentral()
maven {
url "http://repo.akka.io/snapshots/"
}
}
2.3.12 Using Akka with Eclipse
Setup SBT project and then use sbteclipse to generate an Eclipse project.
2.3.13 Using Akka with IntelliJ IDEA
Setup SBT project and then use sbt-idea to generate an IntelliJ IDEA project.
2.3.14 Using Akka with NetBeans
Setup SBT project and then use nbsbt to generate a NetBeans project.
You should also use nbscala for general scala support in the IDE.
2.3.15 Do not use -optimize Scala compiler flag
Warning: Akka has not been compiled or tested with -optimize Scala compiler flag. Strange behavior has
been reported by users that have tried it.
2.3. Getting Started
8
Akka Scala Documentation, Release 2.5.0-RC1
2.3.16 Build from sources
Akka uses Git and is hosted at Github.
• Akka: clone the Akka repository from https://github.com/akka/akka
Continue reading the page on Building Akka
2.3.17 Need help?
If you have questions you can get help on the Akka Mailing List.
You can also ask for commercial support.
Thanks for being a part of the Akka community.
2.4 The Obligatory Hello World
The actor based version of the tough problem of printing a well-known greeting to the console is introduced in a
Lightbend Activator tutorial named Akka Main in Scala.
The tutorial illustrates the generic launcher class akka.Main which expects only one command line argument:
the class name of the application’s main actor. This main method will then create the infrastructure needed for
running the actors, start the given main actor and arrange for the whole application to shut down once the main
actor terminates.
There is also another Lightbend Activator tutorial in the same problem domain that is named Hello Akka!. It
describes the basics of Akka in more depth.
2.5 Use-case and Deployment Scenarios
2.5.1 How can I use and deploy Akka?
Akka can be used in different ways:
• As a library: used as a regular JAR on the classpath and/or in a web app, to be put into WEB-INF/lib
• As an application packaged with sbt-native-packager
• As an application packaged and deployed using Lightbend ConductR.
2.5.2 Native Packager
sbt-native-packager is a tool for creating distributions of any type of application, including Akka applications.
Define sbt version in project/build.properties file:
sbt.version=0.13.13
Add sbt-native-packager in project/plugins.sbt file:
addSbtPlugin("com.typesafe.sbt" % "sbt-native-packager" % "1.1.5")
Follow the instructions for the JavaAppPackaging in the sbt-native-packager plugin documentation.
2.4. The Obligatory Hello World
9
Akka Scala Documentation, Release 2.5.0-RC1
2.5.3 In a Docker container
You can use both Akka remoting and Akka Cluster inside of Docker containers. But note that you will need to
take special care with the network configuration when using Docker, described here: Akka behind NAT or in a
Docker container
For an example of how to set up a project using Akka Cluster and Docker take a look at the “akka-docker-cluster”
activator template.
2.6 Examples of use-cases for Akka
We see Akka being adopted by many large organizations in a big range of industries all from investment and
merchant banking, retail and social media, simulation, gaming and betting, automobile and traffic systems, health
care, data analytics and much more. Any system that have the need for high-throughput and low latency is a good
candidate for using Akka.
There is a great discussion on use-cases for Akka with some good write-ups by production users here
2.6.1 Here are some of the areas where Akka is being deployed into production
Transaction processing (Online Gaming, Finance/Banking, Trading, Statistics, Betting, Social
Media, Telecom)
Scale up, scale out, fault-tolerance / HA
Service backend (any industry, any app)
Service REST, SOAP, Cometd, WebSockets etc Act as message hub / integration layer Scale up, scale
out, fault-tolerance / HA
Concurrency/parallelism (any app)
Correct Simple to work with and understand Just add the jars to your existing JVM project (use Scala,
Java, Groovy or JRuby)
Simulation
Master/Worker, Compute Grid, MapReduce etc.
Batch processing (any industry)
Camel integration to hook up with batch data sources Actors divide and conquer the batch workloads
Communications Hub (Telecom, Web media, Mobile media)
Scale up, scale out, fault-tolerance / HA
Gaming and Betting (MOM, online gaming, betting)
Scale up, scale out, fault-tolerance / HA
2.6. Examples of use-cases for Akka
10
Akka Scala Documentation, Release 2.5.0-RC1
Business Intelligence/Data Mining/general purpose crunching
Scale up, scale out, fault-tolerance / HA
Complex Event Stream Processing
Scale up, scale out, fault-tolerance / HA
2.6. Examples of use-cases for Akka
11
CHAPTER
THREE
GENERAL
3.1 Terminology, Concepts
In this chapter we attempt to establish a common terminology to define a solid ground for communicating about
concurrent, distributed systems which Akka targets. Please note that, for many of these terms, there is no single agreed definition. We simply seek to give working definitions that will be used in the scope of the Akka
documentation.
3.1.1 Concurrency vs. Parallelism
Concurrency and parallelism are related concepts, but there are small differences. Concurrency means that two or
more tasks are making progress even though they might not be executing simultaneously. This can for example
be realized with time slicing where parts of tasks are executed sequentially and mixed with parts of other tasks.
Parallelism on the other hand arise when the execution can be truly simultaneous.
3.1.2 Asynchronous vs. Synchronous
A method call is considered synchronous if the caller cannot make progress until the method returns a value or
throws an exception. On the other hand, an asynchronous call allows the caller to progress after a finite number of
steps, and the completion of the method may be signalled via some additional mechanism (it might be a registered
callback, a Future, or a message).
A synchronous API may use blocking to implement synchrony, but this is not a necessity. A very CPU intensive
task might give a similar behavior as blocking. In general, it is preferred to use asynchronous APIs, as they
guarantee that the system is able to progress. Actors are asynchronous by nature: an actor can progress after a
message send without waiting for the actual delivery to happen.
3.1.3 Non-blocking vs. Blocking
We talk about blocking if the delay of one thread can indefinitely delay some of the other threads. A good example
is a resource which can be used exclusively by one thread using mutual exclusion. If a thread holds on to the
resource indefinitely (for example accidentally running an infinite loop) other threads waiting on the resource can
not progress. In contrast, non-blocking means that no thread is able to indefinitely delay others.
Non-blocking operations are preferred to blocking ones, as the overall progress of the system is not trivially
guaranteed when it contains blocking operations.
3.1.4 Deadlock vs. Starvation vs. Live-lock
Deadlock arises when several participants are waiting on each other to reach a specific state to be able to progress.
As none of them can progress without some other participant to reach a certain state (a “Catch-22” problem) all
affected subsystems stall. Deadlock is closely related to blocking, as it is necessary that a participant thread be
able to delay the progression of other threads indefinitely.
12
Akka Scala Documentation, Release 2.5.0-RC1
In the case of deadlock, no participants can make progress, while in contrast Starvation happens, when there are
participants that can make progress, but there might be one or more that cannot. Typical scenario is the case
of a naive scheduling algorithm that always selects high-priority tasks over low-priority ones. If the number of
incoming high-priority tasks is constantly high enough, no low-priority ones will be ever finished.
Livelock is similar to deadlock as none of the participants make progress. The difference though is that instead
of being frozen in a state of waiting for others to progress, the participants continuously change their state. An
example scenario when two participants have two identical resources available. They each try to get the resource,
but they also check if the other needs the resource, too. If the resource is requested by the other participant, they
try to get the other instance of the resource. In the unfortunate case it might happen that the two participants
“bounce” between the two resources, never acquiring it, but always yielding to the other.
3.1.5 Race Condition
We call it a Race condition when an assumption about the ordering of a set of events might be violated by external
non-deterministic effects. Race conditions often arise when multiple threads have a shared mutable state, and the
operations of thread on the state might be interleaved causing unexpected behavior. While this is a common case,
shared state is not necessary to have race conditions. One example could be a client sending unordered packets
(e.g UDP datagrams) P1, P2 to a server. As the packets might potentially travel via different network routes, it
is possible that the server receives P2 first and P1 afterwards. If the messages contain no information about their
sending order it is impossible to determine by the server that they were sent in a different order. Depending on the
meaning of the packets this can cause race conditions.
Note: The only guarantee that Akka provides about messages sent between a given pair of actors is that their
order is always preserved. see Message Delivery Reliability
3.1.6 Non-blocking Guarantees (Progress Conditions)
As discussed in the previous sections blocking is undesirable for several reasons, including the dangers of deadlocks and reduced throughput in the system. In the following sections we discuss various non-blocking properties
with different strength.
Wait-freedom
A method is wait-free if every call is guaranteed to finish in a finite number of steps. If a method is bounded
wait-free then the number of steps has a finite upper bound.
From this definition it follows that wait-free methods are never blocking, therefore deadlock can not happen.
Additionally, as each participant can progress after a finite number of steps (when the call finishes), wait-free
methods are free of starvation.
Lock-freedom
Lock-freedom is a weaker property than wait-freedom. In the case of lock-free calls, infinitely often some method
finishes in a finite number of steps. This definition implies that no deadlock is possible for lock-free calls. On the
other hand, the guarantee that some call finishes in a finite number of steps is not enough to guarantee that all of
them eventually finish. In other words, lock-freedom is not enough to guarantee the lack of starvation.
Obstruction-freedom
Obstruction-freedom is the weakest non-blocking guarantee discussed here. A method is called obstruction-free if
there is a point in time after which it executes in isolation (other threads make no steps, e.g.: become suspended),
it finishes in a bounded number of steps. All lock-free objects are obstruction-free, but the opposite is generally
not true.
3.1. Terminology, Concepts
13
Akka Scala Documentation, Release 2.5.0-RC1
Optimistic concurrency control (OCC) methods are usually obstruction-free. The OCC approach is that every
participant tries to execute its operation on the shared object, but if a participant detects conflicts from others, it
rolls back the modifications, and tries again according to some schedule. If there is a point in time, where one of
the participants is the only one trying, the operation will succeed.
3.1.7 Recommended literature
• The Art of Multiprocessor Programming, M. Herlihy and N Shavit, 2008. ISBN 978-0123705914
• Java Concurrency in Practice, B. Goetz, T. Peierls, J. Bloch, J. Bowbeer, D. Holmes and D. Lea, 2006.
ISBN 978-0321349606
3.2 Actor Systems
Actors are objects which encapsulate state and behavior, they communicate exclusively by exchanging messages
which are placed into the recipient’s mailbox. In a sense, actors are the most stringent form of object-oriented
programming, but it serves better to view them as persons: while modeling a solution with actors, envision a group
of people and assign sub-tasks to them, arrange their functions into an organizational structure and think about
how to escalate failure (all with the benefit of not actually dealing with people, which means that we need not
concern ourselves with their emotional state or moral issues). The result can then serve as a mental scaffolding for
building the software implementation.
Note: An ActorSystem is a heavyweight structure that will allocate 1. . . N Threads, so create one per logical
application.
3.2.1 Hierarchical Structure
Like in an economic organization, actors naturally form hierarchies. One actor, which is to oversee a certain
function in the program might want to split up its task into smaller, more manageable pieces. For this purpose it
starts child actors which it supervises. While the details of supervision are explained here, we shall concentrate on
the underlying concepts in this section. The only prerequisite is to know that each actor has exactly one supervisor,
which is the actor that created it.
The quintessential feature of actor systems is that tasks are split up and delegated until they become small enough
to be handled in one piece. In doing so, not only is the task itself clearly structured, but the resulting actors can
be reasoned about in terms of which messages they should process, how they should react normally and how
failure should be handled. If one actor does not have the means for dealing with a certain situation, it sends a
corresponding failure message to its supervisor, asking for help. The recursive structure then allows to handle
failure at the right level.
Compare this to layered software design which easily devolves into defensive programming with the aim of not
leaking any failure out: if the problem is communicated to the right person, a better solution can be found than if
trying to keep everything “under the carpet”.
Now, the difficulty in designing such a system is how to decide who should supervise what. There is of course no
single best solution, but there are a few guidelines which might be helpful:
• If one actor manages the work another actor is doing, e.g. by passing on sub-tasks, then the manager should
supervise the child. The reason is that the manager knows which kind of failures are expected and how to
handle them.
• If one actor carries very important data (i.e. its state shall not be lost if avoidable), this actor should source
out any possibly dangerous sub-tasks to children it supervises and handle failures of these children as appropriate. Depending on the nature of the requests, it may be best to create a new child for each request,
which simplifies state management for collecting the replies. This is known as the “Error Kernel Pattern”
from Erlang.
3.2. Actor Systems
14
Akka Scala Documentation, Release 2.5.0-RC1
• If one actor depends on another actor for carrying out its duty, it should watch that other actor’s liveness
and act upon receiving a termination notice. This is different from supervision, as the watching party has
no influence on the supervisor strategy, and it should be noted that a functional dependency alone is not a
criterion for deciding where to place a certain child actor in the hierarchy.
There are of course always exceptions to these rules, but no matter whether you follow the rules or break them,
you should always have a reason.
3.2.2 Configuration Container
The actor system as a collaborating ensemble of actors is the natural unit for managing shared facilities like
scheduling services, configuration, logging, etc. Several actor systems with different configuration may co-exist
within the same JVM without problems, there is no global shared state within Akka itself. Couple this with the
transparent communication between actor systems—within one node or across a network connection—to see that
actor systems themselves can be used as building blocks in a functional hierarchy.
3.2.3 Actor Best Practices
1. Actors should be like nice co-workers: do their job efficiently without bothering everyone else needlessly
and avoid hogging resources. Translated to programming this means to process events and generate responses (or more requests) in an event-driven manner. Actors should not block (i.e. passively wait while
occupying a Thread) on some external entity—which might be a lock, a network socket, etc.—unless it is
unavoidable; in the latter case see below.
2. Do not pass mutable objects between actors. In order to ensure that, prefer immutable messages. If the
encapsulation of actors is broken by exposing their mutable state to the outside, you are back in normal Java
concurrency land with all the drawbacks.
3. Actors are made to be containers for behavior and state, embracing this means to not routinely send behavior
within messages (which may be tempting using Scala closures). One of the risks is to accidentally share
mutable state between actors, and this violation of the actor model unfortunately breaks all the properties
which make programming in actors such a nice experience.
4. Top-level actors are the innermost part of your Error Kernel, so create them sparingly and prefer truly
hierarchical systems. This has benefits with respect to fault-handling (both considering the granularity of
configuration and the performance) and it also reduces the strain on the guardian actor, which is a single
point of contention if over-used.
3.2.4 Blocking Needs Careful Management
In some cases it is unavoidable to do blocking operations, i.e. to put a thread to sleep for an indeterminate
time, waiting for an external event to occur. Examples are legacy RDBMS drivers or messaging APIs, and the
underlying reason is typically that (network) I/O occurs under the covers. When facing this, you may be tempted
to just wrap the blocking call inside a Future and work with that instead, but this strategy is too simple: you are
quite likely to find bottlenecks or run out of memory or threads when the application runs under increased load.
The non-exhaustive list of adequate solutions to the “blocking problem” includes the following suggestions:
• Do the blocking call within an actor (or a set of actors managed by a router [Java, Scala]), making sure to
configure a thread pool which is either dedicated for this purpose or sufficiently sized.
• Do the blocking call within a Future, ensuring an upper bound on the number of such calls at any point in
time (submitting an unbounded number of tasks of this nature will exhaust your memory or thread limits).
• Do the blocking call within a Future, providing a thread pool with an upper limit on the number of threads
which is appropriate for the hardware on which the application runs.
• Dedicate a single thread to manage a set of blocking resources (e.g. a NIO selector driving multiple channels) and dispatch events as they occur as actor messages.
3.2. Actor Systems
15
Akka Scala Documentation, Release 2.5.0-RC1
The first possibility is especially well-suited for resources which are single-threaded in nature, like database handles which traditionally can only execute one outstanding query at a time and use internal synchronization to
ensure this. A common pattern is to create a router for N actors, each of which wraps a single DB connection and
handles queries as sent to the router. The number N must then be tuned for maximum throughput, which will vary
depending on which DBMS is deployed on what hardware.
Note: Configuring thread pools is a task best delegated to Akka, simply configure in the application.conf
and instantiate through an ActorSystem [Java, Scala]
3.2.5 What you should not concern yourself with
An actor system manages the resources it is configured to use in order to run the actors which it contains. There
may be millions of actors within one such system, after all the mantra is to view them as abundant and they
weigh in at an overhead of only roughly 300 bytes per instance. Naturally, the exact order in which messages are
processed in large systems is not controllable by the application author, but this is also not intended. Take a step
back and relax while Akka does the heavy lifting under the hood.
3.3 What is an Actor?
The previous section about Actor Systems explained how actors form hierarchies and are the smallest unit when
building an application. This section looks at one such actor in isolation, explaining the concepts you encounter
while implementing it. For a more in depth reference with all the details please refer to Actors (Scala) and Actors
(Java).
An actor is a container for State, Behavior, a Mailbox, Child Actors and a Supervisor Strategy. All of this is
encapsulated behind an Actor Reference. One noteworthy aspect is that actors have an explicit lifecycle, they are
not automatically destroyed when no longer referenced; after having created one, it is your responsibility to make
sure that it will eventually be terminated as well—which also gives you control over how resources are released
When an Actor Terminates.
3.3.1 Actor Reference
As detailed below, an actor object needs to be shielded from the outside in order to benefit from the actor model.
Therefore, actors are represented to the outside using actor references, which are objects that can be passed around
freely and without restriction. This split into inner and outer object enables transparency for all the desired
operations: restarting an actor without needing to update references elsewhere, placing the actual actor object on
remote hosts, sending messages to actors in completely different applications. But the most important aspect is
that it is not possible to look inside an actor and get hold of its state from the outside, unless the actor unwisely
publishes this information itself.
3.3.2 State
Actor objects will typically contain some variables which reflect possible states the actor may be in. This can be
an explicit state machine (e.g. using the FSM module), or it could be a counter, set of listeners, pending requests,
etc. These data are what make an actor valuable, and they must be protected from corruption by other actors. The
good news is that Akka actors conceptually each have their own light-weight thread, which is completely shielded
from the rest of the system. This means that instead of having to synchronize access using locks you can just write
your actor code without worrying about concurrency at all.
Behind the scenes Akka will run sets of actors on sets of real threads, where typically many actors share one
thread, and subsequent invocations of one actor may end up being processed on different threads. Akka ensures
that this implementation detail does not affect the single-threadedness of handling the actor’s state.
3.3. What is an Actor?
16
Akka Scala Documentation, Release 2.5.0-RC1
Because the internal state is vital to an actor’s operations, having inconsistent state is fatal. Thus, when the actor
fails and is restarted by its supervisor, the state will be created from scratch, like upon first creating the actor. This
is to enable the ability of self-healing of the system.
Optionally, an actor’s state can be automatically recovered to the state before a restart by persisting received
messages and replaying them after restart (see Persistence).
3.3.3 Behavior
Every time a message is processed, it is matched against the current behavior of the actor. Behavior means a
function which defines the actions to be taken in reaction to the message at that point in time, say forward a
request if the client is authorized, deny it otherwise. This behavior may change over time, e.g. because different
clients obtain authorization over time, or because the actor may go into an “out-of-service” mode and later come
back. These changes are achieved by either encoding them in state variables which are read from the behavior
logic, or the function itself may be swapped out at runtime, see the become and unbecome operations. However,
the initial behavior defined during construction of the actor object is special in the sense that a restart of the actor
will reset its behavior to this initial one.
3.3.4 Mailbox
An actor’s purpose is the processing of messages, and these messages were sent to the actor from other actors (or
from outside the actor system). The piece which connects sender and receiver is the actor’s mailbox: each actor
has exactly one mailbox to which all senders enqueue their messages. Enqueuing happens in the time-order of
send operations, which means that messages sent from different actors may not have a defined order at runtime
due to the apparent randomness of distributing actors across threads. Sending multiple messages to the same target
from the same actor, on the other hand, will enqueue them in the same order.
There are different mailbox implementations to choose from, the default being a FIFO: the order of the messages
processed by the actor matches the order in which they were enqueued. This is usually a good default, but
applications may need to prioritize some messages over others. In this case, a priority mailbox will enqueue not
always at the end but at a position as given by the message priority, which might even be at the front. While using
such a queue, the order of messages processed will naturally be defined by the queue’s algorithm and in general
not be FIFO.
An important feature in which Akka differs from some other actor model implementations is that the current
behavior must always handle the next dequeued message, there is no scanning the mailbox for the next matching
one. Failure to handle a message will typically be treated as a failure, unless this behavior is overridden.
3.3.5 Child Actors
Each actor is potentially a supervisor: if it creates children for delegating sub-tasks, it will automatically supervise
them. The list of children is maintained within the actor’s context and the actor has access to it. Modifications to
the list are done by creating (context.actorOf(...)) or stopping (context.stop(child)) children
and these actions are reflected immediately. The actual creation and termination actions happen behind the scenes
in an asynchronous way, so they do not “block” their supervisor.
3.3.6 Supervisor Strategy
The final piece of an actor is its strategy for handling faults of its children. Fault handling is then done transparently
by Akka, applying one of the strategies described in Supervision and Monitoring for each incoming failure. As
this strategy is fundamental to how an actor system is structured, it cannot be changed once an actor has been
created.
Considering that there is only one such strategy for each actor, this means that if different strategies apply to
the various children of an actor, the children should be grouped beneath intermediate supervisors with matching
strategies, preferring once more the structuring of actor systems according to the splitting of tasks into sub-tasks.
3.3. What is an Actor?
17
Akka Scala Documentation, Release 2.5.0-RC1
3.3.7 When an Actor Terminates
Once an actor terminates, i.e. fails in a way which is not handled by a restart, stops itself or is stopped by its
supervisor, it will free up its resources, draining all remaining messages from its mailbox into the system’s “dead
letter mailbox” which will forward them to the EventStream as DeadLetters. The mailbox is then replaced within
the actor reference with a system mailbox, redirecting all new messages to the EventStream as DeadLetters. This
is done on a best effort basis, though, so do not rely on it in order to construct “guaranteed delivery”.
The reason for not just silently dumping the messages was inspired by our tests: we register the TestEventListener on the event bus to which the dead letters are forwarded, and that will log a warning for every dead letter
received—this has been very helpful for deciphering test failures more quickly. It is conceivable that this feature
may also be of use for other purposes.
3.4 Supervision and Monitoring
This chapter outlines the concept behind supervision, the primitives offered and their semantics. For details on
how that translates into real code, please refer to the corresponding chapters for Scala and Java APIs.
3.4.1 What Supervision Means
As described in Actor Systems supervision describes a dependency relationship between actors: the supervisor
delegates tasks to subordinates and therefore must respond to their failures. When a subordinate detects a failure
(i.e. throws an exception), it suspends itself and all its subordinates and sends a message to its supervisor, signaling
failure. Depending on the nature of the work to be supervised and the nature of the failure, the supervisor has a
choice of the following four options:
1. Resume the subordinate, keeping its accumulated internal state
2. Restart the subordinate, clearing out its accumulated internal state
3. Stop the subordinate permanently
4. Escalate the failure, thereby failing itself
It is important to always view an actor as part of a supervision hierarchy, which explains the existence of the fourth
choice (as a supervisor also is subordinate to another supervisor higher up) and has implications on the first three:
resuming an actor resumes all its subordinates, restarting an actor entails restarting all its subordinates (but see
below for more details), similarly terminating an actor will also terminate all its subordinates. It should be noted
that the default behavior of the preRestart hook of the Actor class is to terminate all its children before
restarting, but this hook can be overridden; the recursive restart applies to all children left after this hook has been
executed.
Each supervisor is configured with a function translating all possible failure causes (i.e. exceptions) into one of
the four choices given above; notably, this function does not take the failed actor’s identity as an input. It is quite
easy to come up with examples of structures where this might not seem flexible enough, e.g. wishing for different
strategies to be applied to different subordinates. At this point it is vital to understand that supervision is about
forming a recursive fault handling structure. If you try to do too much at one level, it will become hard to reason
about, hence the recommended way in this case is to add a level of supervision.
Akka implements a specific form called “parental supervision”. Actors can only be created by other actors—where
the top-level actor is provided by the library—and each created actor is supervised by its parent. This restriction
makes the formation of actor supervision hierarchies implicit and encourages sound design decisions. It should
be noted that this also guarantees that actors cannot be orphaned or attached to supervisors from the outside,
which might otherwise catch them unawares. In addition, this yields a natural and clean shutdown procedure for
(sub-trees of) actor applications.
3.4. Supervision and Monitoring
18
Akka Scala Documentation, Release 2.5.0-RC1
Warning: Supervision related parent-child communication happens by special system messages that have
their own mailboxes separate from user messages. This implies that supervision related events are not deterministically ordered relative to ordinary messages. In general, the user cannot influence the order of normal
messages and failure notifications. For details and example see the Discussion: Message Ordering section.
3.4.2 The Top-Level Supervisors
An actor system will during its creation start at least three actors, shown in the image above. For more information
about the consequences for actor paths see Top-Level Scopes for Actor Paths.
/user: The Guardian Actor
The actor which is probably most interacted with is the parent of all user-created actors, the guardian named
"/user". Actors created using system.actorOf() are children of this actor. This means that when this
guardian terminates, all normal actors in the system will be shutdown, too. It also means that this guardian’s
supervisor strategy determines how the top-level normal actors are supervised. Since Akka 2.1 it is possible to
configure this using the setting akka.actor.guardian-supervisor-strategy, which takes the fullyqualified class-name of a SupervisorStrategyConfigurator. When the guardian escalates a failure, the
root guardian’s response will be to terminate the guardian, which in effect will shut down the whole actor system.
/system: The System Guardian
This special guardian has been introduced in order to achieve an orderly shut-down sequence where logging remains active while all normal actors terminate, even though logging itself is implemented using actors. This
is realized by having the system guardian watch the user guardian and initiate its own shut-down upon reception of the Terminated message. The top-level system actors are supervised using a strategy which
will restart indefinitely upon all types of Exception except for ActorInitializationException and
ActorKilledException, which will terminate the child in question. All other throwables are escalated,
which will shut down the whole actor system.
3.4. Supervision and Monitoring
19
Akka Scala Documentation, Release 2.5.0-RC1
/: The Root Guardian
The root guardian is the grand-parent of all so-called “top-level” actors and supervises all the special actors
mentioned in Top-Level Scopes for Actor Paths using the SupervisorStrategy.stoppingStrategy,
whose purpose is to terminate the child upon any type of Exception. All other throwables will be escalated
. . . but to whom? Since every real actor has a supervisor, the supervisor of the root guardian cannot be a real
actor. And because this means that it is “outside of the bubble”, it is called the “bubble-walker”. This is a
synthetic ActorRef which in effect stops its child upon the first sign of trouble and sets the actor system’s
isTerminated status to true as soon as the root guardian is fully terminated (all children recursively stopped).
3.4.3 What Restarting Means
When presented with an actor which failed while processing a certain message, causes for the failure fall into three
categories:
• Systematic (i.e. programming) error for the specific message received
• (Transient) failure of some external resource used during processing the message
• Corrupt internal state of the actor
Unless the failure is specifically recognizable, the third cause cannot be ruled out, which leads to the conclusion
that the internal state needs to be cleared out. If the supervisor decides that its other children or itself is not
affected by the corruption—e.g. because of conscious application of the error kernel pattern—it is therefore best
to restart the child. This is carried out by creating a new instance of the underlying Actor class and replacing
the failed instance with the fresh one inside the child’s ActorRef; the ability to do this is one of the reasons for
encapsulating actors within special references. The new actor then resumes processing its mailbox, meaning that
the restart is not visible outside of the actor itself with the notable exception that the message during which the
failure occurred is not re-processed.
The precise sequence of events during a restart is the following:
1. suspend the actor (which means that it will not process normal messages until resumed), and recursively
suspend all children
2. call the old instance’s preRestart hook (defaults to sending termination requests to all children and
calling postStop)
3. wait for all children which were requested to terminate (using context.stop()) during preRestart
to actually terminate; this—like all actor operations—is non-blocking, the termination notice from the last
killed child will effect the progression to the next step
4. create new actor instance by invoking the originally provided factory again
5. invoke postRestart on the new instance (which by default also calls preStart)
6. send restart request to all children which were not killed in step 3; restarted children will follow the same
process recursively, from step 2
7. resume the actor
3.4.4 What Lifecycle Monitoring Means
Note: Lifecycle Monitoring in Akka is usually referred to as DeathWatch
In contrast to the special relationship between parent and child described above, each actor may monitor any other
actor. Since actors emerge from creation fully alive and restarts are not visible outside of the affected supervisors,
the only state change available for monitoring is the transition from alive to dead. Monitoring is thus used to tie
one actor to another so that it may react to the other actor’s termination, in contrast to supervision which reacts to
failure.
3.4. Supervision and Monitoring
20
Akka Scala Documentation, Release 2.5.0-RC1
Lifecycle monitoring is implemented using a Terminated message to be received by the monitoring actor,
where the default behavior is to throw a special DeathPactException if not otherwise handled. In order to
start listening for Terminated messages, invoke ActorContext.watch(targetActorRef). To stop
listening, invoke ActorContext.unwatch(targetActorRef). One important property is that the message will be delivered irrespective of the order in which the monitoring request and target’s termination occur, i.e.
you still get the message even if at the time of registration the target is already dead.
Monitoring is particularly useful if a supervisor cannot simply restart its children and has to terminate them, e.g.
in case of errors during actor initialization. In that case it should monitor those children and re-create them or
schedule itself to retry this at a later time.
Another common use case is that an actor needs to fail in the absence of an external resource, which may also be
one of its own children. If a third party terminates a child by way of the system.stop(child) method or
sending a PoisonPill, the supervisor might well be affected.
Delayed restarts with the BackoffSupervisor pattern
Provided as a built-in pattern the akka.pattern.BackoffSupervisor implements the so-called exponential backoff supervision strategy, starting a child actor again when it fails, each time with a growing time delay
between restarts.
This pattern is useful when the started actor fails1 because some external resource is not available, and we need to
give it some time to start-up again. One of the prime examples when this is useful is when a PersistentActor fails
(by stopping) with a persistence failure - which indicates that the database may be down or overloaded, in such
situations it makes most sense to give it a little bit of time to recover before the peristent actor is started.
The following Scala snippet shows how to create a backoff supervisor which will start the given echo actor after
it has stopped because of a failure, in increasing intervals of 3, 6, 12, 24 and finally 30 seconds:
val childProps = Props(classOf[EchoActor])
val supervisor = BackoffSupervisor.props(
Backoff.onStop(
childProps,
childName = "myEcho",
minBackoff = 3.seconds,
maxBackoff = 30.seconds,
randomFactor = 0.2 // adds 20% "noise" to vary the intervals slightly
))
system.actorOf(supervisor, name = "echoSupervisor")
The above is equivalent to this Java code:
import scala.concurrent.duration.Duration;
final Props childProps = Props.create(EchoActor.class);
final Props supervisorProps = BackoffSupervisor.props(
Backoff.onStop(
childProps,
"myEcho",
Duration.create(3, TimeUnit.SECONDS),
Duration.create(30, TimeUnit.SECONDS),
0.2)); // adds 20% "noise" to vary the intervals slightly
system.actorOf(supervisorProps, "echoSupervisor");
Using a randomFactor to add a little bit of additional variance to the backoff intervals is highly recommended,
in order to avoid multiple actors re-start at the exact same point in time, for example because they were stopped
1
A failure can be indicated in two different ways; by an actor stopping or crashing.
3.4. Supervision and Monitoring
21
Akka Scala Documentation, Release 2.5.0-RC1
due to a shared resource such as a database going down and re-starting after the same configured interval. By
adding additional randomness to the re-start intervals the actors will start in slightly different points in time, thus
avoiding large spikes of traffic hitting the recovering shared database or other resource that they all need to contact.
The akka.pattern.BackoffSupervisor actor can also be configured to restart the actor after a delay
when the actor crashes and the supervision strategy decides that it should restart.
The following Scala snippet shows how to create a backoff supervisor which will start the given echo actor after
it has crashed because of some exception, in increasing intervals of 3, 6, 12, 24 and finally 30 seconds:
val childProps = Props(classOf[EchoActor])
val supervisor = BackoffSupervisor.props(
Backoff.onFailure(
childProps,
childName = "myEcho",
minBackoff = 3.seconds,
maxBackoff = 30.seconds,
randomFactor = 0.2 // adds 20% "noise" to vary the intervals slightly
))
system.actorOf(supervisor, name = "echoSupervisor")
The above is equivalent to this Java code:
import scala.concurrent.duration.Duration;
final Props childProps = Props.create(EchoActor.class);
final Props supervisorProps = BackoffSupervisor.props(
Backoff.onFailure(
childProps,
"myEcho",
Duration.create(3, TimeUnit.SECONDS),
Duration.create(30, TimeUnit.SECONDS),
0.2)); // adds 20% "noise" to vary the intervals slightly
system.actorOf(supervisorProps, "echoSupervisor");
The akka.pattern.BackoffOptions can be used to customize the behavior of the back-off supervisor
actor, below are some examples:
val supervisor = BackoffSupervisor.props(
Backoff.onStop(
childProps,
childName = "myEcho",
minBackoff = 3.seconds,
maxBackoff = 30.seconds,
randomFactor = 0.2 // adds 20% "noise" to vary the intervals slightly
).withManualReset // the child must send BackoffSupervisor.Reset to its parent
.withDefaultStoppingStrategy // Stop at any Exception thrown
)
The above code sets up a back-off supervisor that requires the child actor to send a akka.pattern.
BackoffSupervisor.Reset message to its parent when a message is successfully processed, resetting the
back-off. It also uses a default stopping strategy, any exception will cause the child to stop.
val supervisor = BackoffSupervisor.props(
Backoff.onFailure(
childProps,
childName = "myEcho",
minBackoff = 3.seconds,
maxBackoff = 30.seconds,
3.4. Supervision and Monitoring
22
Akka Scala Documentation, Release 2.5.0-RC1
randomFactor = 0.2 // adds 20% "noise" to vary the intervals slightly
).withAutoReset(10.seconds) // the child must send BackoffSupervisor.Reset to
˓→its parent
.withSupervisorStrategy(
OneForOneStrategy() {
case _: MyException => SupervisorStrategy.Restart
case _
=> SupervisorStrategy.Escalate
}))
The above code sets up a back-off supervisor that restarts the child after back-off if MyException is thrown, any
other exception will be escalated. The back-off is automatically reset if the child does not throw any errors within
10 seconds.
3.4.5 One-For-One Strategy vs. All-For-One Strategy
There are two classes of supervision strategies which come with Akka: OneForOneStrategy and
AllForOneStrategy. Both are configured with a mapping from exception type to supervision directive (see
above) and limits on how often a child is allowed to fail before terminating it. The difference between them is that
the former applies the obtained directive only to the failed child, whereas the latter applies it to all siblings as well.
Normally, you should use the OneForOneStrategy, which also is the default if none is specified explicitly.
The AllForOneStrategy is applicable in cases where the ensemble of children has such tight dependencies
among them, that a failure of one child affects the function of the others, i.e. they are inextricably linked. Since
a restart does not clear out the mailbox, it often is best to terminate the children upon failure and re-create them
explicitly from the supervisor (by watching the children’s lifecycle); otherwise you have to make sure that it is no
problem for any of the actors to receive a message which was queued before the restart but processed afterwards.
Normally stopping a child (i.e. not in response to a failure) will not automatically terminate the other children
in an all-for-one strategy; this can easily be done by watching their lifecycle: if the Terminated message is
not handled by the supervisor, it will throw a DeathPactException which (depending on its supervisor) will
restart it, and the default preRestart action will terminate all children. Of course this can be handled explicitly
as well.
Please note that creating one-off actors from an all-for-one supervisor entails that failures escalated by the temporary actor will affect all the permanent ones. If this is not desired, install an intermediate supervisor; this can very
easily be done by declaring a router of size 1 for the worker, see Routing or routing-java.
3.5 Actor References, Paths and Addresses
This chapter describes how actors are identified and located within a possibly distributed actor system. It ties into
the central idea that Actor Systems form intrinsic supervision hierarchies as well as that communication between
actors is transparent with respect to their placement across multiple network nodes.
3.5. Actor References, Paths and Addresses
23
Akka Scala Documentation, Release 2.5.0-RC1
The above image displays the relationship between the most important entities within an actor system, please read
on for the details.
3.5.1 What is an Actor Reference?
An actor reference is a subtype of ActorRef, whose foremost purpose is to support sending messages to the
actor it represents. Each actor has access to its canonical (local) reference through the self field; this reference
is also included as sender reference by default for all messages sent to other actors. Conversely, during message
processing the actor has access to a reference representing the sender of the current message through the sender
method.
There are several different types of actor references that are supported depending on the configuration of the actor
system:
• Purely local actor references are used by actor systems which are not configured to support networking
functions. These actor references will not function if sent across a network connection to a remote JVM.
• Local actor references when remoting is enabled are used by actor systems which support networking functions for those references which represent actors within the same JVM. In order to also be reachable when
sent to other network nodes, these references include protocol and remote addressing information.
• There is a subtype of local actor references which is used for routers (i.e. actors mixing in the Router
trait). Its logical structure is the same as for the aforementioned local references, but sending a message to
them dispatches to one of their children directly instead.
• Remote actor references represent actors which are reachable using remote communication, i.e. sending
messages to them will serialize the messages transparently and send them to the remote JVM.
• There are several special types of actor references which behave like local actor references for all practical
purposes:
– PromiseActorRef is the special representation of a Promise for the purpose of being completed
by the response from an actor. akka.pattern.ask creates this actor reference.
– DeadLetterActorRef is the default implementation of the dead letters service to which Akka
routes all messages whose destinations are shut down or non-existent.
– EmptyLocalActorRef is what Akka returns when looking up a non-existent local actor path: it
is equivalent to a DeadLetterActorRef, but it retains its path so that Akka can send it over the
network and compare it to other existing actor references for that path, some of which might have been
obtained before the actor died.
• And then there are some one-off internal implementations which you should never really see:
3.5. Actor References, Paths and Addresses
24
Akka Scala Documentation, Release 2.5.0-RC1
– There is an actor reference which does not represent an actor but acts only as a pseudo-supervisor for
the root guardian, we call it “the one who walks the bubbles of space-time”.
– The first logging service started before actually firing up actor creation facilities is a fake actor
reference which accepts log events and prints them directly to standard output; it is Logging.
StandardOutLogger.
3.5.2 What is an Actor Path?
Since actors are created in a strictly hierarchical fashion, there exists a unique sequence of actor names given by
recursively following the supervision links between child and parent down towards the root of the actor system.
This sequence can be seen as enclosing folders in a file system, hence we adopted the name “path” to refer to it,
although actor hierarchy has some fundamental difference from file system hierarchy.
An actor path consists of an anchor, which identifies the actor system, followed by the concatenation of the path
elements, from root guardian to the designated actor; the path elements are the names of the traversed actors and
are separated by slashes.
What is the Difference Between Actor Reference and Path?
An actor reference designates a single actor and the life-cycle of the reference matches that actor’s life-cycle; an
actor path represents a name which may or may not be inhabited by an actor and the path itself does not have a
life-cycle, it never becomes invalid. You can create an actor path without creating an actor, but you cannot create
an actor reference without creating corresponding actor.
You can create an actor, terminate it, and then create a new actor with the same actor path. The newly created
actor is a new incarnation of the actor. It is not the same actor. An actor reference to the old incarnation is not
valid for the new incarnation. Messages sent to the old actor reference will not be delivered to the new incarnation
even though they have the same path.
Actor Path Anchors
Each actor path has an address component, describing the protocol and location by which the corresponding actor
is reachable, followed by the names of the actors in the hierarchy from the root up. Examples are:
"akka://my-sys/user/service-a/worker1"
// purely local
"akka.tcp://my-sys@host.example.com:5678/user/service-b" // remote
Here, akka.tcp is the default remote transport for the 2.4 release; other transports are pluggable. The interpretation of the host and port part (i.e. host.example.com:5678 in the example) depends on the transport
mechanism used, but it must abide by the URI structural rules.
Logical Actor Paths
The unique path obtained by following the parental supervision links towards the root guardian is called the logical
actor path. This path matches exactly the creation ancestry of an actor, so it is completely deterministic as soon as
the actor system’s remoting configuration (and with it the address component of the path) is set.
Physical Actor Paths
While the logical actor path describes the functional location within one actor system, configuration-based remote
deployment means that an actor may be created on a different network host than its parent, i.e. within a different
actor system. In this case, following the actor path from the root guardian up entails traversing the network, which
is a costly operation. Therefore, each actor also has a physical path, starting at the root guardian of the actor
system where the actual actor object resides. Using this path as sender reference when querying other actors will
let them reply directly to this actor, minimizing delays incurred by routing.
3.5. Actor References, Paths and Addresses
25
Akka Scala Documentation, Release 2.5.0-RC1
One important aspect is that a physical actor path never spans multiple actor systems or JVMs. This means that
the logical path (supervision hierarchy) and the physical path (actor deployment) of an actor may diverge if one
of its ancestors is remotely supervised.
Actor path alias or symbolic link?
As in some real file-systems you might think of a “path alias” or “symbolic link” for an actor, i.e. one actor
may be reachable using more than one path. However, you should note that actor hierarchy is different from file
system hierarchy. You cannot freely create actor paths like symbolic links to refer to arbitrary actors. As described
in the above logical and physical actor path sections, an actor path must be either logical path which represents
supervision hierarchy, or physical path which represents actor deployment.
3.5.3 How are Actor References obtained?
There are two general categories to how actor references may be obtained: by creating actors or by looking them
up, where the latter functionality comes in the two flavours of creating actor references from concrete actor paths
and querying the logical actor hierarchy.
Creating Actors
An actor system is typically started by creating actors beneath the guardian actor using the ActorSystem.
actorOf method and then using ActorContext.actorOf from within the created actors to spawn the actor
tree. These methods return a reference to the newly created actor. Each actor has direct access (through its
ActorContext) to references for its parent, itself and its children. These references may be sent within messages to other actors, enabling those to reply directly.
Looking up Actors by Concrete Path
In addition, actor references may be looked up using the ActorSystem.actorSelection method. The
selection can be used for communicating with said actor and the actor corresponding to the selection is looked up
when delivering each message.
To acquire an ActorRef that is bound to the life-cycle of a specific actor you need to send a message, such as
the built-in Identify message, to the actor and use the sender() reference of a reply from the actor.
Absolute vs. Relative Paths
In addition to ActorSystem.actorSelection there is also ActorContext.actorSelection, which
is available inside any actor as context.actorSelection. This yields an actor selection much like its twin
on ActorSystem, but instead of looking up the path starting from the root of the actor tree it starts out on the
current actor. Path elements consisting of two dots ("..") may be used to access the parent actor. You can for
example send a message to a specific sibling:
context.actorSelection("../brother") ! msg
Absolute paths may of course also be looked up on context in the usual way, i.e.
context.actorSelection("/user/serviceA") ! msg
will work as expected.
3.5. Actor References, Paths and Addresses
26
Akka Scala Documentation, Release 2.5.0-RC1
Querying the Logical Actor Hierarchy
Since the actor system forms a file-system like hierarchy, matching on paths is possible in the same way as supported by Unix shells: you may replace (parts of) path element names with wildcards («*» and «?») to formulate
a selection which may match zero or more actual actors. Because the result is not a single actor reference, it has a
different type ActorSelection and does not support the full set of operations an ActorRef does. Selections
may be formulated using the ActorSystem.actorSelection and ActorContext.actorSelection
methods and do support sending messages:
context.actorSelection("../*") ! msg
will send msg to all siblings including the current actor. As for references obtained using actorSelection, a traversal
of the supervision hierarchy is done in order to perform the message send. As the exact set of actors which match
a selection may change even while a message is making its way to the recipients, it is not possible to watch a
selection for liveliness changes. In order to do that, resolve the uncertainty by sending a request and gathering all
answers, extracting the sender references, and then watch all discovered concrete actors. This scheme of resolving
a selection may be improved upon in a future release.
Summary: actorOf vs. actorSelection
Note: What the above sections described in some detail can be summarized and memorized easily as follows:
• actorOf only ever creates a new actor, and it creates it as a direct child of the context on which this method
is invoked (which may be any actor or actor system).
• actorSelection only ever looks up existing actors when messages are delivered, i.e. does not create
actors, or verify existence of actors when the selection is created.
3.5.4 Actor Reference and Path Equality
Equality of ActorRef match the intention that an ActorRef corresponds to the target actor incarnation. Two
actor references are compared equal when they have the same path and point to the same actor incarnation. A
reference pointing to a terminated actor does not compare equal to a reference pointing to another (re-created)
actor with the same path. Note that a restart of an actor caused by a failure still means that it is the same actor
incarnation, i.e. a restart is not visible for the consumer of the ActorRef.
If you need to keep track of actor references in a collection and do not care about the exact actor incarnation you
can use the ActorPath as key, because the identifier of the target actor is not taken into account when comparing
actor paths.
3.5.5 Reusing Actor Paths
When an actor is terminated, its reference will point to the dead letter mailbox, DeathWatch will publish its
final transition and in general it is not expected to come back to life again (since the actor life cycle does not
allow this). While it is possible to create an actor at a later time with an identical path—simply due to it being
impossible to enforce the opposite without keeping the set of all actors ever created available—this is not good
practice: messages sent with actorSelection to an actor which “died” suddenly start to work again, but
without any guarantee of ordering between this transition and any other event, hence the new inhabitant of the
path may receive messages which were destined for the previous tenant.
It may be the right thing to do in very specific circumstances, but make sure to confine the handling of this precisely
to the actor’s supervisor, because that is the only actor which can reliably detect proper deregistration of the name,
before which creation of the new child will fail.
It may also be required during testing, when the test subject depends on being instantiated at a specific path. In
that case it is best to mock its supervisor so that it will forward the Terminated message to the appropriate point in
the test procedure, enabling the latter to await proper deregistration of the name.
3.5. Actor References, Paths and Addresses
27
Akka Scala Documentation, Release 2.5.0-RC1
3.5.6 The Interplay with Remote Deployment
When an actor creates a child, the actor system’s deployer will decide whether the new actor resides in the same
JVM or on another node. In the second case, creation of the actor will be triggered via a network connection to
happen in a different JVM and consequently within a different actor system. The remote system will place the
new actor below a special path reserved for this purpose and the supervisor of the new actor will be a remote actor
reference (representing that actor which triggered its creation). In this case, context.parent (the supervisor
reference) and context.path.parent (the parent node in the actor’s path) do not represent the same actor.
However, looking up the child’s name within the supervisor will find it on the remote node, preserving logical
structure e.g. when sending to an unresolved actor reference.
3.5.7 What is the Address part used for?
When sending an actor reference across the network, it is represented by its path. Hence, the path must fully
encode all information necessary to send messages to the underlying actor. This is achieved by encoding protocol,
host and port in the address part of the path string. When an actor system receives an actor path from a remote
node, it checks whether that path’s address matches the address of this actor system, in which case it will be
resolved to the actor’s local reference. Otherwise, it will be represented by a remote actor reference.
3.5.8 Top-Level Scopes for Actor Paths
At the root of the path hierarchy resides the root guardian above which all other actors are found; its name is "/".
The next level consists of the following:
• "/user" is the guardian actor for all user-created top-level actors; actors created using ActorSystem.
actorOf are found below this one.
3.5. Actor References, Paths and Addresses
28
Akka Scala Documentation, Release 2.5.0-RC1
• "/system" is the guardian actor for all system-created top-level actors, e.g. logging listeners or actors
automatically deployed by configuration at the start of the actor system.
• "/deadLetters" is the dead letter actor, which is where all messages sent to stopped or non-existing
actors are re-routed (on a best-effort basis: messages may be lost even within the local JVM).
• "/temp" is the guardian for all short-lived system-created actors, e.g. those which are used in the implementation of ActorRef.ask.
• "/remote" is an artificial path below which all actors reside whose supervisors are remote actor references
The need to structure the name space for actors like this arises from a central and very simple design goal:
everything in the hierarchy is an actor, and all actors function in the same way. Hence you can not only look
up the actors you created, you can also look up the system guardian and send it a message (which it will dutifully
discard in this case). This powerful principle means that there are no quirks to remember, it makes the whole
system more uniform and consistent.
If you want to read more about the top-level structure of an actor system, have a look at The Top-Level Supervisors.
3.6 Location Transparency
The previous section describes how actor paths are used to enable location transparency. This special feature
deserves some extra explanation, because the related term “transparent remoting” was used quite differently in the
context of programming languages, platforms and technologies.
3.6.1 Distributed by Default
Everything in Akka is designed to work in a distributed setting: all interactions of actors use purely message
passing and everything is asynchronous. This effort has been undertaken to ensure that all functions are available
equally when running within a single JVM or on a cluster of hundreds of machines. The key for enabling this
is to go from remote to local by way of optimization instead of trying to go from local to remote by way of
generalization. See this classic paper for a detailed discussion on why the second approach is bound to fail.
3.6.2 Ways in which Transparency is Broken
What is true of Akka need not be true of the application which uses it, since designing for distributed execution
poses some restrictions on what is possible. The most obvious one is that all messages sent over the wire must be
serializable. While being a little less obvious this includes closures which are used as actor factories (i.e. within
Props) if the actor is to be created on a remote node.
Another consequence is that everything needs to be aware of all interactions being fully asynchronous, which in
a computer network might mean that it may take several minutes for a message to reach its recipient (depending
on configuration). It also means that the probability for a message to be lost is much higher than within one JVM,
where it is close to zero (still: no hard guarantee!).
3.6.3 How is Remoting Used?
We took the idea of transparency to the limit in that there is nearly no API for the remoting layer of Akka: it is
purely driven by configuration. Just write your application according to the principles outlined in the previous
sections, then specify remote deployment of actor sub-trees in the configuration file. This way, your application
can be scaled out without having to touch the code. The only piece of the API which allows programmatic
influence on remote deployment is that Props contain a field which may be set to a specific Deploy instance; this
has the same effect as putting an equivalent deployment into the configuration file (if both are given, configuration
file wins).
3.6. Location Transparency
29
Akka Scala Documentation, Release 2.5.0-RC1
3.6.4 Peer-to-Peer vs. Client-Server
Akka Remoting is a communication module for connecting actor systems in a peer-to-peer fashion, and it is the
foundation for Akka Clustering. The design of remoting is driven by two (related) design decisions:
1. Communication between involved systems is symmetric: if a system A can connect to a system B then
system B must also be able to connect to system A independently.
2. The role of the communicating systems are symmetric in regards to connection patterns: there is no system
that only accepts connections, and there is no system that only initiates connections.
The consequence of these decisions is that it is not possible to safely create pure client-server setups with predefined roles (violates assumption 2). For client-server setups it is better to use HTTP or Akka I/O.
Important: Using setups involving Network Address Translation, Load Balancers or Docker containers violates
assumption 1, unless additional steps are taken in the network configuration to allow symmetric communication
between involved systems. In such situations Akka can be configured to bind to a different network address than
the one used for establishing connections between Akka nodes. See Akka behind NAT or in a Docker container.
3.6.5 Marking Points for Scaling Up with Routers
In addition to being able to run different parts of an actor system on different nodes of a cluster, it is also possible
to scale up onto more cores by multiplying actor sub-trees which support parallelization (think for example a
search engine processing different queries in parallel). The clones can then be routed to in different fashions, e.g.
round-robin. The only thing necessary to achieve this is that the developer needs to declare a certain actor as
“withRouter”, then—in its stead—a router actor will be created which will spawn up a configurable number of
children of the desired type and route to them in the configured fashion. Once such a router has been declared, its
configuration can be freely overridden from the configuration file, including mixing it with the remote deployment
of (some of) the children. Read more about this in Routing (Scala) and Routing (Java).
3.7 Akka and the Java Memory Model
A major benefit of using the Lightbend Platform, including Scala and Akka, is that it simplifies the process of writing concurrent software. This article discusses how the Lightbend Platform, and Akka in particular, approaches
shared memory in concurrent applications.
3.7.1 The Java Memory Model
Prior to Java 5, the Java Memory Model (JMM) was ill defined. It was possible to get all kinds of strange results
when shared memory was accessed by multiple threads, such as:
• a thread not seeing values written by other threads: a visibility problem
• a thread observing ‘impossible’ behavior of other threads, caused by instructions not being executed in the
order expected: an instruction reordering problem.
With the implementation of JSR 133 in Java 5, a lot of these issues have been resolved. The JMM is a set of rules
based on the “happens-before” relation, which constrain when one memory access must happen before another,
and conversely, when they are allowed to happen out of order. Two examples of these rules are:
• The monitor lock rule: a release of a lock happens before every subsequent acquire of the same lock.
• The volatile variable rule: a write of a volatile variable happens before every subsequent read of the same
volatile variable
Although the JMM can seem complicated, the specification tries to find a balance between ease of use and the
ability to write performant and scalable concurrent data structures.
3.7. Akka and the Java Memory Model
30
Akka Scala Documentation, Release 2.5.0-RC1
3.7.2 Actors and the Java Memory Model
With the Actors implementation in Akka, there are two ways multiple threads can execute actions on shared
memory:
• if a message is sent to an actor (e.g. by another actor). In most cases messages are immutable, but if
that message is not a properly constructed immutable object, without a “happens before” rule, it would be
possible for the receiver to see partially initialized data structures and possibly even values out of thin air
(longs/doubles).
• if an actor makes changes to its internal state while processing a message, and accesses that state while
processing another message moments later. It is important to realize that with the actor model you don’t get
any guarantee that the same thread will be executing the same actor for different messages.
To prevent visibility and reordering problems on actors, Akka guarantees the following two “happens before”
rules:
• The actor send rule: the send of the message to an actor happens before the receive of that message by the
same actor.
• The actor subsequent processing rule: processing of one message happens before processing of the next
message by the same actor.
Note: In layman’s terms this means that changes to internal fields of the actor are visible when the next message
is processed by that actor. So fields in your actor need not be volatile or equivalent.
Both rules only apply for the same actor instance and are not valid if different actors are used.
3.7.3 Futures and the Java Memory Model
The completion of a Future “happens before” the invocation of any callbacks registered to it are executed.
We recommend not to close over non-final fields (final in Java and val in Scala), and if you do choose to close
over non-final fields, they must be marked volatile in order for the current value of the field to be visible to the
callback.
If you close over a reference, you must also ensure that the instance that is referred to is thread safe. We highly
recommend staying away from objects that use locking, since it can introduce performance problems and in the
worst case, deadlocks. Such are the perils of synchronized.
3.7.4 Actors and shared mutable state
Since Akka runs on the JVM there are still some rules to be followed.
• Closing over internal Actor state and exposing it to other threads
import
import
import
import
import
import
import
akka.actor.{ Actor, ActorRef }
akka.pattern.ask
akka.util.Timeout
scala.concurrent.Future
scala.concurrent.duration._
scala.language.postfixOps
scala.collection.mutable
case class Message(msg: String)
class EchoActor extends Actor {
def receive = {
case msg => sender() ! msg
}
}
3.7. Akka and the Java Memory Model
31
Akka Scala Documentation, Release 2.5.0-RC1
class CleanUpActor extends Actor {
def receive = {
case set: mutable.Set[_] => set.clear()
}
}
class MyActor(echoActor: ActorRef, cleanUpActor: ActorRef) extends Actor {
var state = ""
val mySet = mutable.Set[String]()
def expensiveCalculation(actorRef: ActorRef): String = {
// this is a very costly operation
"Meaning of live is 42"
}
def expensiveCalculation(): String = {
// this is a very costly operation
"Meaning of live is 42"
}
def receive = {
case _ =>
//Wrong ways
implicit val ec = context.dispatcher
implicit val timeout = Timeout(5 seconds) // needed for `?` below
// Very bad, shared mutable state,
// will break your application in weird ways
Future { state = "This will race" }
((echoActor ? Message("With this other one")).mapTo[Message])
.foreach { received => state = received.msg }
// Very bad,
// the other
// or worse,
cleanUpActor
shared mutable object,
actor cand mutate your own state,
you might get weird race conditions
! mySet
// Very bad, "sender" changes for every message,
// shared mutable state bug
Future { expensiveCalculation(sender()) }
//Right ways
// Completely safe, "self" is OK to close over
// and it's an ActorRef, which is thread-safe
Future { expensiveCalculation() } foreach { self ! _ }
// Completely safe, we close over a fixed value
// and it's an ActorRef, which is thread-safe
val currentSender = sender()
Future { expensiveCalculation(currentSender) }
}
}
• Messages should be immutable, this is to avoid the shared mutable state trap.
3.7. Akka and the Java Memory Model
32
Akka Scala Documentation, Release 2.5.0-RC1
3.8 Message Delivery Reliability
Akka helps you build reliable applications which make use of multiple processor cores in one machine (“scaling
up”) or distributed across a computer network (“scaling out”). The key abstraction to make this work is that all
interactions between your code units—actors—happen via message passing, which is why the precise semantics
of how messages are passed between actors deserve their own chapter.
In order to give some context to the discussion below, consider an application which spans multiple network hosts.
The basic mechanism for communication is the same whether sending to an actor on the local JVM or to a remote
actor, but of course there will be observable differences in the latency of delivery (possibly also depending on the
bandwidth of the network link and the message size) and the reliability. In case of a remote message send there
are obviously more steps involved which means that more can go wrong. Another aspect is that local sending will
just pass a reference to the message inside the same JVM, without any restrictions on the underlying object which
is sent, whereas a remote transport will place a limit on the message size.
Writing your actors such that every interaction could possibly be remote is the safe, pessimistic bet. It means to
only rely on those properties which are always guaranteed and which are discussed in detail below. This has of
course some overhead in the actor’s implementation. If you are willing to sacrifice full location transparency—for
example in case of a group of closely collaborating actors—you can place them always on the same JVM and
enjoy stricter guarantees on message delivery. The details of this trade-off are discussed further below.
As a supplementary part we give a few pointers at how to build stronger reliability on top of the built-in ones. The
chapter closes by discussing the role of the “Dead Letter Office”.
3.8.1 The General Rules
These are the rules for message sends (i.e. the tell or ! method, which also underlies the ask pattern):
• at-most-once delivery, i.e. no guaranteed delivery
• message ordering per sender–receiver pair
The first rule is typically found also in other actor implementations while the second is specific to Akka.
Discussion: What does “at-most-once” mean?
When it comes to describing the semantics of a delivery mechanism, there are three basic categories:
• at-most-once delivery means that for each message handed to the mechanism, that message is delivered
zero or one times; in more casual terms it means that messages may be lost.
• at-least-once delivery means that for each message handed to the mechanism potentially multiple attempts
are made at delivering it, such that at least one succeeds; again, in more casual terms this means that
messages may be duplicated but not lost.
• exactly-once delivery means that for each message handed to the mechanism exactly one delivery is made
to the recipient; the message can neither be lost nor duplicated.
The first one is the cheapest—highest performance, least implementation overhead—because it can be done in
a fire-and-forget fashion without keeping state at the sending end or in the transport mechanism. The second
one requires retries to counter transport losses, which means keeping state at the sending end and having an
acknowledgement mechanism at the receiving end. The third is most expensive—and has consequently worst
performance—because in addition to the second it requires state to be kept at the receiving end in order to filter
out duplicate deliveries.
Discussion: Why No Guaranteed Delivery?
At the core of the problem lies the question what exactly this guarantee shall mean:
1. The message is sent out on the network?
3.8. Message Delivery Reliability
33
Akka Scala Documentation, Release 2.5.0-RC1
2. The message is received by the other host?
3. The message is put into the target actor’s mailbox?
4. The message is starting to be processed by the target actor?
5. The message is processed successfully by the target actor?
Each one of these have different challenges and costs, and it is obvious that there are conditions under which
any message passing library would be unable to comply; think for example about configurable mailbox types
and how a bounded mailbox would interact with the third point, or even what it would mean to decide upon the
“successfully” part of point five.
Along those same lines goes the reasoning in Nobody Needs Reliable Messaging. The only meaningful way for a
sender to know whether an interaction was successful is by receiving a business-level acknowledgement message,
which is not something Akka could make up on its own (neither are we writing a “do what I mean” framework
nor would you want us to).
Akka embraces distributed computing and makes the fallibility of communication explicit through message passing, therefore it does not try to lie and emulate a leaky abstraction. This is a model that has been used with great
success in Erlang and requires the users to design their applications around it. You can read more about this
approach in the Erlang documentation (section 10.9 and 10.10), Akka follows it closely.
Another angle on this issue is that by providing only basic guarantees those use cases which do not need stronger
reliability do not pay the cost of their implementation; it is always possible to add stronger reliability on top of
basic ones, but it is not possible to retro-actively remove reliability in order to gain more performance.
Discussion: Message Ordering
The rule more specifically is that for a given pair of actors, messages sent directly from the first to the second will
not be received out-of-order. The word directly emphasizes that this guarantee only applies when sending with
the tell operator to the final destination, not when employing mediators or other message dissemination features
(unless stated otherwise).
The guarantee is illustrated in the following:
Actor A1 sends messages M1, M2, M3 to A2
Actor A3 sends messages M4, M5, M6 to A2
This means that:
1. If M1 is delivered it must be delivered before M2 and M3
2. If M2 is delivered it must be delivered before M3
3. If M4 is delivered it must be delivered before M5 and M6
4. If M5 is delivered it must be delivered before M6
5. A2 can see messages from A1 interleaved with messages from A3
6. Since there is no guaranteed delivery, any of the messages may be dropped, i.e. not arrive
at A2
Note: It is important to note that Akka’s guarantee applies to the order in which messages are enqueued into the
recipient’s mailbox. If the mailbox implementation does not respect FIFO order (e.g. a PriorityMailbox),
then the order of processing by the actor can deviate from the enqueueing order.
Please note that this rule is not transitive:
Actor A sends message M1 to actor C
Actor A then sends message M2 to actor B
Actor B forwards message M2 to actor C
3.8. Message Delivery Reliability
34
Akka Scala Documentation, Release 2.5.0-RC1
Actor C may receive M1 and M2 in any order
Causal transitive ordering would imply that M2 is never received before M1 at actor C (though any of them might
be lost). This ordering can be violated due to different message delivery latencies when A, B and C reside on
different network hosts, see more below.
Note: Actor creation is treated as a message sent from the parent to the child, with the same semantics as discussed
above. Sending a message to an actor in a way which could be reordered with this initial creation message means
that the message might not arrive because the actor does not exist yet. An example where the message might arrive
too early would be to create a remote-deployed actor R1, send its reference to another remote actor R2 and have
R2 send a message to R1. An example of well-defined ordering is a parent which creates an actor and immediately
sends a message to it.
Communication of failure
Please note, that the ordering guarantees discussed above only hold for user messages between actors. Failure
of a child of an actor is communicated by special system messages that are not ordered relative to ordinary user
messages. In particular:
Child actor C sends message M to its parent P
Child actor fails with failure F
Parent actor P might receive the two events either in order M, F or F, M
The reason for this is that internal system messages has their own mailboxes therefore the ordering of enqueue
calls of a user and system message cannot guarantee the ordering of their dequeue times.
3.8.2 The Rules for In-JVM (Local) Message Sends
Be careful what you do with this section!
Relying on the stronger reliability in this section is not recommended since it will bind your application to localonly deployment: an application may have to be designed differently (as opposed to just employing some message
exchange patterns local to some actors) in order to be fit for running on a cluster of machines. Our credo is “design
once, deploy any way you wish”, and to achieve this you should only rely on The General Rules.
Reliability of Local Message Sends
The Akka test suite relies on not losing messages in the local context (and for non-error condition tests also for
remote deployment), meaning that we actually do apply the best effort to keep our tests stable. A local tell
operation can however fail for the same reasons as a normal method call can on the JVM:
• StackOverflowError
• OutOfMemoryError
• other VirtualMachineError
In addition, local sends can fail in Akka-specific ways:
• if the mailbox does not accept the message (e.g. full BoundedMailbox)
• if the receiving actor fails while processing the message or is already terminated
While the first is clearly a matter of configuration the second deserves some thought: the sender of a message does
not get feedback if there was an exception while processing, that notification goes to the supervisor instead. This
is in general not distinguishable from a lost message for an outside observer.
3.8. Message Delivery Reliability
35
Akka Scala Documentation, Release 2.5.0-RC1
Ordering of Local Message Sends
Assuming strict FIFO mailboxes the aforementioned caveat of non-transitivity of the message ordering guarantee
is eliminated under certain conditions. As you will note, these are quite subtle as it stands, and it is even possible
that future performance optimizations will invalidate this whole paragraph. The possibly non-exhaustive list of
counter-indications is:
• Before receiving the first reply from a top-level actor, there is a lock which protects an internal interim
queue, and this lock is not fair; the implication is that enqueue requests from different senders which arrive
during the actor’s construction (figuratively, the details are more involved) may be reordered depending on
low-level thread scheduling. Since completely fair locks do not exist on the JVM this is unfixable.
• The same mechanism is used during the construction of a Router, more precisely the routed ActorRef, hence
the same problem exists for actors deployed with Routers.
• As mentioned above, the problem occurs anywhere a lock is involved during enqueueing, which may also
apply to custom mailboxes.
This list has been compiled carefully, but other problematic scenarios may have escaped our analysis.
How does Local Ordering relate to Network Ordering
The rule that for a given pair of actors, messages sent directly from the first to the second will not be received
out-of-order holds for messages sent over the network with the TCP based Akka remote transport protocol.
As explained in the previous section local message sends obey transitive causal ordering under certain conditions.
This ordering can be violated due to different message delivery latencies. For example:
Actor A on node-1 sends message M1 to actor C on node-3
Actor A on node-1 then sends message M2 to actor B on node-2
Actor B on node-2 forwards message M2 to actor C on node-3
Actor C may receive M1 and M2 in any order
It might take longer time for M1 to “travel” to node-3 than it takes for M2 to “travel” to node-3 via node-2.
3.8.3 Higher-level abstractions
Based on a small and consistent tool set in Akka’s core, Akka also provides powerful, higher-level abstractions on
top it.
Messaging Patterns
As discussed above a straight-forward answer to the requirement of reliable delivery is an explicit ACK–RETRY
protocol. In its simplest form this requires
• a way to identify individual messages to correlate message with acknowledgement
• a retry mechanism which will resend messages if not acknowledged in time
• a way for the receiver to detect and discard duplicates
The third becomes necessary by virtue of the acknowledgements not being guaranteed to arrive either. An ACKRETRY protocol with business-level acknowledgements is supported by At-Least-Once Delivery of the Akka
Persistence module. Duplicates can be detected by tracking the identifiers of messages sent via At-Least-Once
Delivery. Another way of implementing the third part would be to make processing the messages idempotent on
the level of the business logic.
Another example of implementing all three requirements is shown at Reliable Proxy Pattern (which is now superseded by At-Least-Once Delivery).
3.8. Message Delivery Reliability
36
Akka Scala Documentation, Release 2.5.0-RC1
Event Sourcing
Event sourcing (and sharding) is what makes large websites scale to billions of users, and the idea is quite simple:
when a component (think actor) processes a command it will generate a list of events representing the effect of
the command. These events are stored in addition to being applied to the component’s state. The nice thing about
this scheme is that events only ever are appended to the storage, nothing is ever mutated; this enables perfect
replication and scaling of consumers of this event stream (i.e. other components may consume the event stream as
a means to replicate the component’s state on a different continent or to react to changes). If the component’s state
is lost—due to a machine failure or by being pushed out of a cache—it can easily be reconstructed by replaying
the event stream (usually employing snapshots to speed up the process). Event sourcing is supported by Akka
Persistence.
Mailbox with Explicit Acknowledgement
By implementing a custom mailbox type it is possible to retry message processing at the receiving actor’s end
in order to handle temporary failures. This pattern is mostly useful in the local communication context where
delivery guarantees are otherwise sufficient to fulfill the application’s requirements.
Please note that the caveats for The Rules for In-JVM (Local) Message Sends do apply.
An example implementation of this pattern is shown at Mailbox with Explicit Acknowledgement.
3.8.4 Dead Letters
Messages which cannot be delivered (and for which this can be ascertained) will be delivered to a synthetic actor
called /deadLetters. This delivery happens on a best-effort basis; it may fail even within the local JVM (e.g.
during actor termination). Messages sent via unreliable network transports will be lost without turning up as dead
letters.
What Should I Use Dead Letters For?
The main use of this facility is for debugging, especially if an actor send does not arrive consistently (where
usually inspecting the dead letters will tell you that the sender or recipient was set wrong somewhere along the
way). In order to be useful for this purpose it is good practice to avoid sending to deadLetters where possible, i.e.
run your application with a suitable dead letter logger (see more below) from time to time and clean up the log
output. This exercise—like all else—requires judicious application of common sense: it may well be that avoiding
to send to a terminated actor complicates the sender’s code more than is gained in debug output clarity.
The dead letter service follows the same rules with respect to delivery guarantees as all other message sends, hence
it cannot be used to implement guaranteed delivery.
How do I Receive Dead Letters?
An actor can subscribe to class akka.actor.DeadLetter on the event stream, see event-stream-java (Java)
or Event Stream (Scala) for how to do that. The subscribed actor will then receive all dead letters published in
the (local) system from that point onwards. Dead letters are not propagated over the network, if you want to
collect them in one place you will have to subscribe one actor per network node and forward them manually. Also
consider that dead letters are generated at that node which can determine that a send operation is failed, which for
a remote send can be the local system (if no network connection can be established) or the remote one (if the actor
you are sending to does not exist at that point in time).
Dead Letters Which are (Usually) not Worrisome
Every time an actor does not terminate by its own decision, there is a chance that some messages which it sends
to itself are lost. There is one which happens quite easily in complex shutdown scenarios that is usually benign:
seeing a akka.dispatch.Terminate message dropped means that two termination requests were given, but
3.8. Message Delivery Reliability
37
Akka Scala Documentation, Release 2.5.0-RC1
of course only one can succeed. In the same vein, you might see akka.actor.Terminated messages from
children while stopping a hierarchy of actors turning up in dead letters if the parent is still watching the child when
the parent terminates.
3.9 Configuration
You can start using Akka without defining any configuration, since sensible default values are provided. Later on
you might need to amend the settings to change the default behavior or adapt for specific runtime environments.
Typical examples of settings that you might amend:
• log level and logger backend
• enable remoting
• message serializers
• definition of routers
• tuning of dispatchers
Akka uses the Typesafe Config Library, which might also be a good choice for the configuration of your own application or library built with or without Akka. This library is implemented in Java with no external dependencies;
you should have a look at its documentation (in particular about ConfigFactory), which is only summarized in the
following.
Warning: If you use Akka from the Scala REPL from the 2.9.x series, and you do not provide your own
ClassLoader to the ActorSystem, start the REPL with “-Yrepl-sync” to work around a deficiency in the REPLs
provided Context ClassLoader.
3.9.1 Where configuration is read from
All configuration for Akka is held within instances of ActorSystem, or put differently, as viewed from the
outside, ActorSystem is the only consumer of configuration information. While constructing an actor system,
you can either pass in a Config object or not, where the second case is equivalent to passing ConfigFactory.
load() (with the right class loader). This means roughly that the default is to parse all application.conf,
application.json and application.properties found at the root of the class path—please refer to
the aforementioned documentation for details. The actor system then merges in all reference.conf resources
found at the root of the class path to form the fallback configuration, i.e. it internally uses
appConfig.withFallback(ConfigFactory.defaultReference(classLoader))
The philosophy is that code never contains default values, but instead relies upon their presence in the
reference.conf supplied with the library in question.
Highest precedence is given to overrides given as system properties, see the HOCON specification (near the
bottom). Also noteworthy is that the application configuration—which defaults to application—may be
overridden using the config.resource property (there are more, please refer to the Config docs).
Note: If you are writing an Akka application, keep you configuration in application.conf at the root of
the class path. If you are writing an Akka-based library, keep its configuration in reference.conf at the root
of the JAR file.
3.9. Configuration
38
Akka Scala Documentation, Release 2.5.0-RC1
3.9.2 When using JarJar, OneJar, Assembly or any jar-bundler
Warning: Akka’s configuration approach relies heavily on the notion of every module/jar having its own
reference.conf file, all of these will be discovered by the configuration and loaded. Unfortunately this also
means that if you put/merge multiple jars into the same jar, you need to merge all the reference.confs as well.
Otherwise all defaults will be lost and Akka will not function.
If you are using Maven to package your application, you can also make use of the Apache Maven Shade Plugin
support for Resource Transformers to merge all the reference.confs on the build classpath into one.
The plugin configuration might look like this:
<plugin>
<groupId>org.apache.maven.plugins</groupId>
<artifactId>maven-shade-plugin</artifactId>
<version>1.5</version>
<executions>
<execution>
<phase>package</phase>
<goals>
<goal>shade</goal>
</goals>
<configuration>
<shadedArtifactAttached>true</shadedArtifactAttached>
<shadedClassifierName>allinone</shadedClassifierName>
<artifactSet>
<includes>
<include>*:*</include>
</includes>
</artifactSet>
<transformers>
<transformer
implementation="org.apache.maven.plugins.shade.resource.AppendingTransformer
˓→">
<resource>reference.conf</resource>
</transformer>
<transformer
implementation="org.apache.maven.plugins.shade.resource.
˓→ManifestResourceTransformer">
<manifestEntries>
<Main-Class>akka.Main</Main-Class>
</manifestEntries>
</transformer>
</transformers>
</configuration>
</execution>
</executions>
</plugin>
3.9.3 Custom application.conf
A custom application.conf might look like this:
# In this file you can override any option defined in the reference files.
# Copy in parts of the reference files and modify as you please.
akka {
# Loggers to register at boot time (akka.event.Logging$DefaultLogger logs
3.9. Configuration
39
Akka Scala Documentation, Release 2.5.0-RC1
# to STDOUT)
loggers = ["akka.event.slf4j.Slf4jLogger"]
# Log level used by the configured loggers (see "loggers") as soon
# as they have been started; before that, see "stdout-loglevel"
# Options: OFF, ERROR, WARNING, INFO, DEBUG
loglevel = "DEBUG"
# Log level for the very basic logger activated during ActorSystem startup.
# This logger prints the log messages to stdout (System.out).
# Options: OFF, ERROR, WARNING, INFO, DEBUG
stdout-loglevel = "DEBUG"
# Filter of log events that is used by the LoggingAdapter before
# publishing log events to the eventStream.
logging-filter = "akka.event.slf4j.Slf4jLoggingFilter"
actor {
provider = "cluster"
default-dispatcher {
# Throughput for default Dispatcher, set to 1 for as fair as possible
throughput = 10
}
}
remote {
# The port clients should connect to. Default is 2552.
netty.tcp.port = 4711
}
}
3.9.4 Including files
Sometimes it can be useful to include another configuration file, for example if you have one application.
conf with all environment independent settings and then override some settings for specific environments.
Specifying system property with -Dconfig.resource=/dev.conf will load the dev.conf file, which
includes the application.conf
dev.conf:
include "application"
akka {
loglevel = "DEBUG"
}
More advanced include and substitution mechanisms are explained in the HOCON specification.
3.9.5 Logging of Configuration
If the system or config property akka.log-config-on-start is set to on, then the complete configuration is
logged at INFO level when the actor system is started. This is useful when you are uncertain of what configuration
is used.
If in doubt, you can also easily and nicely inspect configuration objects before or after using them to construct an
actor system:
3.9. Configuration
40
Akka Scala Documentation, Release 2.5.0-RC1
Welcome to Scala version 2.11.8 (Java HotSpot(TM) 64-Bit Server VM, Java 1.8.0).
Type in expressions to have them evaluated.
Type :help for more information.
scala> import com.typesafe.config._
import com.typesafe.config._
scala> ConfigFactory.parseString("a.b=12")
res0: com.typesafe.config.Config = Config(SimpleConfigObject({"a" : {"b" : 12}}))
scala> res0.root.render
res1: java.lang.String =
{
# String: 1
"a" : {
# String: 1
"b" : 12
}
}
The comments preceding every item give detailed information about the origin of the setting (file & line number)
plus possible comments which were present, e.g. in the reference configuration. The settings as merged with the
reference and parsed by the actor system can be displayed like this:
final ActorSystem system = ActorSystem.create();
System.out.println(system.settings());
// this is a shortcut for system.settings().config().root().render()
3.9.6 A Word About ClassLoaders
In several places of the configuration file it is possible to specify the fully-qualified class name of something to be
instantiated by Akka. This is done using Java reflection, which in turn uses a ClassLoader. Getting the right
one in challenging environments like application containers or OSGi bundles is not always trivial, the current
approach of Akka is that each ActorSystem implementation stores the current thread’s context class loader
(if available, otherwise just its own loader as in this.getClass.getClassLoader) and uses that for all
reflective accesses. This implies that putting Akka on the boot class path will yield NullPointerException
from strange places: this is simply not supported.
3.9.7 Application specific settings
The configuration can also be used for application specific settings. A good practice is to place those settings in
an Extension, as described in:
• Scala API: Application specific settings
• Java API: extending-akka-java.settings
3.9.8 Configuring multiple ActorSystem
If you have more than one ActorSystem (or you’re writing a library and have an ActorSystem that may be
separate from the application’s) you may want to separate the configuration for each system.
Given that ConfigFactory.load() merges all resources with matching name from the whole class path, it
is easiest to utilize that functionality and differentiate actor systems within the hierarchy of the configuration:
myapp1 {
akka.loglevel = "WARNING"
my.own.setting = 43
3.9. Configuration
41
Akka Scala Documentation, Release 2.5.0-RC1
}
myapp2 {
akka.loglevel = "ERROR"
app2.setting = "appname"
}
my.own.setting = 42
my.other.setting = "hello"
val config = ConfigFactory.load()
val app1 = ActorSystem("MyApp1", config.getConfig("myapp1").withFallback(config))
val app2 = ActorSystem("MyApp2",
config.getConfig("myapp2").withOnlyPath("akka").withFallback(config))
These two samples demonstrate different variations of the “lift-a-subtree” trick: in the first case, the configuration
accessible from within the actor system is this
akka.loglevel = "WARNING"
my.own.setting = 43
my.other.setting = "hello"
// plus myapp1 and myapp2 subtrees
while in the second one, only the “akka” subtree is lifted, with the following result
akka.loglevel = "ERROR"
my.own.setting = 42
my.other.setting = "hello"
// plus myapp1 and myapp2 subtrees
Note: The configuration library is really powerful, explaining all features exceeds the scope affordable here.
In particular not covered are how to include other configuration files within other files (see a small example at
Including files) and copying parts of the configuration tree by way of path substitutions.
You may also specify and parse the configuration programmatically in other ways when instantiating the
ActorSystem.
import akka.actor.ActorSystem
import com.typesafe.config.ConfigFactory
val customConf = ConfigFactory.parseString("""
akka.actor.deployment {
/my-service {
router = round-robin-pool
nr-of-instances = 3
}
}
""")
// ConfigFactory.load sandwiches customConfig between default reference
// config and default overrides, and then resolves it.
val system = ActorSystem("MySystem", ConfigFactory.load(customConf))
3.9.9 Reading configuration from a custom location
You can replace or supplement application.conf either in code or using system properties.
If you’re using ConfigFactory.load() (which Akka does by default) you can replace application.
conf by defining -Dconfig.resource=whatever, -Dconfig.file=whatever, or -Dconfig.
url=whatever.
From inside your replacement file specified with -Dconfig.resource and friends, you can include
"application" if you still want to use application.{conf,json,properties} as well. Settings
3.9. Configuration
42
Akka Scala Documentation, Release 2.5.0-RC1
specified before include "application" would be overridden by the included file, while those after would
override the included file.
In code, there are many customization options.
There are several overloads of ConfigFactory.load(); these allow you to specify something to be sandwiched between system properties (which override) and the defaults (from reference.conf), replacing the
usual application.{conf,json,properties} and replacing -Dconfig.file and friends.
The simplest variant of ConfigFactory.load() takes a resource basename (instead of application);
myname.conf, myname.json, and myname.properties would then be used instead of application.
{conf,json,properties}.
The most flexible variant takes a Config object, which you can load using any method in ConfigFactory.
For example you could put a config string in code using ConfigFactory.parseString() or you could
make a map and ConfigFactory.parseMap(), or you could load a file.
You can also combine your custom config with the usual config, that might look like:
// make a Config with just your special setting
Config myConfig =
ConfigFactory.parseString("something=somethingElse");
// load the normal config stack (system props,
// then application.conf, then reference.conf)
Config regularConfig =
ConfigFactory.load();
// override regular stack with myConfig
Config combined =
myConfig.withFallback(regularConfig);
// put the result in between the overrides
// (system props) and defaults again
Config complete =
ConfigFactory.load(combined);
// create ActorSystem
ActorSystem system =
ActorSystem.create("myname", complete);
When working with Config objects, keep in mind that there are three “layers” in the cake:
• ConfigFactory.defaultOverrides() (system properties)
• the app’s settings
• ConfigFactory.defaultReference() (reference.conf)
The normal goal is to customize the middle layer while leaving the other two alone.
• ConfigFactory.load() loads the whole stack
• the overloads of ConfigFactory.load() let you specify a different middle layer
• the ConfigFactory.parse() variations load single files or resources
To stack two layers, use override.withFallback(fallback); try to keep system props
(defaultOverrides()) on top and reference.conf (defaultReference()) on the bottom.
Do keep in mind, you can often just add another include statement in application.conf rather than
writing code. Includes at the top of application.conf will be overridden by the rest of application.
conf, while those at the bottom will override the earlier stuff.
3.9.10 Actor Deployment Configuration
Deployment settings for specific actors can be defined in the akka.actor.deployment section of the configuration. In the deployment section it is possible to define things like dispatcher, mailbox, router settings, and
remote deployment. Configuration of these features are described in the chapters detailing corresponding topics.
An example may look like this:
3.9. Configuration
43
Akka Scala Documentation, Release 2.5.0-RC1
akka.actor.deployment {
# '/user/actorA/actorB' is a remote deployed actor
/actorA/actorB {
remote = "akka.tcp://sampleActorSystem@127.0.0.1:2553"
}
# all direct children of '/user/actorC' have a dedicated dispatcher
"/actorC/*" {
dispatcher = my-dispatcher
}
˓→
# all descendants of '/user/actorC' (direct children, and their children
recursively)
# have a dedicated dispatcher
"/actorC/**" {
dispatcher = my-dispatcher
}
# '/user/actorD/actorE' has a special priority mailbox
/actorD/actorE {
mailbox = prio-mailbox
}
# '/user/actorF/actorG/actorH' is a random pool
/actorF/actorG/actorH {
router = random-pool
nr-of-instances = 5
}
}
my-dispatcher {
fork-join-executor.parallelism-min = 10
fork-join-executor.parallelism-max = 10
}
prio-mailbox {
mailbox-type = "a.b.MyPrioMailbox"
}
Note: The deployment section for a specific actor is identified by the path of the actor relative to /user.
You can use asterisks as wildcard matches for the actor path sections, so you could specify: /*/sampleActor
and that would match all sampleActor on that level in the hierarchy. In addition, please note:
• you can also use wildcards in the last position to match all actors at a certain level: /someParent/*
• you can use double-wildcards in the last position to match all child actors and their children recursively:
/someParent/**
• non-wildcard matches always have higher priority to match than wildcards, and single wildcard matches
have higher priority than double-wildcards, so: /foo/bar is considered more specific than /foo/*,
which is considered more specific than /foo/**. Only the highest priority match is used
• wildcards cannot be used to partially match section, like this: /foo*/bar, /f*o/bar etc.
Note: Double-wildcards can only be placed in the last position.
3.9. Configuration
44
Akka Scala Documentation, Release 2.5.0-RC1
3.9.11 Listing of the Reference Configuration
Each Akka module has a reference configuration file with the default values.
akka-actor
####################################
# Akka Actor Reference Config File #
####################################
# This is the reference config file that contains all the default settings.
# Make your edits/overrides in your application.conf.
# Akka version, checked against the runtime version of Akka. Loaded from generated
˓→conf file.
include "version"
akka {
# Home directory of Akka, modules in the deploy directory will be loaded
home = ""
# Loggers to register at boot time (akka.event.Logging$DefaultLogger logs
# to STDOUT)
loggers = ["akka.event.Logging$DefaultLogger"]
# Filter of log events that is used by the LoggingAdapter before
# publishing log events to the eventStream. It can perform
# fine grained filtering based on the log source. The default
# implementation filters on the `loglevel`.
# FQCN of the LoggingFilter. The Class of the FQCN must implement
# akka.event.LoggingFilter and have a public constructor with
# (akka.actor.ActorSystem.Settings, akka.event.EventStream) parameters.
logging-filter = "akka.event.DefaultLoggingFilter"
# Specifies the default loggers dispatcher
loggers-dispatcher = "akka.actor.default-dispatcher"
# Loggers are created and registered synchronously during ActorSystem
# start-up, and since they are actors, this timeout is used to bound the
# waiting time
logger-startup-timeout = 5s
# Log level used by the configured loggers (see "loggers") as soon
# as they have been started; before that, see "stdout-loglevel"
# Options: OFF, ERROR, WARNING, INFO, DEBUG
loglevel = "INFO"
# Log level for the very basic logger activated during ActorSystem startup.
# This logger prints the log messages to stdout (System.out).
# Options: OFF, ERROR, WARNING, INFO, DEBUG
stdout-loglevel = "WARNING"
# Log the complete configuration at INFO level when the actor system is started.
# This is useful when you are uncertain of what configuration is used.
log-config-on-start = off
# Log at info level when messages are sent to dead letters.
# Possible values:
# on: all dead letters are logged
# off: no logging of dead letters
# n: positive integer, number of dead letters that will be logged
log-dead-letters = 10
3.9. Configuration
45
Akka Scala Documentation, Release 2.5.0-RC1
# Possibility to turn off logging of dead letters while the actor system
# is shutting down. Logging is only done when enabled by 'log-dead-letters'
# setting.
log-dead-letters-during-shutdown = on
# List FQCN of extensions which shall be loaded at actor system startup.
# Library extensions are regular extensions that are loaded at startup and are
# available for third party library authors to enable auto-loading of extensions
˓→when
# present on the classpath. This is done by appending entries:
# 'library-extensions += "Extension"' in the library `reference.conf`.
#
# Should not be set by end user applications in 'application.conf', use the
˓→extensions property for that
#
library-extensions = ${?akka.library-extensions} []
# List FQCN of extensions which shall be loaded at actor system startup.
# Should be on the format: 'extensions = ["foo", "bar"]' etc.
# See the Akka Documentation for more info about Extensions
extensions = []
# Toggles whether threads created by this ActorSystem should be daemons or not
daemonic = off
# JVM shutdown, System.exit(-1), in case of a fatal error,
# such as OutOfMemoryError
jvm-exit-on-fatal-error = on
actor {
# Either one of "local", "remote" or "cluster" or the
# FQCN of the ActorRefProvider to be used; the below is the built-in default,
# note that "remote" and "cluster" requires the akka-remote and akka-cluster
# artifacts to be on the classpath.
provider = "local"
# The guardian "/user" will use this class to obtain its supervisorStrategy.
# It needs to be a subclass of akka.actor.SupervisorStrategyConfigurator.
# In addition to the default there is akka.actor.StoppingSupervisorStrategy.
guardian-supervisor-strategy = "akka.actor.DefaultSupervisorStrategy"
# Timeout for ActorSystem.actorOf
creation-timeout = 20s
# Serializes and deserializes (non-primitive) messages to ensure immutability,
# this is only intended for testing.
serialize-messages = off
# Serializes and deserializes creators (in Props) to ensure that they can be
# sent over the network, this is only intended for testing. Purely local
˓→deployments
# as marked with deploy.scope == LocalScope are exempt from verification.
serialize-creators = off
# Timeout for send operations to top-level actors which are in the process
# of being started. This is only relevant if using a bounded mailbox or the
# CallingThreadDispatcher for a top-level actor.
unstarted-push-timeout = 10s
typed {
# Default timeout for typed actor methods with non-void return type
3.9. Configuration
46
Akka Scala Documentation, Release 2.5.0-RC1
timeout = 5s
}
˓→
# Mapping between ´deployment.router' short names to fully qualified class
names
router.type-mapping {
from-code = "akka.routing.NoRouter"
round-robin-pool = "akka.routing.RoundRobinPool"
round-robin-group = "akka.routing.RoundRobinGroup"
random-pool = "akka.routing.RandomPool"
random-group = "akka.routing.RandomGroup"
balancing-pool = "akka.routing.BalancingPool"
smallest-mailbox-pool = "akka.routing.SmallestMailboxPool"
broadcast-pool = "akka.routing.BroadcastPool"
broadcast-group = "akka.routing.BroadcastGroup"
scatter-gather-pool = "akka.routing.ScatterGatherFirstCompletedPool"
scatter-gather-group = "akka.routing.ScatterGatherFirstCompletedGroup"
tail-chopping-pool = "akka.routing.TailChoppingPool"
tail-chopping-group = "akka.routing.TailChoppingGroup"
consistent-hashing-pool = "akka.routing.ConsistentHashingPool"
consistent-hashing-group = "akka.routing.ConsistentHashingGroup"
}
deployment {
# deployment id pattern - on the format: /parent/child etc.
default {
# The id of the dispatcher to use for this actor.
# If undefined or empty the dispatcher specified in code
# (Props.withDispatcher) is used, or default-dispatcher if not
# specified at all.
dispatcher = ""
# The id of the mailbox to use for this actor.
# If undefined or empty the default mailbox of the configured dispatcher
# is used or if there is no mailbox configuration the mailbox specified
# in code (Props.withMailbox) is used.
# If there is a mailbox defined in the configured dispatcher then that
# overrides this setting.
mailbox = ""
# routing (load-balance) scheme to use
# - available: "from-code", "round-robin", "random", "smallest-mailbox",
#
"scatter-gather", "broadcast"
# - or:
Fully qualified class name of the router class.
#
The class must extend akka.routing.CustomRouterConfig and
#
have a public constructor with com.typesafe.config.Config
#
and optional akka.actor.DynamicAccess parameter.
# - default is "from-code";
# Whether or not an actor is transformed to a Router is decided in code
# only (Props.withRouter). The type of router can be overridden in the
# configuration; specifying "from-code" means that the values specified
# in the code shall be used.
# In case of routing, the actors to be routed to can be specified
# in several ways:
# - nr-of-instances: will create that many children
# - routees.paths: will route messages to these paths using ActorSelection,
#
i.e. will not create children
# - resizer: dynamically resizable number of routees as specified in
#
resizer below
router = "from-code"
3.9. Configuration
47
Akka Scala Documentation, Release 2.5.0-RC1
# number of children to create in case of a router;
# this setting is ignored if routees.paths is given
nr-of-instances = 1
# within is the timeout used for routers containing future calls
within = 5 seconds
# number of virtual nodes per node for consistent-hashing router
virtual-nodes-factor = 10
tail-chopping-router {
# interval is duration between sending message to next routee
interval = 10 milliseconds
}
routees {
# Alternatively to giving nr-of-instances you can specify the full
# paths of those actors which should be routed to. This setting takes
# precedence over nr-of-instances
paths = []
}
#
#
#
#
#
#
#
#
To use a dedicated dispatcher for the routees of the pool you can
define the dispatcher configuration inline with the property name
'pool-dispatcher' in the deployment section of the router.
For example:
pool-dispatcher {
fork-join-executor.parallelism-min = 5
fork-join-executor.parallelism-max = 5
}
# Routers with dynamically resizable number of routees; this feature is
# enabled by including (parts of) this section in the deployment
resizer {
enabled = off
# The fewest number of routees the router should ever have.
lower-bound = 1
# The most number of routees the router should ever have.
# Must be greater than or equal to lower-bound.
upper-bound = 10
# Threshold used to evaluate if a routee is considered to be busy
# (under pressure). Implementation depends on this value (default is 1).
# 0:
number of routees currently processing a message.
# 1:
number of routees currently processing a message has
#
some messages in mailbox.
# > 1: number of routees with at least the configured pressure-threshold
#
messages in their mailbox. Note that estimating mailbox size of
#
default UnboundedMailbox is O(N) operation.
pressure-threshold = 1
# Percentage to increase capacity whenever all routees are busy.
# For example, 0.2 would increase 20% (rounded up), i.e. if current
# capacity is 6 it will request an increase of 2 more routees.
rampup-rate = 0.2
#
#
#
#
Minimum fraction of busy routees before backing off.
For example, if this is 0.3, then we'll remove some routees only when
less than 30% of routees are busy, i.e. if current capacity is 10 and
3 are busy then the capacity is unchanged, but if 2 or less are busy
3.9. Configuration
48
Akka Scala Documentation, Release 2.5.0-RC1
# the capacity is decreased.
# Use 0.0 or negative to avoid removal of routees.
backoff-threshold = 0.3
# Fraction of routees to be removed when the resizer reaches the
# backoffThreshold.
# For example, 0.1 would decrease 10% (rounded up), i.e. if current
# capacity is 9 it will request an decrease of 1 routee.
backoff-rate = 0.1
# Number of messages between resize operation.
# Use 1 to resize before each message.
messages-per-resize = 10
}
# Routers with dynamically resizable number of routees based on
# performance metrics.
# This feature is enabled by including (parts of) this section in
# the deployment, cannot be enabled together with default resizer.
optimal-size-exploring-resizer {
enabled = off
# The fewest number of routees the router should ever have.
lower-bound = 1
# The most number of routees the router should ever have.
# Must be greater than or equal to lower-bound.
upper-bound = 10
# probability of doing a ramping down when all routees are busy
# during exploration.
chance-of-ramping-down-when-full = 0.2
# Interval between each resize attempt
action-interval = 5s
# If the routees have not been fully utilized (i.e. all routees busy)
# for such length, the resizer will downsize the pool.
downsize-after-underutilized-for = 72h
# Duration exploration, the ratio between the largest step size and
# current pool size. E.g. if the current pool size is 50, and the
# explore-step-size is 0.1, the maximum pool size change during
# exploration will be +- 5
explore-step-size = 0.1
# Probabily of doing an exploration v.s. optmization.
chance-of-exploration = 0.4
# When downsizing after a long streak of underutilization, the resizer
# will downsize the pool to the highest utiliziation multiplied by a
# a downsize rasio. This downsize ratio determines the new pools size
# in comparison to the highest utilization.
# E.g. if the highest utilization is 10, and the down size ratio
# is 0.8, the pool will be downsized to 8
downsize-ratio = 0.8
# When optimizing, the resizer only considers the sizes adjacent to the
# current size. This number indicates how many adjacent sizes to
˓→consider.
optimization-range = 16
3.9. Configuration
49
Akka Scala Documentation, Release 2.5.0-RC1
# The weight of the latest metric over old metrics when collecting
# performance metrics.
# E.g. if the last processing speed is 10 millis per message at pool
# size 5, and if the new processing speed collected is 6 millis per
# message at pool size 5. Given a weight of 0.3, the metrics
# representing pool size 5 will be 6 * 0.3 + 10 * 0.7, i.e. 8.8 millis
# Obviously, this number should be between 0 and 1.
weight-of-latest-metric = 0.5
}
}
/IO-DNS/inet-address {
mailbox = "unbounded"
router = "consistent-hashing-pool"
nr-of-instances = 4
}
"/IO-DNS/inet-address/*" {
dispatcher = "akka.actor.default-blocking-io-dispatcher"
}
}
default-dispatcher {
# Must be one of the following
# Dispatcher, PinnedDispatcher, or a FQCN to a class inheriting
# MessageDispatcherConfigurator with a public constructor with
# both com.typesafe.config.Config parameter and
# akka.dispatch.DispatcherPrerequisites parameters.
# PinnedDispatcher must be used together with executor=thread-pool-executor.
type = "Dispatcher"
# Which kind of ExecutorService to use for this dispatcher
# Valid options:
# - "default-executor" requires a "default-executor" section
# - "fork-join-executor" requires a "fork-join-executor" section
# - "thread-pool-executor" requires a "thread-pool-executor" section
# - A FQCN of a class extending ExecutorServiceConfigurator
executor = "default-executor"
# This will be used if you have set "executor = "default-executor"".
# If an ActorSystem is created with a given ExecutionContext, this
# ExecutionContext will be used as the default executor for all
# dispatchers in the ActorSystem configured with
# executor = "default-executor". Note that "default-executor"
# is the default value for executor, and therefore used if not
# specified otherwise. If no ExecutionContext is given,
# the executor configured in "fallback" will be used.
default-executor {
fallback = "fork-join-executor"
}
# This will be used if you have set "executor = "fork-join-executor""
# Underlying thread pool implementation is akka.dispatch.forkjoin.
˓→ForkJoinPool
fork-join-executor {
# Min number of threads to cap factor-based parallelism number to
parallelism-min = 8
# The parallelism factor is used to determine thread pool size using the
# following formula: ceil(available processors * factor). Resulting size
# is then bounded by the parallelism-min and parallelism-max values.
parallelism-factor = 3.0
# Max number of threads to cap factor-based parallelism number to
3.9. Configuration
50
Akka Scala Documentation, Release 2.5.0-RC1
parallelism-max = 64
˓→
# Setting to "FIFO" to use queue like peeking mode which "poll" or "LIFO"
to use stack
# like peeking mode which "pop".
task-peeking-mode = "FIFO"
}
# This will be used if you have set "executor = "thread-pool-executor""
# Underlying thread pool implementation is java.util.concurrent.
˓→ThreadPoolExecutor
thread-pool-executor {
# Keep alive time for threads
keep-alive-time = 60s
# Define a fixed thread pool size with this property. The corePoolSize
# and the maximumPoolSize of the ThreadPoolExecutor will be set to this
# value, if it is defined. Then the other pool-size properties will not
# be used.
#
# Valid values are: `off` or a positive integer.
fixed-pool-size = off
# Min number of threads to cap factor-based corePoolSize number to
core-pool-size-min = 8
# The core-pool-size-factor is used to determine corePoolSize of the
# ThreadPoolExecutor using the following formula:
# ceil(available processors * factor).
# Resulting size is then bounded by the core-pool-size-min and
# core-pool-size-max values.
core-pool-size-factor = 3.0
# Max number of threads to cap factor-based corePoolSize number to
core-pool-size-max = 64
# Minimum number of threads to cap factor-based maximumPoolSize number to
max-pool-size-min = 8
# The max-pool-size-factor is used to determine maximumPoolSize of the
# ThreadPoolExecutor using the following formula:
# ceil(available processors * factor)
# The maximumPoolSize will not be less than corePoolSize.
# It is only used if using a bounded task queue.
max-pool-size-factor = 3.0
# Max number of threads to cap factor-based maximumPoolSize number to
max-pool-size-max = 64
# Specifies the bounded capacity of the task queue (< 1 == unbounded)
task-queue-size = -1
# Specifies which type of task queue will be used, can be "array" or
# "linked" (default)
task-queue-type = "linked"
# Allow core threads to time out
allow-core-timeout = on
}
# How long time the dispatcher will wait for new actors until it shuts down
shutdown-timeout = 1s
3.9. Configuration
51
Akka Scala Documentation, Release 2.5.0-RC1
# Throughput defines the number of messages that are processed in a batch
# before the thread is returned to the pool. Set to 1 for as fair as
˓→possible.
throughput = 5
# Throughput deadline for Dispatcher, set to 0 or negative for no deadline
throughput-deadline-time = 0ms
# For BalancingDispatcher: If the balancing dispatcher should attempt to
# schedule idle actors using the same dispatcher when a message comes in,
# and the dispatchers ExecutorService is not fully busy already.
attempt-teamwork = on
# If this dispatcher requires a specific type of mailbox, specify the
# fully-qualified class name here; the actually created mailbox will
# be a subtype of this type. The empty string signifies no requirement.
mailbox-requirement = ""
}
default-blocking-io-dispatcher {
type = "Dispatcher"
executor = "thread-pool-executor"
throughput = 1
thread-pool-executor {
fixed-pool-size = 16
}
}
default-mailbox {
# FQCN of the MailboxType. The Class of the FQCN must have a public
# constructor with
# (akka.actor.ActorSystem.Settings, com.typesafe.config.Config) parameters.
mailbox-type = "akka.dispatch.UnboundedMailbox"
# If the mailbox is bounded then it uses this setting to determine its
# capacity. The provided value must be positive.
# NOTICE:
# Up to version 2.1 the mailbox type was determined based on this setting;
# this is no longer the case, the type must explicitly be a bounded mailbox.
mailbox-capacity = 1000
# If the mailbox is bounded then this is the timeout for enqueueing
# in case the mailbox is full. Negative values signify infinite
# timeout, which should be avoided as it bears the risk of dead-lock.
mailbox-push-timeout-time = 10s
# For Actor with Stash: The default capacity of the stash.
# If negative (or zero) then an unbounded stash is used (default)
# If positive then a bounded stash is used and the capacity is set using
# the property
stash-capacity = -1
}
mailbox {
# Mapping between message queue semantics and mailbox configurations.
# Used by akka.dispatch.RequiresMessageQueue[T] to enforce different
# mailbox types on actors.
# If your Actor implements RequiresMessageQueue[T], then when you create
# an instance of that actor its mailbox type will be decided by looking
# up a mailbox configuration via T in this mapping
requirements {
"akka.dispatch.UnboundedMessageQueueSemantics" =
3.9. Configuration
52
Akka Scala Documentation, Release 2.5.0-RC1
akka.actor.mailbox.unbounded-queue-based
"akka.dispatch.BoundedMessageQueueSemantics" =
akka.actor.mailbox.bounded-queue-based
"akka.dispatch.DequeBasedMessageQueueSemantics" =
akka.actor.mailbox.unbounded-deque-based
"akka.dispatch.UnboundedDequeBasedMessageQueueSemantics" =
akka.actor.mailbox.unbounded-deque-based
"akka.dispatch.BoundedDequeBasedMessageQueueSemantics" =
akka.actor.mailbox.bounded-deque-based
"akka.dispatch.MultipleConsumerSemantics" =
akka.actor.mailbox.unbounded-queue-based
"akka.dispatch.ControlAwareMessageQueueSemantics" =
akka.actor.mailbox.unbounded-control-aware-queue-based
"akka.dispatch.UnboundedControlAwareMessageQueueSemantics" =
akka.actor.mailbox.unbounded-control-aware-queue-based
"akka.dispatch.BoundedControlAwareMessageQueueSemantics" =
akka.actor.mailbox.bounded-control-aware-queue-based
"akka.event.LoggerMessageQueueSemantics" =
akka.actor.mailbox.logger-queue
}
unbounded-queue-based {
# FQCN of the MailboxType, The Class of the FQCN must have a public
# constructor with (akka.actor.ActorSystem.Settings,
# com.typesafe.config.Config) parameters.
mailbox-type = "akka.dispatch.UnboundedMailbox"
}
bounded-queue-based {
# FQCN of the MailboxType, The Class of the FQCN must have a public
# constructor with (akka.actor.ActorSystem.Settings,
# com.typesafe.config.Config) parameters.
mailbox-type = "akka.dispatch.BoundedMailbox"
}
unbounded-deque-based {
# FQCN of the MailboxType, The Class of the FQCN must have a public
# constructor with (akka.actor.ActorSystem.Settings,
# com.typesafe.config.Config) parameters.
mailbox-type = "akka.dispatch.UnboundedDequeBasedMailbox"
}
bounded-deque-based {
# FQCN of the MailboxType, The Class of the FQCN must have a public
# constructor with (akka.actor.ActorSystem.Settings,
# com.typesafe.config.Config) parameters.
mailbox-type = "akka.dispatch.BoundedDequeBasedMailbox"
}
unbounded-control-aware-queue-based {
# FQCN of the MailboxType, The Class of the FQCN must have a public
# constructor with (akka.actor.ActorSystem.Settings,
# com.typesafe.config.Config) parameters.
mailbox-type = "akka.dispatch.UnboundedControlAwareMailbox"
}
bounded-control-aware-queue-based {
# FQCN of the MailboxType, The Class of the FQCN must have a public
# constructor with (akka.actor.ActorSystem.Settings,
# com.typesafe.config.Config) parameters.
mailbox-type = "akka.dispatch.BoundedControlAwareMailbox"
}
3.9. Configuration
53
Akka Scala Documentation, Release 2.5.0-RC1
# The LoggerMailbox will drain all messages in the mailbox
# when the system is shutdown and deliver them to the StandardOutLogger.
# Do not change this unless you know what you are doing.
logger-queue {
mailbox-type = "akka.event.LoggerMailboxType"
}
}
debug {
# enable function of Actor.loggable(), which is to log any received message
# at DEBUG level, see the “Testing Actor Systems” section of the Akka
# Documentation at http://akka.io/docs
receive = off
# enable DEBUG logging of all AutoReceiveMessages (Kill, PoisonPill et.c.)
autoreceive = off
# enable DEBUG logging of actor lifecycle changes
lifecycle = off
# enable DEBUG logging of all LoggingFSMs for events, transitions and timers
fsm = off
# enable DEBUG logging of subscription changes on the eventStream
event-stream = off
# enable DEBUG logging of unhandled messages
unhandled = off
# enable WARN logging of misconfigured routers
router-misconfiguration = off
}
# SECURITY BEST-PRACTICE is to disable java serialization for its multiple
# known attack surfaces.
#
# This setting is a short-cut to
# - using DisabledJavaSerializer instead of JavaSerializer
# - enable-additional-serialization-bindings = on
#
# Completely disable the use of `akka.serialization.JavaSerialization` by the
# Akka Serialization extension, instead DisabledJavaSerializer will
# be inserted which will fail explicitly if attempts to use java serialization
˓→are made.
#
# The log messages emitted by such serializer SHOULD be be treated as potential
# attacks which the serializer prevented, as they MAY indicate an external
˓→operator
# attempting to send malicious messages intending to use java serialization as
˓→attack vector.
# The attempts are logged with the SECURITY marker.
#
# Please note that this option does not stop you from manually invoking java
˓→serialization
#
# The default value for this might be changed to off in future versions of
˓→Akka.
allow-java-serialization = on
# Entries for pluggable serializers and their bindings.
serializers {
java = "akka.serialization.JavaSerializer"
bytes = "akka.serialization.ByteArraySerializer"
3.9. Configuration
54
Akka Scala Documentation, Release 2.5.0-RC1
}
# Class to Serializer binding. You only need to specify the name of an
# interface or abstract base class of the messages. In case of ambiguity it
# is using the most specific configured class, or giving a warning and
# choosing the “first” one.
#
# To disable one of the default serializers, assign its class to "none", like
# "java.io.Serializable" = none
serialization-bindings {
"[B" = bytes
"java.io.Serializable" = java
}
# Additional serialization-bindings that are replacing Java serialization are
# defined in this section for backwards compatibility reasons. They are
˓→included
# by default but can be excluded for backwards compatibility with Akka 2.4.x.
# They can be disabled with enable-additional-serialization-bindings=off.
#
# This should only be needed for backwards compatibility reasons.
enable-additional-serialization-bindings = on
# Additional serialization-bindings that are replacing Java serialization are
# defined in this section for backwards compatibility reasons. They are
˓→included
# by default but can be excluded for backwards compatibility with Akka 2.4.x.
# They can be disabled with enable-additional-serialization-bindings=off.
additional-serialization-bindings {
}
# Log warnings when the default Java serialization is used to serialize
messages.
# The default serializer uses Java serialization which is not very performant
˓→and should not
# be used in production environments unless you don't care about performance.
˓→In that case
# you can turn this off.
warn-about-java-serializer-usage = on
˓→
# To be used with the above warn-about-java-serializer-usage
# When warn-about-java-serializer-usage = on, and this warn-on-no˓→serialization-verification = off,
# warnings are suppressed for classes extending
˓→NoSerializationVerificationNeeded
# to reduce noize.
warn-on-no-serialization-verification = on
# Configuration namespace of serialization identifiers.
# Each serializer implementation must have an entry in the following format:
# `akka.actor.serialization-identifiers."FQCN" = ID`
# where `FQCN` is fully qualified class name of the serializer implementation
# and `ID` is globally unique serializer identifier number.
# Identifier values from 0 to 40 are reserved for Akka internal usage.
serialization-identifiers {
"akka.serialization.JavaSerializer" = 1
"akka.serialization.ByteArraySerializer" = 4
}
# Configuration items which are used by the akka.actor.ActorDSL._ methods
dsl {
# Maximum queue size of the actor created by newInbox(); this protects
# against faulty programs which use select() and consistently miss messages
3.9. Configuration
55
Akka Scala Documentation, Release 2.5.0-RC1
inbox-size = 1000
# Default timeout to assume for operations like Inbox.receive et al
default-timeout = 5s
}
}
# Used to set the behavior of the scheduler.
# Changing the default values may change the system behavior drastically so make
# sure you know what you're doing! See the Scheduler section of the Akka
# Documentation for more details.
scheduler {
# The LightArrayRevolverScheduler is used as the default scheduler in the
# system. It does not execute the scheduled tasks on exact time, but on every
# tick, it will run everything that is (over)due. You can increase or decrease
# the accuracy of the execution timing by specifying smaller or larger tick
# duration. If you are scheduling a lot of tasks you should consider increasing
# the ticks per wheel.
# Note that it might take up to 1 tick to stop the Timer, so setting the
# tick-duration to a high value will make shutting down the actor system
# take longer.
tick-duration = 10ms
# The timer uses a circular wheel of buckets to store the timer tasks.
# This should be set such that the majority of scheduled timeouts (for high
# scheduling frequency) will be shorter than one rotation of the wheel
# (ticks-per-wheel * ticks-duration)
# THIS MUST BE A POWER OF TWO!
ticks-per-wheel = 512
# This setting selects the timer implementation which shall be loaded at
# system start-up.
# The class given here must implement the akka.actor.Scheduler interface
# and offer a public constructor which takes three arguments:
# 1) com.typesafe.config.Config
# 2) akka.event.LoggingAdapter
# 3) java.util.concurrent.ThreadFactory
implementation = akka.actor.LightArrayRevolverScheduler
# When shutting down the scheduler, there will typically be a thread which
# needs to be stopped, and this timeout determines how long to wait for
# that to happen. In case of timeout the shutdown of the actor system will
# proceed without running possibly still enqueued tasks.
shutdown-timeout = 5s
}
io {
# By default the select loops run on dedicated threads, hence using a
# PinnedDispatcher
pinned-dispatcher {
type = "PinnedDispatcher"
executor = "thread-pool-executor"
thread-pool-executor.allow-core-timeout = off
}
tcp {
# The number of selectors to stripe the served channels over; each of
# these will use one select loop on the selector-dispatcher.
nr-of-selectors = 1
# Maximum number of open channels supported by this TCP module; there is
3.9. Configuration
56
Akka Scala Documentation, Release 2.5.0-RC1
# no intrinsic general limit, this setting is meant to enable DoS
# protection by limiting the number of concurrently connected clients.
# Also note that this is a "soft" limit; in certain cases the implementation
# will accept a few connections more or a few less than the number configured
# here. Must be an integer > 0 or "unlimited".
max-channels = 256000
# When trying to assign a new connection to a selector and the chosen
# selector is at full capacity, retry selector choosing and assignment
# this many times before giving up
selector-association-retries = 10
# The maximum number of connection that are accepted in one go,
# higher numbers decrease latency, lower numbers increase fairness on
# the worker-dispatcher
batch-accept-limit = 10
# The number of bytes per direct buffer in the pool used to read or write
# network data from the kernel.
direct-buffer-size = 128 KiB
# The maximal number of direct buffers kept in the direct buffer pool for
# reuse.
direct-buffer-pool-limit = 1000
# The duration a connection actor waits for a `Register` message from
# its commander before aborting the connection.
register-timeout = 5s
# The maximum number of bytes delivered by a `Received` message. Before
# more data is read from the network the connection actor will try to
# do other work.
# The purpose of this setting is to impose a smaller limit than the
# configured receive buffer size. When using value 'unlimited' it will
# try to read all from the receive buffer.
max-received-message-size = unlimited
# Enable fine grained logging of what goes on inside the implementation.
# Be aware that this may log more than once per message sent to the actors
# of the tcp implementation.
trace-logging = off
# Fully qualified config path which holds the dispatcher configuration
# to be used for running the select() calls in the selectors
selector-dispatcher = "akka.io.pinned-dispatcher"
# Fully qualified config path which holds the dispatcher configuration
# for the read/write worker actors
worker-dispatcher = "akka.actor.default-dispatcher"
# Fully qualified config path which holds the dispatcher configuration
# for the selector management actors
management-dispatcher = "akka.actor.default-dispatcher"
# Fully qualified config path which holds the dispatcher configuration
# on which file IO tasks are scheduled
file-io-dispatcher = "akka.actor.default-blocking-io-dispatcher"
#
#
#
#
#
The maximum number of bytes (or "unlimited") to transfer in one batch
when using `WriteFile` command which uses `FileChannel.transferTo` to
pipe files to a TCP socket. On some OS like Linux `FileChannel.transferTo`
may block for a long time when network IO is faster than file IO.
Decreasing the value may improve fairness while increasing may improve
3.9. Configuration
57
Akka Scala Documentation, Release 2.5.0-RC1
# throughput.
file-io-transferTo-limit = 512 KiB
# The number of times to retry the `finishConnect` call after being notified
about
# OP_CONNECT. Retries are needed if the OP_CONNECT notification doesn't
˓→imply that
# `finishConnect` will succeed, which is the case on Android.
finish-connect-retries = 5
˓→
# On Windows connection aborts are not reliably detected unless an OP_READ is
# registered on the selector _after_ the connection has been reset. This
# workaround enables an OP_CONNECT which forces the abort to be visible on
˓→Windows.
# Enabling this setting on other platforms than Windows will cause various
˓→failures
# and undefined behavior.
# Possible values of this key are on, off and auto where auto will enable the
# workaround if Windows is detected automatically.
windows-connection-abort-workaround-enabled = off
}
udp {
# The number of selectors to stripe the served channels over; each of
# these will use one select loop on the selector-dispatcher.
nr-of-selectors = 1
# Maximum number of open channels supported by this UDP module Generally
# UDP does not require a large number of channels, therefore it is
# recommended to keep this setting low.
max-channels = 4096
# The select loop can be used in two modes:
# - setting "infinite" will select without a timeout, hogging a thread
# - setting a positive timeout will do a bounded select call,
#
enabling sharing of a single thread between multiple selectors
#
(in this case you will have to use a different configuration for the
#
selector-dispatcher, e.g. using "type=Dispatcher" with size 1)
# - setting it to zero means polling, i.e. calling selectNow()
select-timeout = infinite
# When trying to assign a new connection to a selector and the chosen
# selector is at full capacity, retry selector choosing and assignment
# this many times before giving up
selector-association-retries = 10
# The maximum number of datagrams that are read in one go,
# higher numbers decrease latency, lower numbers increase fairness on
# the worker-dispatcher
receive-throughput = 3
# The number of bytes per direct buffer in the pool used to read or write
# network data from the kernel.
direct-buffer-size = 128 KiB
# The maximal number of direct buffers kept in the direct buffer pool for
# reuse.
direct-buffer-pool-limit = 1000
# Enable fine grained logging of what goes on inside the implementation.
# Be aware that this may log more than once per message sent to the actors
# of the tcp implementation.
3.9. Configuration
58
Akka Scala Documentation, Release 2.5.0-RC1
trace-logging = off
# Fully qualified config path which holds the dispatcher configuration
# to be used for running the select() calls in the selectors
selector-dispatcher = "akka.io.pinned-dispatcher"
# Fully qualified config path which holds the dispatcher configuration
# for the read/write worker actors
worker-dispatcher = "akka.actor.default-dispatcher"
# Fully qualified config path which holds the dispatcher configuration
# for the selector management actors
management-dispatcher = "akka.actor.default-dispatcher"
}
udp-connected {
# The number of selectors to stripe the served channels over; each of
# these will use one select loop on the selector-dispatcher.
nr-of-selectors = 1
# Maximum number of open channels supported by this UDP module Generally
# UDP does not require a large number of channels, therefore it is
# recommended to keep this setting low.
max-channels = 4096
# The select loop can be used in two modes:
# - setting "infinite" will select without a timeout, hogging a thread
# - setting a positive timeout will do a bounded select call,
#
enabling sharing of a single thread between multiple selectors
#
(in this case you will have to use a different configuration for the
#
selector-dispatcher, e.g. using "type=Dispatcher" with size 1)
# - setting it to zero means polling, i.e. calling selectNow()
select-timeout = infinite
# When trying to assign a new connection to a selector and the chosen
# selector is at full capacity, retry selector choosing and assignment
# this many times before giving up
selector-association-retries = 10
# The maximum number of datagrams that are read in one go,
# higher numbers decrease latency, lower numbers increase fairness on
# the worker-dispatcher
receive-throughput = 3
# The number of bytes per direct buffer in the pool used to read or write
# network data from the kernel.
direct-buffer-size = 128 KiB
# The maximal number of direct buffers kept in the direct buffer pool for
# reuse.
direct-buffer-pool-limit = 1000
# Enable fine grained logging of what goes on inside the implementation.
# Be aware that this may log more than once per message sent to the actors
# of the tcp implementation.
trace-logging = off
# Fully qualified config path which holds the dispatcher configuration
# to be used for running the select() calls in the selectors
selector-dispatcher = "akka.io.pinned-dispatcher"
# Fully qualified config path which holds the dispatcher configuration
3.9. Configuration
59
Akka Scala Documentation, Release 2.5.0-RC1
# for the read/write worker actors
worker-dispatcher = "akka.actor.default-dispatcher"
# Fully qualified config path which holds the dispatcher configuration
# for the selector management actors
management-dispatcher = "akka.actor.default-dispatcher"
}
dns {
# Fully qualified config path which holds the dispatcher configuration
# for the manager and resolver router actors.
# For actual router configuration see akka.actor.deployment./IO-DNS/*
dispatcher = "akka.actor.default-dispatcher"
# Name of the subconfig at path akka.io.dns, see inet-address below
resolver = "inet-address"
inet-address {
# Must implement akka.io.DnsProvider
provider-object = "akka.io.InetAddressDnsProvider"
# To set the time to cache name resolutions
# Possible values:
# default: sun.net.InetAddressCachePolicy.get() and getNegative()
# forever: cache forever
# never: no caching
# n [time unit]: positive timeout with unit, for example "30 s"
positive-ttl = default
negative-ttl = default
# How often to sweep out expired cache entries.
# Note that this interval has nothing to do with TTLs
cache-cleanup-interval = 120s
}
}
}
# CoordinatedShutdown is an extension that will perform registered
# tasks in the order that is defined by the phases. It is started
# by calling CoordinatedShutdown(system).run(). This can be triggered
# by different things, for example:
# - JVM shutdown hook will by default run CoordinatedShutdown
# - Cluster node will automatically run CoordinatedShutdown when it
#
sees itself as Exiting
# - A management console or other application specific command can
#
run CoordinatedShutdown
coordinated-shutdown {
# The timeout that will be used for a phase if not specified with
# 'timeout' in the phase
default-phase-timeout = 5 s
# Terminate the ActorSystem in the last phase actor-system-terminate.
terminate-actor-system = on
# Exit the JVM (System.exit(0)) in the last phase actor-system-terminate
# if this is set to 'on'. It is done after termination of the
# ActorSystem if terminate-actor-system=on, otherwise it is done
# immediately when the last phase is reached.
exit-jvm = off
# Run the coordinated shutdown when the JVM process exits, e.g.
# via kill SIGTERM signal (SIGINT ctrl-c doesn't work).
3.9. Configuration
60
Akka Scala Documentation, Release 2.5.0-RC1
run-by-jvm-shutdown-hook = on
#//#coordinated-shutdown-phases
# CoordinatedShutdown will run the tasks that are added to these
# phases. The phases can be ordered as a DAG by defining the
# dependencies between the phases.
# Each phase is defined as a named config section with the
# following optional properties:
# - timeout=15s: Override the default-phase-timeout for this phase.
# - recover=off: If the phase fails the shutdown is aborted
#
and depending phases will not be executed.
# depends-on=[]: Run the phase after the given phases
phases {
# The first pre-defined phase that applications can add tasks to.
# Note that more phases can be be added in the application's
# configuration by overriding this phase with an additional
# depends-on.
before-service-unbind {
}
# Stop accepting new incoming requests in for example HTTP.
service-unbind {
depends-on = [before-service-unbind]
}
# Wait for requests that are in progress to be completed.
service-requests-done {
depends-on = [service-unbind]
}
# Final shutdown of service endpoints.
service-stop {
depends-on = [service-requests-done]
}
# Phase for custom application tasks that are to be run
# after service shutdown and before cluster shutdown.
before-cluster-shutdown {
depends-on = [service-stop]
}
# Graceful shutdown of the Cluster Sharding regions.
cluster-sharding-shutdown-region {
timeout = 10 s
depends-on = [before-cluster-shutdown]
}
# Emit the leave command for the node that is shutting down.
cluster-leave {
depends-on = [cluster-sharding-shutdown-region]
}
# Shutdown cluster singletons
cluster-exiting {
timeout = 10 s
depends-on = [cluster-leave]
}
# Wait until exiting has been completed
cluster-exiting-done {
depends-on = [cluster-exiting]
}
3.9. Configuration
61
Akka Scala Documentation, Release 2.5.0-RC1
# Shutdown the cluster extension
cluster-shutdown {
depends-on = [cluster-exiting-done]
}
# Phase for custom application tasks that are to be run
# after cluster shutdown and before ActorSystem termination.
before-actor-system-terminate {
depends-on = [cluster-shutdown]
}
# Last phase. See terminate-actor-system and exit-jvm above.
# Don't add phases that depends on this phase because the
# dispatcher and scheduler of the ActorSystem have been shutdown.
actor-system-terminate {
timeout = 10 s
depends-on = [before-actor-system-terminate]
}
}
#//#coordinated-shutdown-phases
}
}
akka-agent
####################################
# Akka Agent Reference Config File #
####################################
# This is the reference config file that contains all the default settings.
# Make your edits/overrides in your application.conf.
akka {
agent {
# The dispatcher used for agent-send-off actor
send-off-dispatcher {
executor = thread-pool-executor
type = PinnedDispatcher
}
# The dispatcher used for agent-alter-off actor
alter-off-dispatcher {
executor = thread-pool-executor
type = PinnedDispatcher
}
}
}
akka-camel
####################################
# Akka Camel Reference Config File #
####################################
# This is the reference config file that contains all the default settings.
# Make your edits/overrides in your application.conf.
3.9. Configuration
62
Akka Scala Documentation, Release 2.5.0-RC1
akka {
camel {
# FQCN of the ContextProvider to be used to create or locate a CamelContext
# it must implement akka.camel.ContextProvider and have a no-arg constructor
# the built-in default create a fresh DefaultCamelContext
context-provider = akka.camel.DefaultContextProvider
# Whether JMX should be enabled or disabled for the Camel Context
jmx = off
# enable/disable streaming cache on the Camel Context
streamingCache = on
consumer {
# Configured setting which determines whether one-way communications
# between an endpoint and this consumer actor
# should be auto-acknowledged or application-acknowledged.
# This flag has only effect when exchange is in-only.
auto-ack = on
# When endpoint is out-capable (can produce responses) reply-timeout is the
# maximum time the endpoint can take to send the response before the message
# exchange fails. This setting is used for out-capable, in-only,
# manually acknowledged communication.
reply-timeout = 1m
# The duration of time to await activation of an endpoint.
activation-timeout = 10s
}
#Scheme to FQCN mappings for CamelMessage body conversions
conversions {
"file" = "java.io.InputStream"
}
}
}
akka-cluster
######################################
# Akka Cluster Reference Config File #
######################################
# This is the reference config file that contains all the default settings.
# Make your edits/overrides in your application.conf.
akka {
cluster {
# Initial contact points of the cluster.
# The nodes to join automatically at startup.
# Comma separated full URIs defined by a string on the form of
# "akka.tcp://system@hostname:port"
# Leave as empty if the node is supposed to be joined manually.
seed-nodes = []
# how long to wait for one of the seed nodes to reply to initial join request
seed-node-timeout = 5s
# If a join request fails it will be retried after this period.
# Disable join retry by specifying "off".
retry-unsuccessful-join-after = 10s
3.9. Configuration
63
Akka Scala Documentation, Release 2.5.0-RC1
# Should the 'leader' in the cluster be allowed to automatically mark
# unreachable nodes as DOWN after a configured time of unreachability?
# Using auto-down implies that two separate clusters will automatically be
# formed in case of network partition.
#
# Don't enable this in production, see 'Auto-downing (DO NOT USE)' section
# of Akka Cluster documentation.
#
# Disable with "off" or specify a duration to enable auto-down.
# If a downing-provider-class is configured this setting is ignored.
auto-down-unreachable-after = off
# Time margin after which shards or singletons that belonged to a downed/
removed
# partition are created in surviving partition. The purpose of this margin is
˓→that
# in case of a network partition the persistent actors in the non-surviving
˓→partitions
# must be stopped before corresponding persistent actors are started somewhere
˓→else.
# This is useful if you implement downing strategies that handle network
˓→partitions,
# e.g. by keeping the larger side of the partition and shutting down the
˓→smaller side.
# It will not add any extra safety for auto-down-unreachable-after, since that
˓→is not
# handling network partitions.
# Disable with "off" or specify a duration to enable.
down-removal-margin = off
˓→
# Pluggable support for downing of nodes in the cluster.
# If this setting is left empty behaviour will depend on 'auto-down-unreachable
˓→' in the following ways:
# * if it is 'off' the `NoDowning` provider is used and no automatic downing
˓→will be performed
# * if it is set to a duration the `AutoDowning` provider is with the
˓→configured downing duration
#
# If specified the value must be the fully qualified class name of a subclass
˓→of
# `akka.cluster.DowningProvider` having a public one argument constructor
˓→accepting an `ActorSystem`
downing-provider-class = ""
# Artery only setting
# When a node has been gracefully removed, let this time pass (to allow for
˓→example
# cluster singleton handover to complete) and then quarantine the removed node.
quarantine-removed-node-after=30s
# If this is set to "off", the leader will not move 'Joining' members to 'Up'
during a network
# split. This feature allows the leader to accept 'Joining' members to be
˓→'WeaklyUp'
# so they become part of the cluster even during a network split. The leader
˓→will
# move 'WeaklyUp' members to 'Up' status once convergence has been reached.
allow-weakly-up-members = on
˓→
# The roles of this member. List of strings, e.g. roles = ["A", "B"].
# The roles are part of the membership information and can be used by
# routers or other services to distribute work to certain member types,
3.9. Configuration
64
Akka Scala Documentation, Release 2.5.0-RC1
# e.g. front-end and back-end nodes.
roles = []
# Run the coordinated shutdown from phase 'cluster-shutdown' when the cluster
# is shutdown for other reasons than when leaving, e.g. when downing. This
# will terminate the ActorSystem when the cluster extension is shutdown.
run-coordinated-shutdown-when-down = on
role {
# Minimum required number of members of a certain role before the leader
# changes member status of 'Joining' members to 'Up'. Typically used together
# with 'Cluster.registerOnMemberUp' to defer some action, such as starting
# actors, until the cluster has reached a certain size.
# E.g. to require 2 nodes with role 'frontend' and 3 nodes with role 'backend
˓→':
#
frontend.min-nr-of-members = 2
#
backend.min-nr-of-members = 3
#<role-name>.min-nr-of-members = 1
}
# Minimum required number of members before the leader changes member status
# of 'Joining' members to 'Up'. Typically used together with
# 'Cluster.registerOnMemberUp' to defer some action, such as starting actors,
# until the cluster has reached a certain size.
min-nr-of-members = 1
# Enable/disable info level logging of cluster events
log-info = on
# Enable or disable JMX MBeans for management of the cluster
jmx.enabled = on
# Enable or disable multiple JMX MBeans in the same JVM
# If this is disabled, the MBean Object name is "akka:type=Cluster"
# If this is enabled, them MBean Object names become "akka:type=Cluster,port=
˓→$clusterPortNumber"
jmx.multi-mbeans-in-same-jvm = off
# how long should the node wait before starting the periodic tasks
# maintenance tasks?
periodic-tasks-initial-delay = 1s
# how often should the node send out gossip information?
gossip-interval = 1s
# discard incoming gossip messages if not handled within this duration
gossip-time-to-live = 2s
# how often should the leader perform maintenance tasks?
leader-actions-interval = 1s
# how often should the node move nodes, marked as unreachable by the failure
# detector, out of the membership ring?
unreachable-nodes-reaper-interval = 1s
# How often the current internal stats should be published.
# A value of 0s can be used to always publish the stats, when it happens.
# Disable with "off".
publish-stats-interval = off
# The id of the dispatcher to use for cluster actors. If not specified
# default dispatcher is used.
# If specified you need to define the settings of the actual dispatcher.
3.9. Configuration
65
Akka Scala Documentation, Release 2.5.0-RC1
use-dispatcher = ""
# Gossip to random node with newer or older state information, if any with
# this probability. Otherwise Gossip to any random live node.
# Probability value is between 0.0 and 1.0. 0.0 means never, 1.0 means always.
gossip-different-view-probability = 0.8
# Reduced the above probability when the number of nodes in the cluster
# greater than this value.
reduce-gossip-different-view-probability = 400
# Settings for the Phi accrual failure detector (http://www.jaist.ac.jp/~
defago/files/pdf/IS_RR_2004_010.pdf
# [Hayashibara et al]) used by the cluster subsystem to detect unreachable
# members.
# The default PhiAccrualFailureDetector will trigger if there are no
˓→heartbeats within
# the duration heartbeat-interval + acceptable-heartbeat-pause + threshold_
˓→adjustment,
# i.e. around 5.5 seconds with default settings.
failure-detector {
˓→
# FQCN of the failure detector implementation.
# It must implement akka.remote.FailureDetector and have
# a public constructor with a com.typesafe.config.Config and
# akka.actor.EventStream parameter.
implementation-class = "akka.remote.PhiAccrualFailureDetector"
# How often keep-alive heartbeat messages should be sent to each connection.
heartbeat-interval = 1 s
# Defines the failure detector threshold.
# A low threshold is prone to generate many wrong suspicions but ensures
# a quick detection in the event of a real crash. Conversely, a high
# threshold generates fewer mistakes but needs more time to detect
# actual crashes.
threshold = 8.0
# Number of the samples of inter-heartbeat arrival times to adaptively
# calculate the failure timeout for connections.
max-sample-size = 1000
# Minimum standard deviation to use for the normal distribution in
# AccrualFailureDetector. Too low standard deviation might result in
# too much sensitivity for sudden, but normal, deviations in heartbeat
# inter arrival times.
min-std-deviation = 100 ms
# Number of potentially lost/delayed heartbeats that will be
# accepted before considering it to be an anomaly.
# This margin is important to be able to survive sudden, occasional,
# pauses in heartbeat arrivals, due to for example garbage collect or
# network drop.
acceptable-heartbeat-pause = 3 s
# Number of member nodes that each member will send heartbeat messages to,
# i.e. each node will be monitored by this number of other nodes.
monitored-by-nr-of-members = 5
# After the heartbeat request has been sent the first failure detection
# will start after this period, even though no heartbeat message has
# been received.
expected-response-after = 1 s
3.9. Configuration
66
Akka Scala Documentation, Release 2.5.0-RC1
}
# If the tick-duration of the default scheduler is longer than the
# tick-duration configured here a dedicated scheduler will be used for
# periodic tasks of the cluster, otherwise the default scheduler is used.
# See akka.scheduler settings for more details.
scheduler {
tick-duration = 33ms
ticks-per-wheel = 512
}
debug {
# log heartbeat events (very verbose, useful mostly when debugging
˓→heartbeating issues)
verbose-heartbeat-logging = off
}
}
actor.deployment.default.cluster {
# enable cluster aware router that deploys to nodes in the cluster
enabled = off
# Maximum number of routees that will be deployed on each cluster
# member node.
# Note that max-total-nr-of-instances defines total number of routees, but
# number of routees per node will not be exceeded, i.e. if you
# define max-total-nr-of-instances = 50 and max-nr-of-instances-per-node = 2
# it will deploy 2 routees per new member in the cluster, up to
# 25 members.
max-nr-of-instances-per-node = 1
# Maximum number of routees that will be deployed, in total
# on all nodes. See also description of max-nr-of-instances-per-node.
# For backwards compatibility reasons, nr-of-instances
# has the same purpose as max-total-nr-of-instances for cluster
# aware routers and nr-of-instances (if defined by user) takes
# precedence over max-total-nr-of-instances.
max-total-nr-of-instances = 10000
# Defines if routees are allowed to be located on the same node as
# the head router actor, or only on remote nodes.
# Useful for master-worker scenario where all routees are remote.
allow-local-routees = on
# Use members with specified role, or all members if undefined or empty.
use-role = ""
}
# Protobuf serializer for cluster messages
actor {
serializers {
akka-cluster = "akka.cluster.protobuf.ClusterMessageSerializer"
}
serialization-bindings {
"akka.cluster.ClusterMessage" = akka-cluster
"akka.cluster.routing.ClusterRouterPool" = akka-cluster
}
serialization-identifiers {
3.9. Configuration
67
Akka Scala Documentation, Release 2.5.0-RC1
"akka.cluster.protobuf.ClusterMessageSerializer" = 5
}
}
}
akka-multi-node-testkit
#############################################
# Akka Remote Testing Reference Config File #
#############################################
# This is the reference config file that contains all the default settings.
# Make your edits/overrides in your application.conf.
akka {
testconductor {
# Timeout for joining a barrier: this is the maximum time any participants
# waits for everybody else to join a named barrier.
barrier-timeout = 30s
# Timeout for interrogation of TestConductor’s Controller actor
query-timeout = 10s
# Threshold for packet size in time unit above which the failure injector will
# split the packet and deliver in smaller portions; do not give value smaller
# than HashedWheelTimer resolution (would not make sense)
packet-split-threshold = 100ms
# amount of time for the ClientFSM to wait for the connection to the conductor
# to be successful
connect-timeout = 20s
# Number of connect attempts to be made to the conductor controller
client-reconnects = 30
# minimum time interval which is to be inserted between reconnect attempts
reconnect-backoff = 1s
netty {
# (I&O) Used to configure the number of I/O worker threads on server sockets
server-socket-worker-pool {
# Min number of threads to cap factor-based number to
pool-size-min = 1
# The pool size factor is used to determine thread pool size
# using the following formula: ceil(available processors * factor).
# Resulting size is then bounded by the pool-size-min and
# pool-size-max values.
pool-size-factor = 1.0
# Max number of threads to cap factor-based number to
pool-size-max = 2
}
# (I&O) Used to configure the number of I/O worker threads on client sockets
client-socket-worker-pool {
# Min number of threads to cap factor-based number to
pool-size-min = 1
3.9. Configuration
68
Akka Scala Documentation, Release 2.5.0-RC1
# The pool size factor is used to determine thread pool size
# using the following formula: ceil(available processors * factor).
# Resulting size is then bounded by the pool-size-min and
# pool-size-max values.
pool-size-factor = 1.0
# Max number of threads to cap factor-based number to
pool-size-max = 2
}
}
}
}
akka-persistence
###########################################################
# Akka Persistence Extension Reference Configuration File #
###########################################################
# This is the reference config file that contains all the default settings.
# Make your edits in your application.conf in order to override these settings.
# Directory of persistence journal and snapshot store plugins is available at the
# Akka Community Projects page http://akka.io/community/
# Default persistence extension settings.
akka.persistence {
# Fully qualified class name providing a default internal stash overflow
˓→strategy.
# It needs to be a subclass of akka.persistence.
˓→StashOverflowStrategyConfigurator.
# The default strategy throws StashOverflowException.
internal-stash-overflow-strategy = "akka.persistence.ThrowExceptionConfigurator
˓→"
journal {
# Absolute path to the journal plugin configuration entry used by
# persistent actor or view by default.
# Persistent actor or view can override `journalPluginId` method
# in order to rely on a different journal plugin.
plugin = ""
# List of journal plugins to start automatically. Use "" for the default
˓→journal plugin.
auto-start-journals = []
}
snapshot-store {
# Absolute path to the snapshot plugin configuration entry used by
# persistent actor or view by default.
# Persistent actor or view can override `snapshotPluginId` method
# in order to rely on a different snapshot plugin.
# It is not mandatory to specify a snapshot store plugin.
# If you don't use snapshots you don't have to configure it.
# Note that Cluster Sharding is using snapshots, so if you
# use Cluster Sharding you need to define a snapshot store plugin.
plugin = ""
# List of snapshot stores to start automatically. Use "" for the default
˓→snapshot store.
auto-start-snapshot-stores = []
}
# used as default-snapshot store if no plugin configured
# (see `akka.persistence.snapshot-store`)
3.9. Configuration
69
Akka Scala Documentation, Release 2.5.0-RC1
no-snapshot-store {
class = "akka.persistence.snapshot.NoSnapshotStore"
}
# Default persistent view settings.
view {
# Automated incremental view update.
auto-update = on
# Interval between incremental updates.
auto-update-interval = 5s
# Maximum number of messages to replay per incremental view update.
# Set to -1 for no upper limit.
auto-update-replay-max = -1
}
# Default reliable delivery settings.
at-least-once-delivery {
# Interval between re-delivery attempts.
redeliver-interval = 5s
# Maximum number of unconfirmed messages that will be sent in one
# re-delivery burst.
redelivery-burst-limit = 10000
# After this number of delivery attempts a
# `ReliableRedelivery.UnconfirmedWarning`, message will be sent to the
˓→actor.
warn-after-number-of-unconfirmed-attempts = 5
# Maximum number of unconfirmed messages that an actor with
# AtLeastOnceDelivery is allowed to hold in memory.
max-unconfirmed-messages = 100000
}
# Default persistent extension thread pools.
dispatchers {
# Dispatcher used by every plugin which does not declare explicit
# `plugin-dispatcher` field.
default-plugin-dispatcher {
type = PinnedDispatcher
executor = "thread-pool-executor"
}
# Default dispatcher for message replay.
default-replay-dispatcher {
type = Dispatcher
executor = "fork-join-executor"
fork-join-executor {
parallelism-min = 2
parallelism-max = 8
}
}
# Default dispatcher for streaming snapshot IO
default-stream-dispatcher {
type = Dispatcher
executor = "fork-join-executor"
fork-join-executor {
parallelism-min = 2
parallelism-max = 8
}
}
}
# Fallback settings for journal plugin configurations.
# These settings are used if they are not defined in plugin config section.
journal-plugin-fallback {
# Fully qualified class name providing journal plugin api implementation.
# It is mandatory to specify this property.
# The class must have a constructor without parameters or constructor with
3.9. Configuration
70
Akka Scala Documentation, Release 2.5.0-RC1
# one `com.typesafe.config.Config` parameter.
class = ""
# Dispatcher for the plugin actor.
plugin-dispatcher = "akka.persistence.dispatchers.default-plugin-dispatcher"
# Dispatcher for message replay.
replay-dispatcher = "akka.persistence.dispatchers.default-replay-dispatcher"
# Removed: used to be the Maximum size of a persistent message batch written
to the journal.
# Now this setting is without function, PersistentActor will write as many
˓→messages
# as it has accumulated since the last write.
max-message-batch-size = 200
˓→
# If there is more time in between individual events gotten from the journal
# recovery than this the recovery will fail.
# Note that it also affects reading the snapshot before replaying events on
# top of it, even though it is configured for the journal.
recovery-event-timeout = 30s
circuit-breaker {
max-failures = 10
call-timeout = 10s
reset-timeout = 30s
}
# The replay filter can detect a corrupt event stream by inspecting
# sequence numbers and writerUuid when replaying events.
replay-filter {
# What the filter should do when detecting invalid events.
# Supported values:
# `repair-by-discard-old` : discard events from old writers,
#
warning is logged
# `fail` : fail the replay, error is logged
# `warn` : log warning but emit events untouched
# `off` : disable this feature completely
mode = repair-by-discard-old
# It uses a look ahead buffer for analyzing the events.
# This defines the size (in number of events) of the buffer.
window-size = 100
# How many old writerUuid to remember
max-old-writers = 10
# Set this to `on` to enable detailed debug logging of each
# replayed event.
debug = off
}
}
# Fallback settings for snapshot store plugin configurations
# These settings are used if they are not defined in plugin config section.
snapshot-store-plugin-fallback {
# Fully qualified class name providing snapshot store plugin api
# implementation. It is mandatory to specify this property if
# snapshot store is enabled.
# The class must have a constructor without parameters or constructor with
# one `com.typesafe.config.Config` parameter.
class = ""
3.9. Configuration
71
Akka Scala Documentation, Release 2.5.0-RC1
# Dispatcher for the plugin actor.
plugin-dispatcher = "akka.persistence.dispatchers.default-plugin-dispatcher"
circuit-breaker {
max-failures = 5
call-timeout = 20s
reset-timeout = 60s
}
}
}
# Protobuf serialization for the persistent extension messages.
akka.actor {
serializers {
akka-persistence-message = "akka.persistence.serialization.
˓→MessageSerializer"
akka-persistence-snapshot = "akka.persistence.serialization.
˓→SnapshotSerializer"
}
serialization-bindings {
"akka.persistence.serialization.Message" = akka-persistence-message
"akka.persistence.serialization.Snapshot" = akka-persistence-snapshot
}
serialization-identifiers {
"akka.persistence.serialization.MessageSerializer" = 7
"akka.persistence.serialization.SnapshotSerializer" = 8
}
}
###################################################
# Persistence plugins included with the extension #
###################################################
# In-memory journal plugin.
akka.persistence.journal.inmem {
# Class name of the plugin.
class = "akka.persistence.journal.inmem.InmemJournal"
# Dispatcher for the plugin actor.
plugin-dispatcher = "akka.actor.default-dispatcher"
}
# Local file system snapshot store plugin.
akka.persistence.snapshot-store.local {
# Class name of the plugin.
class = "akka.persistence.snapshot.local.LocalSnapshotStore"
# Dispatcher for the plugin actor.
plugin-dispatcher = "akka.persistence.dispatchers.default-plugin-dispatcher"
# Dispatcher for streaming snapshot IO.
stream-dispatcher = "akka.persistence.dispatchers.default-stream-dispatcher"
# Storage location of snapshot files.
dir = "snapshots"
# Number load attempts when recovering from the latest snapshot fails
# yet older snapshot files are available. Each recovery attempt will try
# to recover using an older than previously failed-on snapshot file
# (if any are present). If all attempts fail the recovery will fail and
# the persistent actor will be stopped.
max-load-attempts = 3
}
# LevelDB journal plugin.
# Note: this plugin requires explicit LevelDB dependency, see below.
3.9. Configuration
72
Akka Scala Documentation, Release 2.5.0-RC1
akka.persistence.journal.leveldb {
# Class name of the plugin.
class = "akka.persistence.journal.leveldb.LeveldbJournal"
# Dispatcher for the plugin actor.
plugin-dispatcher = "akka.persistence.dispatchers.default-plugin-dispatcher"
# Dispatcher for message replay.
replay-dispatcher = "akka.persistence.dispatchers.default-replay-dispatcher"
# Storage location of LevelDB files.
dir = "journal"
# Use fsync on write.
fsync = on
# Verify checksum on read.
checksum = off
# Native LevelDB (via JNI) or LevelDB Java port.
native = on
}
# Shared LevelDB journal plugin (for testing only).
# Note: this plugin requires explicit LevelDB dependency, see below.
akka.persistence.journal.leveldb-shared {
# Class name of the plugin.
class = "akka.persistence.journal.leveldb.SharedLeveldbJournal"
# Dispatcher for the plugin actor.
plugin-dispatcher = "akka.actor.default-dispatcher"
# Timeout for async journal operations.
timeout = 10s
store {
# Dispatcher for shared store actor.
store-dispatcher = "akka.persistence.dispatchers.default-plugin-dispatcher"
# Dispatcher for message replay.
replay-dispatcher = "akka.persistence.dispatchers.default-replay-dispatcher
˓→"
# Storage location of LevelDB files.
dir = "journal"
# Use fsync on write.
fsync = on
# Verify checksum on read.
checksum = off
# Native LevelDB (via JNI) or LevelDB Java port.
native = on
}
}
akka.persistence.journal.proxy {
# Class name of the plugin.
class = "akka.persistence.journal.PersistencePluginProxy"
# Dispatcher for the plugin actor.
plugin-dispatcher = "akka.actor.default-dispatcher"
# Set this to on in the configuration of the ActorSystem
# that will host the target journal
start-target-journal = off
# The journal plugin config path to use for the target journal
target-journal-plugin = ""
# The address of the proxy to connect to from other nodes. Optional setting.
target-journal-address = ""
# Initialization timeout of target lookup
init-timeout = 10s
}
akka.persistence.snapshot-store.proxy {
# Class name of the plugin.
class = "akka.persistence.journal.PersistencePluginProxy"
# Dispatcher for the plugin actor.
3.9. Configuration
73
Akka Scala Documentation, Release 2.5.0-RC1
plugin-dispatcher = "akka.actor.default-dispatcher"
# Set this to on in the configuration of the ActorSystem
# that will host the target snapshot-store
start-target-snapshot-store = off
# The journal plugin config path to use for the target snapshot-store
target-snapshot-store-plugin = ""
# The address of the proxy to connect to from other nodes. Optional setting.
target-snapshot-store-address = ""
# Initialization timeout of target lookup
init-timeout = 10s
}
# LevelDB persistence requires the following dependency declarations:
#
# SBT:
#
"org.iq80.leveldb"
% "leveldb"
% "0.7"
#
"org.fusesource.leveldbjni"
% "leveldbjni-all"
% "1.8"
#
# Maven:
#
<dependency>
#
<groupId>org.iq80.leveldb</groupId>
#
<artifactId>leveldb</artifactId>
#
<version>0.7</version>
#
</dependency>
#
<dependency>
#
<groupId>org.fusesource.leveldbjni</groupId>
#
<artifactId>leveldbjni-all</artifactId>
#
<version>1.8</version>
#
</dependency>
akka-remote
#####################################
# Akka Remote Reference Config File #
#####################################
# This is the reference config file that contains all the default settings.
# Make your edits/overrides in your application.conf.
#
#
#
#
#
comments about akka.actor settings left out where they are already in akkaactor.jar, because otherwise they would be repeated in config rendering.
For the configuration of the new remoting implementation (Artery) please look
at the bottom section of this file as it is listed separately.
akka {
actor {
serializers {
akka-containers = "akka.remote.serialization.MessageContainerSerializer"
akka-misc = "akka.remote.serialization.MiscMessageSerializer"
artery = "akka.remote.serialization.ArteryMessageSerializer"
proto = "akka.remote.serialization.ProtobufSerializer"
daemon-create = "akka.remote.serialization.DaemonMsgCreateSerializer"
primitive-long = "akka.remote.serialization.LongSerializer"
primitive-int = "akka.remote.serialization.IntSerializer"
primitive-string = "akka.remote.serialization.StringSerializer"
primitive-bytestring = "akka.remote.serialization.ByteStringSerializer"
akka-system-msg = "akka.remote.serialization.SystemMessageSerializer"
}
3.9. Configuration
74
Akka Scala Documentation, Release 2.5.0-RC1
serialization-bindings {
"akka.actor.ActorSelectionMessage" = akka-containers
"akka.remote.DaemonMsgCreate" = daemon-create
"akka.remote.artery.ArteryMessage" = artery
# Since akka.protobuf.Message does not extend Serializable but
# GeneratedMessage does, need to use the more specific one here in order
# to avoid ambiguity.
"akka.protobuf.GeneratedMessage" = proto
# Since com.google.protobuf.Message does not extend Serializable but
# GeneratedMessage does, need to use the more specific one here in order
# to avoid ambiguity.
# This com.google.protobuf serialization binding is only used if the class
˓→can be loaded,
# i.e. com.google.protobuf dependency has been added in the application
˓→project.
"com.google.protobuf.GeneratedMessage" = proto
"java.util.Optional" = akka-misc
}
# Additional serialization-bindings that are replacing Java serialization are
# defined in this section for backwards compatibility reasons. They are
˓→included
# by default but can be excluded for backwards compatibility with Akka 2.4.x.
# They can be disabled with enable-additional-serialization-bindings=off.
additional-serialization-bindings {
"akka.actor.Identify" = akka-misc
"akka.actor.ActorIdentity" = akka-misc
"scala.Some" = akka-misc
"scala.None$" = akka-misc
"akka.actor.Status$Success" = akka-misc
"akka.actor.Status$Failure" = akka-misc
"akka.actor.ActorRef" = akka-misc
"akka.actor.PoisonPill$" = akka-misc
"akka.actor.Kill$" = akka-misc
"akka.remote.RemoteWatcher$Heartbeat$" = akka-misc
"akka.remote.RemoteWatcher$HeartbeatRsp" = akka-misc
"akka.actor.ActorInitializationException" = akka-misc
"akka.dispatch.sysmsg.SystemMessage" = akka-system-msg
"java.lang.String" = primitive-string
"akka.util.ByteString$ByteString1C" = primitive-bytestring
"akka.util.ByteString$ByteString1" = primitive-bytestring
"akka.util.ByteString$ByteStrings" = primitive-bytestring
"java.lang.Long" = primitive-long
"scala.Long" = primitive-long
"java.lang.Integer" = primitive-int
"scala.Int" = primitive-int
# Java Serializer is by default used for exceptions.
# It's recommended that you implement custom serializer for exceptions that
˓→
are
˓→
add
# sent remotely, e.g. in akka.actor.Status.Failure for ask replies. You can
# binding to akka-misc (MiscMessageSerializerSpec) for the exceptions that
have
# a constructor with single message String or constructor with message
˓→String as
˓→
3.9. Configuration
75
Akka Scala Documentation, Release 2.5.0-RC1
# first parameter and cause Throwable as second parameter. Note that it's not
# safe to add this binding for general exceptions such as
˓→IllegalArgumentException
# because it may have a subclass without required constructor.
"java.lang.Throwable" = java
"akka.actor.IllegalActorStateException" = akka-misc
"akka.actor.ActorKilledException" = akka-misc
"akka.actor.InvalidActorNameException" = akka-misc
"akka.actor.InvalidMessageException" = akka-misc
"akka.actor.LocalScope$" = akka-misc
"akka.remote.RemoteScope" = akka-misc
"com.typesafe.config.impl.SimpleConfig" = akka-misc
"com.typesafe.config.Config" = akka-misc
"akka.routing.FromConfig" = akka-misc
"akka.routing.DefaultResizer" = akka-misc
"akka.routing.BalancingPool" = akka-misc
"akka.routing.BroadcastGroup" = akka-misc
"akka.routing.BroadcastPool" = akka-misc
"akka.routing.RandomGroup" = akka-misc
"akka.routing.RandomPool" = akka-misc
"akka.routing.RoundRobinGroup" = akka-misc
"akka.routing.RoundRobinPool" = akka-misc
"akka.routing.ScatterGatherFirstCompletedGroup" = akka-misc
"akka.routing.ScatterGatherFirstCompletedPool" = akka-misc
"akka.routing.SmallestMailboxPool" = akka-misc
"akka.routing.TailChoppingGroup" = akka-misc
"akka.routing.TailChoppingPool" = akka-misc
"akka.remote.routing.RemoteRouterConfig" = akka-misc
}
serialization-identifiers {
"akka.remote.serialization.ProtobufSerializer" = 2
"akka.remote.serialization.DaemonMsgCreateSerializer" = 3
"akka.remote.serialization.MessageContainerSerializer" = 6
"akka.remote.serialization.MiscMessageSerializer" = 16
"akka.remote.serialization.ArteryMessageSerializer" = 17
"akka.remote.serialization.LongSerializer" = 18
"akka.remote.serialization.IntSerializer" = 19
"akka.remote.serialization.StringSerializer" = 20
"akka.remote.serialization.ByteStringSerializer" = 21
"akka.remote.serialization.SystemMessageSerializer" = 22
}
deployment {
default {
# if this is set to a valid remote address, the named actor will be
# deployed at that node e.g. "akka.tcp://sys@host:port"
remote = ""
target {
# A list of hostnames and ports for instantiating the children of a
# router
#
The format should be on "akka.tcp://sys@host:port", where:
#
- sys is the remote actor system name
#
- hostname can be either hostname or IP address the remote actor
#
should connect to
#
- port should be the port for the remote server on the other node
3.9. Configuration
76
Akka Scala Documentation, Release 2.5.0-RC1
# The number of actor instances to be spawned is still taken from the
# nr-of-instances setting as for local routers; the instances will be
# distributed round-robin among the given nodes.
nodes = []
}
}
}
}
remote {
### Settings shared by classic remoting and Artery (the new implementation of
˓→remoting)
# If set to a nonempty string remoting will use the given dispatcher for
# its internal actors otherwise the default dispatcher is used. Please note
# that since remoting can load arbitrary 3rd party drivers (see
# "enabled-transport" and "adapters" entries) it is not guaranteed that
# every module will respect this setting.
use-dispatcher = "akka.remote.default-remote-dispatcher"
# Settings for the failure detector to monitor connections.
# For TCP it is not important to have fast failure detection, since
# most connection failures are captured by TCP itself.
# The default DeadlineFailureDetector will trigger if there are no heartbeats
˓→within
# the duration heartbeat-interval + acceptable-heartbeat-pause, i.e. 20 seconds
# with the default settings.
transport-failure-detector {
# FQCN of the failure detector implementation.
# It must implement akka.remote.FailureDetector and have
# a public constructor with a com.typesafe.config.Config and
# akka.actor.EventStream parameter.
implementation-class = "akka.remote.DeadlineFailureDetector"
# How often keep-alive heartbeat messages should be sent to each connection.
heartbeat-interval = 4 s
# Number of potentially lost/delayed heartbeats that will be
# accepted before considering it to be an anomaly.
# A margin to the `heartbeat-interval` is important to be able to survive
˓→sudden,
# occasional, pauses in heartbeat arrivals, due to for example garbage
˓→collect or
# network drop.
acceptable-heartbeat-pause = 16 s
}
# Settings for the Phi accrual failure detector (http://www.jaist.ac.jp/~
defago/files/pdf/IS_RR_2004_010.pdf
# [Hayashibara et al]) used for remote death watch.
# The default PhiAccrualFailureDetector will trigger if there are no
˓→heartbeats within
# the duration heartbeat-interval + acceptable-heartbeat-pause + threshold_
˓→adjustment,
# i.e. around 12.5 seconds with default settings.
watch-failure-detector {
˓→
#
#
#
#
FQCN of the failure detector implementation.
It must implement akka.remote.FailureDetector and have
a public constructor with a com.typesafe.config.Config and
akka.actor.EventStream parameter.
3.9. Configuration
77
Akka Scala Documentation, Release 2.5.0-RC1
implementation-class = "akka.remote.PhiAccrualFailureDetector"
# How often keep-alive heartbeat messages should be sent to each connection.
heartbeat-interval = 1 s
# Defines the failure detector threshold.
# A low threshold is prone to generate many wrong suspicions but ensures
# a quick detection in the event of a real crash. Conversely, a high
# threshold generates fewer mistakes but needs more time to detect
# actual crashes.
threshold = 10.0
# Number of the samples of inter-heartbeat arrival times to adaptively
# calculate the failure timeout for connections.
max-sample-size = 200
# Minimum standard deviation to use for the normal distribution in
# AccrualFailureDetector. Too low standard deviation might result in
# too much sensitivity for sudden, but normal, deviations in heartbeat
# inter arrival times.
min-std-deviation = 100 ms
# Number of potentially lost/delayed heartbeats that will be
# accepted before considering it to be an anomaly.
# This margin is important to be able to survive sudden, occasional,
# pauses in heartbeat arrivals, due to for example garbage collect or
# network drop.
acceptable-heartbeat-pause = 10 s
# How often to check for nodes marked as unreachable by the failure
# detector
unreachable-nodes-reaper-interval = 1s
# After the heartbeat request has been sent the first failure detection
# will start after this period, even though no heartbeat mesage has
# been received.
expected-response-after = 1 s
}
# remote deployment configuration section
deployment {
# If true, will only allow specific classes to be instanciated on this
˓→system via remote deployment
enable-whitelist = off
whitelist = []
}
### Configuration for classic remoting
# Timeout after which the startup of the remoting subsystem is considered
# to be failed. Increase this value if your transport drivers (see the
# enabled-transports section) need longer time to be loaded.
startup-timeout = 10 s
# Timout after which the graceful shutdown of the remoting subsystem is
# considered to be failed. After the timeout the remoting system is
# forcefully shut down. Increase this value if your transport drivers
# (see the enabled-transports section) need longer time to stop properly.
shutdown-timeout = 10 s
3.9. Configuration
78
Akka Scala Documentation, Release 2.5.0-RC1
# Before shutting down the drivers, the remoting subsystem attempts to flush
# all pending writes. This setting controls the maximum time the remoting is
# willing to wait before moving on to shut down the drivers.
flush-wait-on-shutdown = 2 s
# Reuse inbound connections for outbound messages
use-passive-connections = on
# Controls the backoff interval after a refused write is reattempted.
# (Transports may refuse writes if their internal buffer is full)
backoff-interval = 5 ms
# Acknowledgment timeout of management commands sent to the transport stack.
command-ack-timeout = 30 s
# The timeout for outbound associations to perform the handshake.
# If the transport is akka.remote.netty.tcp or akka.remote.netty.ssl
# the configured connection-timeout for the transport will be used instead.
handshake-timeout = 15 s
### Security settings
# Enable untrusted mode for full security of server managed actors, prevents
# system messages to be send by clients, e.g. messages like 'Create',
# 'Suspend', 'Resume', 'Terminate', 'Supervise', 'Link' etc.
untrusted-mode = off
# When 'untrusted-mode=on' inbound actor selections are by default discarded.
# Actors with paths defined in this white list are granted permission to
˓→receive actor
# selections messages.
# E.g. trusted-selection-paths = ["/user/receptionist", "/user/namingService"]
trusted-selection-paths = []
# Should the remote server require that its peers share the same
# secure-cookie (defined in the 'remote' section)? Secure cookies are passed
# between during the initial handshake. Connections are refused if the initial
# message contains a mismatching cookie or the cookie is missing.
require-cookie = off
# Deprecated since 2.4-M1
secure-cookie = ""
### Logging
# If this is "on", Akka will log all inbound messages at DEBUG level,
# if off then they are not logged
log-received-messages = off
# If this is "on", Akka will log all outbound messages at DEBUG level,
# if off then they are not logged
log-sent-messages = off
# Sets the log granularity level at which Akka logs remoting events. This
setting
# can take the values OFF, ERROR, WARNING, INFO, DEBUG, or ON. For
˓→compatibility
# reasons the setting "on" will default to "debug" level. Please note that the
˓→effective
# logging level is still determined by the global logging level of the actor
˓→system:
# for example debug level remoting events will be only logged if the system
# is running with debug level logging.
˓→
3.9. Configuration
79
Akka Scala Documentation, Release 2.5.0-RC1
# Failures to deserialize received messages also fall under this flag.
log-remote-lifecycle-events = on
# Logging of message types with payload size in bytes larger than
# this value. Maximum detected size per message type is logged once,
# with an increase threshold of 10%.
# By default this feature is turned off. Activate it by setting the property to
# a value in bytes, such as 1000b. Note that for all messages larger than this
# limit there will be extra performance and scalability cost.
log-frame-size-exceeding = off
# Log warning if the number of messages in the backoff buffer in the endpoint
# writer exceeds this limit. It can be disabled by setting the value to off.
log-buffer-size-exceeding = 50000
# After failed to establish an outbound connection, the remoting will mark the
# address as failed. This configuration option controls how much time should
# be elapsed before reattempting a new connection. While the address is
# gated, all messages sent to the address are delivered to dead-letters.
# Since this setting limits the rate of reconnects setting it to a
# very short interval (i.e. less than a second) may result in a storm of
# reconnect attempts.
retry-gate-closed-for = 5 s
# After catastrophic communication failures that result in the loss of system
# messages or after the remote DeathWatch triggers the remote system gets
# quarantined to prevent inconsistent behavior.
# This setting controls how long the Quarantine marker will be kept around
# before being removed to avoid long-term memory leaks.
# WARNING: DO NOT change this to a small value to re-enable communication with
# quarantined nodes. Such feature is not supported and any behavior between
# the affected systems after lifting the quarantine is undefined.
prune-quarantine-marker-after = 5 d
# If system messages have been exchanged between two systems (i.e. remote death
# watch or remote deployment has been used) a remote system will be marked as
# quarantined after the two system has no active association, and no
# communication happens during the time configured here.
# The only purpose of this setting is to avoid storing system message
˓→redelivery
# data (sequence number state, etc.) for an undefined amount of time leading
˓→to long
# term memory leak. Instead, if a system has been gone for this period,
# or more exactly
# - there is no association between the two systems (TCP connection, if TCP
˓→transport is used)
# - neither side has been attempting to communicate with the other
# - there are no pending system messages to deliver
# for the amount of time configured here, the remote system will be
˓→quarantined and all state
# associated with it will be dropped.
quarantine-after-silence = 5 d
# This setting defines the maximum number of unacknowledged system messages
# allowed for a remote system. If this limit is reached the remote system is
# declared to be dead and its UID marked as tainted.
system-message-buffer-size = 20000
# This setting defines the maximum idle time after an individual
# acknowledgement for system messages is sent. System message delivery
# is guaranteed by explicit acknowledgement messages. These acks are
3.9. Configuration
80
Akka Scala Documentation, Release 2.5.0-RC1
# piggybacked on ordinary traffic messages. If no traffic is detected
# during the time period configured here, the remoting will send out
# an individual ack.
system-message-ack-piggyback-timeout = 0.3 s
# This setting defines the time after internal management signals
# between actors (used for DeathWatch and supervision) that have not been
# explicitly acknowledged or negatively acknowledged are resent.
# Messages that were negatively acknowledged are always immediately
# resent.
resend-interval = 2 s
# Maximum number of unacknowledged system messages that will be resent
# each 'resend-interval'. If you watch many (> 1000) remote actors you can
# increase this value to for example 600, but a too large limit (e.g. 10000)
# may flood the connection and might cause false failure detection to trigger.
# Test such a configuration by watching all actors at the same time and stop
# all watched actors at the same time.
resend-limit = 200
# WARNING: this setting should not be not changed unless all of its
consequences
# are properly understood which assumes experience with remoting internals
# or expert advice.
# This setting defines the time after redelivery attempts of internal
˓→management
# signals are stopped to a remote system that has been not confirmed to be
˓→alive by
# this system before.
initial-system-message-delivery-timeout = 3 m
˓→
### Transports and adapters
# List of the transport drivers that will be loaded by the remoting.
# A list of fully qualified config paths must be provided where
# the given configuration path contains a transport-class key
# pointing to an implementation class of the Transport interface.
# If multiple transports are provided, the address of the first
# one will be used as a default address.
enabled-transports = ["akka.remote.netty.tcp"]
# Transport drivers can be augmented with adapters by adding their
# name to the applied-adapters setting in the configuration of a
# transport. The available adapters should be configured in this
# section by providing a name, and the fully qualified name of
# their corresponding implementation. The class given here
# must implement akka.akka.remote.transport.TransportAdapterProvider
# and have public constructor without parameters.
adapters {
gremlin = "akka.remote.transport.FailureInjectorProvider"
trttl = "akka.remote.transport.ThrottlerProvider"
}
### Default configuration for the Netty based transport drivers
netty.tcp {
# The class given here must implement the akka.remote.transport.Transport
# interface and offer a public constructor which takes two arguments:
# 1) akka.actor.ExtendedActorSystem
# 2) com.typesafe.config.Config
transport-class = "akka.remote.transport.netty.NettyTransport"
# Transport drivers can be augmented with adapters by adding their
3.9. Configuration
81
Akka Scala Documentation, Release 2.5.0-RC1
# name to the applied-adapters list. The last adapter in the
# list is the adapter immediately above the driver, while
# the first one is the top of the stack below the standard
# Akka protocol
applied-adapters = []
transport-protocol = tcp
# The default remote server port clients should connect to.
# Default is 2552 (AKKA), use 0 if you want a random available port
# This port needs to be unique for each actor system on the same machine.
port = 2552
# The hostname or ip clients should connect to.
# InetAddress.getLocalHost.getHostAddress is used if empty
hostname = ""
# Use this setting to bind a network interface to a different port
# than remoting protocol expects messages at. This may be used
# when running akka nodes in a separated networks (under NATs or docker
˓→containers).
# Use 0 if you want a random available port. Examples:
#
# akka.remote.netty.tcp.port = 2552
# akka.remote.netty.tcp.bind-port = 2553
# Network interface will be bound to the 2553 port, but remoting protocol
˓→will
# expect messages sent to port 2552.
#
# akka.remote.netty.tcp.port = 0
# akka.remote.netty.tcp.bind-port = 0
# Network interface will be bound to a random port, and remoting protocol
˓→will
# expect messages sent to the bound port.
#
# akka.remote.netty.tcp.port = 2552
# akka.remote.netty.tcp.bind-port = 0
# Network interface will be bound to a random port, but remoting protocol
˓→will
# expect messages sent to port 2552.
#
# akka.remote.netty.tcp.port = 0
# akka.remote.netty.tcp.bind-port = 2553
# Network interface will be bound to the 2553 port, and remoting protocol
˓→will
# expect messages sent to the bound port.
#
# akka.remote.netty.tcp.port = 2552
# akka.remote.netty.tcp.bind-port = ""
# Network interface will be bound to the 2552 port, and remoting protocol
˓→will
# expect messages sent to the bound port.
#
# akka.remote.netty.tcp.port if empty
bind-port = ""
# Use this setting to bind a network interface to a different hostname or ip
# than remoting protocol expects messages at.
# Use "0.0.0.0" to bind to all interfaces.
# akka.remote.netty.tcp.hostname if empty
bind-hostname = ""
# Enables SSL support on this transport
3.9. Configuration
82
Akka Scala Documentation, Release 2.5.0-RC1
enable-ssl = false
# Sets the connectTimeoutMillis of all outbound connections,
# i.e. how long a connect may take until it is timed out
connection-timeout = 15 s
# If set to "<id.of.dispatcher>" then the specified dispatcher
# will be used to accept inbound connections, and perform IO. If "" then
# dedicated threads will be used.
# Please note that the Netty driver only uses this configuration and does
# not read the "akka.remote.use-dispatcher" entry. Instead it has to be
# configured manually to point to the same dispatcher if needed.
use-dispatcher-for-io = ""
# Sets the high water mark for the in and outbound sockets,
# set to 0b for platform default
write-buffer-high-water-mark = 0b
# Sets the low water mark for the in and outbound sockets,
# set to 0b for platform default
write-buffer-low-water-mark = 0b
# Sets the send buffer size of the Sockets,
# set to 0b for platform default
send-buffer-size = 256000b
# Sets the receive buffer size of the Sockets,
# set to 0b for platform default
receive-buffer-size = 256000b
# Maximum message size the transport will accept, but at least
# 32000 bytes.
# Please note that UDP does not support arbitrary large datagrams,
# so this setting has to be chosen carefully when using UDP.
# Both send-buffer-size and receive-buffer-size settings has to
# be adjusted to be able to buffer messages of maximum size.
maximum-frame-size = 128000b
# Sets the size of the connection backlog
backlog = 4096
# Enables the TCP_NODELAY flag, i.e. disables Nagle’s algorithm
tcp-nodelay = on
# Enables TCP Keepalive, subject to the O/S kernel’s configuration
tcp-keepalive = on
# Enables SO_REUSEADDR, which determines when an ActorSystem can open
# the specified listen port (the meaning differs between *nix and Windows)
# Valid values are "on", "off" and "off-for-windows"
# due to the following Windows bug: http://bugs.sun.com/bugdatabase/view_bug.
˓→do?bug_id=4476378
# "off-for-windows" of course means that it's "on" for all other platforms
tcp-reuse-addr = off-for-windows
# Used to configure the number of I/O worker threads on server sockets
server-socket-worker-pool {
# Min number of threads to cap factor-based number to
pool-size-min = 2
# The pool size factor is used to determine thread pool size
# using the following formula: ceil(available processors * factor).
# Resulting size is then bounded by the pool-size-min and
3.9. Configuration
83
Akka Scala Documentation, Release 2.5.0-RC1
# pool-size-max values.
pool-size-factor = 1.0
# Max number of threads to cap factor-based number to
pool-size-max = 2
}
# Used to configure the number of I/O worker threads on client sockets
client-socket-worker-pool {
# Min number of threads to cap factor-based number to
pool-size-min = 2
# The pool size factor is used to determine thread pool size
# using the following formula: ceil(available processors * factor).
# Resulting size is then bounded by the pool-size-min and
# pool-size-max values.
pool-size-factor = 1.0
# Max number of threads to cap factor-based number to
pool-size-max = 2
}
}
# DEPRECATED, since 2.5.0
#
The netty.udp transport is deprecated, please use Artery instead.
#
See: http://doc.akka.io/docs/akka/2.4/scala/remoting-artery.html
netty.udp = ${akka.remote.netty.tcp}
netty.udp {
transport-protocol = udp
}
netty.ssl = ${akka.remote.netty.tcp}
netty.ssl = {
# Enable SSL/TLS encryption.
# This must be enabled on both the client and server to work.
enable-ssl = true
security {
# This is the Java Key Store used by the server connection
key-store = "keystore"
# This password is used for decrypting the key store
key-store-password = "changeme"
# This password is used for decrypting the key
key-password = "changeme"
# This is the Java Key Store used by the client connection
trust-store = "truststore"
# This password is used for decrypting the trust store
trust-store-password = "changeme"
# Protocol to use for SSL encryption, choose from:
# TLS 1.2 is available since JDK7, and default since JDK8:
# https://blogs.oracle.com/java-platform-group/entry/java_8_will_use_tls
protocol = "TLSv1.2"
# Example: ["TLS_RSA_WITH_AES_128_CBC_SHA", "TLS_RSA_WITH_AES_256_CBC_SHA"]
# You need to install the JCE Unlimited Strength Jurisdiction Policy
# Files to use AES 256.
3.9. Configuration
84
Akka Scala Documentation, Release 2.5.0-RC1
# More info here:
# http://docs.oracle.com/javase/7/docs/technotes/guides/security/
˓→SunProviders.html#SunJCEProvider
enabled-algorithms = ["TLS_RSA_WITH_AES_128_CBC_SHA"]
# There are three options, in increasing order of security:
# "" or SecureRandom => (default)
# "SHA1PRNG" => Can be slow because of blocking issues on Linux
# "AES128CounterSecureRNG" => fastest startup and based on AES encryption
# algorithm
# "AES256CounterSecureRNG" (Install JCE Unlimited Strength Jurisdiction
# Policy Files first)
#
# Setting a value here may require you to supply the appropriate cipher
# suite (see enabled-algorithms section above)
random-number-generator = ""
# Require mutual authentication between TLS peers
#
# Without mutual authentication only the peer that actively establishes a
˓→connection (TLS client side)
# checks if the passive side (TLS server side) sends over a trusted
˓→certificate. With the flag turned on,
# the passive side will also request and verify a certificate from the
˓→connecting peer.
#
# To prevent man-in-the-middle attacks this setting is enabled by default.
#
# Note: Nodes that are configured with this setting to 'on' might not be
˓→able to receive messages from nodes that
# run on older versions of akka-remote. This is because in versions of
˓→Akka < 2.4.12 the active side of the remoting
# connection will not send over certificates even if asked.
#
# However, starting with Akka 2.4.12, even with this setting "off", the
˓→active side (TLS client side)
# will use the given key-store to send over a certificate if asked. A
˓→rolling upgrade from versions of
# Akka < 2.4.12 can therefore work like this:
#
- upgrade all nodes to an Akka version >= 2.4.12, in the best case the
˓→latest version, but keep this setting at "off"
#
- then switch this flag to "on" and do again a rolling upgrade of all
˓→nodes
# The first step ensures that all nodes will send over a certificate when
˓→asked to. The second
# step will ensure that all nodes finally enforce the secure checking of
˓→client certificates.
require-mutual-authentication = on
}
}
### Default configuration for the failure injector transport adapter
gremlin {
# Enable debug logging of the failure injector transport adapter
debug = off
}
### Default dispatcher for the remoting subsystem
default-remote-dispatcher {
type = Dispatcher
executor = "fork-join-executor"
3.9. Configuration
85
Akka Scala Documentation, Release 2.5.0-RC1
fork-join-executor {
parallelism-min = 2
parallelism-factor = 0.5
parallelism-max = 16
}
throughput = 10
}
backoff-remote-dispatcher {
type = Dispatcher
executor = "fork-join-executor"
fork-join-executor {
# Min number of threads to cap factor-based parallelism number to
parallelism-min = 2
parallelism-max = 2
}
}
}
}
akka-remote (artery)
#####################################
# Akka Remote Reference Config File #
#####################################
# This is the reference config file that contains all the default settings.
# Make your edits/overrides in your application.conf.
#
#
#
#
#
comments about akka.actor settings left out where they are already in akkaactor.jar, because otherwise they would be repeated in config rendering.
For the configuration of the new remoting implementation (Artery) please look
at the bottom section of this file as it is listed separately.
akka {
actor {
serializers {
akka-containers = "akka.remote.serialization.MessageContainerSerializer"
akka-misc = "akka.remote.serialization.MiscMessageSerializer"
artery = "akka.remote.serialization.ArteryMessageSerializer"
proto = "akka.remote.serialization.ProtobufSerializer"
daemon-create = "akka.remote.serialization.DaemonMsgCreateSerializer"
primitive-long = "akka.remote.serialization.LongSerializer"
primitive-int = "akka.remote.serialization.IntSerializer"
primitive-string = "akka.remote.serialization.StringSerializer"
primitive-bytestring = "akka.remote.serialization.ByteStringSerializer"
akka-system-msg = "akka.remote.serialization.SystemMessageSerializer"
}
serialization-bindings {
"akka.actor.ActorSelectionMessage" = akka-containers
"akka.remote.DaemonMsgCreate" = daemon-create
"akka.remote.artery.ArteryMessage" = artery
# Since akka.protobuf.Message does not extend Serializable but
# GeneratedMessage does, need to use the more specific one here in order
3.9. Configuration
86
Akka Scala Documentation, Release 2.5.0-RC1
# to avoid ambiguity.
"akka.protobuf.GeneratedMessage" = proto
# Since com.google.protobuf.Message does not extend Serializable but
# GeneratedMessage does, need to use the more specific one here in order
# to avoid ambiguity.
# This com.google.protobuf serialization binding is only used if the class
˓→can be loaded,
# i.e. com.google.protobuf dependency has been added in the application
˓→project.
"com.google.protobuf.GeneratedMessage" = proto
"java.util.Optional" = akka-misc
}
# Additional serialization-bindings that are replacing Java serialization are
# defined in this section for backwards compatibility reasons. They are
˓→included
# by default but can be excluded for backwards compatibility with Akka 2.4.x.
# They can be disabled with enable-additional-serialization-bindings=off.
additional-serialization-bindings {
"akka.actor.Identify" = akka-misc
"akka.actor.ActorIdentity" = akka-misc
"scala.Some" = akka-misc
"scala.None$" = akka-misc
"akka.actor.Status$Success" = akka-misc
"akka.actor.Status$Failure" = akka-misc
"akka.actor.ActorRef" = akka-misc
"akka.actor.PoisonPill$" = akka-misc
"akka.actor.Kill$" = akka-misc
"akka.remote.RemoteWatcher$Heartbeat$" = akka-misc
"akka.remote.RemoteWatcher$HeartbeatRsp" = akka-misc
"akka.actor.ActorInitializationException" = akka-misc
"akka.dispatch.sysmsg.SystemMessage" = akka-system-msg
"java.lang.String" = primitive-string
"akka.util.ByteString$ByteString1C" = primitive-bytestring
"akka.util.ByteString$ByteString1" = primitive-bytestring
"akka.util.ByteString$ByteStrings" = primitive-bytestring
"java.lang.Long" = primitive-long
"scala.Long" = primitive-long
"java.lang.Integer" = primitive-int
"scala.Int" = primitive-int
# Java Serializer is by default used for exceptions.
# It's recommended that you implement custom serializer for exceptions that
˓→
are
˓→
add
# sent remotely, e.g. in akka.actor.Status.Failure for ask replies. You can
# binding to akka-misc (MiscMessageSerializerSpec) for the exceptions that
have
# a constructor with single message String or constructor with message
˓→String as
# first parameter and cause Throwable as second parameter. Note that it's not
# safe to add this binding for general exceptions such as
˓→IllegalArgumentException
# because it may have a subclass without required constructor.
"java.lang.Throwable" = java
"akka.actor.IllegalActorStateException" = akka-misc
"akka.actor.ActorKilledException" = akka-misc
"akka.actor.InvalidActorNameException" = akka-misc
"akka.actor.InvalidMessageException" = akka-misc
˓→
3.9. Configuration
87
Akka Scala Documentation, Release 2.5.0-RC1
"akka.actor.LocalScope$" = akka-misc
"akka.remote.RemoteScope" = akka-misc
"com.typesafe.config.impl.SimpleConfig" = akka-misc
"com.typesafe.config.Config" = akka-misc
"akka.routing.FromConfig" = akka-misc
"akka.routing.DefaultResizer" = akka-misc
"akka.routing.BalancingPool" = akka-misc
"akka.routing.BroadcastGroup" = akka-misc
"akka.routing.BroadcastPool" = akka-misc
"akka.routing.RandomGroup" = akka-misc
"akka.routing.RandomPool" = akka-misc
"akka.routing.RoundRobinGroup" = akka-misc
"akka.routing.RoundRobinPool" = akka-misc
"akka.routing.ScatterGatherFirstCompletedGroup" = akka-misc
"akka.routing.ScatterGatherFirstCompletedPool" = akka-misc
"akka.routing.SmallestMailboxPool" = akka-misc
"akka.routing.TailChoppingGroup" = akka-misc
"akka.routing.TailChoppingPool" = akka-misc
"akka.remote.routing.RemoteRouterConfig" = akka-misc
}
serialization-identifiers {
"akka.remote.serialization.ProtobufSerializer" = 2
"akka.remote.serialization.DaemonMsgCreateSerializer" = 3
"akka.remote.serialization.MessageContainerSerializer" = 6
"akka.remote.serialization.MiscMessageSerializer" = 16
"akka.remote.serialization.ArteryMessageSerializer" = 17
"akka.remote.serialization.LongSerializer" = 18
"akka.remote.serialization.IntSerializer" = 19
"akka.remote.serialization.StringSerializer" = 20
"akka.remote.serialization.ByteStringSerializer" = 21
"akka.remote.serialization.SystemMessageSerializer" = 22
}
deployment {
default {
# if this is set to a valid remote address, the named actor will be
# deployed at that node e.g. "akka.tcp://sys@host:port"
remote = ""
target {
# A list of hostnames and ports for instantiating the children of a
# router
#
The format should be on "akka.tcp://sys@host:port", where:
#
- sys is the remote actor system name
#
- hostname can be either hostname or IP address the remote actor
#
should connect to
#
- port should be the port for the remote server on the other node
# The number of actor instances to be spawned is still taken from the
# nr-of-instances setting as for local routers; the instances will be
# distributed round-robin among the given nodes.
nodes = []
}
}
}
}
3.9. Configuration
88
Akka Scala Documentation, Release 2.5.0-RC1
remote {
### Settings shared by classic remoting and Artery (the new implementation of
˓→remoting)
# If set to a nonempty string remoting will use the given dispatcher for
# its internal actors otherwise the default dispatcher is used. Please note
# that since remoting can load arbitrary 3rd party drivers (see
# "enabled-transport" and "adapters" entries) it is not guaranteed that
# every module will respect this setting.
use-dispatcher = "akka.remote.default-remote-dispatcher"
# Settings for the failure detector to monitor connections.
# For TCP it is not important to have fast failure detection, since
# most connection failures are captured by TCP itself.
# The default DeadlineFailureDetector will trigger if there are no heartbeats
˓→within
# the duration heartbeat-interval + acceptable-heartbeat-pause, i.e. 20 seconds
# with the default settings.
transport-failure-detector {
# FQCN of the failure detector implementation.
# It must implement akka.remote.FailureDetector and have
# a public constructor with a com.typesafe.config.Config and
# akka.actor.EventStream parameter.
implementation-class = "akka.remote.DeadlineFailureDetector"
# How often keep-alive heartbeat messages should be sent to each connection.
heartbeat-interval = 4 s
# Number of potentially lost/delayed heartbeats that will be
# accepted before considering it to be an anomaly.
# A margin to the `heartbeat-interval` is important to be able to survive
˓→sudden,
# occasional, pauses in heartbeat arrivals, due to for example garbage
˓→collect or
# network drop.
acceptable-heartbeat-pause = 16 s
}
# Settings for the Phi accrual failure detector (http://www.jaist.ac.jp/~
defago/files/pdf/IS_RR_2004_010.pdf
# [Hayashibara et al]) used for remote death watch.
# The default PhiAccrualFailureDetector will trigger if there are no
˓→heartbeats within
# the duration heartbeat-interval + acceptable-heartbeat-pause + threshold_
˓→adjustment,
# i.e. around 12.5 seconds with default settings.
watch-failure-detector {
˓→
# FQCN of the failure detector implementation.
# It must implement akka.remote.FailureDetector and have
# a public constructor with a com.typesafe.config.Config and
# akka.actor.EventStream parameter.
implementation-class = "akka.remote.PhiAccrualFailureDetector"
# How often keep-alive heartbeat messages should be sent to each connection.
heartbeat-interval = 1 s
#
#
#
#
Defines the failure detector threshold.
A low threshold is prone to generate many wrong suspicions but ensures
a quick detection in the event of a real crash. Conversely, a high
threshold generates fewer mistakes but needs more time to detect
3.9. Configuration
89
Akka Scala Documentation, Release 2.5.0-RC1
# actual crashes.
threshold = 10.0
# Number of the samples of inter-heartbeat arrival times to adaptively
# calculate the failure timeout for connections.
max-sample-size = 200
# Minimum standard deviation to use for the normal distribution in
# AccrualFailureDetector. Too low standard deviation might result in
# too much sensitivity for sudden, but normal, deviations in heartbeat
# inter arrival times.
min-std-deviation = 100 ms
# Number of potentially lost/delayed heartbeats that will be
# accepted before considering it to be an anomaly.
# This margin is important to be able to survive sudden, occasional,
# pauses in heartbeat arrivals, due to for example garbage collect or
# network drop.
acceptable-heartbeat-pause = 10 s
# How often to check for nodes marked as unreachable by the failure
# detector
unreachable-nodes-reaper-interval = 1s
# After the heartbeat request has been sent the first failure detection
# will start after this period, even though no heartbeat mesage has
# been received.
expected-response-after = 1 s
}
# remote deployment configuration section
deployment {
# If true, will only allow specific classes to be instanciated on this
˓→system via remote deployment
enable-whitelist = off
whitelist = []
}
### Configuration for Artery, the reimplementation of remoting
artery {
# Enable the new remoting with this flag
enabled = off
# Canonical address is the address other clients should connect to.
# Artery transport will expect messages to this address.
canonical {
# The default remote server port clients should connect to.
# Default is 25520, use 0 if you want a random available port
# This port needs to be unique for each actor system on the same machine.
port = 25520
# Hostname clients should connect to. Can be set to an ip, hostname
# or one of the following special values:
#
"<getHostAddress>"
InetAddress.getLocalHost.getHostAddress
#
"<getHostName>"
InetAddress.getLocalHost.getHostName
#
hostname = "<getHostAddress>"
}
3.9. Configuration
90
Akka Scala Documentation, Release 2.5.0-RC1
# Use these settings to bind a network interface to a different address
# than artery expects messages at. This may be used when running Akka
# nodes in a separated networks (under NATs or in containers). If canonical
# and bind addresses are different, then network configuration that relays
# communications from canonical to bind addresses is expected.
bind {
# Port to bind a network interface to. Can be set to a port number
# of one of the following special values:
#
0
random available port
#
""
akka.remote.artery.canonical.port
#
port = ""
# Hostname to bind a network interface to. Can be set to an ip, hostname
# or one of the following special values:
#
"0.0.0.0"
all interfaces
#
""
akka.remote.artery.canonical.hostname
#
"<getHostAddress>"
InetAddress.getLocalHost.getHostAddress
#
"<getHostName>"
InetAddress.getLocalHost.getHostName
#
hostname = ""
}
# Actor paths to use the large message stream for when a message
# is sent to them over remoting. The large message stream dedicated
# is separate from "normal" and system messages so that sending a
# large message does not interfere with them.
# Entries should be the full path to the actor. Wildcards in the form of "*"
# can be supplied at any place and matches any name at that segment # "/user/supervisor/actor/*" will match any direct child to actor,
# while "/supervisor/*/child" will match any grandchild to "supervisor" that
# has the name "child"
# Entries have to be specified on both the sending and receiving side.
# Messages sent to ActorSelections will not be passed through the large
˓→message
# stream, to pass such messages through the large message stream the
˓→selections
# but must be resolved to ActorRefs first.
large-message-destinations = []
# Enable untrusted mode, which discards inbound system messages,
PossiblyHarmful and
# ActorSelection messages. E.g. remote watch and remote deployment will not
˓→work.
# ActorSelection messages can be enabled for specific paths with the trusted˓→selection-paths
untrusted-mode = off
˓→
# When 'untrusted-mode=on' inbound actor selections are by default discarded.
# Actors with paths defined in this white list are granted permission to
˓→receive actor
# selections messages.
# E.g. trusted-selection-paths = ["/user/receptionist", "/user/namingService
˓→"]
trusted-selection-paths = []
# If this is "on", all inbound remote messages will be logged at DEBUG level,
# if off then they are not logged
log-received-messages = off
˓→
# If this is "on", all outbound remote messages will be logged at DEBUG
level,
3.9. Configuration
91
Akka Scala Documentation, Release 2.5.0-RC1
# if off then they are not logged
log-sent-messages = off
advanced {
# Maximum serialized message size, including header data.
maximum-frame-size = 256 KiB
# Direct byte buffers are reused in a pool with this maximum size.
# Each buffer has the size of 'maximum-frame-size'.
# This is not a hard upper limit on number of created buffers. Additional
# buffers will be created if needed, e.g. when using many outbound
# associations at the same time. Such additional buffers will be garbage
# collected, which is not as efficient as reusing buffers in the pool.
buffer-pool-size = 128
# Maximum serialized message size for the large messages, including header
˓→
data.
# It is currently restricted to 1/8th the size of a term buffer that can be
# configured by setting the 'aeron.term.buffer.length' system property.
# See 'large-message-destinations'.
maximum-large-frame-size = 2 MiB
˓→
# Direct byte buffers for the large messages are reused in a pool with
this maximum size.
# Each buffer has the size of 'maximum-large-frame-size'.
# See 'large-message-destinations'.
# This is not a hard upper limit on number of created buffers. Additional
# buffers will be created if needed, e.g. when using many outbound
# associations at the same time. Such additional buffers will be garbage
# collected, which is not as efficient as reusing buffers in the pool.
large-buffer-pool-size = 32
# For enabling testing features, such as blackhole in akka-remote-testkit.
test-mode = off
# Settings for the materializer that is used for the remote streams.
materializer = ${akka.stream.materializer}
# If set to a nonempty string artery will use the given dispatcher for
# the ordinary and large message streams, otherwise the default dispatcher
˓→is used.
use-dispatcher = "akka.remote.default-remote-dispatcher"
# If set to a nonempty string remoting will use the given dispatcher for
# the control stream, otherwise the default dispatcher is used.
# It can be good to not use the same dispatcher for the control stream as
# the dispatcher for the ordinary message stream so that heartbeat messages
# are not disturbed.
use-control-stream-dispatcher = ""
# Controls whether to start the Aeron media driver in the same JVM or use
external
# process. Set to 'off' when using external media driver, and then also
˓→set the
# 'aeron-dir'.
embedded-media-driver = on
˓→
# Directory used by the Aeron media driver. It's mandatory to define the
'aeron-dir'
# if using external media driver, i.e. when 'embedded-media-driver = off'.
# Embedded media driver will use a this directory, or a temporary
˓→directory if this
˓→
3.9. Configuration
92
Akka Scala Documentation, Release 2.5.0-RC1
# property is not defined (empty).
aeron-dir = ""
# Whether to delete aeron embeded driver directory upon driver stop.
delete-aeron-dir = yes
# Level of CPU time used, on a scale between 1 and 10, during backoff/idle.
# The tradeoff is that to have low latency more CPU time must be used to be
# able to react quickly on incoming messages or send as fast as possible
˓→
after
# backoff backpressure.
# Level 1 strongly prefer low CPU consumption over low latency.
# Level 10 strongly prefer low latency over low CPU consumption.
idle-cpu-level = 5
# WARNING: This feature is not supported yet. Don't use other value than 1.
# It requires more hardening and performance optimizations.
# Number of outbound lanes for each outbound association. A value greater
˓→than 1
# means that serialization can be performed in parallel for different
˓→destination
# actors. The selection of lane is based on consistent hashing of the
˓→recipient
# ActorRef to preserve message ordering per receiver.
outbound-lanes = 1
# WARNING: This feature is not supported yet. Don't use other value than 1.
# It requires more hardening and performance optimizations.
# Total number of inbound lanes, shared among all inbound associations. A
˓→
value
˓→
for
# greater than 1 means that deserialization can be performed in parallel
# different destination actors. The selection of lane is based on
consistent
# hashing of the recipient ActorRef to preserve message ordering per
˓→receiver.
inbound-lanes = 1
˓→
# Size of the send queue for outgoing messages. Messages will be dropped if
# the queue becomes full. This may happen if you send a burst of many
˓→messages
# without end-to-end flow control. Note that there is one such queue per
# outbound association. The trade-off of using a larger queue size is that
# it consumes more memory, since the queue is based on preallocated array
˓→with
# fixed size.
outbound-message-queue-size = 3072
# Size of the send queue for outgoing control messages, such as system
messages.
# If this limit is reached the remote system is declared to be dead and
˓→its UID
# marked as quarantined.
# The trade-off of using a larger queue size is that it consumes more
˓→memory,
# since the queue is based on preallocated array with fixed size.
outbound-control-queue-size = 3072
˓→
# Size of the send queue for outgoing large messages. Messages will be
dropped if
# the queue becomes full. This may happen if you send a burst of many
˓→messages
# without end-to-end flow control. Note that there is one such queue per
˓→
3.9. Configuration
93
Akka Scala Documentation, Release 2.5.0-RC1
# outbound association. The trade-off of using a larger queue size is that
# it consumes more memory, since the queue is based on preallocated array
˓→
with
# fixed size.
outbound-large-message-queue-size = 256
# This setting defines the maximum number of unacknowledged system messages
# allowed for a remote system. If this limit is reached the remote system
˓→
is
# declared to be dead and its UID marked as quarantined.
system-message-buffer-size = 20000
# unacknowledged system messages are re-delivered with this interval
system-message-resend-interval = 1 second
# The timeout for outbound associations to perform the handshake.
# This timeout must be greater than the 'image-liveness-timeout'.
handshake-timeout = 20 s
# incomplete handshake attempt is retried with this interval
handshake-retry-interval = 1 second
# handshake requests are performed periodically with this interval,
# also after the handshake has been completed to be able to establish
# a new session with a restarted destination system
inject-handshake-interval = 1 second
˓→
# messages that are not accepted by Aeron are dropped after retrying for
this period
give-up-message-after = 60 seconds
# System messages that are not acknowledged after re-sending for this
period are
# dropped and will trigger quarantine. The value should be longer than the
˓→length
# of a network partition that you need to survive.
give-up-system-message-after = 6 hours
˓→
# during ActorSystem termination the remoting will wait this long for
# an acknowledgment by the destination system that flushing of outstanding
# remote messages has been completed
shutdown-flush-timeout = 1 second
# See 'inbound-max-restarts'
inbound-restart-timeout = 5 seconds
˓→
# Max number of restarts within 'inbound-restart-timeout' for the inbound
streams.
# If more restarts occurs the ActorSystem will be terminated.
inbound-max-restarts = 5
# See 'outbound-max-restarts'
outbound-restart-timeout = 5 seconds
˓→
˓→
# Max number of restarts within 'outbound-restart-timeout' for the
outbound streams.
# If more restarts occurs the ActorSystem will be terminated.
outbound-max-restarts = 5
# Stop outbound stream of a quarantined association after this idle
timeout, i.e.
# when not used any more.
stop-quarantined-after-idle = 3 seconds
3.9. Configuration
94
Akka Scala Documentation, Release 2.5.0-RC1
# Timeout after which aeron driver has not had keepalive messages
# from a client before it considers the client dead.
client-liveness-timeout = 20 seconds
# Timeout for each the
# will be retained for
# This timeout must be
image-liveness-timeout
INACTIVE and LINGER stages an aeron image
when it is no longer referenced.
less than the 'handshake-timeout'.
= 10 seconds
# Timeout after which the aeron driver is considered dead
# if it does not update its C'n'C timestamp.
driver-timeout = 20 seconds
flight-recorder {
// FIXME it should be enabled by default when we have a good solution
˓→for naming the files
enabled = off
# Controls where the flight recorder file will be written. There are
˓→three options:
# 1. Empty: a file will be generated in the temporary directory of the OS
# 2. A relative or absolute path ending with ".afr": this file will be
˓→used
# 3. A relative or absolute path: this directory will be used, the file
˓→will get a random file name
destination = ""
}
˓→
# compression of common strings in remoting messages, like actor
destinations, serializers etc
compression {
actor-refs {
# Max number of compressed actor-refs
# Note that compression tables are "rolling" (i.e. a new table
˓→replaces the old
# compression table once in a while), and this setting is only about
˓→the total number
# of compressions within a single such table.
# Must be a positive natural number.
max = 256
# interval between new table compression advertisements.
# this means the time during which we collect heavy-hitter data and
˓→then turn it into a compression table.
advertisement-interval = 1 minute
}
manifests {
# Max number of compressed manifests
# Note that compression tables are "rolling" (i.e. a new table
˓→replaces the old
# compression table once in a while), and this setting is only about
˓→the total number
# of compressions within a single such table.
# Must be a positive natural number.
max = 256
# interval between new table compression advertisements.
# this means the time during which we collect heavy-hitter data and
˓→then turn it into a compression table.
advertisement-interval = 1 minute
}
}
3.9. Configuration
95
Akka Scala Documentation, Release 2.5.0-RC1
# List of fully qualified class names of remote instruments which should
# be initialized and used for monitoring of remote messages.
# The class must extend akka.remote.artery.RemoteInstrument and
# have a public constructor with empty parameters or one
˓→ExtendedActorSystem
# parameter.
# A new instance of RemoteInstrument will be created for each encoder and
˓→decoder.
# It's only called from the stage, so if it dosn't delegate to any shared
˓→instance
# it doesn't have to be thread-safe.
# Refer to `akka.remote.artery.RemoteInstrument` for more information.
instruments = ${?akka.remote.artery.advanced.instruments} []
}
}
}
}
akka-testkit
######################################
# Akka Testkit Reference Config File #
######################################
# This is the reference config file that contains all the default settings.
# Make your edits/overrides in your application.conf.
akka {
test {
# factor by which to scale timeouts during tests, e.g. to account for shared
# build system load
timefactor = 1.0
# duration of EventFilter.intercept waits after the block is finished until
# all required messages are received
filter-leeway = 3s
# duration to wait in expectMsg and friends outside of within() block
# by default
single-expect-default = 3s
# The timeout that is added as an implicit by DefaultTimeout trait
default-timeout = 5s
calling-thread-dispatcher {
type = akka.testkit.CallingThreadDispatcherConfigurator
}
}
actor {
serializers {
test-message-serializer = "akka.testkit.TestMessageSerializer"
}
serialization-identifiers {
"akka.testkit.TestMessageSerializer" = 23
}
serialization-bindings {
3.9. Configuration
96
Akka Scala Documentation, Release 2.5.0-RC1
"akka.testkit.JavaSerializable" = java
}
}
}
akka-cluster-metrics ~~~~~~~~~~~~——–
##############################################
# Akka Cluster Metrics Reference Config File #
##############################################
# This is the reference config file that contains all the default settings.
# Make your edits in your application.conf in order to override these settings.
# Sigar provisioning:
#
# User can provision sigar classes and native library in one of the following
˓→ways:
#
# 1) Use https://github.com/kamon-io/sigar-loader Kamon sigar-loader as a project
˓→dependency for the user project.
# Metrics extension will extract and load sigar library on demand with help of
˓→Kamon sigar provisioner.
#
# 2) Use https://github.com/kamon-io/sigar-loader Kamon sigar-loader as java
˓→agent: `java -javaagent:/path/to/sigar-loader.jar`
# Kamon sigar loader agent will extract and load sigar library during JVM start.
#
# 3) Place `sigar.jar` on the `classpath` and sigar native library for the o/s on
˓→the `java.library.path`
# User is required to manage both project dependency and library deployment
˓→manually.
# Cluster metrics extension.
# Provides periodic statistics collection and publication throughout the cluster.
akka.cluster.metrics {
# Full path of dispatcher configuration key.
# Use "" for default key `akka.actor.default-dispatcher`.
dispatcher = ""
# How long should any actor wait before starting the periodic tasks.
periodic-tasks-initial-delay = 1s
# Sigar native library extract location.
# Use per-application-instance scoped location, such as program working
˓→directory.
native-library-extract-folder = ${user.dir}"/native"
# Metrics supervisor actor.
supervisor {
# Actor name. Example name space: /system/cluster-metrics
name = "cluster-metrics"
# Supervision strategy.
strategy {
#
# FQCN of class providing `akka.actor.SupervisorStrategy`.
# Must have a constructor with signature `<init>(com.typesafe.config.
˓→Config)`.
# Default metrics strategy provider is a configurable extension of
˓→`OneForOneStrategy`.
provider = "akka.cluster.metrics.ClusterMetricsStrategy"
#
# Configuration of the default strategy provider.
# Replace with custom settings when overriding the provider.
configuration = {
# Log restart attempts.
3.9. Configuration
97
Akka Scala Documentation, Release 2.5.0-RC1
loggingEnabled = true
# Child actor restart-on-failure window.
withinTimeRange = 3s
# Maximum number of restart attempts before child actor is stopped.
maxNrOfRetries = 3
}
}
}
# Metrics collector actor.
collector {
# Enable or disable metrics collector for load-balancing nodes.
# Metrics collection can also be controlled at runtime by sending control
˓→messages
# to /system/cluster-metrics actor: `akka.cluster.metrics.
˓→{CollectionStartMessage,CollectionStopMessage}`
enabled = on
# FQCN of the metrics collector implementation.
# It must implement `akka.cluster.metrics.MetricsCollector` and
# have public constructor with akka.actor.ActorSystem parameter.
# Will try to load in the following order of priority:
# 1) configured custom collector 2) internal `SigarMetricsCollector` 3)
˓→internal `JmxMetricsCollector`
provider = ""
# Try all 3 available collector providers, or else fail on the configured
˓→custom collector provider.
fallback = true
# How often metrics are sampled on a node.
# Shorter interval will collect the metrics more often.
# Also controls frequency of the metrics publication to the node system event
˓→bus.
sample-interval = 3s
# How often a node publishes metrics information to the other nodes in the
˓→cluster.
# Shorter interval will publish the metrics gossip more often.
gossip-interval = 3s
# How quickly the exponential weighting of past data is decayed compared to
# new data. Set lower to increase the bias toward newer values.
# The relevance of each data sample is halved for every passing half-life
# duration, i.e. after 4 times the half-life, a data sample’s relevance is
# reduced to 6% of its original relevance. The initial relevance of a data
# sample is given by 1 - 0.5 ^ (collect-interval / half-life).
# See http://en.wikipedia.org/wiki/Moving_average#Exponential_moving_average
moving-average-half-life = 12s
}
}
# Cluster metrics extension serializers and routers.
akka.actor {
# Protobuf serializer for remote cluster metrics messages.
serializers {
akka-cluster-metrics = "akka.cluster.metrics.protobuf.MessageSerializer"
}
# Interface binding for remote cluster metrics messages.
serialization-bindings {
"akka.cluster.metrics.ClusterMetricsMessage" = akka-cluster-metrics
"akka.cluster.metrics.AdaptiveLoadBalancingPool" = akka-cluster-metrics
"akka.cluster.metrics.MixMetricsSelector" = akka-cluster-metrics
"akka.cluster.metrics.CpuMetricsSelector$" = akka-cluster-metrics
"akka.cluster.metrics.HeapMetricsSelector$" = akka-cluster-metrics
"akka.cluster.metrics.SystemLoadAverageMetricsSelector$" = akka-cluster-metrics
}
# Globally unique metrics extension serializer identifier.
serialization-identifiers {
3.9. Configuration
98
Akka Scala Documentation, Release 2.5.0-RC1
"akka.cluster.metrics.protobuf.MessageSerializer" = 10
}
# Provide routing of messages based on cluster metrics.
router.type-mapping {
cluster-metrics-adaptive-pool = "akka.cluster.metrics.
˓→AdaptiveLoadBalancingPool"
cluster-metrics-adaptive-group = "akka.cluster.metrics.
˓→AdaptiveLoadBalancingGroup"
}
}
akka-cluster-tools ~~~~~~~~~~~~——
############################################
# Akka Cluster Tools Reference Config File #
############################################
# This is the reference config file that contains all the default settings.
# Make your edits/overrides in your application.conf.
# //#pub-sub-ext-config
# Settings for the DistributedPubSub extension
akka.cluster.pub-sub {
# Actor name of the mediator actor, /system/distributedPubSubMediator
name = distributedPubSubMediator
# Start the mediator on members tagged with this role.
# All members are used if undefined or empty.
role = ""
# The routing logic to use for 'Send'
# Possible values: random, round-robin, broadcast
routing-logic = random
# How often the DistributedPubSubMediator should send out gossip information
gossip-interval = 1s
# Removed entries are pruned after this duration
removed-time-to-live = 120s
˓→
# Maximum number of elements to transfer in one message when synchronizing the
registries.
# Next chunk will be transferred in next round of gossip.
max-delta-elements = 3000
# The id of the dispatcher to use for DistributedPubSubMediator actors.
# If not specified default dispatcher is used.
# If specified you need to define the settings of the actual dispatcher.
use-dispatcher = ""
}
# //#pub-sub-ext-config
# Protobuf serializer for cluster DistributedPubSubMeditor messages
akka.actor {
serializers {
akka-pubsub = "akka.cluster.pubsub.protobuf.DistributedPubSubMessageSerializer"
}
serialization-bindings {
"akka.cluster.pubsub.DistributedPubSubMessage" = akka-pubsub
}
serialization-identifiers {
"akka.cluster.pubsub.protobuf.DistributedPubSubMessageSerializer" = 9
}
3.9. Configuration
99
Akka Scala Documentation, Release 2.5.0-RC1
# adds the protobuf serialization of pub sub messages to groups
additional-serialization-bindings {
"akka.cluster.pubsub.DistributedPubSubMediator$Internal$SendToOneSubscriber" =
˓→akka-pubsub
}
}
# //#receptionist-ext-config
# Settings for the ClusterClientReceptionist extension
akka.cluster.client.receptionist {
# Actor name of the ClusterReceptionist actor, /system/receptionist
name = receptionist
# Start the receptionist on members tagged with this role.
# All members are used if undefined or empty.
role = ""
# The receptionist will send this number of contact points to the client
number-of-contacts = 3
# The actor that tunnel response messages to the client will be stopped
# after this time of inactivity.
response-tunnel-receive-timeout = 30s
# The id of the dispatcher to use for ClusterReceptionist actors.
# If not specified default dispatcher is used.
# If specified you need to define the settings of the actual dispatcher.
use-dispatcher = ""
# How often failure detection heartbeat messages should be received for
# each ClusterClient
heartbeat-interval = 2s
# Number of potentially lost/delayed heartbeats that will be
# accepted before considering it to be an anomaly.
# The ClusterReceptionist is using the akka.remote.DeadlineFailureDetector, which
# will trigger if there are no heartbeats within the duration
# heartbeat-interval + acceptable-heartbeat-pause, i.e. 15 seconds with
# the default settings.
acceptable-heartbeat-pause = 13s
# Failure detection checking interval for checking all ClusterClients
failure-detection-interval = 2s
}
# //#receptionist-ext-config
# //#cluster-client-config
# Settings for the ClusterClient
akka.cluster.client {
# Actor paths of the ClusterReceptionist actors on the servers (cluster nodes)
# that the client will try to contact initially. It is mandatory to specify
# at least one initial contact.
# Comma separated full actor paths defined by a string on the form of
# "akka.tcp://system@hostname:port/system/receptionist"
initial-contacts = []
# Interval at which the client retries to establish contact with one of
# ClusterReceptionist on the servers (cluster nodes)
establishing-get-contacts-interval = 3s
# Interval at which the client will ask the ClusterReceptionist for
# new contact points to be used for next reconnect.
3.9. Configuration
100
Akka Scala Documentation, Release 2.5.0-RC1
refresh-contacts-interval = 60s
# How often failure detection heartbeat messages should be sent
heartbeat-interval = 2s
# Number of potentially lost/delayed heartbeats that will be
# accepted before considering it to be an anomaly.
# The ClusterClient is using the akka.remote.DeadlineFailureDetector, which
# will trigger if there are no heartbeats within the duration
# heartbeat-interval + acceptable-heartbeat-pause, i.e. 15 seconds with
# the default settings.
acceptable-heartbeat-pause = 13s
# If connection to the receptionist is not established the client will buffer
# this number of messages and deliver them the connection is established.
# When the buffer is full old messages will be dropped when new messages are sent
# via the client. Use 0 to disable buffering, i.e. messages will be dropped
# immediately if the location of the singleton is unknown.
# Maximum allowed buffer size is 10000.
buffer-size = 1000
# If connection to the receiptionist is lost and the client has not been
# able to acquire a new connection for this long the client will stop itself.
# This duration makes it possible to watch the cluster client and react on a
˓→more permanent
# loss of connection with the cluster, for example by accessing some kind of
# service registry for an updated set of initial contacts to start a new cluster
˓→client with.
# If this is not wanted it can be set to "off" to disable the timeout and retry
# forever.
reconnect-timeout = off
}
# //#cluster-client-config
# Protobuf serializer for ClusterClient messages
akka.actor {
serializers {
akka-cluster-client = "akka.cluster.client.protobuf.
˓→ClusterClientMessageSerializer"
}
serialization-bindings {
"akka.cluster.client.ClusterClientMessage" = akka-cluster-client
}
serialization-identifiers {
"akka.cluster.client.protobuf.ClusterClientMessageSerializer" = 15
}
}
# //#singleton-config
akka.cluster.singleton {
# The actor name of the child singleton actor.
singleton-name = "singleton"
# Singleton among the nodes tagged with specified role.
# If the role is not specified it's a singleton among all nodes in the cluster.
role = ""
# When a node is becoming oldest it sends hand-over request to previous oldest,
# that might be leaving the cluster. This is retried with this interval until
# the previous oldest confirms that the hand over has started or the previous
# oldest member is removed from the cluster (+ akka.cluster.down-removal-margin).
hand-over-retry-interval = 1s
3.9. Configuration
101
Akka Scala Documentation, Release 2.5.0-RC1
# The number of retries are derived from hand-over-retry-interval and
# akka.cluster.down-removal-margin (or ClusterSingletonManagerSettings.
˓→removalMargin),
# but it will never be less than this property.
min-number-of-hand-over-retries = 10
}
# //#singleton-config
# //#singleton-proxy-config
akka.cluster.singleton-proxy {
# The actor name of the singleton actor that is started by the
˓→ClusterSingletonManager
singleton-name = ${akka.cluster.singleton.singleton-name}
# The role of the cluster nodes where the singleton can be deployed.
# If the role is not specified then any node will do.
role = ""
# Interval at which the proxy will try to resolve the singleton instance.
singleton-identification-interval = 1s
# If the location of the singleton is unknown the proxy will buffer this
# number of messages and deliver them when the singleton is identified.
# When the buffer is full old messages will be dropped when new messages are
# sent via the proxy.
# Use 0 to disable buffering, i.e. messages will be dropped immediately if
# the location of the singleton is unknown.
# Maximum allowed buffer size is 10000.
buffer-size = 1000
}
# //#singleton-proxy-config
# Serializer for cluster ClusterSingleton messages
akka.actor {
serializers {
akka-singleton = "akka.cluster.singleton.protobuf.
˓→ClusterSingletonMessageSerializer"
}
serialization-bindings {
"akka.cluster.singleton.ClusterSingletonMessage" = akka-singleton
}
serialization-identifiers {
"akka.cluster.singleton.protobuf.ClusterSingletonMessageSerializer" = 14
}
}
akka-cluster-sharding ~~~~~~~~~~~~———
###############################################
# Akka Cluster Sharding Reference Config File #
###############################################
# This is the reference config file that contains all the default settings.
# Make your edits/overrides in your application.conf.
# //#sharding-ext-config
# Settings for the ClusterShardingExtension
akka.cluster.sharding {
˓→
# The extension creates a top level actor with this name in top level system
scope,
# e.g. '/system/sharding'
3.9. Configuration
102
Akka Scala Documentation, Release 2.5.0-RC1
guardian-name = sharding
# Specifies that entities runs on cluster nodes with a specific role.
# If the role is not specified (or empty) all nodes in the cluster are used.
role = ""
˓→
# When this is set to 'on' the active entity actors will automatically be
restarted
# upon Shard restart. i.e. if the Shard is started on a different ShardRegion
# due to rebalance or crash.
remember-entities = off
# If the coordinator can't store state changes it will be stopped
# and started again after this duration, with an exponential back-off
# of up to 5 times this duration.
coordinator-failure-backoff = 5 s
# The ShardRegion retries registration and shard location requests to the
# ShardCoordinator with this interval if it does not reply.
retry-interval = 2 s
# Maximum number of messages that are buffered by a ShardRegion actor.
buffer-size = 100000
# Timeout of the shard rebalancing process.
handoff-timeout = 60 s
# Time given to a region to acknowledge it's hosting a shard.
shard-start-timeout = 10 s
# If the shard is remembering entities and can't store state changes
# will be stopped and then started again after this duration. Any messages
# sent to an affected entity may be lost in this process.
shard-failure-backoff = 10 s
# If the shard is remembering entities and an entity stops itself without
# using passivate. The entity will be restarted after this duration or when
# the next message for it is received, which ever occurs first.
entity-restart-backoff = 10 s
# Rebalance check is performed periodically with this interval.
rebalance-interval = 10 s
# Absolute path to the journal plugin configuration entity that is to be
# used for the internal persistence of ClusterSharding. If not defined
# the default journal plugin is used. Note that this is not related to
# persistence used by the entity actors.
# Only used when state-store-mode=persistence
journal-plugin-id = ""
# Absolute path to the snapshot plugin configuration entity that is to be
# used for the internal persistence of ClusterSharding. If not defined
# the default snapshot plugin is used. Note that this is not related to
# persistence used by the entity actors.
# Only used when state-store-mode=persistence
snapshot-plugin-id = ""
# Defines how the coordinator stores its state. Same is also used by the
# shards for rememberEntities.
# Valid values are "ddata" or "persistence".
state-store-mode = "ddata"
# The shard saves persistent snapshots after this number of persistent
3.9. Configuration
103
Akka Scala Documentation, Release 2.5.0-RC1
# events. Snapshots are used to reduce recovery times.
# Only used when state-store-mode=persistence
snapshot-after = 1000
# The shard deletes persistent events (messages and snapshots) after doing
snapshot
# keeping this number of old persistent batches.
# Batch is of size `snapshot-after`.
# When set to 0 after snapshot is successfully done all messages with equal or
˓→lower sequence number will be deleted.
# Default value of 2 leaves last maximum 2*`snapshot-after` messages and 3
˓→snapshots (2 old ones + fresh snapshot)
keep-nr-of-batches = 2
˓→
# Setting for the default shard allocation strategy
least-shard-allocation-strategy {
# Threshold of how large the difference between most and least number of
# allocated shards must be to begin the rebalancing.
rebalance-threshold = 10
# The number of ongoing rebalancing processes is limited to this number.
max-simultaneous-rebalance = 3
}
˓→
˓→
# Timeout of waiting the initial distributed state (an initial state will be
queried again if the timeout happened)
# Only used when state-store-mode=ddata
waiting-for-state-timeout = 5 s
# Timeout of waiting for update the distributed state (update will be retried if
the timeout happened)
# Only used when state-store-mode=ddata
updating-state-timeout = 5 s
# The shard uses this strategy to determines how to recover the underlying
entity actors. The strategy is only used
# by the persistent shard when rebalancing or restarting. The value can either
˓→be "all" or "constant". The "all"
# strategy start all the underlying entity actors at the same time. The constant
˓→strategy will start the underlying
# entity actors at a fix rate. The default strategy "all".
entity-recovery-strategy = "all"
˓→
# Default settings for the constant rate entity recovery strategy
entity-recovery-constant-rate-strategy {
# Sets the frequency at which a batch of entity actors is started.
frequency = 100 ms
# Sets the number of entity actors to be restart at a particular interval
number-of-entities = 5
}
# Settings for the coordinator singleton. Same layout as akka.cluster.singleton.
# The "role" of the singleton configuration is not used. The singleton role will
# be the same as "akka.cluster.sharding.role".
coordinator-singleton = ${akka.cluster.singleton}
# Settings for the Distributed Data replicator.
# Same layout as akka.cluster.distributed-data.
# The "role" of the distributed-data configuration is not used. The distributed˓→data
# role will be the same as "akka.cluster.sharding.role".
# Note that there is one Replicator per role and it's not possible
# to have different distributed-data settings for different sharding entity
˓→types.
3.9. Configuration
104
Akka Scala Documentation, Release 2.5.0-RC1
# Only used when state-store-mode=ddata
distributed-data = ${akka.cluster.distributed-data}
distributed-data {
# minCap parameter to MajorityWrite and MajorityRead consistency level.
majority-min-cap = 5
durable.keys = ["shard-*"]
# When using many entities with "remember entities" the Gossip message
# can become to large if including to many in same message. Limit to
# the same number as the number of ORSet per shard.
akka.cluster.sharding.distributed-data.max-delta-elements = 5
}
# The id of the dispatcher to use for ClusterSharding actors.
# If not specified default dispatcher is used.
# If specified you need to define the settings of the actual dispatcher.
# This dispatcher for the entity actors is defined by the user provided
# Props, i.e. this dispatcher is not used for the entity actors.
use-dispatcher = ""
}
# //#sharding-ext-config
# Protobuf serializer for Cluster Sharding messages
akka.actor {
serializers {
akka-sharding = "akka.cluster.sharding.protobuf.
˓→ClusterShardingMessageSerializer"
}
serialization-bindings {
"akka.cluster.sharding.ClusterShardingSerializable" = akka-sharding
}
serialization-identifiers {
"akka.cluster.sharding.protobuf.ClusterShardingMessageSerializer" = 13
}
}
akka-distributed-data ~~~~~~~~~~~~———
##############################################
# Akka Distributed DataReference Config File #
##############################################
# This is the reference config file that contains all the default settings.
# Make your edits/overrides in your application.conf.
#//#distributed-data
# Settings for the DistributedData extension
akka.cluster.distributed-data {
# Actor name of the Replicator actor, /system/ddataReplicator
name = ddataReplicator
# Replicas are running on members tagged with this role.
# All members are used if undefined or empty.
role = ""
# How often the Replicator should send out gossip information
gossip-interval = 2 s
# How often the subscribers will be notified of changes, if any
notify-subscribers-interval = 500 ms
3.9. Configuration
105
Akka Scala Documentation, Release 2.5.0-RC1
# Maximum number of entries to transfer in one gossip message when synchronizing
# the replicas. Next chunk will be transferred in next round of gossip.
max-delta-elements = 1000
# The id of the dispatcher to use for Replicator actors. If not specified
# default dispatcher is used.
# If specified you need to define the settings of the actual dispatcher.
use-dispatcher = ""
# How often the Replicator checks for pruning of data associated with
# removed cluster nodes. If this is set to 'off' the pruning feature will
# be completely disabled.
pruning-interval = 120 s
# How long time it takes to spread the data to all other replica nodes.
# This is used when initiating and completing the pruning process of data
˓→associated
# with removed cluster nodes. The time measurement is stopped when any replica is
# unreachable, but it's still recommended to configure this with certain margin.
# It should be in the magnitude of minutes even though typical dissemination time
# is shorter (grows logarithmic with number of nodes). There is no advantage of
# setting this too low. Setting it to large value will delay the pruning process.
max-pruning-dissemination = 300 s
# The markers of that pruning has been performed for a removed node are kept for
this
# time and thereafter removed. If and old data entry that was never pruned is
˓→somehow
# injected and merged with existing data after this time the value will not be
˓→correct.
# This would be possible (although unlikely) in the case of a long network
˓→partition.
# It should be in the magnitude of hours. For durable data it is configured by
# 'akka.cluster.distributed-data.durable.pruning-marker-time-to-live'.
pruning-marker-time-to-live = 6 h
˓→
# Serialized Write and Read messages are cached when they are sent to
# several nodes. If no further activity they are removed from the cache
# after this duration.
serializer-cache-time-to-live = 10s
# Settings for delta-CRDT
delta-crdt {
# enable or disable delta-CRDT replication
enabled = on
}
durable {
# List of keys that are durable. Prefix matching is supported by using * at the
# end of a key.
keys = []
# The markers of that pruning has been performed for a removed node are kept
for this
# time and thereafter removed. If and old data entry that was never pruned is
# injected and merged with existing data after this time the value will not be
˓→correct.
# This would be possible if replica with durable data didn't participate in
˓→the pruning
# (e.g. it was shutdown) and later started after this time. A durable replica
˓→should not
# be stopped for longer time than this duration and if it is joining again
˓→after this
˓→
3.9. Configuration
106
Akka Scala Documentation, Release 2.5.0-RC1
# duration its data should first be manually removed (from the lmdb directory).
# It should be in the magnitude of days. Note that there is a corresponding
˓→setting
# for non-durable data: 'akka.cluster.distributed-data.pruning-marker-time-to˓→live'.
pruning-marker-time-to-live = 10 d
# Fully qualified class name of the durable store actor. It must be a subclass
# of akka.actor.Actor and handle the protocol defined in
# akka.cluster.ddata.DurableStore. The class must have a constructor with
# com.typesafe.config.Config parameter.
store-actor-class = akka.cluster.ddata.LmdbDurableStore
use-dispatcher = akka.cluster.distributed-data.durable.pinned-store
pinned-store {
executor = thread-pool-executor
type = PinnedDispatcher
}
# Config for the LmdbDurableStore
lmdb {
# Directory of LMDB file. There are two options:
# 1. A relative or absolute path to a directory that ends with 'ddata'
#
the full name of the directory will contain name of the ActorSystem
#
and its remote port.
# 2. Otherwise the path is used as is, as a relative or absolute path to
#
a directory.
#
# When running in production you may want to configure this to a specific
# path (alt 2), since the default directory contains the remote port of the
# actor system to make the name unique. If using a dynamically assigned
# port (0) it will be different each time and the previously stored data
# will not be loaded.
dir = "ddata"
# Size in bytes of the memory mapped file.
map-size = 100 MiB
# Accumulate changes before storing improves performance with the
# risk of losing the last writes if the JVM crashes.
# The interval is by default set to 'off' to write each update immediately.
# Enabling write behind by specifying a duration, e.g. 200ms, is especially
# efficient when performing many writes to the same key, because it is only
# the last value for each key that will be serialized and stored.
# write-behind-interval = 200 ms
write-behind-interval = off
}
}
}
#//#distributed-data
# Protobuf serializer for cluster DistributedData messages
akka.actor {
serializers {
akka-data-replication = "akka.cluster.ddata.protobuf.
˓→ReplicatorMessageSerializer"
akka-replicated-data = "akka.cluster.ddata.protobuf.ReplicatedDataSerializer"
}
serialization-bindings {
"akka.cluster.ddata.Replicator$ReplicatorMessage" = akka-data-replication
"akka.cluster.ddata.ReplicatedDataSerialization" = akka-replicated-data
3.9. Configuration
107
Akka Scala Documentation, Release 2.5.0-RC1
}
serialization-identifiers {
"akka.cluster.ddata.protobuf.ReplicatedDataSerializer" = 11
"akka.cluster.ddata.protobuf.ReplicatorMessageSerializer" = 12
}
}
3.9. Configuration
108
CHAPTER
FOUR
ACTORS
4.1 Actors
The Actor Model provides a higher level of abstraction for writing concurrent and distributed systems. It alleviates
the developer from having to deal with explicit locking and thread management, making it easier to write correct
concurrent and parallel systems. Actors were defined in the 1973 paper by Carl Hewitt but have been popularized
by the Erlang language, and used for example at Ericsson with great success to build highly concurrent and reliable
telecom systems.
The API of Akka’s Actors is similar to Scala Actors which has borrowed some of its syntax from Erlang.
4.1.1 Creating Actors
Note: Since Akka enforces parental supervision every actor is supervised and (potentially) the supervisor of its
children, it is advisable that you familiarize yourself with Actor Systems and Supervision and Monitoring and it
may also help to read Actor References, Paths and Addresses.
Defining an Actor class
Actors are implemented by extending the Actor base trait and implementing the receive method. The
receive method should define a series of case statements (which has the type PartialFunction[Any,
Unit]) that defines which messages your Actor can handle, using standard Scala pattern matching, along with
the implementation of how the messages should be processed.
Here is an example:
import akka.actor.Actor
import akka.actor.Props
import akka.event.Logging
class MyActor extends Actor {
val log = Logging(context.system, this)
def receive = {
case "test" => log.info("received test")
case _
=> log.info("received unknown message")
}
}
Please note that the Akka Actor receive message loop is exhaustive, which is different compared to Erlang and
the late Scala Actors. This means that you need to provide a pattern match for all messages that it can accept
and if you want to be able to handle unknown messages then you need to have a default case as in the example
above. Otherwise an akka.actor.UnhandledMessage(message, sender, recipient) will be
published to the ActorSystem‘s EventStream.
109
Akka Scala Documentation, Release 2.5.0-RC1
Note further that the return type of the behavior defined above is Unit; if the actor shall reply to the received
message then this must be done explicitly as explained below.
The result of the receive method is a partial function object, which is stored within the actor as its “initial behavior”, see Become/Unbecome for further information on changing the behavior of an actor after its construction.
Props
Props is a configuration class to specify options for the creation of actors, think of it as an immutable and thus
freely shareable recipe for creating an actor including associated deployment information (e.g. which dispatcher
to use, see more below). Here are some examples of how to create a Props instance.
import akka.actor.Props
val props1 = Props[MyActor]
val props2 = Props(new ActorWithArgs("arg")) // careful, see below
val props3 = Props(classOf[ActorWithArgs], "arg") // no support for value class
˓→arguments
The second variant shows how to pass constructor arguments to the Actor being created, but it should only be
used outside of actors as explained below.
The last line shows a possibility to pass constructor arguments regardless of the context it is being used in.
The presence of a matching constructor is verified during construction of the Props object, resulting in an
IllegalArgumentException if no or multiple matching constructors are found.
Note: The recommended approach to create the actor Props is not supported for cases when the actor constructor
takes value classes as arguments.
Dangerous Variants
// NOT RECOMMENDED within another actor:
// encourages to close over enclosing class
val props7 = Props(new MyActor)
This method is not recommended to be used within another actor because it encourages to close over the enclosing scope, resulting in non-serializable Props and possibly race conditions (breaking the actor encapsulation).
On the other hand using this variant in a Props factory in the actor’s companion object as documented under
“Recommended Practices” below is completely fine.
There were two use-cases for these methods: passing constructor arguments to the actor—which is solved by the
newly introduced Props.apply(clazz, args) method above or the recommended practice below—and
creating actors “on the spot” as anonymous classes. The latter should be solved by making these actors named
classes instead (if they are not declared within a top-level object then the enclosing instance’s this reference
needs to be passed as the first argument).
Warning: Declaring one actor within another is very dangerous and breaks actor encapsulation. Never pass
an actor’s this reference into Props!
Edge cases
There are two edge cases in actor creation with Props:
• An actor with AnyVal arguments.
4.1. Actors
110
Akka Scala Documentation, Release 2.5.0-RC1
case class MyValueClass(v: Int) extends AnyVal
class ValueActor(value: MyValueClass) extends Actor {
def receive = {
case multiplier: Long => sender() ! (value.v * multiplier)
}
}
val valueClassProp = Props(classOf[ValueActor], MyValueClass(5)) // Unsupported
• An actor with default constructor values.
class DefaultValueActor(a: Int, b: Int = 5) extends Actor {
def receive = {
case x: Int => sender() ! ((a + x) * b)
}
}
val defaultValueProp1 = Props(classOf[DefaultValueActor], 2.0) // Unsupported
class DefaultValueActor2(b: Int = 5) extends Actor {
def receive = {
case x: Int => sender() ! (x * b)
}
}
val defaultValueProp2 = Props[DefaultValueActor2] // Unsupported
val defaultValueProp3 = Props(classOf[DefaultValueActor2]) // Unsupported
In both cases an IllegalArgumentException will be thrown stating no matching constructor could be
found.
The next section explains the recommended ways to create Actor props in a way, which simultaneously safeguards against these edge cases.
Recommended Practices
It is a good idea to provide factory methods on the companion object of each Actor which help keeping the
creation of suitable Props as close to the actor definition as possible. This also avoids the pitfalls associated with
using the Props.apply(...) method which takes a by-name argument, since within a companion object the
given code block will not retain a reference to its enclosing scope:
object DemoActor {
/**
* Create Props for an actor of this type.
*
* @param magicNumber The magic number to be passed to this actor’s constructor.
* @return a Props for creating this actor, which can then be further configured
(e.g. calling `.withDispatcher()` on it)
*
*/
def props(magicNumber: Int): Props = Props(new DemoActor(magicNumber))
}
class DemoActor(magicNumber: Int) extends Actor {
def receive = {
case x: Int => sender() ! (x + magicNumber)
}
}
class SomeOtherActor extends Actor {
// Props(new DemoActor(42)) would not be safe
context.actorOf(DemoActor.props(42), "demo")
4.1. Actors
111
Akka Scala Documentation, Release 2.5.0-RC1
// ...
}
Another good practice is to declare what messages an Actor can receive in the companion object of the Actor,
which makes easier to know what it can receive:
object MyActor {
case class Greeting(from: String)
case object Goodbye
}
class MyActor extends Actor with ActorLogging {
import MyActor._
def receive = {
case Greeting(greeter) => log.info(s"I was greeted by $greeter.")
case Goodbye
=> log.info("Someone said goodbye to me.")
}
}
Creating Actors with Props
Actors are created by passing a Props instance into the actorOf factory method which is available on
ActorSystem and ActorContext.
import akka.actor.ActorSystem
// ActorSystem is a heavy object: create only one per application
val system = ActorSystem("mySystem")
val myActor = system.actorOf(Props[MyActor], "myactor2")
Using the ActorSystem will create top-level actors, supervised by the actor system’s provided guardian actor,
while using an actor’s context will create a child actor.
class FirstActor extends Actor {
val child = context.actorOf(Props[MyActor], name = "myChild")
// plus some behavior ...
}
It is recommended to create a hierarchy of children, grand-children and so on such that it fits the logical failurehandling structure of the application, see Actor Systems.
The call to actorOf returns an instance of ActorRef. This is a handle to the actor instance and the only way to
interact with it. The ActorRef is immutable and has a one to one relationship with the Actor it represents. The
ActorRef is also serializable and network-aware. This means that you can serialize it, send it over the wire and
use it on a remote host and it will still be representing the same Actor on the original node, across the network.
The name parameter is optional, but you should preferably name your actors, since that is used in log messages
and for identifying actors. The name must not be empty or start with $, but it may contain URL encoded characters (eg. %20 for a blank space). If the given name is already in use by another child to the same parent an
InvalidActorNameException is thrown.
Actors are automatically started asynchronously when created.
Value classes as constructor arguments
The recommended way to instantiate actor props uses reflection at runtime to determine the correct actor constructor to be invoked and due to technical limitations is not supported when said constructor takes arguments that are
value classes. In these cases you should either unpack the arguments or create the props by calling the constructor
manually:
4.1. Actors
112
Akka Scala Documentation, Release 2.5.0-RC1
class Argument(val value: String) extends AnyVal
class ValueClassActor(arg: Argument) extends Actor {
def receive = { case _ => () }
}
object ValueClassActor {
def props1(arg: Argument) = Props(classOf[ValueClassActor], arg) // fails at
˓→runtime
def props2(arg: Argument) = Props(classOf[ValueClassActor], arg.value) // ok
def props3(arg: Argument) = Props(new ValueClassActor(arg)) // ok
}
Dependency Injection
If your Actor has a constructor that takes parameters then those need to be part of the Props as well, as described above. But there are cases when a factory method must be used, for example when the actual constructor
arguments are determined by a dependency injection framework.
import akka.actor.IndirectActorProducer
class DependencyInjector(applicationContext: AnyRef, beanName: String)
extends IndirectActorProducer {
override def actorClass = classOf[Actor]
override def produce =
// obtain fresh Actor instance from DI framework ...
}
val actorRef = system.actorOf(
Props(classOf[DependencyInjector], applicationContext, "hello"),
"helloBean")
Warning: You might be tempted at times to offer an IndirectActorProducer which always returns
the same instance, e.g. by using a lazy val. This is not supported, as it goes against the meaning of an
actor restart, which is described here: What Restarting Means.
When using a dependency injection framework, actor beans MUST NOT have singleton scope.
Techniques for dependency injection and integration with dependency injection frameworks are described in more
depth in the Using Akka with Dependency Injection guideline and the Akka Java Spring tutorial in Lightbend
Activator.
The Inbox
When writing code outside of actors which shall communicate with actors, the ask pattern can be a solution (see
below), but there are two things it cannot do: receiving multiple replies (e.g. by subscribing an ActorRef to a
notification service) and watching other actors’ lifecycle. For these purposes there is the Inbox class:
implicit val i = inbox()
echo ! "hello"
i.receive() should ===("hello")
There is an implicit conversion from inbox to actor reference which means that in this example the sender reference
will be that of the actor hidden away within the inbox. This allows the reply to be received on the last line.
Watching an actor is quite simple as well:
4.1. Actors
113
Akka Scala Documentation, Release 2.5.0-RC1
val target = // some actor
val i = inbox()
i watch target
4.1.2 Actor API
The Actor trait defines only one abstract method, the above mentioned receive, which implements the behavior of the actor.
If the current actor behavior does not match a received message, unhandled is called, which by default publishes
an akka.actor.UnhandledMessage(message, sender, recipient) on the actor system’s event
stream (set configuration item akka.actor.debug.unhandled to on to have them converted into actual
Debug messages).
In addition, it offers:
• self reference to the ActorRef of the actor
• sender reference sender Actor of the last received message, typically used as described in Reply to messages
• supervisorStrategy user overridable definition the strategy to use for supervising child actors
This strategy is typically declared inside the actor in order to have access to the actor’s internal state within
the decider function: since failure is communicated as a message sent to the supervisor and processed like
other messages (albeit outside of the normal behavior), all values and variables within the actor are available,
as is the sender reference (which will be the immediate child reporting the failure; if the original failure
occurred within a distant descendant it is still reported one level up at a time).
• context exposes contextual information for the actor and the current message, such as:
– factory methods to create child actors (actorOf)
– system that the actor belongs to
– parent supervisor
– supervised children
– lifecycle monitoring
– hotswap behavior stack as described in Become/Unbecome
You can import the members in the context to avoid prefixing access with context.
class FirstActor extends Actor {
import context._
val myActor = actorOf(Props[MyActor], name = "myactor")
def receive = {
case x => myActor ! x
}
}
The remaining visible methods are user-overridable life-cycle hooks which are described in the following:
def preStart(): Unit = ()
def postStop(): Unit = ()
def preRestart(reason: Throwable, message: Option[Any]): Unit = {
context.children foreach { child ⇒
context.unwatch(child)
context.stop(child)
}
postStop()
4.1. Actors
114
Akka Scala Documentation, Release 2.5.0-RC1
}
def postRestart(reason: Throwable): Unit = {
preStart()
}
The implementations shown above are the defaults provided by the Actor trait.
Actor Lifecycle
A path in an actor system represents a “place” which might be occupied by a living actor. Initially (apart from
system initialized actors) a path is empty. When actorOf() is called it assigns an incarnation of the actor
described by the passed Props to the given path. An actor incarnation is identified by the path and a UID. A
restart only swaps the Actor instance defined by the Props but the incarnation and hence the UID remains the
same.
4.1. Actors
115
Akka Scala Documentation, Release 2.5.0-RC1
The lifecycle of an incarnation ends when the actor is stopped. At that point the appropriate lifecycle events are
called and watching actors are notified of the termination. After the incarnation is stopped, the path can be reused
again by creating an actor with actorOf(). In this case the name of the new incarnation will be the same
as the previous one but the UIDs will differ. An actor can be stopped by the actor itself, another actor or the
ActorSystem (see Stopping actors).
Note: It is important to note that Actors do not stop automatically when no longer referenced, every Actor that
is created must also explicitly be destroyed. The only simplification is that stopping a parent Actor will also
recursively stop all the child Actors that this parent has created.
An ActorRef always represents an incarnation (path and UID) not just a given path. Therefore if an actor is
stopped and a new one with the same name is created an ActorRef of the old incarnation will not point to the
new one.
ActorSelection on the other hand points to the path (or multiple paths if wildcards are used) and is completely
oblivious to which incarnation is currently occupying it. ActorSelection cannot be watched for this reason.
It is possible to resolve the current incarnation’s ActorRef living under the path by sending an Identify
message to the ActorSelection which will be replied to with an ActorIdentity containing the correct
reference (see Identifying Actors via Actor Selection). This can also be done with the resolveOne method of
the ActorSelection, which returns a Future of the matching ActorRef.
Lifecycle Monitoring aka DeathWatch
In order to be notified when another actor terminates (i.e. stops permanently, not temporary failure and restart), an
actor may register itself for reception of the Terminated message dispatched by the other actor upon termination
(see Stopping Actors). This service is provided by the DeathWatch component of the actor system.
Registering a monitor is easy:
import akka.actor.{ Actor, Props, Terminated }
class WatchActor extends Actor {
val child = context.actorOf(Props.empty, "child")
context.watch(child) // <-- this is the only call needed for registration
var lastSender = context.system.deadLetters
def receive = {
case "kill" =>
context.stop(child); lastSender = sender()
case Terminated(`child`) => lastSender ! "finished"
}
}
It should be noted that the Terminated message is generated independent of the order in which registration and
termination occur. In particular, the watching actor will receive a Terminated message even if the watched
actor has already been terminated at the time of registration.
Registering multiple times does not necessarily lead to multiple messages being generated, but there is no guarantee that only exactly one such message is received: if termination of the watched actor has generated and queued
the message, and another registration is done before this message has been processed, then a second message will
be queued, because registering for monitoring of an already terminated actor leads to the immediate generation of
the Terminated message.
It is also possible to deregister from watching another actor’s liveliness using context.unwatch(target).
This works even if the Terminated message has already been enqueued in the mailbox; after calling unwatch
no Terminated message for that actor will be processed anymore.
4.1. Actors
116
Akka Scala Documentation, Release 2.5.0-RC1
Start Hook
Right after starting the actor, its preStart method is invoked.
override def preStart() {
child = context.actorOf(Props[MyActor], "child")
}
This method is called when the actor is first created. During restarts it is called by the default implementation of
postRestart, which means that by overriding that method you can choose whether the initialization code in
this method is called only exactly once for this actor or for every restart. Initialization code which is part of the
actor’s constructor will always be called when an instance of the actor class is created, which happens at every
restart.
Restart Hooks
All actors are supervised, i.e. linked to another actor with a fault handling strategy. Actors may be restarted in
case an exception is thrown while processing a message (see Supervision and Monitoring). This restart involves
the hooks mentioned above:
1. The old actor is informed by calling preRestart with the exception which caused the restart and the
message which triggered that exception; the latter may be None if the restart was not caused by processing
a message, e.g. when a supervisor does not trap the exception and is restarted in turn by its supervisor, or if
an actor is restarted due to a sibling’s failure. If the message is available, then that message’s sender is also
accessible in the usual way (i.e. by calling sender).
This method is the best place for cleaning up, preparing hand-over to the fresh actor instance, etc. By default
it stops all children and calls postStop.
2. The initial factory from the actorOf call is used to produce the fresh instance.
3. The new actor’s postRestart method is invoked with the exception which caused the restart. By default
the preStart is called, just as in the normal start-up case.
An actor restart replaces only the actual actor object; the contents of the mailbox is unaffected by the restart,
so processing of messages will resume after the postRestart hook returns. The message that triggered the
exception will not be received again. Any message sent to an actor while it is being restarted will be queued to its
mailbox as usual.
Warning: Be aware that the ordering of failure notifications relative to user messages is not deterministic. In
particular, a parent might restart its child before it has processed the last messages sent by the child before the
failure. See Discussion: Message Ordering for details.
Stop Hook
After stopping an actor, its postStop hook is called, which may be used e.g. for deregistering this actor from
other services. This hook is guaranteed to run after message queuing has been disabled for this actor, i.e. messages
sent to a stopped actor will be redirected to the deadLetters of the ActorSystem.
4.1.3 Identifying Actors via Actor Selection
As described in Actor References, Paths and Addresses, each actor has a unique logical path, which is obtained
by following the chain of actors from child to parent until reaching the root of the actor system, and it has a
physical path, which may differ if the supervision chain includes any remote supervisors. These paths are used
by the system to look up actors, e.g. when a remote message is received and the recipient is searched, but they
are also useful more directly: actors may look up other actors by specifying absolute or relative paths—logical or
physical—and receive back an ActorSelection with the result:
4.1. Actors
117
Akka Scala Documentation, Release 2.5.0-RC1
// will look up this absolute path
context.actorSelection("/user/serviceA/aggregator")
// will look up sibling beneath same supervisor
context.actorSelection("../joe")
Note: It is always preferable to communicate with other Actors using their ActorRef instead of relying upon
ActorSelection. Exceptions are
• sending messages using the At-Least-Once Delivery facility
• initiating first contact with a remote system
In all other cases ActorRefs can be provided during Actor creation or initialization, passing them from parent to
child or introducing Actors by sending their ActorRefs to other Actors within messages.
The supplied path is parsed as a java.net.URI, which basically means that it is split on / into path elements.
If the path starts with /, it is absolute and the look-up starts at the root guardian (which is the parent of "/user");
otherwise it starts at the current actor. If a path element equals .., the look-up will take a step “up” towards the
supervisor of the currently traversed actor, otherwise it will step “down” to the named child. It should be noted
that the .. in actor paths here always means the logical structure, i.e. the supervisor.
The path elements of an actor selection may contain wildcard patterns allowing for broadcasting of messages to
that section:
// will look all children to serviceB with names starting with worker
context.actorSelection("/user/serviceB/worker*")
// will look up all siblings beneath same supervisor
context.actorSelection("../*")
Messages can be sent via the ActorSelection and the path of the ActorSelection is looked up when
delivering each message. If the selection does not match any actors the message will be dropped.
To acquire an ActorRef for an ActorSelection you need to send a message to the selection and use the
sender() reference of the reply from the actor. There is a built-in Identify message that all Actors will
understand and automatically reply to with a ActorIdentity message containing the ActorRef. This message is handled specially by the actors which are traversed in the sense that if a concrete name lookup fails (i.e.
a non-wildcard path element does not correspond to a live actor) then a negative result is generated. Please note
that this does not mean that delivery of that reply is guaranteed, it still is a normal message.
import akka.actor.{ Actor, Props, Identify, ActorIdentity, Terminated }
class Follower extends Actor {
val identifyId = 1
context.actorSelection("/user/another") ! Identify(identifyId)
def receive = {
case ActorIdentity(`identifyId`, Some(ref)) =>
context.watch(ref)
context.become(active(ref))
case ActorIdentity(`identifyId`, None) => context.stop(self)
}
def active(another: ActorRef): Actor.Receive = {
case Terminated(`another`) => context.stop(self)
}
}
You can also acquire an ActorRef for an ActorSelection with the resolveOne method of the
ActorSelection. It returns a Future of the matching ActorRef if such an actor exists. It is completed
4.1. Actors
118
Akka Scala Documentation, Release 2.5.0-RC1
with failure [[akka.actor.ActorNotFound]] if no such actor exists or the identification didn’t complete within the
supplied timeout.
Remote actor addresses may also be looked up, if remoting is enabled:
context.actorSelection("akka.tcp://app@otherhost:1234/user/serviceB")
An example demonstrating actor look-up is given in Remoting Sample.
4.1.4 Messages and immutability
IMPORTANT: Messages can be any kind of object but have to be immutable. Scala can’t enforce immutability
(yet) so this has to be by convention. Primitives like String, Int, Boolean are always immutable. Apart from these
the recommended approach is to use Scala case classes which are immutable (if you don’t explicitly expose the
state) and works great with pattern matching at the receiver side.
Here is an example:
// define the case class
case class Register(user: User)
// create a new case class message
val message = Register(user)
4.1.5 Send messages
Messages are sent to an Actor through one of the following methods.
• ! means “fire-and-forget”, e.g. send a message asynchronously and return immediately. Also known as
tell.
• ? sends a message asynchronously and returns a Future representing a possible reply. Also known as
ask.
Message ordering is guaranteed on a per-sender basis.
Note: There are performance implications of using ask since something needs to keep track of when it times
out, there needs to be something that bridges a Promise into an ActorRef and it also needs to be reachable
through remoting. So always prefer tell for performance, and only ask if you must.
Tell: Fire-forget
This is the preferred way of sending messages. No blocking waiting for a message. This gives the best concurrency
and scalability characteristics.
actorRef ! message
If invoked from within an Actor, then the sending actor reference will be implicitly passed along with the message
and available to the receiving Actor in its sender(): ActorRef member method. The target actor can use
this to reply to the original sender, by using sender() ! replyMsg.
If invoked from an instance that is not an Actor the sender will be deadLetters actor reference by default.
Ask: Send-And-Receive-Future
The ask pattern involves actors as well as futures, hence it is offered as a use pattern rather than a method on
ActorRef:
4.1. Actors
119
Akka Scala Documentation, Release 2.5.0-RC1
import akka.pattern.{ ask, pipe }
import system.dispatcher // The ExecutionContext that will be used
final case class Result(x: Int, s: String, d: Double)
case object Request
implicit val timeout = Timeout(5 seconds) // needed for `?` below
val f: Future[Result] =
for {
x <- ask(actorA, Request).mapTo[Int] // call pattern directly
s <- (actorB ask Request).mapTo[String] // call by implicit conversion
d <- (actorC ? Request).mapTo[Double] // call by symbolic name
} yield Result(x, s, d)
f pipeTo actorD // .. or ..
pipe(f) to actorD
This example demonstrates ask together with the pipeTo pattern on futures, because this is likely to be a common combination. Please note that all of the above is completely non-blocking and asynchronous: ask produces
a Future, three of which are composed into a new future using the for-comprehension and then pipeTo installs
an onComplete-handler on the future to affect the submission of the aggregated Result to another actor.
Using ask will send a message to the receiving Actor as with tell, and the receiving actor must reply with
sender() ! reply in order to complete the returned Future with a value. The ask operation involves
creating an internal actor for handling this reply, which needs to have a timeout after which it is destroyed in order
not to leak resources; see more below.
Warning: To complete the future with an exception you need send a Failure message to the sender. This is
not done automatically when an actor throws an exception while processing a message.
try {
val result = operation()
sender() ! result
} catch {
case e: Exception =>
sender() ! akka.actor.Status.Failure(e)
throw e
}
If the actor does not complete the future, it will expire after the timeout period, completing it with an
AskTimeoutException. The timeout is taken from one of the following locations in order of precedence:
1. explicitly given timeout as in:
import scala.concurrent.duration._
import akka.pattern.ask
val future = myActor.ask("hello")(5 seconds)
2. implicit argument of type akka.util.Timeout, e.g.
import scala.concurrent.duration._
import akka.util.Timeout
import akka.pattern.ask
implicit val timeout = Timeout(5 seconds)
val future = myActor ? "hello"
See Futures for more information on how to await or query a future.
The onComplete, onSuccess, or onFailure methods of the Future can be used to register a callback to
get a notification when the Future completes, giving you a way to avoid blocking.
4.1. Actors
120
Akka Scala Documentation, Release 2.5.0-RC1
Warning: When using future callbacks, such as onComplete, onSuccess, and onFailure, inside actors you need to carefully avoid closing over the containing actor’s reference, i.e. do not call methods or access
mutable state on the enclosing actor from within the callback. This would break the actor encapsulation and
may introduce synchronization bugs and race conditions because the callback will be scheduled concurrently
to the enclosing actor. Unfortunately there is not yet a way to detect these illegal accesses at compile time. See
also: Actors and shared mutable state
Forward message
You can forward a message from one actor to another. This means that the original sender address/reference is
maintained even though the message is going through a ‘mediator’. This can be useful when writing actors that
work as routers, load-balancers, replicators etc.
target forward message
4.1.6 Receive messages
An Actor has to implement the receive method to receive messages:
type Receive = PartialFunction[Any, Unit]
def receive: Actor.Receive
This method returns a PartialFunction, e.g. a ‘match/case’ clause in which the message can be matched
against the different case clauses using Scala pattern matching. Here is an example:
import akka.actor.Actor
import akka.actor.Props
import akka.event.Logging
class MyActor extends Actor {
val log = Logging(context.system, this)
def receive = {
case "test" => log.info("received test")
case _
=> log.info("received unknown message")
}
}
4.1.7 Reply to messages
If you want to have a handle for replying to a message, you can use sender(), which gives you an ActorRef.
You can reply by sending to that ActorRef with sender() ! replyMsg. You can also store the ActorRef for
replying later, or passing on to other actors. If there is no sender (a message was sent without an actor or future
context) then the sender defaults to a ‘dead-letter’ actor ref.
case request =>
val result = process(request)
sender() ! result
// will have dead-letter actor as default
4.1.8 Receive timeout
The ActorContext setReceiveTimeout defines the inactivity timeout after which the sending of a ReceiveTimeout message is triggered. When specified, the receive function should be able to handle an
akka.actor.ReceiveTimeout message. 1 millisecond is the minimum supported timeout.
4.1. Actors
121
Akka Scala Documentation, Release 2.5.0-RC1
Please note that the receive timeout might fire and enqueue the ReceiveTimeout message right after another message was enqueued; hence it is not guaranteed that upon reception of the receive timeout there must have been
an idle period beforehand as configured via this method.
Once set, the receive timeout stays in effect (i.e. continues firing repeatedly after inactivity periods). Pass in
Duration.Undefined to switch off this feature.
import akka.actor.ReceiveTimeout
import scala.concurrent.duration._
class MyActor extends Actor {
// To set an initial delay
context.setReceiveTimeout(30 milliseconds)
def receive = {
case "Hello" =>
// To set in a response to a message
context.setReceiveTimeout(100 milliseconds)
case ReceiveTimeout =>
// To turn it off
context.setReceiveTimeout(Duration.Undefined)
throw new RuntimeException("Receive timed out")
}
}
Messages marked with NotInfluenceReceiveTimeout will not reset the timer. This can be useful when
ReceiveTimeout should be fired by external inactivity but not influenced by internal activity, e.g. scheduled
tick messages.
4.1.9 Stopping actors
Actors are stopped by invoking the stop method of a ActorRefFactory, i.e. ActorContext or
ActorSystem. Typically the context is used for stopping the actor itself or child actors and the system for
stopping top level actors. The actual termination of the actor is performed asynchronously, i.e. stop may return
before the actor is stopped.
class MyActor extends Actor {
val child: ActorRef = ???
def receive = {
case "interrupt-child" =>
context stop child
case "done" =>
context stop self
}
}
Processing of the current message, if any, will continue before the actor is stopped, but additional messages in the
mailbox will not be processed. By default these messages are sent to the deadLetters of the ActorSystem,
but that depends on the mailbox implementation.
Termination of an actor proceeds in two steps: first the actor suspends its mailbox processing and sends a stop
command to all its children, then it keeps processing the internal termination notifications from its children until
the last one is gone, finally terminating itself (invoking postStop, dumping mailbox, publishing Terminated
on the DeathWatch, telling its supervisor). This procedure ensures that actor system sub-trees terminate in an
orderly fashion, propagating the stop command to the leaves and collecting their confirmation back to the stopped
supervisor. If one of the actors does not respond (i.e. processing a message for extended periods of time and
therefore not receiving the stop command), this whole process will be stuck.
Upon ActorSystem.terminate, the system guardian actors will be stopped, and the aforementioned process
will ensure proper termination of the whole system.
4.1. Actors
122
Akka Scala Documentation, Release 2.5.0-RC1
The postStop hook is invoked after an actor is fully stopped. This enables cleaning up of resources:
override def postStop() {
// clean up some resources ...
}
Note: Since stopping an actor is asynchronous, you cannot immediately reuse the name of the child you just
stopped; this will result in an InvalidActorNameException. Instead, watch the terminating actor and
create its replacement in response to the Terminated message which will eventually arrive.
PoisonPill
You can also send an actor the akka.actor.PoisonPill message, which will stop the actor when the message is processed. PoisonPill is enqueued as ordinary messages and will be handled after messages that were
already queued in the mailbox.
Graceful Stop
gracefulStop is useful if you need to wait for termination or compose ordered termination of several actors:
import akka.pattern.gracefulStop
import scala.concurrent.Await
try {
val stopped: Future[Boolean] = gracefulStop(actorRef, 5 seconds, Manager.
˓→Shutdown)
Await.result(stopped, 6 seconds)
// the actor has been stopped
} catch {
// the actor wasn't stopped within 5 seconds
case e: akka.pattern.AskTimeoutException =>
}
object Manager {
case object Shutdown
}
class Manager extends Actor {
import Manager._
val worker = context.watch(context.actorOf(Props[Cruncher], "worker"))
def receive = {
case "job" => worker ! "crunch"
case Shutdown =>
worker ! PoisonPill
context become shuttingDown
}
def shuttingDown: Receive = {
case "job" => sender() ! "service unavailable, shutting down"
case Terminated(`worker`) =>
context stop self
}
}
When gracefulStop() returns successfully, the actor’s postStop() hook will have been executed: there
exists a happens-before edge between the end of postStop() and the return of gracefulStop().
4.1. Actors
123
Akka Scala Documentation, Release 2.5.0-RC1
In the above example a custom Manager.Shutdown message is sent to the target actor to initiate the process
of stopping the actor. You can use PoisonPill for this, but then you have limited possibilities to perform
interactions with other actors before stopping the target actor. Simple cleanup tasks can be handled in postStop.
Warning: Keep in mind that an actor stopping and its name being deregistered are separate events which
happen asynchronously from each other. Therefore it may be that you will find the name still in use after
gracefulStop() returned. In order to guarantee proper deregistration, only reuse names from within a
supervisor you control and only in response to a Terminated message, i.e. not for top-level actors.
Coordinated Shutdown
There is an extension named CoordinatedShutdown that will stop certain actors and services in a specific
order and perform registered tasks during the shutdown process.
The order of the shutdown phases is defined in configuration akka.coordinated-shutdown.phases. The
default phases are defined as:
# CoordinatedShutdown will run the tasks that are added to these
# phases. The phases can be ordered as a DAG by defining the
# dependencies between the phases.
# Each phase is defined as a named config section with the
# following optional properties:
# - timeout=15s: Override the default-phase-timeout for this phase.
# - recover=off: If the phase fails the shutdown is aborted
#
and depending phases will not be executed.
# depends-on=[]: Run the phase after the given phases
phases {
# The first pre-defined phase that applications can add tasks to.
# Note that more phases can be be added in the application's
# configuration by overriding this phase with an additional
# depends-on.
before-service-unbind {
}
# Stop accepting new incoming requests in for example HTTP.
service-unbind {
depends-on = [before-service-unbind]
}
# Wait for requests that are in progress to be completed.
service-requests-done {
depends-on = [service-unbind]
}
# Final shutdown of service endpoints.
service-stop {
depends-on = [service-requests-done]
}
# Phase for custom application tasks that are to be run
# after service shutdown and before cluster shutdown.
before-cluster-shutdown {
depends-on = [service-stop]
}
# Graceful shutdown of the Cluster Sharding regions.
cluster-sharding-shutdown-region {
timeout = 10 s
depends-on = [before-cluster-shutdown]
4.1. Actors
124
Akka Scala Documentation, Release 2.5.0-RC1
}
# Emit the leave command for the node that is shutting down.
cluster-leave {
depends-on = [cluster-sharding-shutdown-region]
}
# Shutdown cluster singletons
cluster-exiting {
timeout = 10 s
depends-on = [cluster-leave]
}
# Wait until exiting has been completed
cluster-exiting-done {
depends-on = [cluster-exiting]
}
# Shutdown the cluster extension
cluster-shutdown {
depends-on = [cluster-exiting-done]
}
# Phase for custom application tasks that are to be run
# after cluster shutdown and before ActorSystem termination.
before-actor-system-terminate {
depends-on = [cluster-shutdown]
}
# Last phase. See terminate-actor-system and exit-jvm above.
# Don't add phases that depends on this phase because the
# dispatcher and scheduler of the ActorSystem have been shutdown.
actor-system-terminate {
timeout = 10 s
depends-on = [before-actor-system-terminate]
}
}
More phases can be be added in the application’s configuration if needed by overriding a phase with an additional depends-on. Especially the phases before-service-unbind, before-cluster-shutdown
and before-actor-system-terminate are intended for application specific phases or tasks.
The default phases are defined in a single linear order, but the phases can be ordered as a directed acyclic graph
(DAG) by defining the dependencies between the phases. The phases are ordered with topological sort of the
DAG.
Tasks can be added to a phase with:
CoordinatedShutdown(system).addTask(
CoordinatedShutdown.PhaseBeforeServiceUnbind, "someTaskName") { () =>
import akka.pattern.ask
import system.dispatcher
implicit val timeout = Timeout(5.seconds)
(someActor ? "stop").map(_ => Done)
}
The returned Future[Done] should be completed when the task is completed. The task name parameter is only
used for debugging/logging.
Tasks added to the same phase are executed in parallel without any ordering assumptions. Next phase will not
start until all tasks of previous phase have been completed.
If tasks are not completed within a configured timeout (see reference.conf ) the next phase will be started anyway.
4.1. Actors
125
Akka Scala Documentation, Release 2.5.0-RC1
It is possible to configure recover=off for a phase to abort the rest of the shutdown process if a task fails or is
not completed within the timeout.
Tasks should typically be registered as early as possible after system startup. When running the coordinated
shutdown tasks that have been registered will be performed but tasks that are added too late will not be run.
To start the coordinated shutdown process you can invoke run on the CoordinatedShutdown extension:
val done: Future[Done] = CoordinatedShutdown(system).run()
It’s safe to call the run method multiple times. It will only run once.
That also means that the ActorSystem will be terminated in the last phase. By default, the JVM is not forcefully
stopped (it will be stopped if all non-daemon threads have been terminated). To enable a hard System.exit as
a final action you can configure:
akka.coordinated-shutdown.exit-jvm = on
When using Akka Cluster the CoordinatedShutdown will automatically run when the cluster node sees itself
as Exiting, i.e. leaving from another node will trigger the shutdown process on the leaving node. Tasks for
graceful leaving of cluster including graceful shutdown of Cluster Singletons and Cluster Sharding are added
automatically when Akka Cluster is used, i.e. running the shutdown process will also trigger the graceful leaving
if it’s not already in progress.
By default, the CoordinatedShutdown will be run when the JVM process exits, e.g. via kill SIGTERM
signal (SIGINT ctrl-c doesn’t work). This behavior can be disabled with:
akka.coordinated-shutdown.run-by-jvm-shutdown-hook=off
If you have application specific JVM shutdown hooks it’s recommended that you register them via the
CoordinatedShutdown so that they are running before Akka internal shutdown hooks, e.g. those shutting
down Akka Remoting (Artery).
CoordinatedShutdown(system).addJvmShutdownHook {
println("custom JVM shutdown hook...")
}
For some tests it might be undesired to terminate the ActorSystem via CoordinatedShutdown. You can
disable that by adding the following to the configuration of the ActorSystem that is used in the test:
# Don't terminate ActorSystem via CoordinatedShutdown in tests
akka.coordinated-shutdown.terminate-actor-system = off
akka.coordinated-shutdown.run-by-jvm-shutdown-hook = off
akka.cluster.run-coordinated-shutdown-when-down = off
4.1.10 Become/Unbecome
Upgrade
Akka supports hotswapping the Actor’s message loop (e.g. its implementation) at runtime: invoke the context.
become method from within the Actor. become takes a PartialFunction[Any, Unit] that implements
the new message handler. The hotswapped code is kept in a Stack which can be pushed and popped.
Warning: Please note that the actor will revert to its original behavior when restarted by its Supervisor.
To hotswap the Actor behavior using become:
4.1. Actors
126
Akka Scala Documentation, Release 2.5.0-RC1
class HotSwapActor extends Actor {
import context._
def angry: Receive = {
case "foo" => sender() ! "I am already angry?"
case "bar" => become(happy)
}
def happy: Receive = {
case "bar" => sender() ! "I am already happy :-)"
case "foo" => become(angry)
}
def receive = {
case "foo" => become(angry)
case "bar" => become(happy)
}
}
This variant of the become method is useful for many different things, such as to implement a Finite State
Machine (FSM, for an example see Dining Hakkers). It will replace the current behavior (i.e. the top of the
behavior stack), which means that you do not use unbecome, instead always the next behavior is explicitly
installed.
The other way of using become does not replace but add to the top of the behavior stack. In this case care must
be taken to ensure that the number of “pop” operations (i.e. unbecome) matches the number of “push” ones in
the long run, otherwise this amounts to a memory leak (which is why this behavior is not the default).
case object Swap
class Swapper extends Actor {
import context._
val log = Logging(system, this)
def receive = {
case Swap =>
log.info("Hi")
become({
case Swap =>
log.info("Ho")
unbecome() // resets the latest 'become' (just for fun)
}, discardOld = false) // push on top instead of replace
}
}
object SwapperApp extends App {
val system = ActorSystem("SwapperSystem")
val swap = system.actorOf(Props[Swapper], name = "swapper")
swap ! Swap // logs Hi
swap ! Swap // logs Ho
swap ! Swap // logs Hi
swap ! Swap // logs Ho
swap ! Swap // logs Hi
swap ! Swap // logs Ho
}
Encoding Scala Actors nested receives without accidentally leaking memory
See this Unnested receive example.
4.1. Actors
127
Akka Scala Documentation, Release 2.5.0-RC1
4.1.11 Stash
The Stash trait enables an actor to temporarily stash away messages that can not or should not be handled using
the actor’s current behavior. Upon changing the actor’s message handler, i.e., right before invoking context.
become or context.unbecome, all stashed messages can be “unstashed”, thereby prepending them to the
actor’s mailbox. This way, the stashed messages can be processed in the same order as they have been received
originally.
Note: The trait Stash extends the marker trait RequiresMessageQueue[DequeBasedMessageQueueSemantics]
which requests the system to automatically choose a deque based mailbox implementation for the actor. If you
want more control over the mailbox, see the documentation on mailboxes: Mailboxes.
Here is an example of the Stash in action:
import akka.actor.Stash
class ActorWithProtocol extends Actor with Stash {
def receive = {
case "open" =>
unstashAll()
context.become({
case "write" => // do writing...
case "close" =>
unstashAll()
context.unbecome()
case msg => stash()
}, discardOld = false) // stack on top instead of replacing
case msg => stash()
}
}
Invoking stash() adds the current message (the message that the actor received last) to the actor’s stash.
It is typically invoked when handling the default case in the actor’s message handler to stash messages that
aren’t handled by the other cases. It is illegal to stash the same message twice; to do so results in an
IllegalStateException being thrown. The stash may also be bounded in which case invoking stash()
may lead to a capacity violation, which results in a StashOverflowException. The capacity of the stash
can be configured using the stash-capacity setting (an Int) of the mailbox’s configuration.
Invoking unstashAll() enqueues messages from the stash to the actor’s mailbox until the capacity of the mailbox (if any) has been reached (note that messages from the stash are prepended to the mailbox). In case a bounded
mailbox overflows, a MessageQueueAppendFailedException is thrown. The stash is guaranteed to be
empty after calling unstashAll().
The stash is backed by a scala.collection.immutable.Vector. As a result, even a very large number
of messages may be stashed without a major impact on performance.
Warning: Note that the Stash trait must be mixed into (a subclass of) the Actor trait before any trait/class
that overrides the preRestart callback. This means it’s not possible to write Actor with MyActor
with Stash if MyActor overrides preRestart.
Note that the stash is part of the ephemeral actor state, unlike the mailbox. Therefore, it should be managed like
other parts of the actor’s state which have the same property. The Stash trait’s implementation of preRestart
will call unstashAll(), which is usually the desired behavior.
Note: If you want to enforce that your actor can only work with an unbounded stash, then you should use the
UnboundedStash trait instead.
4.1. Actors
128
Akka Scala Documentation, Release 2.5.0-RC1
4.1.12 Killing an Actor
You can kill an actor by sending a Kill message.
This will cause the actor to throw a
ActorKilledException, triggering a failure. The actor will suspend operation and its supervisor will be
asked how to handle the failure, which may mean resuming the actor, restarting it or terminating it completely.
See What Supervision Means for more information.
Use Kill like this:
// kill the 'victim' actor
victim ! Kill
4.1.13 Actors and exceptions
It can happen that while a message is being processed by an actor, that some kind of exception is thrown, e.g. a
database exception.
What happens to the Message
If an exception is thrown while a message is being processed (i.e. taken out of its mailbox and handed over to the
current behavior), then this message will be lost. It is important to understand that it is not put back on the mailbox.
So if you want to retry processing of a message, you need to deal with it yourself by catching the exception and
retry your flow. Make sure that you put a bound on the number of retries since you don’t want a system to livelock
(so consuming a lot of cpu cycles without making progress). Another possibility would be to have a look at the
PeekMailbox pattern.
What happens to the mailbox
If an exception is thrown while a message is being processed, nothing happens to the mailbox. If the actor is
restarted, the same mailbox will be there. So all messages on that mailbox will be there as well.
What happens to the actor
If code within an actor throws an exception, that actor is suspended and the supervision process is started (see Supervision and Monitoring). Depending on the supervisor’s decision the actor is resumed (as if nothing happened),
restarted (wiping out its internal state and starting from scratch) or terminated.
4.1.14 Extending Actors using PartialFunction chaining
Sometimes it can be useful to share common behavior among a few actors, or compose one actor’s behavior from
multiple smaller functions. This is possible because an actor’s receive method returns an Actor.Receive,
which is a type alias for PartialFunction[Any,Unit], and partial functions can be chained together using
the PartialFunction#orElse method. You can chain as many functions as you need, however you should
keep in mind that “first match” wins - which may be important when combining functions that both can handle
the same type of message.
For example, imagine you have a set of actors which are either Producers or Consumers, yet sometimes
it makes sense to have an actor share both behaviors. This can be easily achieved without having to duplicate
code by extracting the behaviors to traits and implementing the actor’s receive as combination of these partial
functions.
trait ProducerBehavior {
this: Actor =>
val producerBehavior: Receive = {
4.1. Actors
129
Akka Scala Documentation, Release 2.5.0-RC1
case GiveMeThings =>
sender() ! Give("thing")
}
}
trait ConsumerBehavior {
this: Actor with ActorLogging =>
val consumerBehavior: Receive = {
case ref: ActorRef =>
ref ! GiveMeThings
case Give(thing) =>
log.info("Got a thing! It's {}", thing)
}
}
class Producer extends Actor with ProducerBehavior {
def receive = producerBehavior
}
class Consumer extends Actor with ActorLogging with ConsumerBehavior {
def receive = consumerBehavior
}
class ProducerConsumer extends Actor with ActorLogging
with ProducerBehavior with ConsumerBehavior {
def receive = producerBehavior.orElse[Any, Unit](consumerBehavior)
}
// protocol
case object GiveMeThings
final case class Give(thing: Any)
Instead of inheritance the same pattern can be applied via composition - one would simply compose the receive
method using partial functions from delegates.
4.1.15 Initialization patterns
The rich lifecycle hooks of Actors provide a useful toolkit to implement various initialization patterns. During the
lifetime of an ActorRef, an actor can potentially go through several restarts, where the old instance is replaced
by a fresh one, invisibly to the outside observer who only sees the ActorRef.
One may think about the new instances as “incarnations”. Initialization might be necessary for every incarnation
of an actor, but sometimes one needs initialization to happen only at the birth of the first instance when the
ActorRef is created. The following sections provide patterns for different initialization needs.
Initialization via constructor
Using the constructor for initialization has various benefits. First of all, it makes it possible to use val fields to
store any state that does not change during the life of the actor instance, making the implementation of the actor
more robust. The constructor is invoked for every incarnation of the actor, therefore the internals of the actor can
always assume that proper initialization happened. This is also the drawback of this approach, as there are cases
when one would like to avoid reinitializing internals on restart. For example, it is often useful to preserve child
actors across restarts. The following section provides a pattern for this case.
4.1. Actors
130
Akka Scala Documentation, Release 2.5.0-RC1
Initialization via preStart
The method preStart() of an actor is only called once directly during the initialization of the first instance, that
is, at creation of its ActorRef. In the case of restarts, preStart() is called from postRestart(), therefore
if not overridden, preStart() is called on every incarnation. However, by overriding postRestart() one
can disable this behavior, and ensure that there is only one call to preStart().
One useful usage of this pattern is to disable creation of new ActorRefs for children during restarts. This can
be achieved by overriding preRestart():
override def preStart(): Unit = {
// Initialize children here
}
// Overriding postRestart to disable the call to preStart()
// after restarts
override def postRestart(reason: Throwable): Unit = ()
// The default implementation of preRestart() stops all the children
// of the actor. To opt-out from stopping the children, we
// have to override preRestart()
override def preRestart(reason: Throwable, message: Option[Any]): Unit = {
// Keep the call to postStop(), but no stopping of children
postStop()
}
Please note, that the child actors are still restarted, but no new ActorRef is created. One can recursively apply
the same principles for the children, ensuring that their preStart() method is called only at the creation of
their refs.
For more information see What Restarting Means.
Initialization via message passing
There are cases when it is impossible to pass all the information needed for actor initialization in the constructor,
for example in the presence of circular dependencies. In this case the actor should listen for an initialization
message, and use become() or a finite state-machine state transition to encode the initialized and uninitialized
states of the actor.
var initializeMe: Option[String] = None
override def receive = {
case "init" =>
initializeMe = Some("Up and running")
context.become(initialized, discardOld = true)
}
def initialized: Receive = {
case "U OK?" => initializeMe foreach { sender() ! _ }
}
If the actor may receive messages before it has been initialized, a useful tool can be the Stash to save messages
until the initialization finishes, and replaying them after the actor became initialized.
Warning: This pattern should be used with care, and applied only when none of the patterns above are
applicable. One of the potential issues is that messages might be lost when sent to remote actors. Also,
publishing an ActorRef in an uninitialized state might lead to the condition that it receives a user message
before the initialization has been done.
4.1. Actors
131
Akka Scala Documentation, Release 2.5.0-RC1
4.2 Akka Typed
Warning: This module is currently marked as may change in the sense of being the subject of active research.
This means that API or semantics can change without warning or deprecation period and it is not recommended
to use this module in production just yet—you have been warned.
As discussed in Actor Systems (and following chapters) Actors are about sending messages between independent
units of computation, but how does that look like? In all of the following these imports are assumed:
import
import
import
import
import
import
akka.typed._
akka.typed.scaladsl.Actor._
akka.typed.scaladsl.AskPattern._
scala.concurrent.Future
scala.concurrent.duration._
scala.concurrent.Await
With these in place we can define our first Actor, and of course it will say hello!
object HelloWorld {
final case class Greet(whom: String, replyTo: ActorRef[Greeted])
final case class Greeted(whom: String)
val greeter = Stateless[Greet] { (_, msg) ⇒
println(s"Hello ${msg.whom}!")
msg.replyTo ! Greeted(msg.whom)
}
}
This small piece of code defines two message types, one for commanding the Actor to greet someone and one that
the Actor will use to confirm that it has done so. The Greet type contains not only the information of whom to
greet, it also holds an ActorRef that the sender of the message supplies so that the HelloWorld Actor can
send back the confirmation message.
The behavior of the Actor is defined as the greeter value with the help of the Stateless behavior constructor—there are many different ways of formulating behaviors as we shall see in the following. The “stateless”
behavior is not capable of changing in response to a message, it will stay the same until the Actor is stopped by its
parent.
The type of the messages handled by this behavior is declared to be of class Greet, which implies that the
supplied function’s msg argument is also typed as such. This is why we can access the whom and replyTo
members without needing to use a pattern match.
On the last line we see the HelloWorld Actor send a message to another Actor, which is done using the !
operator (pronounced “tell”). Since the replyTo address is declared to be of type ActorRef[Greeted] the
compiler will only permit us to send messages of this type, other usage will not be accepted.
The accepted message types of an Actor together with all reply types defines the protocol spoken by this Actor; in this case it is a simple request–reply protocol but Actors can model arbitrarily complex protocols when
needed. The protocol is bundled together with the behavior that implements it in a nicely wrapped scope—the
HelloWorld object.
Now we want to try out this Actor, so we must start an ActorSystem to host it:
import HelloWorld._
// using global pool since we want to run tasks after system.terminate
import scala.concurrent.ExecutionContext.Implicits.global
val system: ActorSystem[Greet] = ActorSystem("hello", greeter)
val future: Future[Greeted] = system ? (Greet("world", _))
4.2. Akka Typed
132
Akka Scala Documentation, Release 2.5.0-RC1
for {
greeting ← future.recover { case ex ⇒ ex.getMessage }
done ← { println(s"result: $greeting"); system.terminate() }
} println("system terminated")
After importing the Actor’s protocol definition we start an Actor system from the defined behavior.
As Carl Hewitt said, one Actor is no Actor—it would be quite lonely with nobody to talk to. In this sense the
example is a little cruel because we only give the HelloWorld Actor a fake person to talk to—the “ask” pattern
(represented by the ? operator) can be used to send a message such that the reply fulfills a Promise to which we
get back the corresponding Future.
Note that the Future that is returned by the “ask” operation is properly typed already, no type checks or casts
needed. This is possible due to the type information that is part of the message protocol: the ? operator takes as
argument a function that accepts an ActorRef[U] (which explains the _ hole in the expression on line 7 above)
and the replyTo parameter which we fill in is of type ActorRef[Greeted], which means that the value that
fulfills the Promise can only be of type Greeted.
We use this here to send the Greet command to the Actor and when the reply comes back we will print it
out and tell the actor system to shut down. Once that is done as well we print the "system terminated"
messages and the program ends. The recovery combinator on the original Future is needed in order to ensure
proper system shutdown even in case something went wrong; the flatMap and map combinators that the for
expression gets turned into care only about the “happy path” and if the future failed with a timeout then no
greeting would be extracted and nothing would happen.
This shows that there are aspects of Actor messaging that can be type-checked by the compiler, but this ability
is not unlimited, there are bounds to what we can statically express. Before we go on with a more complex (and
realistic) example we make a small detour to highlight some of the theory behind this.
4.2.1 A Little Bit of Theory
The Actor Model as defined by Hewitt, Bishop and Steiger in 1973 is a computational model that expresses
exactly what it means for computation to be distributed. The processing units—Actors—can only communicate
by exchanging messages and upon reception of a message an Actor can do the following three fundamental actions:
1. send a finite number of messages to Actors it knows
2. create a finite number of new Actors
3. designate the behavior to be applied to the next message
The Akka Typed project expresses these actions using behaviors and addresses. Messages can be sent to an address
and behind this façade there is a behavior that receives the message and acts upon it. The binding between address
and behavior can change over time as per the third point above, but that is not visible on the outside.
With this preamble we can get to the unique property of this project, namely that it introduces static type checking
to Actor interactions: addresses are parameterized and only messages that are of the specified type can be sent to
them. The association between an address and its type parameter must be made when the address (and its Actor) is
created. For this purpose each behavior is also parameterized with the type of messages it is able to process. Since
the behavior can change behind the address façade, designating the next behavior is a constrained operation: the
successor must handle the same type of messages as its predecessor. This is necessary in order to not invalidate
the addresses that refer to this Actor.
What this enables is that whenever a message is sent to an Actor we can statically ensure that the type of the
message is one that the Actor declares to handle—we can avoid the mistake of sending completely pointless
messages. What we cannot statically ensure, though, is that the behavior behind the address will be in a given
state when our message is received. The fundamental reason is that the association between address and behavior
is a dynamic runtime property, the compiler cannot know it while it translates the source code.
This is the same as for normal Java objects with internal variables: when compiling the program we cannot know
what their value will be, and if the result of a method call depends on those variables then the outcome is uncertain
to a degree—we can only be certain that the returned value is of a given type.
4.2. Akka Typed
133
Akka Scala Documentation, Release 2.5.0-RC1
We have seen above that the return type of an Actor command is described by the type of reply-to address that
is contained within the message. This allows a conversation to be described in terms of its types: the reply will
be of type A, but it might also contain an address of type B, which then allows the other Actor to continue the
conversation by sending a message of type B to this new address. While we cannot statically express the “current”
state of an Actor, we can express the current state of a protocol between two Actors, since that is just given by the
last message type that was received or sent.
In the next section we demonstrate this on a more realistic example.
4.2.2 A More Complex Example
Consider an Actor that runs a chat room: client Actors may connect by sending a message that contains their
screen name and then they can post messages. The chat room Actor will disseminate all posted messages to all
currently connected client Actors. The protocol definition could look like the following:
sealed trait Command
final case class GetSession(screenName: String, replyTo: ActorRef[SessionEvent])
extends Command
sealed trait SessionEvent
final case class SessionGranted(handle: ActorRef[PostMessage]) extends SessionEvent
final case class SessionDenied(reason: String) extends SessionEvent
final case class MessagePosted(screenName: String, message: String) extends
˓→SessionEvent
final case class PostMessage(message: String)
Initially the client Actors only get access to an ActorRef[GetSession] which allows them to make the first
step. Once a client’s session has been established it gets a SessionGranted message that contains a handle
to unlock the next protocol step, posting messages. The PostMessage command will need to be sent to this
particular address that represents the session that has been added to the chat room. The other aspect of a session
is that the client has revealed its own address, via the replyTo argument, so that subsequent MessagePosted
events can be sent to it.
This illustrates how Actors can express more than just the equivalent of method calls on Java objects. The declared
message types and their contents describe a full protocol that can involve multiple Actors and that can evolve over
multiple steps. The implementation of the chat room protocol would be as simple as the following:
private final case class PostSessionMessage(screenName: String, message: String)
extends Command
def chatRoom(sessions: List[ActorRef[SessionEvent]] = List.empty):
˓→Behavior[Command] =
Stateful[Command] { (ctx, msg) ⇒
msg match {
case GetSession(screenName, client) ⇒
val wrapper = ctx.spawnAdapter {
p: PostMessage ⇒ PostSessionMessage(screenName, p.message)
}
client ! SessionGranted(wrapper)
chatRoom(client :: sessions)
case PostSessionMessage(screenName, message) ⇒
val mp = MessagePosted(screenName, message)
sessions foreach (_ ! mp)
Same
}
}
The core of this behavior is stateful, the chat room itself does not change into something else when sessions are
established, but we introduce a variable that tracks the opened sessions. Note that by using a method parameter
a var is not needed. When a new GetSession command comes in we add that client to the list that is in
4.2. Akka Typed
134
Akka Scala Documentation, Release 2.5.0-RC1
the returned behavior. Then we also need to create the session’s ActorRef that will be used to post messages.
In this case we want to create a very simple Actor that just repackages the PostMessage command into a
PostSessionMessage command which also includes the screen name. Such a wrapper Actor can be created
by using the spawnAdapter method on the ActorContext, so that we can then go on to reply to the client
with the SessionGranted result.
The behavior that we declare here can handle both subtypes of Command. GetSession has been explained
already and the PostSessionMessage commands coming from the wrapper Actors will trigger the dissemination of the contained chat room message to all connected clients. But we do not want to give the ability
to send PostSessionMessage commands to arbitrary clients, we reserve that right to the wrappers we create—otherwise clients could pose as completely different screen names (imagine the GetSession protocol to
include authentication information to further secure this). Therefore PostSessionMessage has private
visibility and can’t be created outside the actor.
If we did not care about securing the correspondence between a session and a screen name then
we could change the protocol such that PostMessage is removed and all clients just get an
ActorRef[PostSessionMessage] to send to.
In this case no wrapper would be needed and
we could just use ctx.self.
The type-checks work out in that case because ActorRef[-T]
is contravariant in its type parameter, meaning that we can use a ActorRef[Command] wherever an ActorRef[PostSessionMessage] is needed—this makes sense because the former simply speaks more languages than the latter.
The opposite would be problematic, so passing an
ActorRef[PostSessionMessage] where ActorRef[Command] is required will lead to a type error.
Trying it out
In order to see this chat room in action we need to write a client Actor that can use it:
import ChatRoom._
val gabbler =
Stateful[SessionEvent] { (_, msg) ⇒
msg match {
case SessionDenied(reason) ⇒
println(s"cannot start chat room session: $reason")
Stopped
case SessionGranted(handle) ⇒
handle ! PostMessage("Hello World!")
Same
case MessagePosted(screenName, message) ⇒
println(s"message has been posted by '$screenName': $message")
Stopped
}
}
From this behavior we can create an Actor that will accept a chat room session, post a message, wait to see
it published, and then terminate. The last step requires the ability to change behavior, we need to transition
from the normal running behavior into the terminated state. This is why this Actor uses a different behavior
constructor named Total. This constructor takes as argument a function from the handled message type, in this
case SessionEvent, to the next behavior. That next behavior must again be of the same type as we discussed
in the theory section above. Here we either stay in the very same behavior or we terminate, and both of these
cases are so common that there are special values Same and Stopped that can be used. The behavior is named
“total” (as opposed to “partial”) because the declared function must handle all values of its input type. Since
SessionEvent is a sealed trait the Scala compiler will warn us if we forget to handle one of the subtypes; in
this case it reminded us that alternatively to SessionGranted we may also receive a SessionDenied event.
Now to try things out we must start both a chat room and a gabbler and of course we do this inside an Actor
system. Since there can be only one guardian supervisor we could either start the chat room from the gabbler
(which we don’t want—it complicates its logic) or the gabbler from the chat room (which is nonsensical) or we
start both of them from a third Actor—our only sensible choice:
4.2. Akka Typed
135
Akka Scala Documentation, Release 2.5.0-RC1
val main: Behavior[akka.NotUsed] =
Stateful(
behavior = (_, _) ⇒ Unhandled,
signal = { (ctx, sig) ⇒
sig match {
case PreStart ⇒
val chatRoom = ctx.spawn(ChatRoom.chatRoom(), "chatroom")
val gabblerRef = ctx.spawn(gabbler, "gabbler")
ctx.watch(gabblerRef)
chatRoom ! GetSession("ol’ Gabbler", gabblerRef)
Same
case Terminated(ref) ⇒
Stopped
case _ ⇒
Unhandled
}
})
val system = ActorSystem("ChatRoomDemo", main)
Await.result(system.whenTerminated, 1.second)
In good tradition we call the main Actor what it is, it directly corresponds to the main method in a traditional Java
application. This Actor will perform its job on its own accord, we do not need to send messages from the outside,
so we declare it to be of type NotUsed. Actors receive not only external messages, they also are notified of
certain system events, so-called Signals. In order to get access to those we choose to implement this particular one
using the Stateful behavior decorator. The provided signal function will be invoked for signals (subclasses
of Signal) or the mesg function for user messages.
This particular main Actor reacts to two signals: when it is started it will first receive the PreStart signal, upon
which the chat room and the gabbler are created and the session between them is initiated, and when the gabbler is
finished we will receive the Terminated event due to having called ctx.watch for it. This allows us to shut
down the Actor system: when the main Actor terminates there is nothing more to do.
Therefore after creating the Actor system with the main Actor’s Props we just await its termination.
4.2.3 Status of this Project and Relation to Akka Actors
Akka Typed is the result of many years of research and previous attempts (including Typed Channels in the 2.2.x
series) and it is on its way to stabilization, but maturing such a profound change to the core concept of Akka will
take a long time. We expect that this module will stay marked may change for multiple major releases of Akka
and the plain akka.actor.Actor will not be deprecated or go away anytime soon.
Being a research project also entails that the reference documentation is not as detailed as it will be for a final
version, please refer to the API documentation for greater depth and finer detail.
Main Differences
The most prominent difference is the removal of the sender() functionality. This turned out to be the Achilles
heel of the Typed Channels project, it is the feature that makes its type signatures and macros too complex to
be viable. The solution chosen in Akka Typed is to explicitly include the properly typed reply-to address in the
message, which both burdens the user with this task but also places this aspect of protocol design where it belongs.
The other prominent difference is the removal of the Actor trait. In order to avoid closing over unstable references
from different execution contexts (e.g. Future transformations) we turned all remaining methods that were on this
trait into messages: the behavior receives the ActorContext as an argument during processing and the lifecycle
hooks have been converted into Signals.
A side-effect of this is that behaviors can now be tested in isolation without having to be packaged into an Actor,
tests can run fully synchronously without having to worry about timeouts and spurious failures. Another sideeffect is that behaviors can nicely be composed and decorated, see the And, Or, Widened, ContextAware
4.2. Akka Typed
136
Akka Scala Documentation, Release 2.5.0-RC1
combinators; nothing about these is special or internal, new combinators can be written as external libraries or
tailor-made for each project.
4.3 Fault Tolerance
As explained in Actor Systems each actor is the supervisor of its children, and as such each actor defines fault
handling supervisor strategy. This strategy cannot be changed afterwards as it is an integral part of the actor
system’s structure.
4.3.1 Fault Handling in Practice
First, let us look at a sample that illustrates one way to handle data store errors, which is a typical source of failure
in real world applications. Of course it depends on the actual application what is possible to do when the data
store is unavailable, but in this sample we use a best effort re-connect approach.
Read the following source code. The inlined comments explain the different pieces of the fault handling and why
they are added. It is also highly recommended to run this sample as it is easy to follow the log output to understand
what is happening at runtime.
4.3. Fault Tolerance
137
Akka Scala Documentation, Release 2.5.0-RC1
Diagrams of the Fault Tolerance Sample
The above diagram illustrates the normal message flow.
Normal flow:
Step
1
2
3,
4, 5
6, 7
Description
The progress Listener starts the work.
The Worker schedules work by sending Do messages periodically to itself
When receiving Do the Worker tells the CounterService to increment the counter, three times.
The Increment message is forwarded to the Counter, which updates its counter variable and sends
current value to the Storage.
The Worker asks the CounterService of current value of the counter and pipes the result back to
the Listener.
4.3. Fault Tolerance
138
Akka Scala Documentation, Release 2.5.0-RC1
The above diagram illustrates what happens in case of storage failure.
Failure flow:
4.3. Fault Tolerance
139
Akka Scala Documentation, Release 2.5.0-RC1
Step
1
2
3, 4,
5, 6
7
8
9, 10,
11
12
13,
14
15,
16
Description
The Storage throws StorageException.
The CounterService is supervisor of the Storage and restarts the Storage when
StorageException is thrown.
The Storage continues to fail and is restarted.
After 3 failures and restarts within 5 seconds the Storage is stopped by its supervisor, i.e. the
CounterService.
The CounterService is also watching the Storage for termination and receives the
Terminated message when the Storage has been stopped ...
and tells the Counter that there is no Storage.
The CounterService schedules a Reconnect message to itself.
When it receives the Reconnect message it creates a new Storage ...
and tells the Counter to use the new Storage
Full Source Code of the Fault Tolerance Sample
import
import
import
import
import
import
import
akka.actor._
akka.actor.SupervisorStrategy._
scala.concurrent.duration._
akka.util.Timeout
akka.event.LoggingReceive
akka.pattern.{ ask, pipe }
com.typesafe.config.ConfigFactory
/**
* Runs the sample
*/
object FaultHandlingDocSample extends App {
import Worker._
val config = ConfigFactory.parseString("""
akka.loglevel = "DEBUG"
akka.actor.debug {
receive = on
lifecycle = on
}
""")
val system = ActorSystem("FaultToleranceSample", config)
val worker = system.actorOf(Props[Worker], name = "worker")
val listener = system.actorOf(Props[Listener], name = "listener")
// start the work and listen on progress
// note that the listener is used as sender of the tell,
// i.e. it will receive replies from the worker
worker.tell(Start, sender = listener)
}
/**
* Listens on progress from the worker and shuts down the system when enough
* work has been done.
*/
class Listener extends Actor with ActorLogging {
import Worker._
// If we don't get any progress within 15 seconds then the service is unavailable
context.setReceiveTimeout(15 seconds)
4.3. Fault Tolerance
140
Akka Scala Documentation, Release 2.5.0-RC1
def receive = {
case Progress(percent) =>
log.info("Current progress: {} %", percent)
if (percent >= 100.0) {
log.info("That's all, shutting down")
context.system.terminate()
}
case ReceiveTimeout =>
// No progress within 15 seconds, ServiceUnavailable
log.error("Shutting down due to unavailable service")
context.system.terminate()
}
}
object Worker {
case object Start
case object Do
final case class Progress(percent: Double)
}
/**
* Worker performs some work when it receives the `Start` message.
* It will continuously notify the sender of the `Start` message
* of current ``Progress``. The `Worker` supervise the `CounterService`.
*/
class Worker extends Actor with ActorLogging {
import Worker._
import CounterService._
implicit val askTimeout = Timeout(5 seconds)
// Stop the CounterService child if it throws ServiceUnavailable
override val supervisorStrategy = OneForOneStrategy() {
case _: CounterService.ServiceUnavailable => Stop
}
// The sender of the initial Start message will continuously be notified
// about progress
var progressListener: Option[ActorRef] = None
val counterService = context.actorOf(Props[CounterService], name = "counter")
val totalCount = 51
import context.dispatcher // Use this Actors' Dispatcher as ExecutionContext
def receive = LoggingReceive {
case Start if progressListener.isEmpty =>
progressListener = Some(sender())
context.system.scheduler.schedule(Duration.Zero, 1 second, self, Do)
case Do =>
counterService ! Increment(1)
counterService ! Increment(1)
counterService ! Increment(1)
// Send current progress to the initial sender
counterService ? GetCurrentCount map {
case CurrentCount(_, count) => Progress(100.0 * count / totalCount)
} pipeTo progressListener.get
}
}
object CounterService {
final case class Increment(n: Int)
sealed abstract class GetCurrentCount
4.3. Fault Tolerance
141
Akka Scala Documentation, Release 2.5.0-RC1
case object GetCurrentCount extends GetCurrentCount
final case class CurrentCount(key: String, count: Long)
class ServiceUnavailable(msg: String) extends RuntimeException(msg)
private case object Reconnect
}
/**
* Adds the value received in `Increment` message to a persistent
* counter. Replies with `CurrentCount` when it is asked for `CurrentCount`.
* `CounterService` supervise `Storage` and `Counter`.
*/
class CounterService extends Actor {
import CounterService._
import Counter._
import Storage._
// Restart the storage child when StorageException is thrown.
// After 3 restarts within 5 seconds it will be stopped.
override val supervisorStrategy = OneForOneStrategy(
maxNrOfRetries = 3,
withinTimeRange = 5 seconds) {
case _: Storage.StorageException => Restart
}
val
var
var
var
val
key = self.path.name
storage: Option[ActorRef] = None
counter: Option[ActorRef] = None
backlog = IndexedSeq.empty[(ActorRef, Any)]
MaxBacklog = 10000
import context.dispatcher // Use this Actors' Dispatcher as ExecutionContext
override def preStart() {
initStorage()
}
/**
* The child storage is restarted in case of failure, but after 3 restarts,
* and still failing it will be stopped. Better to back-off than continuously
* failing. When it has been stopped we will schedule a Reconnect after a delay.
* Watch the child so we receive Terminated message when it has been terminated.
*/
def initStorage() {
storage = Some(context.watch(context.actorOf(Props[Storage], name = "storage
˓→")))
// Tell the counter, if any, to use the new storage
counter foreach { _ ! UseStorage(storage) }
// We need the initial value to be able to operate
storage.get ! Get(key)
}
def receive = LoggingReceive {
case Entry(k, v) if k == key && counter == None =>
// Reply from Storage of the initial value, now we can create the Counter
val c = context.actorOf(Props(classOf[Counter], key, v))
counter = Some(c)
// Tell the counter to use current storage
c ! UseStorage(storage)
// and send the buffered backlog to the counter
for ((replyTo, msg) <- backlog) c.tell(msg, sender = replyTo)
backlog = IndexedSeq.empty
4.3. Fault Tolerance
142
Akka Scala Documentation, Release 2.5.0-RC1
case msg: Increment
=> forwardOrPlaceInBacklog(msg)
case msg: GetCurrentCount => forwardOrPlaceInBacklog(msg)
case Terminated(actorRef) if Some(actorRef) == storage =>
// After 3 restarts the storage child is stopped.
// We receive Terminated because we watch the child, see initStorage.
storage = None
// Tell the counter that there is no storage for the moment
counter foreach { _ ! UseStorage(None) }
// Try to re-establish storage after while
context.system.scheduler.scheduleOnce(10 seconds, self, Reconnect)
case Reconnect =>
// Re-establish storage after the scheduled delay
initStorage()
}
def forwardOrPlaceInBacklog(msg: Any) {
// We need the initial value from storage before we can start delegate to
// the counter. Before that we place the messages in a backlog, to be sent
// to the counter when it is initialized.
counter match {
case Some(c) => c forward msg
case None =>
if (backlog.size >= MaxBacklog)
throw new ServiceUnavailable(
"CounterService not available, lack of initial value")
backlog :+= (sender() -> msg)
}
}
}
object Counter {
final case class UseStorage(storage: Option[ActorRef])
}
/**
* The in memory count variable that will send current
* value to the `Storage`, if there is any storage
* available at the moment.
*/
class Counter(key: String, initialValue: Long) extends Actor {
import Counter._
import CounterService._
import Storage._
var count = initialValue
var storage: Option[ActorRef] = None
def receive = LoggingReceive {
case UseStorage(s) =>
storage = s
storeCount()
case Increment(n) =>
count += n
storeCount()
case GetCurrentCount =>
sender() ! CurrentCount(key, count)
4.3. Fault Tolerance
143
Akka Scala Documentation, Release 2.5.0-RC1
}
def storeCount() {
// Delegate dangerous work, to protect our valuable state.
// We can continue without storage.
storage foreach { _ ! Store(Entry(key, count)) }
}
}
object Storage {
final case class Store(entry: Entry)
final case class Get(key: String)
final case class Entry(key: String, value: Long)
class StorageException(msg: String) extends RuntimeException(msg)
}
/**
* Saves key/value pairs to persistent storage when receiving `Store` message.
* Replies with current value when receiving `Get` message.
* Will throw StorageException if the underlying data store is out of order.
*/
class Storage extends Actor {
import Storage._
val db = DummyDB
def receive = LoggingReceive {
case Store(Entry(key, count)) => db.save(key, count)
case Get(key)
=> sender() ! Entry(key, db.load(key).
˓→getOrElse(0L))
}
}
object DummyDB {
import Storage.StorageException
private var db = Map[String, Long]()
@throws(classOf[StorageException])
def save(key: String, value: Long): Unit = synchronized {
if (11 <= value && value <= 14)
throw new StorageException("Simulated store failure " + value)
db += (key -> value)
}
@throws(classOf[StorageException])
def load(key: String): Option[Long] = synchronized {
db.get(key)
}
}
4.3.2 Creating a Supervisor Strategy
The following sections explain the fault handling mechanism and alternatives in more depth.
For the sake of demonstration let us consider the following strategy:
import akka.actor.OneForOneStrategy
import akka.actor.SupervisorStrategy._
import scala.concurrent.duration._
4.3. Fault Tolerance
144
Akka Scala Documentation, Release 2.5.0-RC1
override val supervisorStrategy =
OneForOneStrategy(maxNrOfRetries =
case _: ArithmeticException
case _: NullPointerException
case _: IllegalArgumentException
case _: Exception
}
10, withinTimeRange = 1 minute) {
=> Resume
=> Restart
=> Stop
=> Escalate
I have chosen a few well-known exception types in order to demonstrate the application of the fault handling
directives described in Supervision and Monitoring. First off, it is a one-for-one strategy, meaning that each child
is treated separately (an all-for-one strategy works very similarly, the only difference is that any decision is applied
to all children of the supervisor, not only the failing one). There are limits set on the restart frequency, namely
maximum 10 restarts per minute; each of these settings could be left out, which means that the respective limit
does not apply, leaving the possibility to specify an absolute upper limit on the restarts or to make the restarts
work infinitely. The child actor is stopped if the limit is exceeded.
The match statement which forms the bulk of the body is of type Decider, which is a
PartialFunction[Throwable, Directive]. This is the piece which maps child failure types to their
corresponding directives.
Note: If the strategy is declared inside the supervising actor (as opposed to within a companion object) its decider
has access to all internal state of the actor in a thread-safe fashion, including obtaining a reference to the currently
failed child (available as the sender of the failure message).
Default Supervisor Strategy
Escalate is used if the defined strategy doesn’t cover the exception that was thrown.
When the supervisor strategy is not defined for an actor the following exceptions are handled by default:
• ActorInitializationException will stop the failing child actor
• ActorKilledException will stop the failing child actor
• DeathPactException will stop the failing child actor
• Exception will restart the failing child actor
• Other types of Throwable will be escalated to parent actor
If the exception escalate all the way up to the root guardian it will handle it in the same way as the default strategy
defined above.
You can combine your own strategy with the default strategy:
import akka.actor.OneForOneStrategy
import akka.actor.SupervisorStrategy._
import scala.concurrent.duration._
override val supervisorStrategy =
OneForOneStrategy(maxNrOfRetries = 10, withinTimeRange = 1 minute) {
case _: ArithmeticException => Resume
case t =>
super.supervisorStrategy.decider.applyOrElse(t, (_: Any) => Escalate)
}
4.3. Fault Tolerance
145
Akka Scala Documentation, Release 2.5.0-RC1
Stopping Supervisor Strategy
Closer to the Erlang way is the strategy to just stop children when they fail and then take corrective action in the supervisor when DeathWatch signals the loss of the child.
This strategy is
also provided pre-packaged as SupervisorStrategy.stoppingStrategy with an accompanying
StoppingSupervisorStrategy configurator to be used when you want the "/user" guardian to apply it.
Logging of Actor Failures
By default the SupervisorStrategy logs failures unless they are escalated. Escalated failures are supposed
to be handled, and potentially logged, at a level higher in the hierarchy.
You can mute the default logging of a SupervisorStrategy by setting loggingEnabled to false when
instantiating it. Customized logging can be done inside the Decider. Note that the reference to the currently
failed child is available as the sender when the SupervisorStrategy is declared inside the supervising
actor.
You may also customize the logging in your own SupervisorStrategy implementation by overriding the
logFailure method.
4.3.3 Supervision of Top-Level Actors
Toplevel actors means those which are created using system.actorOf(), and they are children of the User
Guardian. There are no special rules applied in this case, the guardian simply applies the configured strategy.
4.3.4 Test Application
The following section shows the effects of the different directives in practice, where a test setup is needed. First
off, we need a suitable supervisor:
import akka.actor.Actor
class Supervisor extends Actor {
import akka.actor.OneForOneStrategy
import akka.actor.SupervisorStrategy._
import scala.concurrent.duration._
override val supervisorStrategy =
OneForOneStrategy(maxNrOfRetries =
case _: ArithmeticException
case _: NullPointerException
case _: IllegalArgumentException
case _: Exception
}
10, withinTimeRange = 1 minute) {
=> Resume
=> Restart
=> Stop
=> Escalate
def receive = {
case p: Props => sender() ! context.actorOf(p)
}
}
This supervisor will be used to create a child, with which we can experiment:
import akka.actor.Actor
class Child extends Actor {
var state = 0
def receive = {
case ex: Exception => throw ex
case x: Int
=> state = x
4.3. Fault Tolerance
146
Akka Scala Documentation, Release 2.5.0-RC1
case "get"
=> sender() ! state
}
}
The test is easier by using the utilities described in Testing Actor Systems.
import com.typesafe.config.{ Config, ConfigFactory }
import org.scalatest.{ FlatSpecLike, Matchers, BeforeAndAfterAll }
import akka.testkit.{ TestActors, TestKit, ImplicitSender, EventFilter }
class FaultHandlingDocSpec(_system: ActorSystem) extends TestKit(_system)
with ImplicitSender with FlatSpecLike with Matchers with BeforeAndAfterAll {
def this() = this(ActorSystem(
"FaultHandlingDocSpec",
ConfigFactory.parseString("""
akka {
loggers = ["akka.testkit.TestEventListener"]
loglevel = "WARNING"
}
""")))
override def afterAll {
TestKit.shutdownActorSystem(system)
}
"A supervisor" must "apply the chosen strategy for its child" in {
// code here
}
}
Let us create actors:
val supervisor = system.actorOf(Props[Supervisor], "supervisor")
supervisor ! Props[Child]
val child = expectMsgType[ActorRef] // retrieve answer from TestKit’s testActor
The first test shall demonstrate the Resume directive, so we try it out by setting some non-initial state in the actor
and have it fail:
child ! 42 // set state to 42
child ! "get"
expectMsg(42)
child ! new ArithmeticException // crash it
child ! "get"
expectMsg(42)
As you can see the value 42 survives the fault handling directive. Now, if we change the failure to a more serious
NullPointerException, that will no longer be the case:
child ! new NullPointerException // crash it harder
child ! "get"
expectMsg(0)
And finally in case of the fatal IllegalArgumentException the child will be terminated by the supervisor:
watch(child) // have testActor watch “child”
child ! new IllegalArgumentException // break it
expectMsgPF() { case Terminated(`child`) => () }
4.3. Fault Tolerance
147
Akka Scala Documentation, Release 2.5.0-RC1
Up to now the supervisor was completely unaffected by the child’s failure, because the directives set did handle it.
In case of an Exception, this is not true anymore and the supervisor escalates the failure.
supervisor ! Props[Child] // create new child
val child2 = expectMsgType[ActorRef]
watch(child2)
child2 ! "get" // verify it is alive
expectMsg(0)
child2 ! new Exception("CRASH") // escalate failure
expectMsgPF() {
case t @ Terminated(`child2`) if t.existenceConfirmed => ()
}
The supervisor itself is supervised by the top-level actor provided by the ActorSystem, which
has the default policy to restart in case of all Exception cases (with the notable exceptions of
ActorInitializationException and ActorKilledException). Since the default directive in case
of a restart is to kill all children, we expected our poor child not to survive this failure.
In case this is not desired (which depends on the use case), we need to use a different supervisor which overrides
this behavior.
class Supervisor2 extends Actor {
import akka.actor.OneForOneStrategy
import akka.actor.SupervisorStrategy._
import scala.concurrent.duration._
override val supervisorStrategy =
OneForOneStrategy(maxNrOfRetries =
case _: ArithmeticException
case _: NullPointerException
case _: IllegalArgumentException
case _: Exception
}
10, withinTimeRange = 1 minute) {
=> Resume
=> Restart
=> Stop
=> Escalate
def receive = {
case p: Props => sender() ! context.actorOf(p)
}
// override default to kill all children during restart
override def preRestart(cause: Throwable, msg: Option[Any]) {}
}
With this parent, the child survives the escalated restart, as demonstrated in the last test:
val supervisor2 = system.actorOf(Props[Supervisor2], "supervisor2")
supervisor2 ! Props[Child]
val child3 = expectMsgType[ActorRef]
child3 ! 23
child3 ! "get"
expectMsg(23)
child3 ! new Exception("CRASH")
child3 ! "get"
expectMsg(0)
4.4 Dispatchers
An Akka MessageDispatcher is what makes Akka Actors “tick”, it is the engine of the machine so to speak.
All MessageDispatcher implementations are also an ExecutionContext, which means that they can be
4.4. Dispatchers
148
Akka Scala Documentation, Release 2.5.0-RC1
used to execute arbitrary code, for instance Futures.
4.4.1 Default dispatcher
Every ActorSystem will have a default dispatcher that will be used in case nothing else is configured for
an Actor. The default dispatcher can be configured, and is by default a Dispatcher with the specified
default-executor. If an ActorSystem is created with an ExecutionContext passed in, this ExecutionContext
will be used as the default executor for all dispatchers in this ActorSystem. If no ExecutionContext is given,
it will fallback to the executor specified in akka.actor.default-dispatcher.default-executor.
fallback. By default this is a “fork-join-executor”, which gives excellent performance in most cases.
4.4.2 Looking up a Dispatcher
Dispatchers implement the ExecutionContext interface and can thus be used to run Future invocations etc.
// for use with Futures, Scheduler, etc.
implicit val executionContext = system.dispatchers.lookup("my-dispatcher")
4.4.3 Setting the dispatcher for an Actor
So in case you want to give your Actor a different dispatcher than the default, you need to do two things, of
which the first is to configure the dispatcher:
my-dispatcher {
# Dispatcher is the name of the event-based dispatcher
type = Dispatcher
# What kind of ExecutionService to use
executor = "fork-join-executor"
# Configuration for the fork join pool
fork-join-executor {
# Min number of threads to cap factor-based parallelism
parallelism-min = 2
# Parallelism (threads) ... ceil(available processors *
parallelism-factor = 2.0
# Max number of threads to cap factor-based parallelism
parallelism-max = 10
}
# Throughput defines the maximum number of messages to be
# processed per actor before the thread jumps to the next
# Set to 1 for as fair as possible.
throughput = 100
}
number to
factor)
number to
actor.
Note: Note that the parallelism-max does not set the upper bound on the total number of threads allocated
by the ForkJoinPool. It is a setting specifically talking about the number of hot threads the pool keep running in
order to reduce the latency of handling a new incoming task. You can read more about parallelism in the JDK’s
ForkJoinPool documentation.
Another example that uses the “thread-pool-executor”:
blocking-io-dispatcher {
type = Dispatcher
executor = "thread-pool-executor"
thread-pool-executor {
fixed-pool-size = 32
}
4.4. Dispatchers
149
Akka Scala Documentation, Release 2.5.0-RC1
throughput = 1
}
Note:
The thread pool executor dispatcher is implemented using by a java.util.concurrent.
ThreadPoolExecutor. You can read more about it in the JDK’s ThreadPoolExecutor documentation.
For more options, see the default-dispatcher section of the Configuration.
Then you create the actor as usual and define the dispatcher in the deployment configuration.
import akka.actor.Props
val myActor = context.actorOf(Props[MyActor], "myactor")
akka.actor.deployment {
/myactor {
dispatcher = my-dispatcher
}
}
An alternative to the deployment configuration is to define the dispatcher in code. If you define the dispatcher
in the deployment configuration then this value will be used instead of programmatically provided parameter.
import akka.actor.Props
val myActor =
context.actorOf(Props[MyActor].withDispatcher("my-dispatcher"), "myactor1")
Note: The dispatcher you specify in withDispatcher and the dispatcher property in the deployment
configuration is in fact a path into your configuration. So in this example it’s a top-level section, but you could
for instance put it as a sub-section, where you’d use periods to denote sub-sections, like this: "foo.bar.
my-dispatcher"
4.4.4 Types of dispatchers
There are 3 different types of message dispatchers:
• Dispatcher
– This is an event-based dispatcher that binds a set of Actors to a thread pool. It is the default dispatcher
used if one is not specified.
– Sharability: Unlimited
– Mailboxes: Any, creates one per Actor
– Use cases: Default dispatcher, Bulkheading
– Driven by: java.util.concurrent.ExecutorService specify using “executor” using
“fork-join-executor”, “thread-pool-executor” or the FQCN of an akka.dispatcher.
ExecutorServiceConfigurator
• PinnedDispatcher
– This dispatcher dedicates a unique thread for each actor using it; i.e. each actor will have its own
thread pool with only one thread in the pool.
– Sharability: None
– Mailboxes: Any, creates one per Actor
– Use cases: Bulkheading
4.4. Dispatchers
150
Akka Scala Documentation, Release 2.5.0-RC1
– Driven by: Any akka.dispatch.ThreadPoolExecutorConfigurator by
“thread-pool-executor”
default
a
• CallingThreadDispatcher
– This dispatcher runs invocations on the current thread only. This dispatcher does not create any new
threads, but it can be used from different threads concurrently for the same actor. See CallingThreadDispatcher for details and restrictions.
– Sharability: Unlimited
– Mailboxes: Any, creates one per Actor per Thread (on demand)
– Use cases: Testing
– Driven by: The calling thread (duh)
More dispatcher configuration examples
Configuring a dispatcher with fixed thread pool size, e.g. for actors that perform blocking IO:
blocking-io-dispatcher {
type = Dispatcher
executor = "thread-pool-executor"
thread-pool-executor {
fixed-pool-size = 32
}
throughput = 1
}
And then using it:
val myActor =
context.actorOf(Props[MyActor].withDispatcher("blocking-io-dispatcher"),
˓→"myactor2")
Another example that uses the thread pool based on the number of cores (e.g. for CPU bound tasks)
my-thread-pool-dispatcher {
# Dispatcher is the name of the event-based dispatcher
type = Dispatcher
# What kind of ExecutionService to use
executor = "thread-pool-executor"
# Configuration for the thread pool
thread-pool-executor {
# minimum number of threads to cap factor-based core number to
core-pool-size-min = 2
# No of core threads ... ceil(available processors * factor)
core-pool-size-factor = 2.0
# maximum number of threads to cap factor-based number to
core-pool-size-max = 10
}
# Throughput defines the maximum number of messages to be
# processed per actor before the thread jumps to the next actor.
# Set to 1 for as fair as possible.
throughput = 100
}
Configuring a PinnedDispatcher:
my-pinned-dispatcher {
executor = "thread-pool-executor"
type = PinnedDispatcher
}
4.4. Dispatchers
151
Akka Scala Documentation, Release 2.5.0-RC1
And then using it:
val myActor =
context.actorOf(Props[MyActor].withDispatcher("my-pinned-dispatcher"), "myactor3
˓→")
Note that thread-pool-executor configuration as per the above my-thread-pool-dispatcher
example is NOT applicable. This is because every actor will have its own thread pool when using
PinnedDispatcher, and that pool will have only one thread.
Note that it’s not guaranteed that the same thread is used over time, since the core pool timeout is used for
PinnedDispatcher to keep resource usage down in case of idle actors. To use the same thread all the
time you need to add thread-pool-executor.allow-core-timeout=off to the configuration of the
PinnedDispatcher.
4.5 Mailboxes
An Akka Mailbox holds the messages that are destined for an Actor. Normally each Actor has its own
mailbox, but with for example a BalancingPool all routees will share a single mailbox instance.
4.5.1 Mailbox Selection
Requiring a Message Queue Type for an Actor
It is possible to require a certain type of message queue for a certain type of actor by having that actor extend the
parameterized trait RequiresMessageQueue. Here is an example:
import akka.dispatch.RequiresMessageQueue
import akka.dispatch.BoundedMessageQueueSemantics
class MyBoundedActor extends MyActor
with RequiresMessageQueue[BoundedMessageQueueSemantics]
The type parameter to the RequiresMessageQueue trait needs to be mapped to a mailbox in configuration
like this:
bounded-mailbox {
mailbox-type = "akka.dispatch.BoundedMailbox"
mailbox-capacity = 1000
mailbox-push-timeout-time = 10s
}
akka.actor.mailbox.requirements {
"akka.dispatch.BoundedMessageQueueSemantics" = bounded-mailbox
}
Now every time you create an actor of type MyBoundedActor it will try to get a bounded mailbox. If the actor
has a different mailbox configured in deployment, either directly or via a dispatcher with a specified mailbox type,
then that will override this mapping.
Note: The type of the queue in the mailbox created for an actor will be checked against the required type in the
trait and if the queue doesn’t implement the required type then actor creation will fail.
4.5. Mailboxes
152
Akka Scala Documentation, Release 2.5.0-RC1
Requiring a Message Queue Type for a Dispatcher
A dispatcher may also have a requirement for the mailbox type used by the actors running on it. An example is
the BalancingDispatcher which requires a message queue that is thread-safe for multiple concurrent consumers.
Such a requirement is formulated within the dispatcher configuration section like this:
my-dispatcher {
mailbox-requirement = org.example.MyInterface
}
The given requirement names a class or interface which will then be ensured to be a supertype of the message
queue’s implementation. In case of a conflict—e.g. if the actor requires a mailbox type which does not satisfy this
requirement—then actor creation will fail.
How the Mailbox Type is Selected
When an actor is created, the ActorRefProvider first determines the dispatcher which will execute it. Then
the mailbox is determined as follows:
1. If the actor’s deployment configuration section contains a mailbox key then that names a configuration
section describing the mailbox type to be used.
2. If the actor’s Props contains a mailbox selection—i.e. withMailbox was called on it—then that names
a configuration section describing the mailbox type to be used.
3. If the dispatcher’s configuration section contains a mailbox-type key the same section will be used to
configure the mailbox type.
4. If the actor requires a mailbox type as described above then the mapping for that requirement will be used
to determine the mailbox type to be used; if that fails then the dispatcher’s requirement—if any—will be
tried instead.
5. If the dispatcher requires a mailbox type as described above then the mapping for that requirement will be
used to determine the mailbox type to be used.
6. The default mailbox akka.actor.default-mailbox will be used.
Default Mailbox
When the mailbox is not specified as described above the default mailbox is used. By default it is an unbounded
mailbox, which is backed by a java.util.concurrent.ConcurrentLinkedQueue.
SingleConsumerOnlyUnboundedMailbox is an even more efficient mailbox, and it can be used as the
default mailbox, but it cannot be used with a BalancingDispatcher.
Configuration of SingleConsumerOnlyUnboundedMailbox as default mailbox:
akka.actor.default-mailbox {
mailbox-type = "akka.dispatch.SingleConsumerOnlyUnboundedMailbox"
}
Which Configuration is passed to the Mailbox Type
Each mailbox type is implemented by a class which extends MailboxType and takes two constructor arguments:
a ActorSystem.Settings object and a Config section. The latter is computed by obtaining the named
configuration section from the actor system’s configuration, overriding its id key with the configuration path of
the mailbox type and adding a fall-back to the default mailbox configuration section.
4.5. Mailboxes
153
Akka Scala Documentation, Release 2.5.0-RC1
4.5.2 Builtin Mailbox Implementations
Akka comes shipped with a number of mailbox implementations:
• UnboundedMailbox (default)
– The default mailbox
– Backed by a java.util.concurrent.ConcurrentLinkedQueue
– Blocking: No
– Bounded: No
– Configuration name: "unbounded" or "akka.dispatch.UnboundedMailbox"
• SingleConsumerOnlyUnboundedMailbox
This queue may or may not be faster than the default one depending on your use-case—be sure to benchmark
properly!
– Backed by a Multiple-Producer
BalancingDispatcher
Single-Consumer
queue,
cannot
be
used
with
– Blocking: No
– Bounded: No
– Configuration name: "akka.dispatch.SingleConsumerOnlyUnboundedMailbox"
• NonBlockingBoundedMailbox
– Backed by a very efficient Multiple-Producer Single-Consumer queue
– Blocking: No (discards overflowing messages into deadLetters)
– Bounded: Yes
– Configuration name: "akka.dispatch.NonBlockingBoundedMailbox"
• UnboundedControlAwareMailbox
– Delivers messages that extend akka.dispatch.ControlMessage with higher priority
– Backed by two java.util.concurrent.ConcurrentLinkedQueue
– Blocking: No
– Bounded: No
– Configuration name: “akka.dispatch.UnboundedControlAwareMailbox”
• UnboundedPriorityMailbox
– Backed by a java.util.concurrent.PriorityBlockingQueue
– Delivery order for messages of equal priority is undefined - contrast with the UnboundedStablePriorityMailbox
– Blocking: No
– Bounded: No
– Configuration name: “akka.dispatch.UnboundedPriorityMailbox”
• UnboundedStablePriorityMailbox
– Backed by a java.util.concurrent.PriorityBlockingQueue wrapped in an akka.
util.PriorityQueueStabilizer
– FIFO order is preserved for messages of equal priority - contrast with the UnboundedPriorityMailbox
– Blocking: No
– Bounded: No
4.5. Mailboxes
154
Akka Scala Documentation, Release 2.5.0-RC1
– Configuration name: “akka.dispatch.UnboundedStablePriorityMailbox”
Other bounded mailbox implementations which will block the sender if the capacity is reached and configured
with non-zero mailbox-push-timeout-time.
Note: The following mailboxes should only be used with zero mailbox-push-timeout-time.
• BoundedMailbox
– Backed by a java.util.concurrent.LinkedBlockingQueue
– Blocking: Yes if used with non-zero mailbox-push-timeout-time, otherwise No
– Bounded: Yes
– Configuration name: “bounded” or “akka.dispatch.BoundedMailbox”
• BoundedPriorityMailbox
– Backed
by
a
java.util.PriorityQueue
BoundedBlockingQueue
– Delivery order for messages of equal
BoundedStablePriorityMailbox
priority
wrapped
is
in
undefined
an
-
akka.util.
contrast
with
the
– Blocking: Yes if used with non-zero mailbox-push-timeout-time, otherwise No
– Bounded: Yes
– Configuration name: "akka.dispatch.BoundedPriorityMailbox"
• BoundedStablePriorityMailbox
– Backed
by
a
java.util.PriorityQueue
wrapped
in
an
akka.util.
PriorityQueueStabilizer and an akka.util.BoundedBlockingQueue
– FIFO order is preserved for messages of equal priority - contrast with the BoundedPriorityMailbox
– Blocking: Yes if used with non-zero mailbox-push-timeout-time, otherwise No
– Bounded: Yes
– Configuration name: “akka.dispatch.BoundedStablePriorityMailbox”
• BoundedControlAwareMailbox
– Delivers messages that extend akka.dispatch.ControlMessage with higher priority
– Backed by two java.util.concurrent.ConcurrentLinkedQueue and blocking on enqueue if capacity has been reached
– Blocking: Yes if used with non-zero mailbox-push-timeout-time, otherwise No
– Bounded: Yes
– Configuration name: “akka.dispatch.BoundedControlAwareMailbox”
4.5.3 Mailbox configuration examples
PriorityMailbox
How to create a PriorityMailbox:
import akka.dispatch.PriorityGenerator
import akka.dispatch.UnboundedStablePriorityMailbox
import com.typesafe.config.Config
// We inherit, in this case, from UnboundedStablePriorityMailbox
4.5. Mailboxes
155
Akka Scala Documentation, Release 2.5.0-RC1
// and seed it with the priority generator
class MyPrioMailbox(settings: ActorSystem.Settings, config: Config)
extends UnboundedStablePriorityMailbox(
// Create a new PriorityGenerator, lower prio means more important
PriorityGenerator {
// 'highpriority messages should be treated first if possible
case 'highpriority => 0
// 'lowpriority messages should be treated last if possible
case 'lowpriority => 2
// PoisonPill when no other left
case PoisonPill
=> 3
// We default to 1, which is in between high and low
case otherwise
=> 1
})
And then add it to the configuration:
prio-dispatcher {
mailbox-type = "docs.dispatcher.DispatcherDocSpec$MyPrioMailbox"
//Other dispatcher configuration goes here
}
And then an example on how you would use it:
// We create a new Actor that just prints out what it processes
class Logger extends Actor {
val log: LoggingAdapter = Logging(context.system, this)
self
self
self
self
self
self
self
self
!
!
!
!
!
!
!
!
'lowpriority
'lowpriority
'highpriority
'pigdog
'pigdog2
'pigdog3
'highpriority
PoisonPill
def receive = {
case x => log.info(x.toString)
}
}
val a = system.actorOf(Props(classOf[Logger], this).withDispatcher(
"prio-dispatcher"))
/*
* Logs:
* 'highpriority
* 'highpriority
* 'pigdog
* 'pigdog2
* 'pigdog3
* 'lowpriority
* 'lowpriority
*/
It is also possible to configure a mailbox type directly like this:
prio-mailbox {
mailbox-type = "docs.dispatcher.DispatcherDocSpec$MyPrioMailbox"
4.5. Mailboxes
156
Akka Scala Documentation, Release 2.5.0-RC1
//Other mailbox configuration goes here
}
akka.actor.deployment {
/priomailboxactor {
mailbox = prio-mailbox
}
}
And then use it either from deployment like this:
import akka.actor.Props
val myActor = context.actorOf(Props[MyActor], "priomailboxactor")
Or code like this:
import akka.actor.Props
val myActor = context.actorOf(Props[MyActor].withMailbox("prio-mailbox"))
ControlAwareMailbox
A ControlAwareMailbox can be very useful if an actor needs to be able to receive control messages immediately no matter how many other messages are already in its mailbox.
It can be configured like this:
control-aware-dispatcher {
mailbox-type = "akka.dispatch.UnboundedControlAwareMailbox"
//Other dispatcher configuration goes here
}
Control messages need to extend the ControlMessage trait:
import akka.dispatch.ControlMessage
case object MyControlMessage extends ControlMessage
And then an example on how you would use it:
// We create a new Actor that just prints out what it processes
class Logger extends Actor {
val log: LoggingAdapter = Logging(context.system, this)
self
self
self
self
!
!
!
!
'foo
'bar
MyControlMessage
PoisonPill
def receive = {
case x => log.info(x.toString)
}
}
val a = system.actorOf(Props(classOf[Logger], this).withDispatcher(
"control-aware-dispatcher"))
/*
* Logs:
* MyControlMessage
* 'foo
* 'bar
*/
4.5. Mailboxes
157
Akka Scala Documentation, Release 2.5.0-RC1
4.5.4 Creating your own Mailbox type
An example is worth a thousand quacks:
import
import
import
import
import
import
import
import
import
akka.actor.ActorRef
akka.actor.ActorSystem
akka.dispatch.Envelope
akka.dispatch.MailboxType
akka.dispatch.MessageQueue
akka.dispatch.ProducesMessageQueue
com.typesafe.config.Config
java.util.concurrent.ConcurrentLinkedQueue
scala.Option
// Marker trait used for mailbox requirements mapping
trait MyUnboundedMessageQueueSemantics
object MyUnboundedMailbox {
// This is the MessageQueue implementation
class MyMessageQueue extends MessageQueue
with MyUnboundedMessageQueueSemantics {
private final val queue = new ConcurrentLinkedQueue[Envelope]()
// these should be implemented; queue used as example
def enqueue(receiver: ActorRef, handle: Envelope): Unit =
queue.offer(handle)
def dequeue(): Envelope = queue.poll()
def numberOfMessages: Int = queue.size
def hasMessages: Boolean = !queue.isEmpty
def cleanUp(owner: ActorRef, deadLetters: MessageQueue) {
while (hasMessages) {
deadLetters.enqueue(owner, dequeue())
}
}
}
}
// This is the Mailbox implementation
class MyUnboundedMailbox extends MailboxType
with ProducesMessageQueue[MyUnboundedMailbox.MyMessageQueue] {
import MyUnboundedMailbox._
// This constructor signature must exist, it will be called by Akka
def this(settings: ActorSystem.Settings, config: Config) = {
// put your initialization code here
this()
}
// The create method is called to create the MessageQueue
final override def create(
owner: Option[ActorRef],
system: Option[ActorSystem]): MessageQueue =
new MyMessageQueue()
}
And then you just specify the FQCN of your MailboxType as the value of the “mailbox-type” in the dispatcher
configuration, or the mailbox configuration.
Note: Make sure to include a constructor which takes akka.actor.ActorSystem.Settings and com.
typesafe.config.Config arguments, as this constructor is invoked reflectively to construct your mailbox
4.5. Mailboxes
158
Akka Scala Documentation, Release 2.5.0-RC1
type. The config passed in as second argument is that section from the configuration which describes the dispatcher
or mailbox setting using this mailbox type; the mailbox type will be instantiated once for each dispatcher or
mailbox setting using it.
You can also use the mailbox as a requirement on the dispatcher like this:
custom-dispatcher {
mailbox-requirement =
"jdocs.dispatcher.MyUnboundedMessageQueueSemantics"
}
akka.actor.mailbox.requirements {
"jdocs.dispatcher.MyUnboundedMessageQueueSemantics" =
custom-dispatcher-mailbox
}
custom-dispatcher-mailbox {
mailbox-type = "jdocs.dispatcher.MyUnboundedMailbox"
}
Or by defining the requirement on your actor class like this:
class MySpecialActor extends Actor
with RequiresMessageQueue[MyUnboundedMessageQueueSemantics] {
// ...
}
4.5.5 Special Semantics of system.actorOf
In order to make system.actorOf both synchronous and non-blocking while keeping the return type
ActorRef (and the semantics that the returned ref is fully functional), special handling takes place for this
case. Behind the scenes, a hollow kind of actor reference is constructed, which is sent to the system’s guardian
actor who actually creates the actor and its context and puts those inside the reference. Until that has happened,
messages sent to the ActorRef will be queued locally, and only upon swapping the real filling in will they be
transferred into the real mailbox. Thus,
val props: Props = ...
// this actor uses MyCustomMailbox, which is assumed to be a singleton
system.actorOf(props.withDispatcher("myCustomMailbox")) ! "bang"
assert(MyCustomMailbox.instance.getLastEnqueuedMessage == "bang")
will probably fail; you will have to allow for some time to pass and retry the check à la TestKit.awaitCond.
4.6 Routing
Messages can be sent via a router to efficiently route them to destination actors, known as its routees. A Router
can be used inside or outside of an actor, and you can manage the routees yourselves or use a self contained router
actor with configuration capabilities.
Different routing strategies can be used, according to your application’s needs. Akka comes with several useful
routing strategies right out of the box. But, as you will see in this chapter, it is also possible to create your own.
4.6.1 A Simple Router
The following example illustrates how to use a Router and manage the routees from within an actor.
4.6. Routing
159
Akka Scala Documentation, Release 2.5.0-RC1
import akka.routing.{ ActorRefRoutee, RoundRobinRoutingLogic, Router }
class Master extends Actor {
var router = {
val routees = Vector.fill(5) {
val r = context.actorOf(Props[Worker])
context watch r
ActorRefRoutee(r)
}
Router(RoundRobinRoutingLogic(), routees)
}
def receive = {
case w: Work =>
router.route(w, sender())
case Terminated(a) =>
router = router.removeRoutee(a)
val r = context.actorOf(Props[Worker])
context watch r
router = router.addRoutee(r)
}
}
We create a Router and specify that it should use RoundRobinRoutingLogic when routing the messages
to the routees.
The routing logic shipped with Akka are:
• akka.routing.RoundRobinRoutingLogic
• akka.routing.RandomRoutingLogic
• akka.routing.SmallestMailboxRoutingLogic
• akka.routing.BroadcastRoutingLogic
• akka.routing.ScatterGatherFirstCompletedRoutingLogic
• akka.routing.TailChoppingRoutingLogic
• akka.routing.ConsistentHashingRoutingLogic
We create the routees as ordinary child actors wrapped in ActorRefRoutee. We watch the routees to be able
to replace them if they are terminated.
Sending messages via the router is done with the route method, as is done for the Work messages in the example
above.
The Router is immutable and the RoutingLogic is thread safe; meaning that they can also be used outside
of actors.
Note: In general, any message sent to a router will be sent onwards to its routees, but there is one exception. The
special Broadcast Messages will send to all of a router’s routees. However, do not use Broadcast Messages when
you use BalancingPool for routees as described in Specially Handled Messages.
4.6.2 A Router Actor
A router can also be created as a self contained actor that manages the routees itself and loads routing logic and
other settings from configuration.
This type of router actor comes in two distinct flavors:
• Pool - The router creates routees as child actors and removes them from the router if they terminate.
4.6. Routing
160
Akka Scala Documentation, Release 2.5.0-RC1
• Group - The routee actors are created externally to the router and the router sends messages to the specified
path using actor selection, without watching for termination.
The settings for a router actor can be defined in configuration or programmatically. In order to make an actor to
make use of an externally configurable router the FromConfig props wrapper must be used to denote that the
actor accepts routing settings from configuration. This is in contrast with Remote Deployment where such marker
props is not necessary. If the props of an actor is NOT wrapped in FromConfig it will ignore the router section of
the deployment configuration.
You send messages to the routees via the router actor in the same way as for ordinary actors, i.e. via its ActorRef.
The router actor forwards messages onto its routees without changing the original sender. When a routee replies
to a routed message, the reply will be sent to the original sender, not to the router actor.
Note: In general, any message sent to a router will be sent onwards to its routees, but there are a few exceptions.
These are documented in the Specially Handled Messages section below.
Pool
The following code and configuration snippets show how to create a round-robin router that forwards messages to
five Worker routees. The routees will be created as the router’s children.
akka.actor.deployment {
/parent/router1 {
router = round-robin-pool
nr-of-instances = 5
}
}
val router1: ActorRef =
context.actorOf(FromConfig.props(Props[Worker]), "router1")
Here is the same example, but with the router configuration provided programmatically instead of from configuration.
val router2: ActorRef =
context.actorOf(RoundRobinPool(5).props(Props[Worker]), "router2")
Remote Deployed Routees
In addition to being able to create local actors as routees, you can instruct the router to deploy its created children
on a set of remote hosts. Routees will be deployed in round-robin fashion. In order to deploy routees remotely,
wrap the router configuration in a RemoteRouterConfig, attaching the remote addresses of the nodes to
deploy to. Remote deployment requires the akka-remote module to be included in the classpath.
import akka.actor.{ Address, AddressFromURIString }
import akka.remote.routing.RemoteRouterConfig
val addresses = Seq(
Address("akka.tcp", "remotesys", "otherhost", 1234),
AddressFromURIString("akka.tcp://othersys@anotherhost:1234"))
val routerRemote = system.actorOf(
RemoteRouterConfig(RoundRobinPool(5), addresses).props(Props[Echo]))
Senders
By default, when a routee sends a message, it will implicitly set itself as the sender.
4.6. Routing
161
Akka Scala Documentation, Release 2.5.0-RC1
sender() ! x // replies will go to this actor
However, it is often useful for routees to set the router as a sender. For example, you might want to set the router
as the sender if you want to hide the details of the routees behind the router. The following code snippet shows
how to set the parent router as sender.
sender().tell("reply", context.parent) // replies will go back to parent
sender().!("reply")(context.parent) // alternative syntax (beware of the parens!)
Supervision
Routees that are created by a pool router will be created as the router’s children. The router is therefore also the
children’s supervisor.
The supervision strategy of the router actor can be configured with the supervisorStrategy property of the
Pool. If no configuration is provided, routers default to a strategy of “always escalate”. This means that errors are
passed up to the router’s supervisor for handling. The router’s supervisor will decide what to do about any errors.
Note the router’s supervisor will treat the error as an error with the router itself. Therefore a directive to stop or
restart will cause the router itself to stop or restart. The router, in turn, will cause its children to stop and restart.
It should be mentioned that the router’s restart behavior has been overridden so that a restart, while still re-creating
the children, will still preserve the same number of actors in the pool.
This means that if you have not specified supervisorStrategy of the router or its parent a failure in a routee
will escalate to the parent of the router, which will by default restart the router, which will restart all routees (it
uses Escalate and does not stop routees during restart). The reason is to make the default behave such that adding
withRouter to a child’s definition does not change the supervision strategy applied to the child. This might be
an inefficiency that you can avoid by specifying the strategy when defining the router.
Setting the strategy is easily done:
val escalator = OneForOneStrategy() {
case e ⇒ testActor ! e; SupervisorStrategy.Escalate
}
val router = system.actorOf(RoundRobinPool(1, supervisorStrategy = escalator).
˓→props(
routeeProps = Props[TestActor]))
Note: If the child of a pool router terminates, the pool router will not automatically spawn a new child. In the
event that all children of a pool router have terminated the router will terminate itself unless it is a dynamic router,
e.g. using a resizer.
Group
Sometimes, rather than having the router actor create its routees, it is desirable to create routees separately and provide them to the router for its use. You can do this by passing an paths of the routees to the router’s configuration.
Messages will be sent with ActorSelection to these paths.
The example below shows how to create a router by providing it with the path strings of three routee actors.
akka.actor.deployment {
/parent/router3 {
router = round-robin-group
routees.paths = ["/user/workers/w1", "/user/workers/w2", "/user/workers/w3"]
}
}
4.6. Routing
162
Akka Scala Documentation, Release 2.5.0-RC1
val router3: ActorRef =
context.actorOf(FromConfig.props(), "router3")
Here is the same example, but with the router configuration provided programmatically instead of from configuration.
val router4: ActorRef =
context.actorOf(RoundRobinGroup(paths).props(), "router4")
The routee actors are created externally from the router:
system.actorOf(Props[Workers], "workers")
class Workers extends Actor {
context.actorOf(Props[Worker], name = "w1")
context.actorOf(Props[Worker], name = "w2")
context.actorOf(Props[Worker], name = "w3")
// ...
The paths may contain protocol and address information for actors running on remote hosts. Remoting requires
the akka-remote module to be included in the classpath.
akka.actor.deployment {
/parent/remoteGroup {
router = round-robin-group
routees.paths = [
"akka.tcp://app@10.0.0.1:2552/user/workers/w1",
"akka.tcp://app@10.0.0.2:2552/user/workers/w1",
"akka.tcp://app@10.0.0.3:2552/user/workers/w1"]
}
}
4.6.3 Router usage
In this section we will describe how to create the different types of router actors.
The router actors in this section are created from within a top level actor named parent. Note that deployment
paths in the configuration starts with /parent/ followed by the name of the router actor.
system.actorOf(Props[Parent], "parent")
RoundRobinPool and RoundRobinGroup
Routes in a round-robin fashion to its routees.
RoundRobinPool defined in configuration:
akka.actor.deployment {
/parent/router1 {
router = round-robin-pool
nr-of-instances = 5
}
}
val router1: ActorRef =
context.actorOf(FromConfig.props(Props[Worker]), "router1")
RoundRobinPool defined in code:
4.6. Routing
163
Akka Scala Documentation, Release 2.5.0-RC1
val router2: ActorRef =
context.actorOf(RoundRobinPool(5).props(Props[Worker]), "router2")
RoundRobinGroup defined in configuration:
akka.actor.deployment {
/parent/router3 {
router = round-robin-group
routees.paths = ["/user/workers/w1", "/user/workers/w2", "/user/workers/w3"]
}
}
val router3: ActorRef =
context.actorOf(FromConfig.props(), "router3")
RoundRobinGroup defined in code:
val paths = List("/user/workers/w1", "/user/workers/w2", "/user/workers/w3")
val router4: ActorRef =
context.actorOf(RoundRobinGroup(paths).props(), "router4")
RandomPool and RandomGroup
This router type selects one of its routees randomly for each message.
RandomPool defined in configuration:
akka.actor.deployment {
/parent/router5 {
router = random-pool
nr-of-instances = 5
}
}
val router5: ActorRef =
context.actorOf(FromConfig.props(Props[Worker]), "router5")
RandomPool defined in code:
val router6: ActorRef =
context.actorOf(RandomPool(5).props(Props[Worker]), "router6")
RandomGroup defined in configuration:
akka.actor.deployment {
/parent/router7 {
router = random-group
routees.paths = ["/user/workers/w1", "/user/workers/w2", "/user/workers/w3"]
}
}
val router7: ActorRef =
context.actorOf(FromConfig.props(), "router7")
RandomGroup defined in code:
val paths = List("/user/workers/w1", "/user/workers/w2", "/user/workers/w3")
val router8: ActorRef =
context.actorOf(RandomGroup(paths).props(), "router8")
4.6. Routing
164
Akka Scala Documentation, Release 2.5.0-RC1
BalancingPool
A Router that will try to redistribute work from busy routees to idle routees. All routees share the same mailbox.
Note: The BalancingPool has the property that its routees do not have truly distinct identity: they have different
names, but talking to them will not end up at the right actor in most cases. Therefore you cannot use it for
workflows that require state to be kept within the routee, you would in this case have to include the whole state in
the messages.
With a SmallestMailboxPool you can have a vertically scaling service that can interact in a stateful fashion with
other services in the back-end before replying to the original client. The other advantage is that it does not place
a restriction on the message queue implementation as BalancingPool does.
Note: Do not use Broadcast Messages when you use BalancingPool for routers. as described in Specially
Handled Messages,
BalancingPool defined in configuration:
akka.actor.deployment {
/parent/router9 {
router = balancing-pool
nr-of-instances = 5
}
}
val router9: ActorRef =
context.actorOf(FromConfig.props(Props[Worker]), "router9")
BalancingPool defined in code:
val router10: ActorRef =
context.actorOf(BalancingPool(5).props(Props[Worker]), "router10")
Addition configuration for the balancing dispatcher, which is used by the pool, can be configured in the
pool-dispatcher section of the router deployment configuration.
akka.actor.deployment {
/parent/router9b {
router = balancing-pool
nr-of-instances = 5
pool-dispatcher {
attempt-teamwork = off
}
}
}
The BalancingPool automatically uses a special BalancingDispatcher for its routees - disregarding
any dispatcher that is set on the routee Props object. This is needed in order to implement the balancing semantics
via sharing the same mailbox by all the routees.
While it is not possible to change the dispatcher used by the routees, it is possible to fine tune the used executor. By default the fork-join-dispatcher is used and can be configured as explained in Dispatchers. In
situations where the routees are expected to perform blocking operations it may be useful to replace it with a
thread-pool-executor hinting the number of allocated threads explicitly:
akka.actor.deployment {
/parent/router10b {
router = balancing-pool
nr-of-instances = 5
4.6. Routing
165
Akka Scala Documentation, Release 2.5.0-RC1
pool-dispatcher {
executor = "thread-pool-executor"
# allocate exactly 5
thread-pool-executor
core-pool-size-min
core-pool-size-max
}
threads for this pool
{
= 5
= 5
}
}
}
It is also possible to change the mailbox used by the balancing dispatcher for scenarios where the default
unbounded mailbox is not well suited. An example of such a scenario could arise whether there exists the need to
manage priority for each message. You can then implement a priority mailbox and configure your dispatcher:
akka.actor.deployment {
/parent/router10c {
router = balancing-pool
nr-of-instances = 5
pool-dispatcher {
mailbox = myapp.myprioritymailbox
}
}
}
Note: Bear in mind that BalancingDispatcher requires a message queue that must be thread-safe for
multiple concurrent consumers. So it is mandatory for the message queue backing a custom mailbox for this kind
of dispatcher to implement akka.dispatch.MultipleConsumerSemantics. See details on how to implement your
custom mailbox in Mailboxes.
There is no Group variant of the BalancingPool.
SmallestMailboxPool
A Router that tries to send to the non-suspended child routee with fewest messages in mailbox. The selection is
done in this order:
• pick any idle routee (not processing message) with empty mailbox
• pick any routee with empty mailbox
• pick routee with fewest pending messages in mailbox
• pick any remote routee, remote actors are consider lowest priority, since their mailbox size is unknown
SmallestMailboxPool defined in configuration:
akka.actor.deployment {
/parent/router11 {
router = smallest-mailbox-pool
nr-of-instances = 5
}
}
val router11: ActorRef =
context.actorOf(FromConfig.props(Props[Worker]), "router11")
SmallestMailboxPool defined in code:
4.6. Routing
166
Akka Scala Documentation, Release 2.5.0-RC1
val router12: ActorRef =
context.actorOf(SmallestMailboxPool(5).props(Props[Worker]), "router12")
There is no Group variant of the SmallestMailboxPool because the size of the mailbox and the internal dispatching
state of the actor is not practically available from the paths of the routees.
BroadcastPool and BroadcastGroup
A broadcast router forwards the message it receives to all its routees.
BroadcastPool defined in configuration:
akka.actor.deployment {
/parent/router13 {
router = broadcast-pool
nr-of-instances = 5
}
}
val router13: ActorRef =
context.actorOf(FromConfig.props(Props[Worker]), "router13")
BroadcastPool defined in code:
val router14: ActorRef =
context.actorOf(BroadcastPool(5).props(Props[Worker]), "router14")
BroadcastGroup defined in configuration:
akka.actor.deployment {
/parent/router15 {
router = broadcast-group
routees.paths = ["/user/workers/w1", "/user/workers/w2", "/user/workers/w3"]
}
}
val router15: ActorRef =
context.actorOf(FromConfig.props(), "router15")
BroadcastGroup defined in code:
val paths = List("/user/workers/w1", "/user/workers/w2", "/user/workers/w3")
val router16: ActorRef =
context.actorOf(BroadcastGroup(paths).props(), "router16")
Note: Broadcast routers always broadcast every message to their routees. If you do not want to broadcast every
message, then you can use a non-broadcasting router and use Broadcast Messages as needed.
ScatterGatherFirstCompletedPool and ScatterGatherFirstCompletedGroup
The ScatterGatherFirstCompletedRouter will send the message on to all its routees. It then waits for first reply it
gets back. This result will be sent back to original sender. Other replies are discarded.
It is expecting at least one reply within a configured duration, otherwise it will reply with akka.pattern.
AskTimeoutException in a akka.actor.Status.Failure.
ScatterGatherFirstCompletedPool defined in configuration:
4.6. Routing
167
Akka Scala Documentation, Release 2.5.0-RC1
akka.actor.deployment {
/parent/router17 {
router = scatter-gather-pool
nr-of-instances = 5
within = 10 seconds
}
}
val router17: ActorRef =
context.actorOf(FromConfig.props(Props[Worker]), "router17")
ScatterGatherFirstCompletedPool defined in code:
val router18: ActorRef =
context.actorOf(ScatterGatherFirstCompletedPool(5, within = 10.seconds).
props(Props[Worker]), "router18")
ScatterGatherFirstCompletedGroup defined in configuration:
akka.actor.deployment {
/parent/router19 {
router = scatter-gather-group
routees.paths = ["/user/workers/w1", "/user/workers/w2", "/user/workers/w3"]
within = 10 seconds
}
}
val router19: ActorRef =
context.actorOf(FromConfig.props(), "router19")
ScatterGatherFirstCompletedGroup defined in code:
val paths = List("/user/workers/w1", "/user/workers/w2", "/user/workers/w3")
val router20: ActorRef =
context.actorOf(ScatterGatherFirstCompletedGroup(
paths,
within = 10.seconds).props(), "router20")
TailChoppingPool and TailChoppingGroup
The TailChoppingRouter will first send the message to one, randomly picked, routee and then after a small delay
to a second routee (picked randomly from the remaining routees) and so on. It waits for first reply it gets back and
forwards it back to original sender. Other replies are discarded.
The goal of this router is to decrease latency by performing redundant queries to multiple routees, assuming that
one of the other actors may still be faster to respond than the initial one.
This optimisation was described nicely in a blog post by Peter Bailis: Doing redundant work to speed up distributed queries.
TailChoppingPool defined in configuration:
akka.actor.deployment {
/parent/router21 {
router = tail-chopping-pool
nr-of-instances = 5
within = 10 seconds
tail-chopping-router.interval = 20 milliseconds
}
}
4.6. Routing
168
Akka Scala Documentation, Release 2.5.0-RC1
val router21: ActorRef =
context.actorOf(FromConfig.props(Props[Worker]), "router21")
TailChoppingPool defined in code:
val router22: ActorRef =
context.actorOf(TailChoppingPool(5, within = 10.seconds, interval = 20.millis).
props(Props[Worker]), "router22")
TailChoppingGroup defined in configuration:
akka.actor.deployment {
/parent/router23 {
router = tail-chopping-group
routees.paths = ["/user/workers/w1", "/user/workers/w2", "/user/workers/w3"]
within = 10 seconds
tail-chopping-router.interval = 20 milliseconds
}
}
val router23: ActorRef =
context.actorOf(FromConfig.props(), "router23")
TailChoppingGroup defined in code:
val paths = List("/user/workers/w1", "/user/workers/w2", "/user/workers/w3")
val router24: ActorRef =
context.actorOf(TailChoppingGroup(
paths,
within = 10.seconds, interval = 20.millis).props(), "router24")
ConsistentHashingPool and ConsistentHashingGroup
The ConsistentHashingPool uses consistent hashing to select a routee based on the sent message. This article
gives good insight into how consistent hashing is implemented.
There is 3 ways to define what data to use for the consistent hash key.
• You can define hashMapping of the router to map incoming messages to their consistent hash key. This
makes the decision transparent for the sender.
• The
messages
may
implement
akka.routing.ConsistentHashingRouter.
ConsistentHashable. The key is part of the message and it’s convenient to define it together
with the message definition.
• The messages can be wrapped in a akka.routing.ConsistentHashingRouter.
ConsistentHashableEnvelope to define what data to use for the consistent hash key. The
sender knows the key to use.
These ways to define the consistent hash key can be use together and at the same time for one router. The
hashMapping is tried first.
Code example:
import akka.actor.Actor
import akka.routing.ConsistentHashingRouter.ConsistentHashable
class Cache extends Actor {
var cache = Map.empty[String, String]
def receive = {
case Entry(key, value) => cache += (key -> value)
4.6. Routing
169
Akka Scala Documentation, Release 2.5.0-RC1
case Get(key)
case Evict(key)
=> sender() ! cache.get(key)
=> cache -= key
}
}
final case class Evict(key: String)
final case class Get(key: String) extends ConsistentHashable {
override def consistentHashKey: Any = key
}
final case class Entry(key: String, value: String)
import
import
import
import
akka.actor.Props
akka.routing.ConsistentHashingPool
akka.routing.ConsistentHashingRouter.ConsistentHashMapping
akka.routing.ConsistentHashingRouter.ConsistentHashableEnvelope
def hashMapping: ConsistentHashMapping = {
case Evict(key) => key
}
val cache: ActorRef =
context.actorOf(ConsistentHashingPool(10, hashMapping = hashMapping).
props(Props[Cache]), name = "cache")
cache ! ConsistentHashableEnvelope(
message = Entry("hello", "HELLO"), hashKey = "hello")
cache ! ConsistentHashableEnvelope(
message = Entry("hi", "HI"), hashKey = "hi")
cache ! Get("hello")
expectMsg(Some("HELLO"))
cache ! Get("hi")
expectMsg(Some("HI"))
cache ! Evict("hi")
cache ! Get("hi")
expectMsg(None)
In the above example you see that the Get message implements ConsistentHashable itself, while the
Entry message is wrapped in a ConsistentHashableEnvelope. The Evict message is handled by
the hashMapping partial function.
ConsistentHashingPool defined in configuration:
akka.actor.deployment {
/parent/router25 {
router = consistent-hashing-pool
nr-of-instances = 5
virtual-nodes-factor = 10
}
}
val router25: ActorRef =
context.actorOf(FromConfig.props(Props[Worker]), "router25")
ConsistentHashingPool defined in code:
val router26: ActorRef =
context.actorOf(
4.6. Routing
170
Akka Scala Documentation, Release 2.5.0-RC1
ConsistentHashingPool(5).props(Props[Worker]),
"router26")
ConsistentHashingGroup defined in configuration:
akka.actor.deployment {
/parent/router27 {
router = consistent-hashing-group
routees.paths = ["/user/workers/w1", "/user/workers/w2", "/user/workers/w3"]
virtual-nodes-factor = 10
}
}
val router27: ActorRef =
context.actorOf(FromConfig.props(), "router27")
ConsistentHashingGroup defined in code:
val paths = List("/user/workers/w1", "/user/workers/w2", "/user/workers/w3")
val router28: ActorRef =
context.actorOf(ConsistentHashingGroup(paths).props(), "router28")
virtual-nodes-factor is the number of virtual nodes per routee that is used in the consistent hash node
ring to make the distribution more uniform.
4.6.4 Specially Handled Messages
Most messages sent to router actors will be forwarded according to the routers’ routing logic. However there are
a few types of messages that have special behavior.
Note that these special messages, except for the Broadcast message, are only handled by self contained router
actors and not by the akka.routing.Router component described in A Simple Router.
Broadcast Messages
A Broadcast message can be used to send a message to all of a router’s routees. When a router receives a
Broadcast message, it will broadcast that message’s payload to all routees, no matter how that router would
normally route its messages.
The example below shows how you would use a Broadcast message to send a very important message to every
routee of a router.
import akka.routing.Broadcast
router ! Broadcast("Watch out for Davy Jones' locker")
In this example the router receives the Broadcast message,
extracts its payload
("Watch out for Davy Jones' locker"), and then sends the payload on to all of the router’s
routees. It is up to each routee actor to handle the received payload message.
Note: Do not use Broadcast Messages when you use BalancingPool for routers. Routees on BalancingPool
shares the same mailbox instance, thus some routees can possibly get the broadcast message multiple times, while
other routees get no broadcast message.
PoisonPill Messages
A PoisonPill message has special handling for all actors, including for routers. When any actor receives a
PoisonPill message, that actor will be stopped. See the PoisonPill documentation for details.
4.6. Routing
171
Akka Scala Documentation, Release 2.5.0-RC1
import akka.actor.PoisonPill
router ! PoisonPill
For a router, which normally passes on messages to routees, it is important to realise that PoisonPill messages
are processed by the router only. PoisonPill messages sent to a router will not be sent on to routees.
However, a PoisonPill message sent to a router may still affect its routees, because it will stop the router and
when the router stops it also stops its children. Stopping children is normal actor behavior. The router will stop
routees that it has created as children. Each child will process its current message and then stop. This may lead to
some messages being unprocessed. See the documentation on Stopping actors for more information.
If you wish to stop a router and its routees, but you would like the routees to first process all the messages
currently in their mailboxes, then you should not send a PoisonPill message to the router. Instead you should
wrap a PoisonPill message inside a Broadcast message so that each routee will receive the PoisonPill
message. Note that this will stop all routees, even if the routees aren’t children of the router, i.e. even routees
programmatically provided to the router.
import akka.actor.PoisonPill
import akka.routing.Broadcast
router ! Broadcast(PoisonPill)
With the code shown above, each routee will receive a PoisonPill message. Each routee will continue to
process its messages as normal, eventually processing the PoisonPill. This will cause the routee to stop. After
all routees have stopped the router will itself be stopped automatically unless it is a dynamic router, e.g. using a
resizer.
Note: Brendan W McAdams’ excellent blog post Distributing Akka Workloads - And Shutting Down Afterwards
discusses in more detail how PoisonPill messages can be used to shut down routers and routees.
Kill Messages
Kill messages are another type of message that has special handling. See Killing an Actor for general information about how actors handle Kill messages.
When a Kill message is sent to a router the router processes the message internally, and does not send it on to its
routees. The router will throw an ActorKilledException and fail. It will then be either resumed, restarted
or terminated, depending how it is supervised.
Routees that are children of the router will also be suspended, and will be affected by the supervision directive
that is applied to the router. Routees that are not the routers children, i.e. those that were created externally to the
router, will not be affected.
import akka.actor.Kill
router ! Kill
As with the PoisonPill message, there is a distinction between killing a router, which indirectly kills its
children (who happen to be routees), and killing routees directly (some of whom may not be children.) To kill
routees directly the router should be sent a Kill message wrapped in a Broadcast message.
import akka.actor.Kill
import akka.routing.Broadcast
router ! Broadcast(Kill)
Management Messages
• Sending akka.routing.GetRoutees to a router actor will make it send back its currently used routees
in a akka.routing.Routees message.
4.6. Routing
172
Akka Scala Documentation, Release 2.5.0-RC1
• Sending akka.routing.AddRoutee to a router actor will add that routee to its collection of routees.
• Sending akka.routing.RemoveRoutee to a router actor will remove that routee to its collection of
routees.
• Sending akka.routing.AdjustPoolSize to a pool router actor will add or remove that number of
routees to its collection of routees.
These management messages may be handled after other messages, so if you send AddRoutee immediately
followed by an ordinary message you are not guaranteed that the routees have been changed when the ordinary
message is routed. If you need to know when the change has been applied you can send AddRoutee followed by
GetRoutees and when you receive the Routees reply you know that the preceding change has been applied.
4.6.5 Dynamically Resizable Pool
Most pools can be used with a fixed number of routees or with a resize strategy to adjust the number of routees
dynamically.
There are two types of resizers: the default Resizer and the OptimalSizeExploringResizer.
Default Resizer
The default resizer ramps up and down pool size based on pressure, measured by the percentage of busy routees
in the pool. It ramps up pool size if the pressure is higher than a certain threshold and backs off if the pressure is
lower than certain threshold. Both thresholds are configurable.
Pool with default resizer defined in configuration:
akka.actor.deployment {
/parent/router29 {
router = round-robin-pool
resizer {
lower-bound = 2
upper-bound = 15
messages-per-resize = 100
}
}
}
val router29: ActorRef =
context.actorOf(FromConfig.props(Props[Worker]), "router29")
Several more configuration options are available and described in akka.actor.deployment.default.
resizer section of the reference Configuration.
Pool with resizer defined in code:
val resizer = DefaultResizer(lowerBound = 2, upperBound = 15)
val router30: ActorRef =
context.actorOf(
RoundRobinPool(5, Some(resizer)).props(Props[Worker]),
"router30")
It is also worth pointing out that if you define the ‘‘router‘‘ in the configuration file then this value will be used
instead of any programmatically sent parameters.
Optimal Size Exploring Resizer
The OptimalSizeExploringResizer resizes the pool to an optimal size that provides the most message
throughput.
4.6. Routing
173
Akka Scala Documentation, Release 2.5.0-RC1
This resizer works best when you expect the pool size to performance function to be a convex function. For
example, when you have a CPU bound tasks, the optimal size is bound to the number of CPU cores. When your
task is IO bound, the optimal size is bound to optimal number of concurrent connections to that IO service - e.g.
a 4 node elastic search cluster may handle 4-8 concurrent requests at optimal speed.
It achieves this by keeping track of message throughput at each pool size and performing the following three
resizing operations (one at a time) periodically:
• Downsize if it hasn’t seen all routees ever fully utilized for a period of time.
• Explore to a random nearby pool size to try and collect throughput metrics.
• Optimize to a nearby pool size with a better (than any other nearby sizes) throughput metrics.
When the pool is fully-utilized (i.e. all routees are busy), it randomly choose between exploring and optimizing.
When the pool has not been fully-utilized for a period of time, it will downsize the pool to the last seen max
utilization multiplied by a configurable ratio.
By constantly exploring and optimizing, the resizer will eventually walk to the optimal size and remain nearby.
When the optimal size changes it will start walking towards the new one.
It keeps a performance log so it’s stateful as well as having a larger memory footprint than the default Resizer.
The memory usage is O(n) where n is the number of sizes you allow, i.e. upperBound - lowerBound.
Pool with OptimalSizeExploringResizer defined in configuration:
akka.actor.deployment {
/parent/router31 {
router = round-robin-pool
optimal-size-exploring-resizer {
enabled = on
action-interval = 5s
downsize-after-underutilized-for = 72h
}
}
}
val router31: ActorRef =
context.actorOf(FromConfig.props(Props[Worker]), "router31")
Several more configuration options are available and described in akka.actor.deployment.default.
optimal-size-exploring-resizer section of the reference Configuration.
Note: Resizing is triggered by sending messages to the actor pool, but it is not completed synchronously; instead
a message is sent to the “head” RouterActor to perform the size change. Thus you cannot rely on resizing
to instantaneously create new workers when all others are busy, because the message just sent will be queued to
the mailbox of a busy actor. To remedy this, configure the pool to use a balancing dispatcher, see Configuring
Dispatchers for more information.
4.6.6 How Routing is Designed within Akka
On the surface routers look like normal actors, but they are actually implemented differently. Routers are designed
to be extremely efficient at receiving messages and passing them quickly on to routees.
A normal actor can be used for routing messages, but an actor’s single-threaded processing can become a bottleneck. Routers can achieve much higher throughput with an optimization to the usual message-processing pipeline
that allows concurrent routing. This is achieved by embedding routers’ routing logic directly in their ActorRef
rather than in the router actor. Messages sent to a router’s ActorRef can be immediately routed to the routee,
bypassing the single-threaded router actor entirely.
4.6. Routing
174
Akka Scala Documentation, Release 2.5.0-RC1
The cost to this is, of course, that the internals of routing code are more complicated than if routers were implemented with normal actors. Fortunately all of this complexity is invisible to consumers of the routing API.
However, it is something to be aware of when implementing your own routers.
4.6.7 Custom Router
You can create your own router should you not find any of the ones provided by Akka sufficient for your needs.
In order to roll your own router you have to fulfill certain criteria which are explained in this section.
Before creating your own router you should consider whether a normal actor with router-like behavior might do
the job just as well as a full-blown router. As explained above, the primary benefit of routers over normal actors
is their higher performance. But they are somewhat more complicated to write than normal actors. Therefore if
lower maximum throughput is acceptable in your application you may wish to stick with traditional actors. This
section, however, assumes that you wish to get maximum performance and so demonstrates how you can create
your own router.
The router created in this example is replicating each message to a few destinations.
Start with the routing logic:
import
import
import
import
import
import
scala.collection.immutable
java.util.concurrent.ThreadLocalRandom
akka.routing.RoundRobinRoutingLogic
akka.routing.RoutingLogic
akka.routing.Routee
akka.routing.SeveralRoutees
class RedundancyRoutingLogic(nbrCopies: Int) extends RoutingLogic {
val roundRobin = RoundRobinRoutingLogic()
def select(message: Any, routees: immutable.IndexedSeq[Routee]): Routee = {
val targets = (1 to nbrCopies).map(_ => roundRobin.select(message, routees))
SeveralRoutees(targets)
}
}
select will be called for each message and in this example pick a few destinations by round-robin, by
reusing the existing RoundRobinRoutingLogic and wrap the result in a SeveralRoutees instance.
SeveralRoutees will send the message to all of the supplied routes.
The implementation of the routing logic must be thread safe, since it might be used outside of actors.
A unit test of the routing logic:
final case class TestRoutee(n: Int) extends Routee {
override def send(message: Any, sender: ActorRef): Unit = ()
}
val logic = new RedundancyRoutingLogic(nbrCopies = 3)
val routees = for (n <- 1 to 7) yield TestRoutee(n)
val r1 = logic.select("msg", routees)
r1.asInstanceOf[SeveralRoutees].routees should be(
Vector(TestRoutee(1), TestRoutee(2), TestRoutee(3)))
val r2 = logic.select("msg", routees)
r2.asInstanceOf[SeveralRoutees].routees should be(
Vector(TestRoutee(4), TestRoutee(5), TestRoutee(6)))
val r3 = logic.select("msg", routees)
r3.asInstanceOf[SeveralRoutees].routees should be(
Vector(TestRoutee(7), TestRoutee(1), TestRoutee(2)))
4.6. Routing
175
Akka Scala Documentation, Release 2.5.0-RC1
You could stop here and use the RedundancyRoutingLogic with a akka.routing.Router as described
in A Simple Router.
Let us continue and make this into a self contained, configurable, router actor.
Create a class that extends Pool, Group or CustomRouterConfig. That class is a factory for the routing
logic and holds the configuration for the router. Here we make it a Group.
import
import
import
import
import
akka.dispatch.Dispatchers
akka.routing.Group
akka.routing.Router
akka.japi.Util.immutableSeq
com.typesafe.config.Config
final case class RedundancyGroup(routeePaths: immutable.Iterable[String],
˓→nbrCopies: Int) extends Group {
def this(config: Config) = this(
routeePaths = immutableSeq(config.getStringList("routees.paths")),
nbrCopies = config.getInt("nbr-copies"))
override def paths(system: ActorSystem): immutable.Iterable[String] = routeePaths
override def createRouter(system: ActorSystem): Router =
new Router(new RedundancyRoutingLogic(nbrCopies))
override val routerDispatcher: String = Dispatchers.DefaultDispatcherId
}
This can be used exactly as the router actors provided by Akka.
for (n <- 1 to 10) system.actorOf(Props[Storage], "s" + n)
val paths = for (n <- 1 to 10) yield ("/user/s" + n)
val redundancy1: ActorRef =
system.actorOf(
RedundancyGroup(paths, nbrCopies = 3).props(),
name = "redundancy1")
redundancy1 ! "important"
Note that we added a constructor in RedundancyGroup that takes a Config parameter. That makes it possible
to define it in configuration.
akka.actor.deployment {
/redundancy2 {
router = "jdocs.routing.RedundancyGroup"
routees.paths = ["/user/s1", "/user/s2", "/user/s3"]
nbr-copies = 5
}
}
Note the fully qualified class name in the router property. The router class must extend akka.
routing.RouterConfig (Pool, Group or CustomRouterConfig) and have constructor with one com.
typesafe.config.Config parameter. The deployment section of the configuration is passed to the constructor.
val redundancy2: ActorRef = system.actorOf(
FromConfig.props(),
name = "redundancy2")
redundancy2 ! "very important"
4.6. Routing
176
Akka Scala Documentation, Release 2.5.0-RC1
4.6.8 Configuring Dispatchers
The dispatcher for created children of the pool will be taken from Props as described in Dispatchers.
To make it easy to define the dispatcher of the routees of the pool you can define the dispatcher inline in the
deployment section of the config.
akka.actor.deployment {
/poolWithDispatcher {
router = random-pool
nr-of-instances = 5
pool-dispatcher {
fork-join-executor.parallelism-min = 5
fork-join-executor.parallelism-max = 5
}
}
}
That is the only thing you need to do enable a dedicated dispatcher for a pool.
Note: If you use a group of actors and route to their paths, then they will still use the same dispatcher that was
configured for them in their Props, it is not possible to change an actors dispatcher after it has been created.
The “head” router cannot always run on the same dispatcher, because it does not process the same type
of messages, hence this special actor does not use the dispatcher configured in Props, but takes the
routerDispatcher from the RouterConfig instead, which defaults to the actor system’s default dispatcher. All standard routers allow setting this property in their constructor or factory method, custom routers
have to implement the method in a suitable way.
val router: ActorRef = system.actorOf(
// “head” router actor will run on "router-dispatcher" dispatcher
// Worker routees will run on "pool-dispatcher" dispatcher
RandomPool(5, routerDispatcher = "router-dispatcher").props(Props[Worker]),
name = "poolWithDispatcher")
Note:
It is not allowed to configure the routerDispatcher to be a akka.dispatch.
BalancingDispatcherConfigurator since the messages meant for the special router actor cannot be
processed by any other actor.
4.7 FSM
4.7.1 Overview
The FSM (Finite State Machine) is available as a mixin for the Akka Actor and is best described in the Erlang
design principles
A FSM can be described as a set of relations of the form:
State(S) x Event(E) -> Actions (A), State(S’)
These relations are interpreted as meaning:
If we are in state S and the event E occurs, we should perform the actions A and make a transition to
the state S’.
4.7. FSM
177
Akka Scala Documentation, Release 2.5.0-RC1
4.7.2 A Simple Example
To demonstrate most of the features of the FSM trait, consider an actor which shall receive and queue messages
while they arrive in a burst and send them on after the burst ended or a flush request is received.
First, consider all of the below to use these import statements:
import akka.actor.{ ActorRef, FSM }
import scala.concurrent.duration._
The contract of our “Buncher” actor is that it accepts or produces the following messages:
// received events
final case class SetTarget(ref: ActorRef)
final case class Queue(obj: Any)
case object Flush
// sent events
final case class Batch(obj: immutable.Seq[Any])
SetTarget is needed for starting it up, setting the destination for the Batches to be passed on; Queue will
add to the internal queue while Flush will mark the end of a burst.
// states
sealed trait State
case object Idle extends State
case object Active extends State
sealed trait Data
case object Uninitialized extends Data
final case class Todo(target: ActorRef, queue: immutable.Seq[Any]) extends Data
The actor can be in two states: no message queued (aka Idle) or some message queued (aka Active). It will
stay in the active state as long as messages keep arriving and no flush is requested. The internal state data of the
actor is made up of the target actor reference to send the batches to and the actual queue of messages.
Now let’s take a look at the skeleton for our FSM actor:
class Buncher extends FSM[State, Data] {
startWith(Idle, Uninitialized)
when(Idle) {
case Event(SetTarget(ref), Uninitialized) =>
stay using Todo(ref, Vector.empty)
}
// transition elided ...
when(Active, stateTimeout = 1 second) {
case Event(Flush | StateTimeout, t: Todo) =>
goto(Idle) using t.copy(queue = Vector.empty)
}
// unhandled elided ...
initialize()
}
The basic strategy is to declare the actor, mixing in the FSM trait and specifying the possible states and data values
as type parameters. Within the body of the actor a DSL is used for declaring the state machine:
• startWith defines the initial state and initial data
4.7. FSM
178
Akka Scala Documentation, Release 2.5.0-RC1
• then there is one when(<state>) { ... } declaration per state to be handled (could potentially be
multiple ones, the passed PartialFunction will be concatenated using orElse)
• finally starting it up using initialize, which performs the transition into the initial state and sets up
timers (if required).
In this case, we start out in the Idle and Uninitialized state, where only the SetTarget() message
is handled; stay prepares to end this event’s processing for not leaving the current state, while the using
modifier makes the FSM replace the internal state (which is Uninitialized at this point) with a fresh Todo()
object containing the target actor reference. The Active state has a state timeout declared, which means that
if no message is received for 1 second, a FSM.StateTimeout message will be generated. This has the same
effect as receiving the Flush command in this case, namely to transition back into the Idle state and resetting
the internal queue to the empty vector. But how do messages get queued? Since this shall work identically in
both states, we make use of the fact that any event which is not handled by the when() block is passed to the
whenUnhandled() block:
whenUnhandled {
// common code for both states
case Event(Queue(obj), t @ Todo(_, v)) =>
goto(Active) using t.copy(queue = v :+ obj)
case Event(e, s) =>
log.warning("received unhandled request {} in state {}/{}", e, stateName, s)
stay
}
The first case handled here is adding Queue() requests to the internal queue and going to the Active state
(this does the obvious thing of staying in the Active state if already there), but only if the FSM data are not
Uninitialized when the Queue() event is received. Otherwise—and in all other non-handled cases—the
second case just logs a warning and does not change the internal state.
The only missing piece is where the Batches are actually sent to the target, for which we use the
onTransition mechanism: you can declare multiple such blocks and all of them will be tried for matching
behavior in case a state transition occurs (i.e. only when the state actually changes).
onTransition {
case Active -> Idle =>
stateData match {
case Todo(ref, queue) => ref ! Batch(queue)
case _
=> // nothing to do
}
}
The transition callback is a partial function which takes as input a pair of states—the current and the next state. The
FSM trait includes a convenience extractor for these in form of an arrow operator, which conveniently reminds
you of the direction of the state change which is being matched. During the state change, the old state data is
available via stateData as shown, and the new state data would be available as nextStateData.
Note: Same-state transitions can be implemented (when currently in state S) using goto(S) or stay().
The difference between those being that goto(S) will emit an event S->S event that can be handled by
onTransition, whereas stay() will not.
To verify that this buncher actually works, it is quite easy to write a test using the Testing Actor Systems, which is
conveniently bundled with ScalaTest traits into AkkaSpec:
import akka.actor.Props
import scala.collection.immutable
object FSMDocSpec {
// messages and data types
}
4.7. FSM
179
Akka Scala Documentation, Release 2.5.0-RC1
class FSMDocSpec extends MyFavoriteTestFrameWorkPlusAkkaTestKit {
import FSMDocSpec._
// fsm code elided ...
"simple finite state machine" must {
"demonstrate NullFunction" in {
class A extends FSM[Int, Null] {
val SomeState = 0
when(SomeState)(FSM.NullFunction)
}
}
"batch correctly" in {
val buncher = system.actorOf(Props(classOf[Buncher], this))
buncher ! SetTarget(testActor)
buncher ! Queue(42)
buncher ! Queue(43)
expectMsg(Batch(immutable.Seq(42, 43)))
buncher ! Queue(44)
buncher ! Flush
buncher ! Queue(45)
expectMsg(Batch(immutable.Seq(44)))
expectMsg(Batch(immutable.Seq(45)))
}
"not batch if uninitialized" in {
val buncher = system.actorOf(Props(classOf[Buncher], this))
buncher ! Queue(42)
expectNoMsg
}
}
}
4.7.3 Reference
The FSM Trait and Object
The FSM trait inherits directly from Actor, when you extend FSM you must be aware that an actor is actually
created:
class Buncher extends FSM[State, Data] {
// fsm body ...
initialize()
}
Note: The FSM trait defines a receive method which handles internal messages and passes everything else
through to the FSM logic (according to the current state). When overriding the receive method, keep in mind
that e.g. state timeout handling depends on actually passing the messages through the FSM logic.
The FSM trait takes two type parameters:
1. the supertype of all state names, usually a sealed trait with case objects extending it,
2. the type of the state data which are tracked by the FSM module itself.
4.7. FSM
180
Akka Scala Documentation, Release 2.5.0-RC1
Note: The state data together with the state name describe the internal state of the state machine; if you stick to
this scheme and do not add mutable fields to the FSM class you have the advantage of making all changes of the
internal state explicit in a few well-known places.
Defining States
A state is defined by one or more invocations of the method
when(<name>[, stateTimeout = <timeout>])(stateFunction).
The given name must be an object which is type-compatible with the first type parameter given to the FSM trait.
This object is used as a hash key, so you must ensure that it properly implements equals and hashCode; in
particular it must not be mutable. The easiest fit for these requirements are case objects.
If the stateTimeout parameter is given, then all transitions into this state, including staying, receive this timeout by default. Initiating the transition with an explicit timeout may be used to override this default, see Initiating
Transitions for more information. The state timeout of any state may be changed during action processing with
setStateTimeout(state, duration). This enables runtime configuration e.g. via external message.
The stateFunction argument is a PartialFunction[Event, State], which is conveniently given
using the partial function literal syntax as demonstrated below:
when(Idle) {
case Event(SetTarget(ref), Uninitialized) =>
stay using Todo(ref, Vector.empty)
}
when(Active, stateTimeout = 1 second) {
case Event(Flush | StateTimeout, t: Todo) =>
goto(Idle) using t.copy(queue = Vector.empty)
}
The Event(msg: Any, data: D) case class is parameterized with the data type held by the FSM for convenient pattern matching.
Warning: It is required that you define handlers for each of the possible FSM states, otherwise there will be
failures when trying to switch to undeclared states.
It is recommended practice to declare the states as objects extending a sealed trait and then verify that there is a
when clause for each of the states. If you want to leave the handling of a state “unhandled” (more below), it still
needs to be declared like this:
when(SomeState)(FSM.NullFunction)
Defining the Initial State
Each FSM needs a starting point, which is declared using
startWith(state, data[, timeout])
The optionally given timeout argument overrides any specification given for the desired initial state. If you want
to cancel a default timeout, use None.
Unhandled Events
If a state doesn’t handle a received event a warning is logged. If you want to do something else in this case you
can specify that with whenUnhandled(stateFunction):
4.7. FSM
181
Akka Scala Documentation, Release 2.5.0-RC1
whenUnhandled {
case Event(x: X, data) =>
log.info("Received unhandled event: " + x)
stay
case Event(msg, _) =>
log.warning("Received unknown event: " + msg)
goto(Error)
}
Within this handler the state of the FSM may be queried using the stateName method.
IMPORTANT: This handler is not stacked, meaning that each invocation of whenUnhandled replaces the
previously installed handler.
Initiating Transitions
The result of any stateFunction must be a definition of the next state unless terminating the FSM, which is
described in Termination from Inside. The state definition can either be the current state, as described by the stay
directive, or it is a different state as given by goto(state). The resulting object allows further qualification by
way of the modifiers described in the following:
• forMax(duration)
This modifier sets a state timeout on the next state. This means that a timer is started which upon expiry
sends a StateTimeout message to the FSM. This timer is canceled upon reception of any other message
in the meantime; you can rely on the fact that the StateTimeout message will not be processed after an
intervening message.
This modifier can also be used to override any default timeout which is specified for the target state. If you
want to cancel the default timeout, use Duration.Inf.
• using(data)
This modifier replaces the old state data with the new data given. If you follow the advice above, this is the
only place where internal state data are ever modified.
• replying(msg)
This modifier sends a reply to the currently processed message and otherwise does not modify the state
transition.
All modifiers can be chained to achieve a nice and concise description:
when(SomeState) {
case Event(msg, _) =>
goto(Processing) using (newData) forMax (5 seconds) replying (WillDo)
}
The parentheses are not actually needed in all cases, but they visually distinguish between modifiers and their
arguments and therefore make the code even more pleasant to read for foreigners.
Note: Please note that the return statement may not be used in when blocks or similar; this is a Scala
restriction. Either refactor your code using if () ... else ... or move it into a method definition.
Monitoring Transitions
Transitions occur “between states” conceptually, which means after any actions you have put into the event handling block; this is obvious since the next state is only defined by the value returned by the event handling logic.
You do not need to worry about the exact order with respect to setting the internal state variable, as everything
within the FSM actor is running single-threaded anyway.
4.7. FSM
182
Akka Scala Documentation, Release 2.5.0-RC1
Internal Monitoring
Up to this point, the FSM DSL has been centered on states and events. The dual view is to describe it as a series
of transitions. This is enabled by the method
onTransition(handler)
which associates actions with a transition instead of with a state and event. The handler is a partial function which
takes a pair of states as input; no resulting state is needed as it is not possible to modify the transition in progress.
onTransition {
case Idle -> Active => setTimer("timeout", Tick, 1 second, repeat = true)
case Active -> _
=> cancelTimer("timeout")
case x -> Idle
=> log.info("entering Idle from " + x)
}
The convenience extractor -> enables decomposition of the pair of states with a clear visual reminder of the
transition’s direction. As usual in pattern matches, an underscore may be used for irrelevant parts; alternatively
you could bind the unconstrained state to a variable, e.g. for logging as shown in the last case.
It is also possible to pass a function object accepting two states to onTransition, in case your transition
handling logic is implemented as a method:
onTransition(handler _)
def handler(from: StateType, to: StateType) {
// handle it here ...
}
The handlers registered with this method are stacked, so you can intersperse onTransition blocks with when
blocks as suits your design. It should be noted, however, that all handlers will be invoked for each transition,
not only the first matching one. This is designed specifically so you can put all transition handling for a certain
aspect into one place without having to worry about earlier declarations shadowing later ones; the actions are still
executed in declaration order, though.
Note: This kind of internal monitoring may be used to structure your FSM according to transitions, so that for
example the cancellation of a timer upon leaving a certain state cannot be forgot when adding new target states.
External Monitoring
External actors may be registered to be notified of state transitions by sending a message SubscribeTransitionCallBack(actorRef).
The named actor will be sent a
CurrentState(self, stateName) message immediately and will receive Transition(actorRef,
oldState, newState) messages whenever a state change is triggered.
Please note that a state change includes the action of performing an goto(S), while already being state S. In
that case the monitoring actor will be notified with an Transition(ref,S,S) message. This may be useful
if your FSM should react on all (also same-state) transitions. In case you’d rather not emit events for same-state
transitions use stay() instead of goto(S).
External monitors may be unregistered by sending UnsubscribeTransitionCallBack(actorRef) to
the FSM actor.
Stopping a listener without unregistering will not remove the listener from the subscription list; use
UnsubscribeTransitionCallback before stopping the listener.
4.7. FSM
183
Akka Scala Documentation, Release 2.5.0-RC1
Transforming State
The partial functions supplied as argument to the when() blocks can be transformed using Scala’s full supplement
of functional programming tools. In order to retain type inference, there is a helper function which may be used
in case some common handling logic shall be applied to different clauses:
when(SomeState)(transform {
case Event(bytes: ByteString, read) => stay using (read + bytes.length)
} using {
case s @ FSM.State(state, read, timeout, stopReason, replies) if read > 1000 =>
goto(Processing)
})
It goes without saying that the arguments to this method may also be stored, to be used several times, e.g. when
applying the same transformation to several when() blocks:
val processingTrigger: PartialFunction[State, State] = {
case s @ FSM.State(state, read, timeout, stopReason, replies) if read > 1000 =>
goto(Processing)
}
when(SomeState)(transform {
case Event(bytes: ByteString, read) => stay using (read + bytes.length)
} using processingTrigger)
Timers
Besides state timeouts, FSM manages timers identified by String names. You may set a timer using
setTimer(name, msg, interval, repeat)
where msg is the message object which will be sent after the duration interval has elapsed. If repeat is
true, then the timer is scheduled at fixed rate given by the interval parameter. Any existing timer with the
same name will automatically be canceled before adding the new timer.
Timers may be canceled using
cancelTimer(name)
which is guaranteed to work immediately, meaning that the scheduled message will not be processed after this call
even if the timer already fired and queued it. The status of any timer may be inquired with
isTimerActive(name)
These named timers complement state timeouts because they are not affected by intervening reception of other
messages.
Termination from Inside
The FSM is stopped by specifying the result state as
stop([reason[, data]])
The reason must be one of Normal (which is the default), Shutdown or Failure(reason), and the second
argument may be given to change the state data which is available during termination handling.
Note: It should be noted that stop does not abort the actions and stop the FSM immediately. The stop action
must be returned from the event handler in the same way as a state transition (but note that the return statement
may not be used within a when block).
4.7. FSM
184
Akka Scala Documentation, Release 2.5.0-RC1
when(Error) {
case Event("stop", _) =>
// do cleanup ...
stop()
}
You can use onTermination(handler) to specify custom code that is executed when the FSM is stopped.
The handler is a partial function which takes a StopEvent(reason, stateName, stateData) as argument:
onTermination {
case StopEvent(FSM.Normal, state, data)
=> // ...
case StopEvent(FSM.Shutdown, state, data)
=> // ...
case StopEvent(FSM.Failure(cause), state, data) => // ...
}
As for the whenUnhandled case, this handler is not stacked, so each invocation of onTermination replaces
the previously installed handler.
Termination from Outside
When an ActorRef associated to a FSM is stopped using the stop method, its postStop hook will be
executed. The default implementation by the FSM trait is to execute the onTermination handler if that is
prepared to handle a StopEvent(Shutdown, ...).
Warning: In case you override postStop and want to have your onTermination handler called, do not
forget to call super.postStop.
4.7.4 Testing and Debugging Finite State Machines
During development and for trouble shooting FSMs need care just as any other actor. There are specialized tools
available as described in Testing Finite State Machines and in the following.
Event Tracing
The setting akka.actor.debug.fsm in Configuration enables logging of an event trace by LoggingFSM
instances:
import akka.actor.LoggingFSM
class MyFSM extends LoggingFSM[StateType, Data] {
// body elided ...
}
This FSM will log at DEBUG level:
• all processed events, including StateTimeout and scheduled timer messages
• every setting and cancellation of named timers
• all state transitions
Life cycle changes and special messages can be logged as described for Actors.
Rolling Event Log
The LoggingFSM trait adds one more feature to the FSM: a rolling event log which may be used during debugging (for tracing how the FSM entered a certain failure state) or for other creative uses:
4.7. FSM
185
Akka Scala Documentation, Release 2.5.0-RC1
import akka.actor.LoggingFSM
class MyFSM extends LoggingFSM[StateType, Data] {
override def logDepth = 12
onTermination {
case StopEvent(FSM.Failure(_), state, data) =>
val lastEvents = getLog.mkString("\n\t")
log.warning("Failure in state " + state + " with data " + data + "\n" +
"Events leading up to this point:\n\t" + lastEvents)
}
// ...
}
The logDepth defaults to zero, which turns off the event log.
Warning: The log buffer is allocated during actor creation, which is why the configuration is done using a
virtual method call. If you want to override with a val, make sure that its initialization happens before the
initializer of LoggingFSM runs, and do not change the value returned by logDepth after the buffer has
been allocated.
The contents of the event log are available using method getLog, which returns an IndexedSeq[LogEntry]
where the oldest entry is at index zero.
4.7.5 Examples
A bigger FSM example contrasted with Actor’s become/unbecome can be found in the Lightbend Activator
template named Akka FSM in Scala
4.8 Persistence
Akka persistence enables stateful actors to persist their internal state so that it can be recovered when an actor is
started, restarted after a JVM crash or by a supervisor, or migrated in a cluster. The key concept behind Akka
persistence is that only changes to an actor’s internal state are persisted but never its current state directly (except
for optional snapshots). These changes are only ever appended to storage, nothing is ever mutated, which allows
for very high transaction rates and efficient replication. Stateful actors are recovered by replaying stored changes
to these actors from which they can rebuild internal state. This can be either the full history of changes or starting
from a snapshot which can dramatically reduce recovery times. Akka persistence also provides point-to-point
communication with at-least-once message delivery semantics.
Akka persistence is inspired by and the official replacement of the eventsourced library. It follows the same
concepts and architecture of eventsourced but significantly differs on API and implementation level. See also
Migration Guide Eventsourced to Akka Persistence 2.3.x
4.8.1 Dependencies
Akka persistence is a separate jar file. Make sure that you have the following dependency in your project:
"com.typesafe.akka" %% "akka-persistence" % "2.5.0-RC1"
The Akka persistence extension comes with few built-in persistence plugins, including in-memory heap based
journal, local file-system based snapshot-store and LevelDB based journal.
LevelDB based plugins will require the following additional dependency declaration:
"org.iq80.leveldb"
"org.fusesource.leveldbjni"
4.8. Persistence
% "leveldb"
% "leveldbjni-all"
% "0.7"
% "1.8"
186
Akka Scala Documentation, Release 2.5.0-RC1
4.8.2 Architecture
• PersistentActor: Is a persistent, stateful actor. It is able to persist events to a journal and can react to them
in a thread-safe manner. It can be used to implement both command as well as event sourced actors. When
a persistent actor is started or restarted, journaled messages are replayed to that actor so that it can recover
internal state from these messages.
• PersistentView: A view is a persistent, stateful actor that receives journaled messages that have been written
by another persistent actor. A view itself does not journal new messages, instead, it updates internal state
only from a persistent actor’s replicated message stream.
• AtLeastOnceDelivery: To send messages with at-least-once delivery semantics to destinations, also in case
of sender and receiver JVM crashes.
• AsyncWriteJournal: A journal stores the sequence of messages sent to a persistent actor. An application
can control which messages are journaled and which are received by the persistent actor without being
journaled. Journal maintains highestSequenceNr that is increased on each message. The storage backend
of a journal is pluggable. The persistence extension comes with a “leveldb” journal plugin, which writes to
the local filesystem. Replicated journals are available as Community plugins.
• Snapshot store: A snapshot store persists snapshots of a persistent actor’s or a view’s internal state. Snapshots are used for optimizing recovery times. The storage backend of a snapshot store is pluggable. The
persistence extension comes with a “local” snapshot storage plugin, which writes to the local filesystem.
Replicated snapshot stores are available as Community plugins.
4.8.3 Event sourcing
The basic idea behind Event Sourcing is quite simple. A persistent actor receives a (non-persistent) command
which is first validated if it can be applied to the current state. Here validation can mean anything, from simple
inspection of a command message’s fields up to a conversation with several external services, for example. If
validation succeeds, events are generated from the command, representing the effect of the command. These
events are then persisted and, after successful persistence, used to change the actor’s state. When the persistent
actor needs to be recovered, only the persisted events are replayed of which we know that they can be successfully
applied. In other words, events cannot fail when being replayed to a persistent actor, in contrast to commands.
Event sourced actors may of course also process commands that do not change application state such as query
commands for example.
Akka persistence supports event sourcing with the PersistentActor trait. An actor that extends this trait
uses the persist method to persist and handle events. The behavior of a PersistentActor is defined by
implementing receiveRecover and receiveCommand. This is demonstrated in the following example.
import akka.actor._
import akka.persistence._
case class Cmd(data: String)
case class Evt(data: String)
case class ExampleState(events: List[String] = Nil) {
def updated(evt: Evt): ExampleState = copy(evt.data :: events)
def size: Int = events.length
override def toString: String = events.reverse.toString
}
class ExamplePersistentActor extends PersistentActor {
override def persistenceId = "sample-id-1"
var state = ExampleState()
def updateState(event: Evt): Unit =
state = state.updated(event)
4.8. Persistence
187
Akka Scala Documentation, Release 2.5.0-RC1
def numEvents =
state.size
val receiveRecover: Receive = {
case evt: Evt
=> updateState(evt)
case SnapshotOffer(_, snapshot: ExampleState) => state = snapshot
}
val snapShotInterval = 1000
val receiveCommand: Receive = {
case Cmd(data) =>
persist(Evt(s"${data}-${numEvents}")) { event =>
updateState(event)
context.system.eventStream.publish(event)
if (lastSequenceNr % snapShotInterval == 0 && lastSequenceNr != 0)
saveSnapshot(state)
}
case "print" => println(state)
}
}
The example defines two data types, Cmd and Evt to represent commands and events, respectively. The state
of the ExamplePersistentActor is a list of persisted event data contained in ExampleState.
The persistent actor’s receiveRecover method defines how state is updated during recovery by handling
Evt and SnapshotOffer messages. The persistent actor’s receiveCommand method is a command handler.
In this example, a command is handled by generating two events which are then persisted and handled. Events are
persisted by calling persist with an event (or a sequence of events) as first argument and an event handler as
second argument.
The persist method persists events asynchronously and the event handler is executed for successfully persisted
events. Successfully persisted events are internally sent back to the persistent actor as individual messages that
trigger event handler executions. An event handler may close over persistent actor state and mutate it. The sender
of a persisted event is the sender of the corresponding command. This allows event handlers to reply to the sender
of a command (not shown).
The main responsibility of an event handler is changing persistent actor state using event data and notifying others
about successful state changes by publishing events.
When persisting events with persist it is guaranteed that the persistent actor will not receive further commands
between the persist call and the execution(s) of the associated event handler. This also holds for multiple
persist calls in context of a single command. Incoming messages are stashed until the persist is completed.
If persistence of an event fails, onPersistFailure will be invoked (logging the error by default), and the actor
will unconditionally be stopped. If persistence of an event is rejected before it is stored, e.g. due to serialization
error, onPersistRejected will be invoked (logging a warning by default) and the actor continues with the
next message.
The easiest way to run this example yourself is to download Lightbend Activator and open the tutorial named
Akka Persistence Samples with Scala. It contains instructions on how to run the PersistentActorExample.
Note: It’s also possible to switch between different command handlers during normal processing and recovery
with context.become() and context.unbecome(). To get the actor into the same state after recovery you need to take special care to perform the same state transitions with become and unbecome in the
receiveRecover method as you would have done in the command handler. Note that when using become
from receiveRecover it will still only use the receiveRecover behavior when replaying the events. When
replay is completed it will use the new behavior.
4.8. Persistence
188
Akka Scala Documentation, Release 2.5.0-RC1
Identifiers
A persistent actor must have an identifier that doesn’t change across different actor incarnations. The identifier
must be defined with the persistenceId method.
override def persistenceId = "my-stable-persistence-id"
Note: persistenceId must be unique to a given entity in the journal (database table/keyspace). When
replaying messages persisted to the journal, you query messages with a persistenceId. So, if two different
entities share the same persistenceId, message-replaying behavior is corrupted.
Recovery
By default, a persistent actor is automatically recovered on start and on restart by replaying journaled messages.
New messages sent to a persistent actor during recovery do not interfere with replayed messages. They are cached
and received by a persistent actor after recovery phase completes.
Note: Accessing the sender() for replayed messages will always result in a deadLetters reference, as the
original sender is presumed to be long gone. If you indeed have to notify an actor during recovery in the future,
store its ActorPath explicitly in your persisted events.
Recovery customization
Applications may also customise how recovery is performed by returning a customised Recovery object in the
recovery method of a PersistentActor,
To skip loading snapshots and replay all events you can use SnapshotSelectionCriteria.None. This
can be useful if snapshot serialization format has changed in an incompatible way. It should typically not be used
when events have been deleted.
override def recovery =
Recovery(fromSnapshot = SnapshotSelectionCriteria.None)
Another example, which can be fun for experiments but probably not in a real application, is setting an upper
bound to the replay which allows the actor to be replayed to a certain point “in the past” instead to its most up
to date state. Note that after that it is a bad idea to persist new events because a later recovery will probably be
confused by the new events that follow the events that were previously skipped.
override def recovery = Recovery(toSequenceNr = 457L)
Recovery can be disabled by returning Recovery.none() in the recovery method of a
PersistentActor:
override def recovery = Recovery.none
Recovery status
A persistent actor can query its own recovery status via the methods
def recoveryRunning: Boolean
def recoveryFinished: Boolean
4.8. Persistence
189
Akka Scala Documentation, Release 2.5.0-RC1
Sometimes there is a need for performing additional initialization when the recovery has completed before processing any other message sent to the persistent actor. The persistent actor will receive a special
RecoveryCompleted message right after recovery and before any other received messages.
override def receiveRecover: Receive = {
case RecoveryCompleted =>
// perform init after recovery, before any other messages
//...
case evt
=> //...
}
override def receiveCommand: Receive = {
case msg => //...
}
The actor will always receive a RecoveryCompleted message, even if there are no events in the journal and
the snapshot store is empty, or if it’s a new persistent actor with a previously unused persistenceId.
If there is a problem with recovering the state of the actor from the journal, onRecoveryFailure is called
(logging the error by default) and the actor will be stopped.
Internal stash
The persistent actor has a private stash for internally caching incoming messages during recovery or the
persist\persistAll method persisting events. You can still use/inherit from the Stash interface. The
internal stash cooperates with the normal stash by hooking into unstashAll method and making sure messages
are unstashed properly to the internal stash to maintain ordering guarantees.
You should be careful to not send more messages to a persistent actor than it can keep up with, otherwise the
number of stashed messages will grow without bounds. It can be wise to protect against OutOfMemoryError
by defining a maximum stash capacity in the mailbox configuration:
akka.actor.default-mailbox.stash-capacity=10000
Note that the stash capacity is per actor. If you have many persistent actors, e.g. when using cluster
sharding, you may need to define a small stash capacity to ensure that the total number of stashed messages in the system doesn’t consume too much memory. Additionally, the persistent actor defines three
strategies to handle failure when the internal stash capacity is exceeded. The default overflow strategy is
the ThrowOverflowExceptionStrategy, which discards the current received message and throws a
StashOverflowException, causing actor restart if the default supervision strategy is used. You can override the internalStashOverflowStrategy method to return DiscardToDeadLetterStrategy or
ReplyToStrategy for any “individual” persistent actor, or define the “default” for all persistent actors by providing FQCN, which must be a subclass of StashOverflowStrategyConfigurator, in the persistence
configuration:
akka.persistence.internal-stash-overflow-strategy=
"akka.persistence.ThrowExceptionConfigurator"
The DiscardToDeadLetterStrategy strategy also has a pre-packaged companion configurator akka.
persistence.DiscardConfigurator.
You can also query the default strategy via the Akka persistence extension singleton:
Persistence(context.system).defaultInternalStashOverflowStrategy
Note: The bounded mailbox should be avoided in the persistent actor, by which the messages come from storage
backends may be discarded. You can use bounded stash instead of it.
4.8. Persistence
190
Akka Scala Documentation, Release 2.5.0-RC1
Relaxed local consistency requirements and high throughput use-cases
If faced with relaxed local consistency requirements and high throughput demands sometimes
PersistentActor and its persist may not be enough in terms of consuming incoming Commands
at a high rate, because it has to wait until all Events related to a given Command are processed in order to start
processing the next Command. While this abstraction is very useful for most cases, sometimes you may be faced
with relaxed requirements about consistency – for example you may want to process commands as fast as you
can, assuming that the Event will eventually be persisted and handled properly in the background, retroactively
reacting to persistence failures if needed.
The persistAsync method provides a tool for implementing high-throughput persistent actors. It will not stash
incoming Commands while the Journal is still working on persisting and/or user code is executing event callbacks.
In the below example, the event callbacks may be called “at any time”, even after the next Command has been
processed. The ordering between events is still guaranteed (“evt-b-1” will be sent after “evt-a-2”, which will be
sent after “evt-a-1” etc.).
class MyPersistentActor extends PersistentActor {
override def persistenceId = "my-stable-persistence-id"
override def receiveRecover: Receive = {
case _ => // handle recovery here
}
override def receiveCommand: Receive = {
case c: String => {
sender() ! c
persistAsync(s"evt-$c-1") { e => sender() ! e }
persistAsync(s"evt-$c-2") { e => sender() ! e }
}
}
}
// usage
persistentActor ! "a"
persistentActor ! "b"
//
//
//
//
//
//
//
possible order of received messages:
a
b
evt-a-1
evt-a-2
evt-b-1
evt-b-2
Note: In order to implement the pattern known as “command sourcing” simply call persistAsync(cmd)(.
..) right away on all incoming messages and handle them in the callback.
Warning: The callback will not be invoked if the actor is restarted (or stopped) in between the call to
persistAsync and the journal has confirmed the write.
Deferring actions until preceding persist handlers have executed
Sometimes when working with persistAsync you may find that it would be nice to define some actions in
terms of ‘’happens-after the previous persistAsync handlers have been invoked’‘. PersistentActor
provides an utility method called deferAsync, which works similarly to persistAsync yet does not persist
4.8. Persistence
191
Akka Scala Documentation, Release 2.5.0-RC1
the passed in event. It is recommended to use it for read operations, and actions which do not have corresponding
events in your domain model.
Using this method is very similar to the persist family of methods, yet it does not persist the passed in event. It
will be kept in memory and used when invoking the handler.
class MyPersistentActor extends PersistentActor {
override def persistenceId = "my-stable-persistence-id"
override def receiveRecover: Receive = {
case _ => // handle recovery here
}
override def receiveCommand: Receive = {
case c: String => {
sender() ! c
persistAsync(s"evt-$c-1") { e => sender() ! e }
persistAsync(s"evt-$c-2") { e => sender() ! e }
deferAsync(s"evt-$c-3") { e => sender() ! e }
}
}
}
Notice that the sender() is safe to access in the handler callback, and will be pointing to the original sender of
the command for which this deferAsync handler was called.
The calling side will get the responses in this (guaranteed) order:
persistentActor ! "a"
persistentActor ! "b"
//
//
//
//
//
//
//
//
//
order of received messages:
a
b
evt-a-1
evt-a-2
evt-a-3
evt-b-1
evt-b-2
evt-b-3
Warning: The callback will not be invoked if the actor is restarted (or stopped) in between the call to
deferAsync and the journal has processed and confirmed all preceding writes.
Nested persist calls
It is possible to call persist and persistAsync inside their respective callback blocks and they will properly
retain both the thread safety (including the right value of sender()) as well as stashing guarantees.
In general it is encouraged to create command handlers which do not need to resort to nested event persisting,
however there are situations where it may be useful. It is important to understand the ordering of callback execution
in those situations, as well as their implication on the stashing behaviour (that persist() enforces). In the
following example two persist calls are issued, and each of them issues another persist inside its callback:
override def receiveCommand: Receive = {
case c: String =>
sender() ! c
persist(s"$c-1-outer") { outer1 =>
4.8. Persistence
192
Akka Scala Documentation, Release 2.5.0-RC1
sender() ! outer1
persist(s"$c-1-inner") { inner1 =>
sender() ! inner1
}
}
persist(s"$c-2-outer") { outer2 =>
sender() ! outer2
persist(s"$c-2-inner") { inner2 =>
sender() ! inner2
}
}
}
When sending two commands to this PersistentActor, the persist handlers will be executed in the following
order:
persistentActor ! "a"
persistentActor ! "b"
//
//
//
//
//
//
//
//
//
//
//
//
order of received messages:
a
a-outer-1
a-outer-2
a-inner-1
a-inner-2
and only then process "b"
b
b-outer-1
b-outer-2
b-inner-1
b-inner-2
First the “outer layer” of persist calls is issued and their callbacks are applied. After these have successfully
completed, the inner callbacks will be invoked (once the events they are persisting have been confirmed to be
persisted by the journal). Only after all these handlers have been successfully invoked will the next command be
delivered to the persistent Actor. In other words, the stashing of incoming commands that is guaranteed by initially
calling persist() on the outer layer is extended until all nested persist callbacks have been handled.
It is also possible to nest persistAsync calls, using the same pattern:
override def receiveCommand: Receive = {
case c: String =>
sender() ! c
persistAsync(c + "-outer-1") { outer =>
sender() ! outer
persistAsync(c + "-inner-1") { inner => sender() ! inner }
}
persistAsync(c + "-outer-2") { outer =>
sender() ! outer
persistAsync(c + "-inner-2") { inner => sender() ! inner }
}
}
In this case no stashing is happening, yet events are still persisted and callbacks are executed in the expected order:
persistentActor ! "a"
persistentActor ! "b"
// order of received messages:
// a
// b
4.8. Persistence
193
Akka Scala Documentation, Release 2.5.0-RC1
//
//
//
//
//
//
//
//
a-outer-1
a-outer-2
b-outer-1
b-outer-2
a-inner-1
a-inner-2
b-inner-1
b-inner-2
// which can be seen as the following causal relationship:
// a -> a-outer-1 -> a-outer-2 -> a-inner-1 -> a-inner-2
// b -> b-outer-1 -> b-outer-2 -> b-inner-1 -> b-inner-2
While it is possible to nest mixed persist and persistAsync with keeping their respective semantics it is
not a recommended practice, as it may lead to overly complex nesting.
Warning: While it is possible to nest persist calls within one another, it is not legal call persist
from any other Thread than the Actors message processing Thread. For example, it is not legal to call
persist from Futures! Doing so will break the guarantees that the persist methods aim to provide. Always call persist and persistAsync from within the Actor’s receive block (or methods synchronously
invoked from there).
Failures
If persistence of an event fails, onPersistFailure will be invoked (logging the error by default), and the
actor will unconditionally be stopped.
The reason that it cannot resume when persist fails is that it is unknown if the event was actually persisted or
not, and therefore it is in an inconsistent state. Restarting on persistent failures will most likely fail anyway since
the journal is probably unavailable. It is better to stop the actor and after a back-off timeout start it again. The
akka.pattern.BackoffSupervisor actor is provided to support such restarts.
val childProps = Props[MyPersistentActor]
val props = BackoffSupervisor.props(
Backoff.onStop(
childProps,
childName = "myActor",
minBackoff = 3.seconds,
maxBackoff = 30.seconds,
randomFactor = 0.2))
context.actorOf(props, name = "mySupervisor")
If persistence of an event is rejected before it is stored, e.g. due to serialization error, onPersistRejected
will be invoked (logging a warning by default), and the actor continues with next message.
If there is a problem with recovering the state of the actor from the journal when the actor is started,
onRecoveryFailure is called (logging the error by default), and the actor will be stopped. Note that failure to load snapshot is also treated like this, but you can disable loading of snapshots if you for example know that
serialization format has changed in an incompatible way, see Recovery customization.
Atomic writes
Each event is of course stored atomically, but it is also possible to store several events atomically by using the
persistAll or persistAllAsync method. That means that all events passed to that method are stored or
none of them are stored if there is an error.
The recovery of a persistent actor will therefore never be done partially with only a subset of events persisted by
persistAll.
4.8. Persistence
194
Akka Scala Documentation, Release 2.5.0-RC1
Some journals may not support atomic writes of several events and they will then reject the
persistAll command, i.e.
onPersistRejected is called with an exception (typically
UnsupportedOperationException).
Batch writes
In order to optimize throughput when using persistAsync, a persistent actor internally batches events to be
stored under high load before writing them to the journal (as a single batch). The batch size is dynamically
determined by how many events are emitted during the time of a journal round-trip: after sending a batch to the
journal no further batch can be sent before confirmation has been received that the previous batch has been written.
Batch writes are never timer-based which keeps latencies at a minimum.
Message deletion
It is possible to delete all messages (journaled by a single persistent actor) up to a specified sequence number;
Persistent actors may call the deleteMessages method to this end.
Deleting messages in event sourcing based applications is typically either not used at all, or used in conjunction
with snapshotting, i.e. after a snapshot has been successfully stored, a deleteMessages(toSequenceNr)
up until the sequence number of the data held by that snapshot can be issued to safely delete the previous events
while still having access to the accumulated state during replays - by loading the snapshot.
Warning: If you are using Persistence Query, query results may be missing deleted messages in a journal,
depending on how deletions are implemented in the journal plugin. Unless you use a plugin which still shows
deleted messages in persistence query results, you have to design your application so that it is not affected by
missing messages.
The result of the deleteMessages request is signaled to the persistent actor with a
DeleteMessagesSuccess message if the delete was successful or a DeleteMessagesFailure
message if it failed.
Message deletion doesn’t affect the highest sequence number of the journal, even if all messages were deleted
from it after deleteMessages invocation.
Persistence status handling
Persisting, deleting, and replaying messages can either succeed or fail.
Method
Success
Failure / Rejection
After failure handler
invoked
onPersistFailure
Actor is stopped.
persist / persistAsync
persist handler invoked
onPersistRejected No automatic actions.
recovery
RecoveryCompleted onRecoveryFailure Actor is stopped.
deleteMessages
DeleteMessagesSuccess
DeleteMessagesFailure
No automatic actions.
The most important operations (persist and recovery) have failure handlers modelled as explicit callbacks
which the user can override in the PersistentActor. The default implementations of these handlers emit
a log message (error for persist/recovery failures, and warning for others), logging the failure cause and
information about which message caused the failure.
For critical failures, such as recovery or persisting events failing, the persistent actor will be stopped after the
failure handler is invoked. This is because if the underlying journal implementation is signalling persistence
failures it is most likely either failing completely or overloaded and restarting right-away and trying to persist the
event again will most likely not help the journal recover – as it would likely cause a Thundering herd problem, as
many persistent actors would restart and try to persist their events again. Instead, using a BackoffSupervisor
(as described in Failures) which implements an exponential-backoff strategy which allows for more breathing
room for the journal to recover between restarts of the persistent actor.
4.8. Persistence
195
Akka Scala Documentation, Release 2.5.0-RC1
Note: Journal implementations may choose to implement a retry mechanism, e.g. such that only after a write
fails N number of times a persistence failure is signalled back to the user. In other words, once a journal returns a
failure, it is considered fatal by Akka Persistence, and the persistent actor which caused the failure will be stopped.
Check the documentation of the journal implementation you are using for details if/how it is using this technique.
Safely shutting down persistent actors
Special care should be given when shutting down persistent actors from the outside. With normal Actors it is often
acceptable to use the special PoisonPill message to signal to an Actor that it should stop itself once it receives
this message – in fact this message is handled automatically by Akka, leaving the target actor no way to refuse
stopping itself when given a poison pill.
This can be dangerous when used with PersistentActor due to the fact that incoming commands are
stashed while the persistent actor is awaiting confirmation from the Journal that events have been written when
persist() was used. Since the incoming commands will be drained from the Actor’s mailbox and put into its
internal stash while awaiting the confirmation (thus, before calling the persist handlers) the Actor may receive
and (auto)handle the PoisonPill before it processes the other messages which have been put into its stash,
causing a pre-mature shutdown of the Actor.
Warning: Consider using explicit shut-down messages instead of PoisonPill when working with persistent actors.
The example below highlights how messages arrive in the Actor’s mailbox and how they interact with its internal
stashing mechanism when persist() is used. Notice the early stop behaviour that occurs when PoisonPill
is used:
/** Explicit shutdown message */
case object Shutdown
class SafePersistentActor extends PersistentActor {
override def persistenceId = "safe-actor"
override def receiveCommand: Receive = {
case c: String =>
println(c)
persist(s"handle-$c") { println(_) }
case Shutdown =>
context.stop(self)
}
override def receiveRecover: Receive = {
case _ => // handle recovery here
}
}
// UN-SAFE, due to PersistentActor's command stashing:
persistentActor ! "a"
persistentActor ! "b"
persistentActor ! PoisonPill
// order of received messages:
// a
//
# b arrives at mailbox, stashing;
internal-stash = [b]
// PoisonPill is an AutoReceivedMessage, is handled automatically
// !! stop !!
// Actor is stopped without handling `b` nor the `a` handler!
4.8. Persistence
196
Akka Scala Documentation, Release 2.5.0-RC1
// SAFE:
persistentActor ! "a"
persistentActor ! "b"
persistentActor ! Shutdown
// order of received messages:
// a
//
# b arrives at mailbox, stashing;
//
# Shutdown arrives at mailbox, stashing;
// handle-a
//
# unstashing;
// b
// handle-b
//
# unstashing;
// Shutdown
// -- stop --
internal-stash = [b]
internal-stash = [b, Shutdown]
internal-stash = [Shutdown]
internal-stash = []
Replay Filter
There could be cases where event streams are corrupted and multiple writers (i.e. multiple persistent actor instances) journaled different messages with the same sequence number. In such a case, you can configure how you
filter replayed messages from multiple writers, upon recovery.
In your configuration, under the akka.persistence.journal.xxx.replay-filter section (where
xxx is your journal plugin id), you can select the replay filter mode from one of the following values:
• repair-by-discard-old
• fail
• warn
• off
For example, if you configure the replay filter for leveldb plugin, it looks like this:
# The replay filter can detect a corrupt event stream by inspecting
# sequence numbers and writerUuid when replaying events.
akka.persistence.journal.leveldb.replay-filter {
# What the filter should do when detecting invalid events.
# Supported values:
# `repair-by-discard-old` : discard events from old writers,
#
warning is logged
# `fail` : fail the replay, error is logged
# `warn` : log warning but emit events untouched
# `off` : disable this feature completely
mode = repair-by-discard-old
}
4.8.4 Snapshots
Snapshots can dramatically reduce recovery times of persistent actors and views. The following discusses snapshots in context of persistent actors but this is also applicable to persistent views.
Persistent actors can save snapshots of internal state by calling the saveSnapshot method. If saving
of a snapshot succeeds, the persistent actor receives a SaveSnapshotSuccess message, otherwise a
SaveSnapshotFailure message
var state: Any = _
val snapShotInterval = 1000
override def receiveCommand: Receive = {
4.8. Persistence
197
Akka Scala Documentation, Release 2.5.0-RC1
case SaveSnapshotSuccess(metadata)
=> // ...
case SaveSnapshotFailure(metadata, reason) => // ...
case cmd: String =>
persist(s"evt-$cmd") { e =>
updateState(e)
if (lastSequenceNr % snapShotInterval == 0 && lastSequenceNr != 0)
saveSnapshot(state)
}
}
where metadata is of type SnapshotMetadata:
final case class SnapshotMetadata(persistenceId: String, sequenceNr: Long,
˓→timestamp: Long = 0L)
During recovery, the persistent actor is offered a previously saved snapshot via a SnapshotOffer message
from which it can initialize internal state.
var state: Any = _
override def receiveRecover: Receive = {
case SnapshotOffer(metadata, offeredSnapshot) => state = offeredSnapshot
case RecoveryCompleted
=>
case event
=> // ...
}
The replayed messages that follow the SnapshotOffer message, if any, are younger than the offered snapshot.
They finally recover the persistent actor to its current (i.e. latest) state.
In general, a persistent actor is only offered a snapshot if that persistent actor has previously saved one or more
snapshots and at least one of these snapshots matches the SnapshotSelectionCriteria that can be specified for recovery.
override def recovery = Recovery(fromSnapshot = SnapshotSelectionCriteria(
maxSequenceNr = 457L,
maxTimestamp = System.currentTimeMillis))
If not specified, they default to SnapshotSelectionCriteria.Latest which selects
the latest (= youngest) snapshot.
To disable snapshot-based recovery, applications should use
SnapshotSelectionCriteria.None. A recovery where no saved snapshot matches the specified
SnapshotSelectionCriteria will replay all journaled messages.
Note: In order to use snapshots, a default snapshot-store (akka.persistence.snapshot-store.
plugin) must be configured, or the PersistentActor can pick a snapshot store explicitly by overriding
def snapshotPluginId: String.
Since it is acceptable for some applications to not use any snapshotting, it is legal to not configure a snapshot store.
However, Akka will log a warning message when this situation is detected and then continue to operate until an actor tries to store a snapshot, at which point the operation will fail (by replying with an SaveSnapshotFailure
for example).
Note that Cluster Sharding is using snapshots, so if you use Cluster Sharding you need to define a snapshot store
plugin.
Snapshot deletion
A persistent actor can delete individual snapshots by calling the deleteSnapshot method with the sequence
number of when the snapshot was taken.
4.8. Persistence
198
Akka Scala Documentation, Release 2.5.0-RC1
To bulk-delete a range of snapshots matching SnapshotSelectionCriteria, persistent actors should use
the deleteSnapshots method.
Snapshot status handling
Saving or deleting snapshots can either succeed or fail – this information is reported back to the persistent actor
via status messages as illustrated in the following table.
Method
Success
Failure message
saveSnapshot(Any)
SaveSnapshotSuccess SaveSnapshotFailure
deleteSnapshot(Long)
DeleteSnapshotSuccess
DeleteSnapshotFailure
deleteSnapshots(SnapshotSelectionCriteria)
DeleteSnapshotsSuccess
DeleteSnapshotsFailure
If failure messages are left unhandled by the actor, a default warning log message will be logged for each incoming
failure message. No default action is performed on the success messages, however you’re free to handle them e.g.
in order to delete an in memory representation of the snapshot, or in the case of failure to attempt save the snapshot
again.
4.8.5 At-Least-Once Delivery
To send messages with at-least-once delivery semantics to destinations you can mix-in
AtLeastOnceDelivery trait to your PersistentActor on the sending side. It takes care of re-sending
messages when they have not been confirmed within a configurable timeout.
The state of the sending actor, including which messages have been sent that have not been confirmed by the recipient must be persistent so that it can survive a crash of the sending actor or JVM. The AtLeastOnceDelivery
trait does not persist anything by itself. It is your responsibility to persist the intent that a message is sent and that
a confirmation has been received.
Note: At-least-once delivery implies that original message sending order is not always preserved, and the destination may receive duplicate messages. Semantics do not match those of a normal ActorRef send operation:
• it is not at-most-once delivery
• message order for the same sender–receiver pair is not preserved due to possible resends
• after a crash and restart of the destination messages are still delivered to the new actor incarnation
These semantics are similar to what an ActorPath represents (see Actor Lifecycle), therefore you need to supply
a path and not a reference when delivering messages. The messages are sent to the path with an actor selection.
Use the deliver method to send a message to a destination. Call the confirmDelivery method when the
destination has replied with a confirmation message.
Relationship between deliver and confirmDelivery
To send messages to the destination path, use the deliver method after you have persisted the intent to send the
message.
The destination actor must send back a confirmation message. When the sending actor receives this confirmation message you should persist the fact that the message was delivered successfully and then call the
confirmDelivery method.
If the persistent actor is not currently recovering, the deliver method will send the message to the destination
actor. When recovering, messages will be buffered until they have been confirmed using confirmDelivery.
Once recovery has completed, if there are outstanding messages that have not been confirmed (during the message
replay), the persistent actor will resend these before sending any other messages.
Deliver requires a deliveryIdToMessage function to pass the provided deliveryId into the message so
that the correlation between deliver and confirmDelivery is possible. The deliveryId must do the
4.8. Persistence
199
Akka Scala Documentation, Release 2.5.0-RC1
round trip. Upon receipt of the message, the destination actor will send the same‘‘deliveryId‘‘ wrapped in a
confirmation message back to the sender. The sender will then use it to call confirmDelivery method to
complete the delivery routine.
import akka.actor.{ Actor, ActorSelection }
import akka.persistence.AtLeastOnceDelivery
case class Msg(deliveryId: Long, s: String)
case class Confirm(deliveryId: Long)
sealed trait Evt
case class MsgSent(s: String) extends Evt
case class MsgConfirmed(deliveryId: Long) extends Evt
class MyPersistentActor(destination: ActorSelection)
extends PersistentActor with AtLeastOnceDelivery {
override def persistenceId: String = "persistence-id"
override def receiveCommand: Receive = {
case s: String
=> persist(MsgSent(s))(updateState)
case Confirm(deliveryId) => persist(MsgConfirmed(deliveryId))(updateState)
}
override def receiveRecover: Receive = {
case evt: Evt => updateState(evt)
}
def updateState(evt: Evt): Unit = evt match {
case MsgSent(s) =>
deliver(destination)(deliveryId => Msg(deliveryId, s))
case MsgConfirmed(deliveryId) => confirmDelivery(deliveryId)
}
}
class MyDestination extends Actor {
def receive = {
case Msg(deliveryId, s) =>
// ...
sender() ! Confirm(deliveryId)
}
}
The deliveryId generated by the persistence module is a strictly monotonically increasing sequence number
without gaps. The same sequence is used for all destinations of the actor, i.e. when sending to multiple destinations the destinations will see gaps in the sequence. It is not possible to use custom deliveryId. However,
you can send a custom correlation identifier in the message to the destination. You must then retain a mapping
between the internal deliveryId (passed into the deliveryIdToMessage function) and your custom correlation id (passed into the message). You can do this by storing such mapping in a Map(correlationId
-> deliveryId) from which you can retrieve the deliveryId to be passed into the confirmDelivery
method once the receiver of your message has replied with your custom correlation id.
The AtLeastOnceDelivery trait has a state consisting of unconfirmed messages and a sequence number. It
does not store this state itself. You must persist events corresponding to the deliver and confirmDelivery
invocations from your PersistentActor so that the state can be restored by calling the same methods during
the recovery phase of the PersistentActor. Sometimes these events can be derived from other business level
events, and sometimes you must create separate events. During recovery, calls to deliver will not send out
messages, those will be sent later if no matching confirmDelivery will have been performed.
Support for snapshots is provided by getDeliverySnapshot and setDeliverySnapshot. The
AtLeastOnceDeliverySnapshot contains the full delivery state, including unconfirmed messages.
If you need a custom snapshot for other parts of the actor state you must also include the
4.8. Persistence
200
Akka Scala Documentation, Release 2.5.0-RC1
AtLeastOnceDeliverySnapshot. It is serialized using protobuf with the ordinary Akka serialization mechanism. It is easiest to include the bytes of the AtLeastOnceDeliverySnapshot as a blob in your custom
snapshot.
The interval between redelivery attempts is defined by the redeliverInterval method. The default value can
be configured with the akka.persistence.at-least-once-delivery.redeliver-interval
configuration key. The method can be overridden by implementation classes to return non-default values.
The maximum number of messages that will be sent at each redelivery burst is defined by the
redeliveryBurstLimit method (burst frequency is half of the redelivery interval). If there’s a lot of unconfirmed messages (e.g. if the destination is not available for a long time), this helps to prevent an overwhelming
amount of messages to be sent at once. The default value can be configured with the akka.persistence.
at-least-once-delivery.redelivery-burst-limit configuration key. The method can be overridden by implementation classes to return non-default values.
After a number of delivery attempts a AtLeastOnceDelivery.UnconfirmedWarning message will be sent to self.
The re-sending will still continue, but you can choose to call
confirmDelivery to cancel the re-sending.
The number of delivery attempts before emitting
the warning is defined by the warnAfterNumberOfUnconfirmedAttempts method.
The
default value can be configured with the akka.persistence.at-least-once-delivery.
warn-after-number-of-unconfirmed-attempts configuration key. The method can be overridden
by implementation classes to return non-default values.
The AtLeastOnceDelivery trait holds messages in memory until their successful delivery has
been confirmed.
The maximum number of unconfirmed messages that the actor is allowed to
hold in memory is defined by the maxUnconfirmedMessages method.
If this limit is exceed
the deliver method will not accept more messages and it will throw AtLeastOnceDelivery.
MaxUnconfirmedMessagesExceededException. The default value can be configured with the akka.
persistence.at-least-once-delivery.max-unconfirmed-messages configuration key. The
method can be overridden by implementation classes to return non-default values.
4.8.6 Event Adapters
In long running projects using event sourcing sometimes the need arises to detach the data model from the domain
model completely.
Event Adapters help in situations where:
• Version Migrations – existing events stored in Version 1 should be “upcasted” to a new Version 2 representation, and the process of doing so involves actual code, not just changes on the serialization layer.
For these scenarios the toJournal function is usually an identity function, however the fromJournal
is implemented as v1.Event=>v2.Event, performing the neccessary mapping inside the fromJournal
method. This technique is sometimes refered to as “upcasting” in other CQRS libraries.
• Separating Domain and Data models – thanks to EventAdapters it is possible to completely separate
the domain model from the model used to persist data in the Journals. For example one may want to
use case classes in the domain model, however persist their protocol-buffer (or any other binary serialization format) counter-parts to the Journal. A simple toJournal:MyModel=>MyDataModel and
fromJournal:MyDataModel=>MyModel adapter can be used to implement this feature.
• Journal Specialized Data Types – exposing data types understood by the underlying Journal, for example
for data stores which understand JSON it is possible to write an EventAdapter toJournal:Any=>JSON
such that the Journal can directly store the json instead of serializing the object to its binary representation.
Implementing an EventAdapter is rather stright forward:
class MyEventAdapter(system: ExtendedActorSystem) extends EventAdapter {
override def manifest(event: Any): String =
"" // when no manifest needed, return ""
override def toJournal(event: Any): Any =
event // identity
4.8. Persistence
201
Akka Scala Documentation, Release 2.5.0-RC1
override def fromJournal(event: Any, manifest: String): EventSeq =
EventSeq.single(event) // identity
}
Then in order for it to be used on events coming to and from the journal you must bind it using the below
configuration syntax:
akka.persistence.journal {
inmem {
event-adapters {
tagging
= "docs.persistence.MyTaggingEventAdapter"
user-upcasting = "docs.persistence.UserUpcastingEventAdapter"
item-upcasting = "docs.persistence.ItemUpcastingEventAdapter"
}
event-adapter-bindings {
"docs.persistence.Item"
= tagging
"docs.persistence.TaggedEvent" = tagging
"docs.persistence.v1.Event"
= [user-upcasting, item-upcasting]
}
}
}
It is possible to bind multiple adapters to one class for recovery, in which case the fromJournal methods of all
bound adapters will be applied to a given matching event (in order of definition in the configuration). Since each
adapter may return from 0 to n adapted events (called as EventSeq), each adapter can investigate the event and if
it should indeed adapt it return the adapted event(s) for it. Other adapters which do not have anything to contribute
during this adaptation simply return EventSeq.empty. The adapted events are then delivered in-order to the
PersistentActor during replay.
Note: For more advanced schema evolution techniques refer to the Persistence - Schema Evolution documentation.
4.8.7 Persistent FSM
PersistentFSM handles the incoming messages in an FSM like fashion. Its internal state is persisted as a
sequence of changes, later referred to as domain events. Relationship between incoming messages, FSM’s states
and transitions, persistence of domain events is defined by a DSL.
A Simple Example
To demonstrate the features of the PersistentFSM trait, consider an actor which represents a Web store customer. The contract of our “WebStoreCustomerFSMActor” is that it accepts the following commands:
sealed trait Command
case class AddItem(item: Item) extends Command
case object Buy extends Command
case object Leave extends Command
case object GetCurrentCart extends Command
AddItem sent when the customer adds an item to a shopping cart Buy - when the customer finishes the purchase
Leave - when the customer leaves the store without purchasing anything GetCurrentCart allows to query
the current state of customer’s shopping cart
The customer can be in one of the following states:
4.8. Persistence
202
Akka Scala Documentation, Release 2.5.0-RC1
sealed trait UserState extends FSMState
case object LookingAround extends UserState {
override def identifier: String = "Looking Around"
}
case object Shopping extends UserState {
override def identifier: String = "Shopping"
}
case object Inactive extends UserState {
override def identifier: String = "Inactive"
}
case object Paid extends UserState {
override def identifier: String = "Paid"
}
LookingAround customer is browsing the site, but hasn’t added anything to the shopping cart Shopping
customer has recently added items to the shopping cart Inactive customer has items in the shopping cart, but
hasn’t added anything recently Paid customer has purchased the items
Note: PersistentFSM states must inherit from trait PersistentFSM.FSMState and implement the def
identifier: String method. This is required in order to simplify the serialization of FSM states. String
identifiers should be unique!
Customer’s actions are “recorded” as a sequence of “domain events” which are persisted. Those events are replayed on an actor’s start in order to restore the latest customer’s state:
sealed trait DomainEvent
case class ItemAdded(item: Item) extends DomainEvent
case object OrderExecuted extends DomainEvent
case object OrderDiscarded extends DomainEvent
Customer state data represents the items in a customer’s shopping cart:
case class Item(id: String, name: String, price: Float)
sealed trait ShoppingCart {
def addItem(item: Item): ShoppingCart
def empty(): ShoppingCart
}
case object EmptyShoppingCart extends ShoppingCart {
def addItem(item: Item) = NonEmptyShoppingCart(item :: Nil)
def empty() = this
}
case class NonEmptyShoppingCart(items: Seq[Item]) extends ShoppingCart {
def addItem(item: Item) = NonEmptyShoppingCart(items :+ item)
def empty() = EmptyShoppingCart
}
Here is how everything is wired together:
startWith(LookingAround, EmptyShoppingCart)
when(LookingAround) {
case Event(AddItem(item), _) ⇒
goto(Shopping) applying ItemAdded(item) forMax (1 seconds)
case Event(GetCurrentCart, data) ⇒
stay replying data
}
when(Shopping) {
case Event(AddItem(item), _) ⇒
stay applying ItemAdded(item) forMax (1 seconds)
4.8. Persistence
203
Akka Scala Documentation, Release 2.5.0-RC1
case Event(Buy, _) ⇒
goto(Paid) applying OrderExecuted andThen {
case NonEmptyShoppingCart(items) ⇒
reportActor ! PurchaseWasMade(items)
saveStateSnapshot()
case EmptyShoppingCart ⇒ saveStateSnapshot()
}
case Event(Leave, _) ⇒
stop applying OrderDiscarded andThen {
case _ ⇒
reportActor ! ShoppingCardDiscarded
saveStateSnapshot()
}
case Event(GetCurrentCart, data) ⇒
stay replying data
case Event(StateTimeout, _) ⇒
goto(Inactive) forMax (2 seconds)
}
when(Inactive) {
case Event(AddItem(item), _) ⇒
goto(Shopping) applying ItemAdded(item) forMax (1 seconds)
case Event(StateTimeout, _) ⇒
stop applying OrderDiscarded andThen {
case _ ⇒ reportActor ! ShoppingCardDiscarded
}
}
when(Paid) {
case Event(Leave, _) ⇒ stop()
case Event(GetCurrentCart, data) ⇒
stay replying data
}
Note: State data can only be modified directly on initialization. Later it’s modified only as a result of applying
domain events. Override the applyEvent method to define how state data is affected by domain events, see the
example below
override def applyEvent(event: DomainEvent, cartBeforeEvent: ShoppingCart):
˓→ShoppingCart = {
event match {
case ItemAdded(item) ⇒ cartBeforeEvent.addItem(item)
case OrderExecuted
⇒ cartBeforeEvent
case OrderDiscarded ⇒ cartBeforeEvent.empty()
}
}
andThen can be used to define actions which will be executed following event’s persistence - convenient for
“side effects” like sending a message or logging. Notice that actions defined in andThen block are not executed
on recovery:
goto(Paid) applying OrderExecuted andThen {
case NonEmptyShoppingCart(items) ⇒
reportActor ! PurchaseWasMade(items)
}
A snapshot of state data can be persisted by calling the saveStateSnapshot() method:
stop applying OrderDiscarded andThen {
case _ ⇒
4.8. Persistence
204
Akka Scala Documentation, Release 2.5.0-RC1
reportActor ! ShoppingCardDiscarded
saveStateSnapshot()
}
On recovery state data is initialized according to the latest available snapshot, then the remaining domain events
are replayed, triggering the applyEvent method.
4.8.8 Storage plugins
Storage backends for journals and snapshot stores are pluggable in the Akka persistence extension.
A directory of persistence journal and snapshot store plugins is available at the Akka Community Projects page,
see Community plugins
Plugins can be selected either by “default” for all persistent actors and views, or “individually”, when a persistent
actor or view defines its own set of plugins.
When a persistent actor or view does NOT override the journalPluginId and snapshotPluginId
methods, the persistence extension will use the “default” journal and snapshot-store plugins configured in
reference.conf:
akka.persistence.journal.plugin = ""
akka.persistence.snapshot-store.plugin = ""
However, these entries are provided as empty “”, and require explicit user configuration via override in the user
application.conf. For an example of a journal plugin which writes messages to LevelDB see Local LevelDB
journal. For an example of a snapshot store plugin which writes snapshots as individual files to the local filesystem
see Local snapshot store.
Applications can provide their own plugins by implementing a plugin API and activating them by configuration.
Plugin development requires the following imports:
import akka.persistence._
import akka.persistence.journal._
import akka.persistence.snapshot._
Eager initialization of persistence plugin
By default, persistence plugins are started on-demand, as they are used. In some case, however, it might be
beneficial to start a certain plugin eagerly. In order to do that, you should first add the akka.persistence.
Persistence under the akka.extensions key. Then, specify the IDs of plugins you wish to start automatically under akka.persistence.journal.auto-start-journals and akka.persistence.
snapshot-store.auto-start-snapshot-stores.
Journal plugin API
A journal plugin extends AsyncWriteJournal.
AsyncWriteJournal is an actor and the methods to be implemented are:
/**
* Plugin API: asynchronously writes a batch (`Seq`) of persistent messages to the
* journal.
*
* The batch is only for performance reasons, i.e. all messages don't have to be
˓→written
* atomically. Higher throughput can typically be achieved by using batch inserts
˓→of many
* records compared to inserting records one-by-one, but this aspect depends on the
4.8. Persistence
205
Akka Scala Documentation, Release 2.5.0-RC1
* underlying data store and a journal implementation can implement it as
efficient as
* possible. Journals should aim to persist events in-order for a given
˓→`persistenceId`
* as otherwise in case of a failure, the persistent state may be end up being
˓→inconsistent.
*
* Each `AtomicWrite` message contains the single `PersistentRepr` that
˓→corresponds to
* the event that was passed to the `persist` method of the `PersistentActor`, or
˓→it
* contains several `PersistentRepr` that corresponds to the events that were
˓→passed
* to the `persistAll` method of the `PersistentActor`. All `PersistentRepr` of the
* `AtomicWrite` must be written to the data store atomically, i.e. all or none
˓→must
* be stored. If the journal (data store) cannot support atomic writes of multiple
* events it should reject such writes with a `Try` `Failure` with an
* `UnsupportedOperationException` describing the issue. This limitation should
* also be documented by the journal plugin.
*
* If there are failures when storing any of the messages in the batch the returned
* `Future` must be completed with failure. The `Future` must only be completed
˓→with
* success when all messages in the batch have been confirmed to be stored
˓→successfully,
* i.e. they will be readable, and visible, in a subsequent replay. If there is
* uncertainty about if the messages were stored or not the `Future` must be
˓→completed
* with failure.
*
* Data store connection problems must be signaled by completing the `Future` with
* failure.
*
* The journal can also signal that it rejects individual messages (`AtomicWrite`)
˓→by
* the returned `immutable.Seq[Try[Unit]]`. It is possible but not mandatory to
˓→reduce
* number of allocations by returning `Future.successful(Nil)` for the happy path,
* i.e. when no messages are rejected. Otherwise the returned `Seq` must have as
˓→many elements
* as the input `messages` `Seq`. Each `Try` element signals if the corresponding
* `AtomicWrite` is rejected or not, with an exception describing the problem.
˓→Rejecting
* a message means it was not stored, i.e. it must not be included in a later
˓→replay.
* Rejecting a message is typically done before attempting to store it, e.g.
˓→because of
* serialization error.
*
* Data store connection problems must not be signaled as rejections.
*
* It is possible but not mandatory to reduce number of allocations by returning
* `Future.successful(Nil)` for the happy path, i.e. when no messages are rejected.
*
* Calls to this method are serialized by the enclosing journal actor. If you spawn
* work in asynchronous tasks it is alright that they complete the futures in any
˓→order,
* but the actual writes for a specific persistenceId should be serialized to avoid
* issues such as events of a later write are visible to consumers (query side, or
˓→replay)
* before the events of an earlier write are visible.
* A PersistentActor will not send a new WriteMessages request before the previous
˓→one
˓→
4.8. Persistence
206
Akka Scala Documentation, Release 2.5.0-RC1
* has been completed.
*
* Please note that the `sender` field of the contained PersistentRepr objects has
˓→been
* nulled out (i.e. set to `ActorRef.noSender`) in order to not use space in the
˓→journal
* for a sender reference that will likely be obsolete during replay.
*
* Please also note that requests for the highest sequence number may be made
˓→concurrently
* to this call executing for the same `persistenceId`, in particular it is
˓→possible that
* a restarting actor tries to recover before its outstanding writes have
˓→completed. In
* the latter case it is highly desirable to defer reading the highest sequence
˓→number
* until all outstanding writes have completed, otherwise the PersistentActor may
˓→reuse
* sequence numbers.
*
* This call is protected with a circuit-breaker.
*/
def asyncWriteMessages(messages: immutable.Seq[AtomicWrite]): Future[immutable.
˓→Seq[Try[Unit]]]
/**
* Plugin API: asynchronously deletes all persistent messages up to `toSequenceNr`
* (inclusive).
*
* This call is protected with a circuit-breaker.
* Message deletion doesn't affect the highest sequence number of messages,
˓→journal must maintain the highest sequence number and never decrease it.
*/
def asyncDeleteMessagesTo(persistenceId: String, toSequenceNr: Long): Future[Unit]
/**
* Plugin API
*
* Allows plugin implementers to use `f pipeTo self` and
* handle additional messages for implementing advanced features
*
*/
def receivePluginInternal: Actor.Receive = Actor.emptyBehavior
If the storage backend API only supports synchronous, blocking writes, the methods should be implemented as:
def asyncWriteMessages(messages: immutable.Seq[AtomicWrite]): Future[immutable.
˓→Seq[Try[Unit]]] =
Future.fromTry(Try {
// blocking call here
???
})
A journal plugin must also implement the methods defined in AsyncRecovery for replays and sequence number
recovery:
/**
* Plugin API: asynchronously replays persistent messages. Implementations replay
* a message by calling `replayCallback`. The returned future must be completed
* when all messages (matching the sequence number bounds) have been replayed.
* The future must be completed with a failure if any of the persistent messages
* could not be replayed.
4.8. Persistence
207
Akka Scala Documentation, Release 2.5.0-RC1
*
*
*
*
*
*
The `replayCallback` must also be called with messages that have been marked
as deleted. In this case a replayed message's `deleted` method must return
`true`.
The `toSequenceNr` is the lowest of what was returned by [[
#asyncReadHighestSequenceNr]]
* and what the user specified as recovery [[akka.persistence.Recovery]] parameter.
* This does imply that this call is always preceded by reading the highest
˓→sequence
* number for the given `persistenceId`.
*
* This call is NOT protected with a circuit-breaker because it may take long time
* to replay all events. The plugin implementation itself must protect against
* an unresponsive backend store and make sure that the returned Future is
* completed with success or failure within reasonable time. It is not allowed
* to ignore completing the future.
*
* @param persistenceId persistent actor id.
* @param fromSequenceNr sequence number where replay should start (inclusive).
* @param toSequenceNr sequence number where replay should end (inclusive).
* @param max maximum number of messages to be replayed.
* @param recoveryCallback called to replay a single message. Can be called from
˓→any
thread.
*
*
* @see [[AsyncWriteJournal]]
*/
def asyncReplayMessages(persistenceId: String, fromSequenceNr: Long, toSequenceNr:
˓→Long,
max: Long)(recoveryCallback: PersistentRepr ⇒ Unit):
˓→Future[Unit]
˓→
/**
* Plugin API: asynchronously reads the highest stored sequence number for the
* given `persistenceId`. The persistent actor will use the highest sequence
* number after recovery as the starting point when persisting new events.
* This sequence number is also used as `toSequenceNr` in subsequent call
* to [[#asyncReplayMessages]] unless the user has specified a lower
˓→`toSequenceNr`.
* Journal must maintain the highest sequence number and never decrease it.
*
* This call is protected with a circuit-breaker.
*
* Please also note that requests for the highest sequence number may be made
˓→concurrently
* to writes executing for the same `persistenceId`, in particular it is possible
˓→that
* a restarting actor tries to recover before its outstanding writes have
˓→completed.
*
* @param persistenceId persistent actor id.
* @param fromSequenceNr hint where to start searching for the highest sequence
number. When a persistent actor is recovering this
*
`fromSequenceNr` will be the sequence number of the used
*
snapshot or `0L` if no snapshot is used.
*
*/
def asyncReadHighestSequenceNr(persistenceId: String, fromSequenceNr: Long):
˓→Future[Long]
A journal plugin can be activated with the following minimal configuration:
4.8. Persistence
208
Akka Scala Documentation, Release 2.5.0-RC1
# Path to the journal plugin to be used
akka.persistence.journal.plugin = "my-journal"
# My custom journal plugin
my-journal {
# Class name of the plugin.
class = "docs.persistence.MyJournal"
# Dispatcher for the plugin actor.
plugin-dispatcher = "akka.actor.default-dispatcher"
}
The journal plugin instance is an actor so the methods corresponding to requests from persistent actors are executed sequentially. It may delegate to asynchronous libraries, spawn futures, or delegate to other actors to achive
parallelism.
The journal plugin class must have a constructor with one of these signatures:
• constructor with one com.typesafe.config.Config parameter and a String parameter for the
config path
• constructor with one com.typesafe.config.Config parameter
• constructor without parameters
The plugin section of the actor system’s config will be passed in the config constructor parameter. The config path
of the plugin is passed in the String parameter.
The plugin-dispatcher is the dispatcher used for the plugin actor. If not specified, it defaults to akka.
persistence.dispatchers.default-plugin-dispatcher.
Don’t run journal tasks/futures on the system default dispatcher, since that might starve other tasks.
Snapshot store plugin API
A snapshot store plugin must extend the SnapshotStore actor and implement the following methods:
/**
* Plugin API: asynchronously loads a snapshot.
*
* If the future `Option` is `None` then all events will be replayed,
* i.e. there was no snapshot. If snapshot could not be loaded the `Future`
* should be completed with failure. That is important because events may
* have been deleted and just replaying the events might not result in a valid
* state.
*
* This call is protected with a circuit-breaker.
*
* @param persistenceId id of the persistent actor.
* @param criteria selection criteria for loading.
*/
def loadAsync(persistenceId: String, criteria: SnapshotSelectionCriteria):
˓→Future[Option[SelectedSnapshot]]
/**
* Plugin API: asynchronously saves a snapshot.
*
* This call is protected with a circuit-breaker.
*
* @param metadata snapshot metadata.
* @param snapshot snapshot.
*/
def saveAsync(metadata: SnapshotMetadata, snapshot: Any): Future[Unit]
4.8. Persistence
209
Akka Scala Documentation, Release 2.5.0-RC1
/**
* Plugin API: deletes the snapshot identified by `metadata`.
*
* This call is protected with a circuit-breaker.
*
* @param metadata snapshot metadata.
*/
def deleteAsync(metadata: SnapshotMetadata): Future[Unit]
/**
* Plugin API: deletes all snapshots matching `criteria`.
*
* This call is protected with a circuit-breaker.
*
* @param persistenceId id of the persistent actor.
* @param criteria selection criteria for deleting.
*/
def deleteAsync(persistenceId: String, criteria: SnapshotSelectionCriteria):
˓→Future[Unit]
/**
* Plugin API
* Allows plugin implementers to use `f pipeTo self` and
* handle additional messages for implementing advanced features
*/
def receivePluginInternal: Actor.Receive = Actor.emptyBehavior
A snapshot store plugin can be activated with the following minimal configuration:
# Path to the snapshot store plugin to be used
akka.persistence.snapshot-store.plugin = "my-snapshot-store"
# My custom snapshot store plugin
my-snapshot-store {
# Class name of the plugin.
class = "docs.persistence.MySnapshotStore"
# Dispatcher for the plugin actor.
plugin-dispatcher = "akka.persistence.dispatchers.default-plugin-dispatcher"
}
The snapshot store instance is an actor so the methods corresponding to requests from persistent actors are executed sequentially. It may delegate to asynchronous libraries, spawn futures, or delegate to other actors to achive
parallelism.
The snapshot store plugin class must have a constructor with one of these signatures:
• constructor with one com.typesafe.config.Config parameter and a String parameter for the
config path
• constructor with one com.typesafe.config.Config parameter
• constructor without parameters
The plugin section of the actor system’s config will be passed in the config constructor parameter. The config path
of the plugin is passed in the String parameter.
The plugin-dispatcher is the dispatcher used for the plugin actor. If not specified, it defaults to akka.
persistence.dispatchers.default-plugin-dispatcher.
Don’t run snapshot store tasks/futures on the system default dispatcher, since that might starve other tasks.
4.8. Persistence
210
Akka Scala Documentation, Release 2.5.0-RC1
Plugin TCK
In order to help developers build correct and high quality storage plugins, we provide a Technology Compatibility
Kit (TCK for short).
The TCK is usable from Java as well as Scala projects. For Scala you need to include the akka-persistence-tck
dependency:
"com.typesafe.akka" %% "akka-persistence-tck" % "2.5.0-RC1" % "test"
To include the Journal TCK tests in your test suite simply extend the provided JournalSpec:
class MyJournalSpec extends JournalSpec(
config = ConfigFactory.parseString(
"""akka.persistence.journal.plugin = "my.journal.plugin"""")) {
override def supportsRejectingNonSerializableObjects: CapabilityFlag =
false // or CapabilityFlag.off
}
Please note that some of the tests are optional, and by overriding the supports... methods you give the TCK
the needed information about which tests to run. You can implement these methods using boolean falues or the
provided CapabilityFlag.on / CapabilityFlag.off values.
We also provide a simple benchmarking class JournalPerfSpec which includes all the tests that
JournalSpec has, and also performs some longer operations on the Journal while printing its performance
stats. While it is NOT aimed to provide a proper benchmarking environment it can be used to get a rough feel
about your journal’s performance in the most typical scenarios.
In order to include the SnapshotStore TCK tests in your test suite simply extend the SnapshotStoreSpec:
class MySnapshotStoreSpec extends SnapshotStoreSpec(
config = ConfigFactory.parseString(
"""
akka.persistence.snapshot-store.plugin = "my.snapshot-store.plugin"
"""))
In case your plugin requires some setting up (starting a mock database, removing temporary files etc.) you can
override the beforeAll and afterAll methods to hook into the tests lifecycle:
class MyJournalSpec extends JournalSpec(
config = ConfigFactory.parseString(
"""
akka.persistence.journal.plugin = "my.journal.plugin"
""")) {
override def supportsRejectingNonSerializableObjects: CapabilityFlag =
true // or CapabilityFlag.on
val storageLocations = List(
new File(system.settings.config.getString("akka.persistence.journal.leveldb.dir
˓→")),
new File(config.getString("akka.persistence.snapshot-store.local.dir")))
override def beforeAll() {
super.beforeAll()
storageLocations foreach FileUtils.deleteRecursively
}
override def afterAll() {
storageLocations foreach FileUtils.deleteRecursively
super.afterAll()
}
4.8. Persistence
211
Akka Scala Documentation, Release 2.5.0-RC1
}
We highly recommend including these specifications in your test suite, as they cover a broad range of cases you
might have otherwise forgotten to test for when writing a plugin from scratch.
4.8.9 Pre-packaged plugins
Local LevelDB journal
The LevelDB journal plugin config entry is akka.persistence.journal.leveldb. It writes messages
to a local LevelDB instance. Enable this plugin by defining config property:
# Path to the journal plugin to be used
akka.persistence.journal.plugin = "akka.persistence.journal.leveldb"
LevelDB based plugins will also require the following additional dependency declaration:
"org.iq80.leveldb"
"org.fusesource.leveldbjni"
% "leveldb"
% "leveldbjni-all"
% "0.7"
% "1.8"
The default location of LevelDB files is a directory named journal in the current working directory. This
location can be changed by configuration where the specified path can be relative or absolute:
akka.persistence.journal.leveldb.dir = "target/journal"
With this plugin, each actor system runs its own private LevelDB instance.
Shared LevelDB journal
A LevelDB instance can also be shared by multiple actor systems (on the same or on different nodes). This, for
example, allows persistent actors to failover to a backup node and continue using the shared journal instance from
the backup node.
Warning: A shared LevelDB instance is a single point of failure and should therefore only be used for testing
purposes. Highly-available, replicated journals are available as Community plugins.
Note: This plugin has been supplanted by Persistence Plugin Proxy.
A shared LevelDB instance is started by instantiating the SharedLeveldbStore actor.
import akka.persistence.journal.leveldb.SharedLeveldbStore
val store = system.actorOf(Props[SharedLeveldbStore], "store")
By default, the shared instance writes journaled messages to a local directory named journal in the current
working directory. The storage location can be changed by configuration:
akka.persistence.journal.leveldb-shared.store.dir = "target/shared"
Actor systems that use a shared LevelDB store must activate the akka.persistence.journal.
leveldb-shared plugin.
akka.persistence.journal.plugin = "akka.persistence.journal.leveldb-shared"
4.8. Persistence
212
Akka Scala Documentation, Release 2.5.0-RC1
This plugin must be initialized by injecting the (remote) SharedLeveldbStore actor reference. Injection is
done by calling the SharedLeveldbJournal.setStore method with the actor reference as argument.
trait SharedStoreUsage extends Actor {
override def preStart(): Unit = {
context.actorSelection("akka.tcp://example@127.0.0.1:2552/user/store") !
˓→Identify(1)
}
def receive = {
case ActorIdentity(1, Some(store)) =>
SharedLeveldbJournal.setStore(store, context.system)
}
}
Internal journal commands (sent by persistent actors) are buffered until injection completes. Injection is idempotent i.e. only the first injection is used.
Local snapshot store
The local snapshot store plugin config entry is akka.persistence.snapshot-store.local. It writes
snapshot files to the local filesystem. Enable this plugin by defining config property:
# Path to the snapshot store plugin to be used
akka.persistence.snapshot-store.plugin = "akka.persistence.snapshot-store.local"
The default storage location is a directory named snapshots in the current working directory. This can be
changed by configuration where the specified path can be relative or absolute:
akka.persistence.snapshot-store.local.dir = "target/snapshots"
Note that it is not mandatory to specify a snapshot store plugin. If you don’t use snapshots you don’t have to
configure it.
Persistence Plugin Proxy
A persistence plugin proxy allows sharing of journals and snapshot stores across multiple actor systems (on the
same or on different nodes). This, for example, allows persistent actors to failover to a backup node and continue
using the shared journal instance from the backup node. The proxy works by forwarding all the journal/snapshot
store messages to a single, shared, persistence plugin instance, and therefor supports any use case supported by
the proxied plugin.
Warning: A shared journal/snapshot store is a single point of failure and should therefore only be used for
testing purposes. Highly-available, replicated persistence plugins are available as Community plugins.
The journal and snapshot store proxies are controlled via the akka.persistence.journal.proxy
and akka.persistence.snapshot-store.proxy configuration entries, respectively.
Set the
target-journal-plugin or target-snapshot-store-plugin keys to the underlying plugin you
wish to use (for example: akka.persistence.journal.leveldb). The start-target-journal
and start-target-snapshot-store keys should be set to on in exactly one actor system this is the system that will instantiate the shared persistence plugin. Next, the proxy needs to be
told how to find the shared plugin. This can be done by setting the target-journal-address
and target-snapshot-store-address configuration keys, or programmatically by calling the
PersistencePluginProxy.setTargetLocation method.
Note:
Akka starts extensions lazily when they are required, and this includes the proxy. This means
that in order for the proxy to work, the persistence plugin on the target node must be instantiated. This
4.8. Persistence
213
Akka Scala Documentation, Release 2.5.0-RC1
can be done by instantiating the PersistencePluginProxyExtension extension, or by calling the
PersistencePluginProxy.start method.
Note: The proxied persistence plugin can (and should) be configured using its original configuration keys.
4.8.10 Custom serialization
Serialization of snapshots and payloads of Persistent messages is configurable with Akka’s Serialization
infrastructure. For example, if an application wants to serialize
• payloads of type MyPayload with a custom MyPayloadSerializer and
• snapshots of type MySnapshot with a custom MySnapshotSerializer
it must add
akka.actor {
serializers {
my-payload = "docs.persistence.MyPayloadSerializer"
my-snapshot = "docs.persistence.MySnapshotSerializer"
}
serialization-bindings {
"docs.persistence.MyPayload" = my-payload
"docs.persistence.MySnapshot" = my-snapshot
}
}
to the application configuration. If not specified, a default serializer is used.
For more advanced schema evolution techniques refer to the Persistence - Schema Evolution documentation.
4.8.11 Testing
When running tests with LevelDB default settings in sbt, make sure to set fork := true in your sbt project.
Otherwise, you’ll see an UnsatisfiedLinkError. Alternatively, you can switch to a LevelDB Java port by
setting
akka.persistence.journal.leveldb.native = off
or
akka.persistence.journal.leveldb-shared.store.native = off
in your Akka configuration. The LevelDB Java port is for testing purposes only.
Warning: It is not possible to test persistence provided classes (i.e. PersistentActor and AtLeastOnceDelivery) using TestActorRef due to its synchronous nature. These traits need to be able to perform asynchronous tasks in the background in order to handle internal persistence related events.
When testing Persistence based projects always rely on asynchronous messaging using the TestKit.
4.8.12 Configuration
There are several configuration properties for the persistence module, please refer to the reference configuration.
4.8. Persistence
214
Akka Scala Documentation, Release 2.5.0-RC1
4.8.13 Multiple persistence plugin configurations
By default, a persistent actor or view will use the “default” journal and snapshot store plugins configured in the
following sections of the reference.conf configuration resource:
# Absolute path to the default journal plugin configuration entry.
akka.persistence.journal.plugin = "akka.persistence.journal.inmem"
# Absolute path to the default snapshot store plugin configuration entry.
akka.persistence.snapshot-store.plugin = "akka.persistence.snapshot-store.local"
Note that in this case the actor or view overrides only the persistenceId method:
trait ActorWithDefaultPlugins extends PersistentActor {
override def persistenceId = "123"
}
When the persistent actor or view overrides the journalPluginId and snapshotPluginId methods, the
actor or view will be serviced by these specific persistence plugins instead of the defaults:
trait ActorWithOverridePlugins extends PersistentActor {
override def persistenceId = "123"
// Absolute path to the journal plugin configuration entry in the `reference.
˓→conf`.
override def journalPluginId = "akka.persistence.chronicle.journal"
// Absolute path to the snapshot store plugin configuration entry in the
˓→`reference.conf`.
override def snapshotPluginId = "akka.persistence.chronicle.snapshot-store"
}
Note that journalPluginId and snapshotPluginId must refer to properly configured reference.
conf plugin entries with a standard class property as well as settings which are specific for those plugins,
i.e.:
# Configuration entry for the custom journal plugin, see `journalPluginId`.
akka.persistence.chronicle.journal {
# Standard persistence extension property: provider FQCN.
class = "akka.persistence.chronicle.ChronicleSyncJournal"
# Custom setting specific for the journal `ChronicleSyncJournal`.
folder = $${user.dir}/store/journal
}
# Configuration entry for the custom snapshot store plugin, see `snapshotPluginId`.
akka.persistence.chronicle.snapshot-store {
# Standard persistence extension property: provider FQCN.
class = "akka.persistence.chronicle.ChronicleSnapshotStore"
# Custom setting specific for the snapshot store `ChronicleSnapshotStore`.
folder = $${user.dir}/store/snapshot
}
4.9 Persistence - Schema Evolution
When working on long running projects using Persistence, or any kind of Event Sourcing architectures, schema
evolution becomes one of the more important technical aspects of developing your application. The requirements
as well as our own understanding of the business domain may (and will) change in time.
In fact, if a project matures to the point where you need to evolve its schema to adapt to changing business
requirements you can view this as first signs of its success – if you wouldn’t need to adapt anything over an apps
lifecycle that could mean that no-one is really using it actively.
In this chapter we will investigate various schema evolution strategies and techniques from which you can pick
and choose the ones that match your domain and challenge at hand.
4.9. Persistence - Schema Evolution
215
Akka Scala Documentation, Release 2.5.0-RC1
Note: This page proposes a number of possible solutions to the schema evolution problem and explains how some
of the utilities Akka provides can be used to achieve this, it is by no means a complete (closed) set of solutions.
Sometimes, based on the capabilities of your serialization formats, you may be able to evolve your schema in
different ways than outlined in the sections below. If you discover useful patterns or techniques for schema
evolution feel free to submit Pull Requests to this page to extend it.
4.9.1 Schema evolution in event-sourced systems
In recent years we have observed a tremendous move towards immutable append-only datastores, with eventsourcing being the prime technique successfully being used in these settings. For an excellent overview why and
how immutable data makes scalability and systems design much simpler you may want to read Pat Helland’s
excellent Immutability Changes Everything whitepaper.
Since with Event Sourcing the events are immutable and usually never deleted – the way schema evolution is
handled differs from how one would go about it in a mutable database setting (e.g. in typical CRUD database
applications).
The system needs to be able to continue to work in the presence of “old” events which were stored under the “old”
schema. We also want to limit complexity in the business logic layer, exposing a consistent view over all of the
events of a given type to PersistentActor s and persistence queries. This allows the business logic layer to
focus on solving business problems instead of having to explicitly deal with different schemas.
In summary, schema evolution in event sourced systems exposes the following characteristics:
• Allow the system to continue operating without large scale migrations to be applied,
• Allow the system to read “old” events from the underlying storage, however present them in a “new”
view to the application logic,
• Transparently promote events to the latest versions during recovery (or queries) such that the business
logic need not consider multiple versions of events
Types of schema evolution
Before we explain the various techniques that can be used to safely evolve the schema of your persistent events
over time, we first need to define what the actual problem is, and what the typical styles of changes are.
Since events are never deleted, we need to have a way to be able to replay (read) old events, in such way that does
not force the PersistentActor to be aware of all possible versions of an event that it may have persisted in
the past. Instead, we want the Actors to work on some form of “latest” version of the event and provide some
means of either converting old “versions” of stored events into this “latest” event type, or constantly evolve the
event definition - in a backwards compatible way - such that the new deserialization code can still read old events.
The most common schema changes you will likely are:
• adding a field to an event type,
• remove or rename field in event type,
• remove event type,
• split event into multiple smaller events.
The following sections will explain some patterns which can be used to safely evolve your schema when facing
those changes.
4.9. Persistence - Schema Evolution
216
Akka Scala Documentation, Release 2.5.0-RC1
4.9.2 Picking the right serialization format
Picking the serialization format is a very important decision you will have to make while building your application.
It affects which kind of evolutions are simple (or hard) to do, how much work is required to add a new datatype,
and, last but not least, serialization performance.
If you find yourself realising you have picked “the wrong” serialization format, it is always possible to change the
format used for storing new events, however you would have to keep the old deserialization code in order to be
able to replay events that were persisted using the old serialization scheme. It is possible to “rebuild” an event-log
from one serialization format to another one, however it may be a more involved process if you need to perform
this on a live system.
Binary serialization formats that we have seen work well for long-lived applications include the very flexible IDL
based: Google Protobuf, Apache Thrift or Apache Avro. Avro schema evolution is more “entire schema” based,
instead of single fields focused like in protobuf or thrift, and usually requires using some kind of schema registry.
Users who want their data to be human-readable directly in the write-side datastore may opt to use plain-old
JSON as the storage format, though that comes at a cost of lacking support for schema evolution and relatively
large marshalling latency.
There are plenty excellent blog posts explaining the various trade-offs between popular serialization formats, one
post we would like to highlight is the very well illustrated Schema evolution in Avro, Protocol Buffers and Thrift
by Martin Kleppmann.
Provided default serializers
Akka Persistence provides Google Protocol Buffers based serializers (using Akka Serialization) for it’s own message types such as PersistentRepr, AtomicWrite and snapshots. Journal plugin implementations may
choose to use those provided serializers, or pick a serializer which suits the underlying database better.
Note: Serialization is NOT handled automatically by Akka Persistence itself. Instead, it only provides the above
described serializers, and in case a AsyncWriteJournal plugin implementation chooses to use them directly,
the above serialization scheme will be used.
Please refer to your write journal’s documentation to learn more about how it handles serialization!
For example, some journals may choose to not use Akka Serialization at all and instead store the data in a format
that is more “native” for the underlying datastore, e.g. using JSON or some other kind of format that the target
datastore understands directly.
The below figure explains how the default serialization scheme works, and how it fits together with serializing the
user provided message itself, which we will from here on refer to as the payload (highlighted in yellow):
The blue colored regions of the PersistentMessage indicate what is serialized using the generated protocol
buffers serializers, and the yellow payload indicates the user provided event (by calling persist(payload)(.
..)). As you can see, the PersistentMessage acts as an envelope around the payload, adding various fields
related to the origin of the event (persistenceId, sequenceNr and more).
More advanced techniques (e.g. Remove event class and ignore events) will dive into using the manifests for
increasing the flexibility of the persisted vs. exposed types even more. However for now we will focus on the
simpler evolution techniques, concerning simply configuring the payload serializers.
By default the payload will be serialized using Java Serialization. This is fine for testing and initial phases
of your development (while you’re still figuring out things and the data will not need to stay persisted forever).
However, once you move to production you should really pick a different serializer for your payloads.
Warning: Do not rely on Java serialization (which will be picked by Akka by default if you don’t specify
any serializers) for serious application development! It does not lean itself well to evolving schemas over
4.9. Persistence - Schema Evolution
217
Akka Scala Documentation, Release 2.5.0-RC1
Fig. 4.1: Akka Persistence provided serializers wrap the user payload in an envelope containing all persistencerelevant information. If the Journal uses provided Protobuf serializers for the wrapper types (e.g. PersistentRepr), then the payload will be serialized using the user configured serializer, and if none is provided
explicitly, Java serialization will be used for it.
long periods of time, and its performance is also not very high (it never was designed for high-throughput
scenarios).
Configuring payload serializers
This section aims to highlight the complete basics on how to define custom serializers using Akka Serialization.
Many journal plugin implementations use Akka Serialization, thus it is tremendously important to understand how
to configure it to work with your event classes.
Note: Read the Akka Serialization docs to learn more about defining custom serializers, to improve performance
and maintainability of your system. Do not depend on Java serialization for production deployments.
The below snippet explains in the minimal amount of lines how a custom serializer can be registered. For more
in-depth explanations on how serialization picks the serializer to use etc, please refer to its documentation.
First we start by defining our domain model class, here representing a person:
final case class Person(name: String, surname: String)
Next we implement a serializer (or extend an existing one to be able to handle the new Person class):
/**
* Simplest possible serializer, uses a string representation of the Person class.
*
* Usually a serializer like this would use a library like:
* protobuf, kryo, avro, cap'n proto, flatbuffers, SBE or some other dedicated
˓→serializer backend
* to perform the actual to/from bytes marshalling.
*/
class SimplestPossiblePersonSerializer extends SerializerWithStringManifest {
val Utf8 = Charset.forName("UTF-8")
val PersonManifest = classOf[Person].getName
// unique identifier of the serializer
def identifier = 1234567
4.9. Persistence - Schema Evolution
218
Akka Scala Documentation, Release 2.5.0-RC1
// extract manifest to be stored together with serialized object
override def manifest(o: AnyRef): String = o.getClass.getName
// serialize the object
override def toBinary(obj: AnyRef): Array[Byte] = obj match {
case p: Person => s"""${p.name}|${p.surname}""".getBytes(Utf8)
case _ => throw new IllegalArgumentException(
s"Unable to serialize to bytes, clazz was: ${obj.getClass}!")
}
// deserialize the object, using the manifest to indicate which logic to apply
override def fromBinary(bytes: Array[Byte], manifest: String): AnyRef =
manifest match {
case PersonManifest =>
val nameAndSurname = new String(bytes, Utf8)
val Array(name, surname) = nameAndSurname.split("[|]")
Person(name, surname)
case _ => throw new IllegalArgumentException(
s"Unable to deserialize from bytes, manifest was: $manifest! Bytes length:
˓→" +
bytes.length)
}
}
And finally we register the serializer and bind it to handle the docs.persistence.Person class:
# application.conf
akka {
actor {
serializers {
person = "docs.persistence.SimplestPossiblePersonSerializer"
}
serialization-bindings {
"docs.persistence.Person" = person
}
}
}
Deserialization will be performed by the same serializer which serialized the message initially because of the
identifier being stored together with the message.
Please refer to the Akka Serialization documentation for more advanced use of serializers, especially the Serializer with String Manifest section since it is very useful for Persistence based applications dealing with schema
evolutions, as we will see in some of the examples below.
4.9.3 Schema evolution in action
In this section we will discuss various schema evolution techniques using concrete examples and explaining some
of the various options one might go about handling the described situation. The list below is by no means a
complete guide, so feel free to adapt these techniques depending on your serializer’s capabilities and/or other
domain specific limitations.
Add fields
Situation:
You need to add a field to an existing message type.
For example, a
SeatReserved(letter:String, row:Int) now needs to have an associated code which indicates if it
is a window or aisle seat.
4.9. Persistence - Schema Evolution
219
Akka Scala Documentation, Release 2.5.0-RC1
Solution: Adding fields is the most common change you’ll need to apply to your messages so make sure the
serialization format you picked for your payloads can handle it apropriately, i.e. such changes should be binary
compatible. This is easily achieved using the right serializer toolkit – we recommend something like Google
Protocol Buffers or Apache Thrift however other tools may fit your needs just as well – picking a serializer
backend is something you should research before picking one to run with. In the following examples we will be
using protobuf, mostly because we are familiar with it, it does its job well and Akka is using it internally as well.
While being able to read messages with missing fields is half of the solution, you also need to deal with the
missing values somehow. This is usually modeled as some kind of default value, or by representing the field as an
Option[T] See below for an example how reading an optional field from a serialized protocol buffers message
might look like.
sealed abstract class SeatType { def code: String }
object SeatType {
def fromString(s: String) = s match {
case Window.code => Window
case Aisle.code => Aisle
case Other.code => Other
case _
=> Unknown
}
case object Window extends SeatType { override val code = "W" }
case object Aisle extends SeatType { override val code = "A" }
case object Other extends SeatType { override val code = "O" }
case object Unknown extends SeatType { override val code = "" }
}
case class SeatReserved(letter: String, row: Int, seatType: SeatType)
Next we prepare an protocol definition using the protobuf Interface Description Language, which we’ll use to
generate the serializer code to be used on the Akka Serialization layer (notice that the schema aproach allows us
to easily rename fields, as long as the numeric identifiers of the fields do not change):
// FlightAppModels.proto
option java_package = "docs.persistence.proto";
option optimize_for = SPEED;
message SeatReserved {
required string letter
= 1;
required uint32 row
= 2;
optional string seatType = 3; // the new field
}
The serializer implementation uses the protobuf generated classes to marshall the payloads. Optional fields can
be handled explicitly or missing values by calling the has... methods on the protobuf object, which we do for
seatType in order to use a Unknown type in case the event was stored before we had introduced the field to
this event type:
/**
* Example serializer impl which uses protocol buffers generated classes (proto.*)
* to perform the to/from binary marshalling.
*/
class AddedFieldsSerializerWithProtobuf extends SerializerWithStringManifest {
override def identifier = 67876
final val SeatReservedManifest = classOf[SeatReserved].getName
override def manifest(o: AnyRef): String = o.getClass.getName
override def fromBinary(bytes: Array[Byte], manifest: String): AnyRef =
manifest match {
case SeatReservedManifest =>
4.9. Persistence - Schema Evolution
220
Akka Scala Documentation, Release 2.5.0-RC1
// use generated protobuf serializer
seatReserved(FlightAppModels.SeatReserved.parseFrom(bytes))
case _ =>
throw new IllegalArgumentException("Unable to handle manifest: " +
˓→manifest)
}
override def toBinary(o: AnyRef): Array[Byte] = o match {
case s: SeatReserved =>
FlightAppModels.SeatReserved.newBuilder
.setRow(s.row)
.setLetter(s.letter)
.setSeatType(s.seatType.code)
.build().toByteArray
}
// -- fromBinary helpers -private def seatReserved(p: FlightAppModels.SeatReserved): SeatReserved =
SeatReserved(p.getLetter, p.getRow, seatType(p))
// handle missing field by assigning "Unknown" value
private def seatType(p: FlightAppModels.SeatReserved): SeatType =
if (p.hasSeatType) SeatType.fromString(p.getSeatType) else SeatType.Unknown
}
Rename fields
Situation: When first designing the system the SeatReserved event featured a code field. After some time
you discover that what was originally called code actually means seatNr, thus the model should be changed to
reflect this concept more accurately.
Solution 1 - using IDL based serializers: First, we will discuss the most efficient way of dealing with such kinds
of schema changes – IDL based serializers.
IDL stands for Interface Description Language, and means that the schema of the messages that will be stored is
based on this description. Most IDL based serializers also generate the serializer / deserializer code so that using
them is not too hard. Examples of such serializers are protobuf or thrift.
Using these libraries rename operations are “free”, because the field name is never actually stored in the binary
representation of the message. This is one of the advantages of schema based serializers, even though that they
add the overhead of having to maintain the schema. When using serializers like this, no additional code change
(except renaming the field and method used during serialization) is needed to perform such evolution:
4.9. Persistence - Schema Evolution
221
Akka Scala Documentation, Release 2.5.0-RC1
This is how such a rename would look in protobuf:
// protobuf message definition, BEFORE:
message SeatReserved {
required string code = 1;
}
// protobuf message definition, AFTER:
message SeatReserved {
required string seatNr = 1; // field renamed, id remains the same
}
It is important to learn about the strengths and limitations of your serializers, in order to be able to move swiftly
and refactor your models fearlessly as you go on with the project.
Note: Learn in-depth about the serialization engine you’re using as it will impact how you can aproach schema
evolution.
Some operations are “free” in certain serialization formats (more often than not: removing/adding optional fields,
sometimes renaming fields etc.), while some other operations are strictly not possible.
Solution 2 - by manually handling the event versions: Another solution, in case your serialization format does
not support renames as easily as the above mentioned formats, is versioning your schema. For example, you could
have made your events carry an additional field called _version which was set to 1 (because it was the initial
schema), and once you change the schema you bump this number to 2, and write an adapter which can perform
the rename.
This approach is popular when your serialization format is something like JSON, where renames can not be
performed automatically by the serializer. You can do these kinds of “promotions” either manually (as shown
in the example below) or using a library like Stamina which helps to create those V1->V2->V3->...->Vn
promotion chains without much boilerplate.
The following snippet showcases how one could apply renames if working with plain JSON (using spray.
json.JsObject):
class JsonRenamedFieldAdapter extends EventAdapter {
val marshaller = new ExampleJsonMarshaller
val V1 = "v1"
val V2 = "v2"
// this could be done independently for each event type
override def manifest(event: Any): String = V2
override def toJournal(event: Any): JsObject =
marshaller.toJson(event)
override def fromJournal(event: Any, manifest: String): EventSeq = event match {
case json: JsObject => EventSeq(marshaller.fromJson(manifest match {
4.9. Persistence - Schema Evolution
222
Akka Scala Documentation, Release 2.5.0-RC1
case V1
=> rename(json, "code", "seatNr")
case V2
=> json // pass-through
case unknown => throw new IllegalArgumentException(s"Unknown manifest:
˓→$unknown")
}))
case _ =>
val c = event.getClass
throw new IllegalArgumentException("Can only work with JSON, was: %s".
˓→format(c))
}
def rename(json: JsObject, from: String, to: String): JsObject = {
val value = json.fields(from)
val withoutOld = json.fields - from
JsObject(withoutOld + (to -> value))
}
}
As you can see, manually handling renames induces some boilerplate onto the EventAdapter, however much of
it you will find is common infrastructure code that can be either provided by an external library (for promotion
management) or put together in a simple helper trait.
Note: The technique of versioning events and then promoting them to the latest version using JSON transformations can of course be applied to more than just field renames – it also applies to adding fields and all kinds of
changes in the message format.
Remove event class and ignore events
Situation: While investigating app performance you notice that insane amounts of CustomerBlinked events
are being stored for every customer each time he/she blinks. Upon investigation you decide that the event does
not add any value and should be deleted. You still have to be able to replay from a journal which contains those
old CustomerBlinked events though.
Naive solution - drop events in EventAdapter:
The problem of removing an event type from the domain model is not as much its removal, as the implications
for the recovery mechanisms that this entails. For example, a naive way of filtering out certain kinds of events
from being delivered to a recovering PersistentActor is pretty simple, as one can simply filter them out in
an EventAdapter:
Fig. 4.2: The EventAdapter can drop old events (O) by emitting an empty EventSeq. Other events can
simply be passed through (E).
This however does not address the underlying cost of having to deserialize all the events during recovery, even
those which will be filtered out by the adapter. In the next section we will improve the above explained mechanism
4.9. Persistence - Schema Evolution
223
Akka Scala Documentation, Release 2.5.0-RC1
to avoid deserializing events which would be filtered out by the adapter anyway, thus allowing to save precious
time during a recovery containing lots of such events (without actually having to delete them).
Improved solution - deserialize into tombstone:
In the just described technique we have saved the PersistentActor from receiving un-wanted events by filtering
them out in the EventAdapter, however the event itself still was deserialized and loaded into memory. This
has two notable downsides:
• first, that the deserialization was actually performed, so we spent some of out time budget on the deserialization, even though the event does not contribute anything to the persistent actors state.
• second, that we are unable to remove the event class from the system – since the serializer still needs to
create the actuall instance of it, as it does not know it will not be used.
The solution to these problems is to use a serializer that is aware of that event being no longer needed, and can
notice this before starting to deserialize the object.
This aproach allows us to remove the original class from our classpath, which makes for less “old” classes lying
around in the project. This can for example be implemented by using an SerializerWithStringManifest
(documented in depth in Serializer with String Manifest). By looking at the string manifest, the serializer can
notice that the type is no longer needed, and skip the deserialization all-together:
Fig. 4.3: The serializer is aware of the old event types that need to be skipped (O), and can skip deserializing them
alltogether by simply returning a “tombstone” (T), which the EventAdapter converts into an empty EventSeq.
Other events (E) can simply be passed through.
The serializer detects that the string manifest points to a removed event type and skips attempting to deserialize it:
case object EventDeserializationSkipped
class RemovedEventsAwareSerializer extends SerializerWithStringManifest {
val utf8 = Charset.forName("UTF-8")
override def identifier: Int = 8337
val SkipEventManifestsEvents = Set(
"docs.persistence.CustomerBlinked" // ...
)
override def manifest(o: AnyRef): String = o.getClass.getName
override def toBinary(o: AnyRef): Array[Byte] = o match {
case _ => o.toString.getBytes(utf8) // example serialization
}
override def fromBinary(bytes: Array[Byte], manifest: String): AnyRef =
manifest match {
case m if SkipEventManifestsEvents.contains(m) =>
EventDeserializationSkipped
case other => new String(bytes, utf8)
}
}
4.9. Persistence - Schema Evolution
224
Akka Scala Documentation, Release 2.5.0-RC1
The EventAdapter we implemented is aware of EventDeserializationSkipped events (our “Tombstones”), and emits and empty EventSeq whenever such object is encoutered:
class SkippedEventsAwareAdapter extends EventAdapter {
override def manifest(event: Any) = ""
override def toJournal(event: Any) = event
override def fromJournal(event: Any, manifest: String) = event match {
case EventDeserializationSkipped => EventSeq.empty
case _
=> EventSeq(event)
}
}
Detach domain model from data model
Situation: You want to separate the application model (often called the “domain model”) completely from the
models used to persist the corresponding events (the “data model”). For example because the data representation
may change independently of the domain model.
Another situation where this technique may be useful is when your serialization tool of choice requires generated
classes to be used for serialization and deserialization of objects, like for example Google Protocol Buffers do,
yet you do not want to leak this implementation detail into the domain model itself, which you’d like to model as
plain Scala case classes.
Solution: In order to detach the domain model, which is often represented using pure scala (case) classes, from
the data model classes which very often may be less user-friendly yet highly optimised for throughput and schema
evolution (like the classes generated by protobuf for example), it is possible to use a simple EventAdapter which
maps between these types in a 1:1 style as illustrated below:
Fig. 4.4: Domain events (A) are adapted to the data model events (D) by the EventAdapter. The data model
can be a format natively understood by the journal, such that it can store it more efficiently or include additional
data for the event (e.g. tags), for ease of later querying.
We will use the following domain and data models to showcase how the separation can be implemented by the
adapter:
/** Domain model - highly optimised for domain language and maybe "fluent" usage */
object DomainModel {
final case class Customer(name: String)
final case class Seat(code: String) {
def bookFor(customer: Customer): SeatBooked = SeatBooked(code, customer)
}
final case class SeatBooked(code: String, customer: Customer)
}
/** Data model - highly optimised for schema evolution and persistence */
object DataModel {
4.9. Persistence - Schema Evolution
225
Akka Scala Documentation, Release 2.5.0-RC1
final case class SeatBooked(code: String, customerName: String)
}
The EventAdapter takes care of converting from one model to the other one (in both directions), alowing the
models to be completely detached from each other, such that they can be optimised independently as long as the
mapping logic is able to convert between them:
class DetachedModelsAdapter extends EventAdapter {
override def manifest(event: Any): String = ""
override def toJournal(event: Any): Any = event match {
case DomainModel.SeatBooked(code, customer) =>
DataModel.SeatBooked(code, customer.name)
}
override def fromJournal(event: Any, manifest: String): EventSeq = event match {
case DataModel.SeatBooked(code, customerName) =>
EventSeq(DomainModel.SeatBooked(code, DomainModel.Customer(customerName)))
}
}
The same technique could also be used directly in the Serializer if the end result of marshalling is bytes. Then the
serializer can simply convert the bytes do the domain object by using the generated protobuf builders.
Store events as human-readable data model
Situation: You want to keep your persisted events in a human-readable format, for example JSON.
Solution: This is a special case of the Detach domain model from data model pattern, and thus requires some
co-operation from the Journal implementation to achieve this.
An example of a Journal which may implement this pattern is MongoDB, however other databases such as PostgreSQL and Cassandra could also do it because of their built-in JSON capabilities.
In this aproach, the EventAdapter is used as the marshalling layer: it serializes the events to/from JSON. The
journal plugin notices that the incoming event type is JSON (for example by performing a match on the incoming
event) and stores the incoming object directly.
class JsonDataModelAdapter extends EventAdapter {
override def manifest(event: Any): String = ""
val marshaller = new ExampleJsonMarshaller
override def toJournal(event: Any): JsObject =
marshaller.toJson(event)
override def fromJournal(event: Any, manifest: String): EventSeq = event match {
case json: JsObject =>
EventSeq(marshaller.fromJson(json))
case _ =>
throw new IllegalArgumentException(
"Unable to fromJournal a non-JSON object! Was: " + event.getClass)
}
}
Note: This technique only applies if the Akka Persistence plugin you are using provides this capability. Check
the documentation of your favourite plugin to see if it supports this style of persistence.
If it doesn’t, you may want to skim the list of existing journal plugins, just in case some other plugin for your
favourite datastore does provide this capability.
4.9. Persistence - Schema Evolution
226
Akka Scala Documentation, Release 2.5.0-RC1
Alternative solution:
In fact, an AsyncWriteJournal implementation could natively decide to not use binary serialization at all, and
always serialize the incoming messages as JSON - in which case the toJournal implementation of the
EventAdapter would be an identity function, and the fromJournal would need to de-serialize messages
from JSON.
Note: If in need of human-readable events on the write-side of your application reconsider whether preparing
materialized views using Persistence Query would not be an efficient way to go about this, without compromising
the write-side’s throughput characteristics.
If indeed you want to use a human-readable representation on the write-side, pick a Persistence plugin that provides
that functionality, or – implement one yourself.
Split large event into fine-grained events
Situation: While refactoring your domain events, you find that one of the events has become too large (coarsegrained) and needs to be split up into multiple fine-grained events.
Solution: Let us consider a situation where an event represents “user details changed”. After some time we
discover that this event is too coarse, and needs to be split into “user name changed” and “user address changed”,
because somehow users keep changing their usernames a lot and we’d like to keep this as a separate event.
The write side change is very simple, we simply persist UserNameChanged or UserAddressChanged
depending on what the user actually intended to change (instead of the composite UserDetailsChanged that
we had in version 1 of our model).
Fig. 4.5: The EventAdapter splits the incoming event into smaller more fine grained events during recovery.
During recovery however, we now need to convert the old V1 model into the V2 representation of the change.
Depending if the old event contains a name change, we either emit the UserNameChanged or we don’t, and the
address change is handled similarily:
trait V1
trait V2
// V1 event:
final case class UserDetailsChanged(name: String, address: String) extends V1
// corresponding V2 events:
final case class UserNameChanged(name: String) extends V2
final case class UserAddressChanged(address: String) extends V2
// event splitting adapter:
class UserEventsAdapter extends EventAdapter {
override def manifest(event: Any): String = ""
override def fromJournal(event: Any, manifest: String): EventSeq = event match {
4.9. Persistence - Schema Evolution
227
Akka Scala Documentation, Release 2.5.0-RC1
case UserDetailsChanged(null, address) => EventSeq(UserAddressChanged(address))
case UserDetailsChanged(name, null)
=> EventSeq(UserNameChanged(name))
case UserDetailsChanged(name, address) =>
EventSeq(
UserNameChanged(name),
UserAddressChanged(address))
case event: V2 => EventSeq(event)
}
override def toJournal(event: Any): Any = event
}
By returning an EventSeq from the event adapter, the recovered event can be converted to multiple events before
being delivered to the persistent actor.
4.10 Persistence Query
Akka persistence query complements Persistence by providing a universal asynchronous stream based query interface that various journal plugins can implement in order to expose their query capabilities.
The most typical use case of persistence query is implementing the so-called query side (also known as “read side”)
in the popular CQRS architecture pattern - in which the writing side of the application (e.g. implemented using
akka persistence) is completely separated from the “query side”. Akka Persistence Query itself is not directly the
query side of an application, however it can help to migrate data from the write side to the query side database. In
very simple scenarios Persistence Query may be powerful enough to fulfill the query needs of your app, however
we highly recommend (in the spirit of CQRS) of splitting up the write/read sides into separate datastores as the
need arises.
4.10.1 Dependencies
Akka persistence query is a separate jar file. Make sure that you have the following dependency in your project:
"com.typesafe.akka" %% "akka-persistence-query" % "2.5.0-RC1"
4.10.2 Design overview
Akka persistence query is purposely designed to be a very loosely specified API. This is in order to keep the
provided APIs general enough for each journal implementation to be able to expose its best features, e.g. a SQL
journal can use complex SQL queries or if a journal is able to subscribe to a live event stream this should also be
possible to expose the same API - a typed stream of events.
Each read journal must explicitly document which types of queries it supports. Refer to your journal’s plugins
documentation for details on which queries and semantics it supports.
While Akka Persistence Query does not provide actual implementations of ReadJournals, it defines a number of
pre-defined query types for the most common query scenarios, that most journals are likely to implement (however
they are not required to).
4.10.3 Read Journals
In order to issue queries one has to first obtain an instance of a ReadJournal. Read journals are implemented
as Community plugins, each targeting a specific datastore (for example Cassandra or JDBC databases). For
example, given a library that provides a akka.persistence.query.my-read-journal obtaining the
related journal is as simple as:
4.10. Persistence Query
228
Akka Scala Documentation, Release 2.5.0-RC1
// obtain read journal by plugin id
val readJournal =
PersistenceQuery(system).readJournalFor[MyScaladslReadJournal](
"akka.persistence.query.my-read-journal")
// issue query to journal
val source: Source[EventEnvelope, NotUsed] =
readJournal.eventsByPersistenceId("user-1337", 0, Long.MaxValue)
// materialize stream, consuming events
implicit val mat = ActorMaterializer()
source.runForeach { event => println("Event: " + event) }
Journal implementers are encouraged to put this identifier in a variable known to the user, such that one can access
it via readJournalFor[NoopJournal](NoopJournal.identifier), however this is not enforced.
Read journal implementations are available as Community plugins.
Predefined queries
Akka persistence query comes with a number of query interfaces built in and suggests Journal implementors to
implement them according to the semantics described below. It is important to notice that while these query types
are very common a journal is not obliged to implement all of them - for example because in a given journal such
query would be significantly inefficient.
Note: Refer to the documentation of the ReadJournal plugin you are using for a specific list of supported
query types. For example, Journal plugins should document their stream completion strategies.
The predefined queries are:
AllPersistenceIdsQuery and CurrentPersistenceIdsQuery
allPersistenceIds which is designed to allow users to subscribe to a stream of all persistent ids in the
system. By default this stream should be assumed to be a “live” stream, which means that the journal should keep
emitting new persistence ids as they come into the system:
readJournal.persistenceIds()
If your usage does not require a live stream, you can use the currentPersistenceIds query:
readJournal.currentPersistenceIds()
EventsByPersistenceIdQuery and CurrentEventsByPersistenceIdQuery
eventsByPersistenceId is a query equivalent to replaying a PersistentActor, however, since it is a stream
it is possible to keep it alive and watch for additional incoming events persisted by the persistent actor identified
by the given persistenceId.
readJournal.eventsByPersistenceId("user-us-1337")
Most journals will have to revert to polling in order to achieve this, which can typically be configured with a
refresh-interval configuration property.
If your usage does not require a live stream, you can use the currentEventsByPersistenceId query.
4.10. Persistence Query
229
Akka Scala Documentation, Release 2.5.0-RC1
EventsByTag and CurrentEventsByTag
eventsByTag allows querying events regardless of which persistenceId they are associated with. This
query is hard to implement in some journals or may need some additional preparation of the used data store to be
executed efficiently. The goal of this query is to allow querying for all events which are “tagged” with a specific
tag. That includes the use case to query all domain events of an Aggregate Root type. Please refer to your read
journal plugin’s documentation to find out if and how it is supported.
Some journals may support tagging of events via an Event Adapters that wraps the events in a akka.
persistence.journal.Tagged with the given tags. The journal may support other ways of doing tagging - again, how exactly this is implemented depends on the used journal. Here is an example of such a tagging
event adapter:
import akka.persistence.journal.WriteEventAdapter
import akka.persistence.journal.Tagged
class MyTaggingEventAdapter extends WriteEventAdapter {
val colors = Set("green", "black", "blue")
override def toJournal(event: Any): Any = event match {
case s: String =>
var tags = colors.foldLeft(Set.empty[String]) { (acc, c) =>
if (s.contains(c)) acc + c else acc
}
if (tags.isEmpty) event
else Tagged(event, tags)
case _ => event
}
override def manifest(event: Any): String = ""
}
Note: A very important thing to keep in mind when using queries spanning multiple persistenceIds, such as
EventsByTag is that the order of events at which the events appear in the stream rarely is guaranteed (or stable
between materializations).
Journals may choose to opt for strict ordering of the events, and should then document explicitly what kind of
ordering guarantee they provide - for example “ordered by timestamp ascending, independently of persistenceId”
is easy to achieve on relational databases, yet may be hard to implement efficiently on plain key-value datastores.
In the example below we query all events which have been tagged (we assume this was performed by the writeside using an EventAdapter, or that the journal is smart enough that it can figure out what we mean by this tag for example if the journal stored the events as json it may try to find those with the field tag set to this value etc.).
// assuming journal is able to work with numeric offsets we can:
val blueThings: Source[EventEnvelope, NotUsed] =
readJournal.eventsByTag("blue")
// find top 10 blue things:
val top10BlueThings: Future[Vector[Any]] =
blueThings
.map(_.event)
.take(10) // cancels the query stream after pulling 10 elements
.runFold(Vector.empty[Any])(_ :+ _)
// start another query, from the known offset
val furtherBlueThings = readJournal.eventsByTag("blue", offset = Sequence(10))
As you can see, we can use all the usual stream combinators available from Streams on the resulting query stream,
including for example taking the first 10 and cancelling the stream. It is worth pointing out that the built-in
4.10. Persistence Query
230
Akka Scala Documentation, Release 2.5.0-RC1
EventsByTag query has an optionally supported offset parameter (of type Long) which the journals can use
to implement resumable-streams. For example a journal may be able to use a WHERE clause to begin the read
starting from a specific row, or in a datastore that is able to order events by insertion time it could treat the Long
as a timestamp and select only older events.
If your usage does not require a live stream, you can use the currentEventsByTag query.
Materialized values of queries
Journals are able to provide additional information related to a query by exposing Materialized values, which are
a feature of Streams that allows to expose additional values at stream materialization time.
More advanced query journals may use this technique to expose information about the character of the materialized
stream, for example if it’s finite or infinite, strictly ordered or not ordered at all. The materialized value type is
defined as the second type parameter of the returned Source, which allows journals to provide users with their
specialised query object, as demonstrated in the sample below:
final case class RichEvent(tags: Set[String], payload: Any)
// a plugin can provide:
case class QueryMetadata(deterministicOrder: Boolean, infinite: Boolean)
def byTagsWithMeta(tags: Set[String]): Source[RichEvent, QueryMetadata] = {
val query: Source[RichEvent, QueryMetadata] =
readJournal.byTagsWithMeta(Set("red", "blue"))
query
.mapMaterializedValue { meta =>
println(s"The query is: " +
s"ordered deterministically: ${meta.deterministicOrder}, " +
s"infinite: ${meta.infinite}")
}
.map { event => println(s"Event payload: ${event.payload}") }
.runWith(Sink.ignore)
4.10.4 Performance and denormalization
When building systems using Event sourcing and CQRS (Command & Query Responsibility Segregation) techniques it is tremendously important to realise that the write-side has completely different needs from the read-side,
and separating those concerns into datastores that are optimised for either side makes it possible to offer the best
experience for the write and read sides independently.
For example, in a bidding system it is important to “take the write” and respond to the bidder that we have accepted
the bid as soon as possible, which means that write-throughput is of highest importance for the write-side – often
this means that data stores which are able to scale to accommodate these requirements have a less expressive query
side.
On the other hand the same application may have some complex statistics view or we may have analysts working
with the data to figure out best bidding strategies and trends – this often requires some kind of expressive query
capabilities like for example SQL or writing Spark jobs to analyse the data. Therefore the data stored in the
write-side needs to be projected into the other read-optimised datastore.
Note: When referring to Materialized Views in Akka Persistence think of it as “some persistent storage of the
result of a Query”. In other words, it means that the view is created once, in order to be afterwards queried multiple
times, as in this format it may be more efficient or interesting to query it (instead of the source events directly).
4.10. Persistence Query
231
Akka Scala Documentation, Release 2.5.0-RC1
Materialize view to Reactive Streams compatible datastore
If the read datastore exposes a Reactive Streams interface then implementing a simple projection is as simple as,
using the read-journal and feeding it into the databases driver interface, for example like so:
implicit val system = ActorSystem()
implicit val mat = ActorMaterializer()
val readJournal =
PersistenceQuery(system).readJournalFor[MyScaladslReadJournal](JournalId)
val dbBatchWriter: Subscriber[immutable.Seq[Any]] =
ReactiveStreamsCompatibleDBDriver.batchWriter
// Using an example (Reactive Streams) Database driver
readJournal
.eventsByPersistenceId("user-1337")
.map(envelope => envelope.event)
.map(convertToReadSideTypes) // convert to datatype
.grouped(20) // batch inserts into groups of 20
.runWith(Sink.fromSubscriber(dbBatchWriter)) // write batches to read-side
˓→database
Materialize view using mapAsync
If the target database does not provide a reactive streams Subscriber that can perform writes, you may have to
implement the write logic using plain functions or Actors instead.
In case your write logic is state-less and you just need to convert the events from one data type to another before
writing into the alternative datastore, then the projection is as simple as:
trait ExampleStore {
def save(event: Any): Future[Unit]
}
val store: ExampleStore = ???
readJournal
.eventsByTag("bid")
.mapAsync(1) { e => store.save(e) }
.runWith(Sink.ignore)
Resumable projections
Sometimes you may need to implement “resumable” projections, that will not start from the beginning of time
each time when run. In this case you will need to store the sequence number (or offset) of the processed event
and use it the next time this projection is started. This pattern is not built-in, however is rather simple to implement
yourself.
The example below additionally highlights how you would use Actors to implement the write side, in case you
need to do some complex logic that would be best handled inside an Actor before persisting the event into the
other datastore:
import akka.pattern.ask
import system.dispatcher
implicit val timeout = Timeout(3.seconds)
val bidProjection = new MyResumableProjection("bid")
val writerProps = Props(classOf[TheOneWhoWritesToQueryJournal], "bid")
val writer = system.actorOf(writerProps, "bid-projection-writer")
4.10. Persistence Query
232
Akka Scala Documentation, Release 2.5.0-RC1
bidProjection.latestOffset.foreach { startFromOffset =>
readJournal
.eventsByTag("bid", Sequence(startFromOffset))
.mapAsync(8) { envelope => (writer ? envelope.event).map(_ => envelope.offset)
˓→}
.mapAsync(1) { offset => bidProjection.saveProgress(offset) }
.runWith(Sink.ignore)
}
class TheOneWhoWritesToQueryJournal(id: String) extends Actor {
val store = new DummyStore()
var state: ComplexState = ComplexState()
def receive = {
case m =>
state = updateState(state, m)
if (state.readyToSave) store.save(Record(state))
}
def updateState(state: ComplexState, msg: Any): ComplexState = {
// some complicated aggregation logic here ...
state
}
}
4.10.5 Query plugins
Query plugins are various (mostly community driven) ReadJournal implementations for all kinds of available
datastores. The complete list of available plugins is maintained on the Akka Persistence Query Community Plugins
page.
The plugin for LevelDB is described in Persistence Query for LevelDB.
This section aims to provide tips and guide plugin developers through implementing a custom query plugin. Most
users will not need to implement journals themselves, except if targeting a not yet supported datastore.
Note: Since different data stores provide different query capabilities journal plugins must extensively document
their exposed semantics as well as handled query scenarios.
ReadJournal plugin API
A read journal plugin must implement akka.persistence.query.ReadJournalProvider which creates instances of akka.persistence.query.scaladsl.ReadJournal and akka.persistence.
query.javaadsl.ReadJournal. The plugin must implement both the scaladsl and the javadsl
traits because the akka.stream.scaladsl.Source and akka.stream.javadsl.Source are different types and even though those types can easily be converted to each other it is most convenient for the end user
to get access to the Java or Scala directly. As illustrated below one of the implementations can delegate to the
other.
Below is a simple journal implementation:
class MyReadJournalProvider(system: ExtendedActorSystem, config: Config)
extends ReadJournalProvider {
override val scaladslReadJournal: MyScaladslReadJournal =
new MyScaladslReadJournal(system, config)
4.10. Persistence Query
233
Akka Scala Documentation, Release 2.5.0-RC1
override val javadslReadJournal: MyJavadslReadJournal =
new MyJavadslReadJournal(scaladslReadJournal)
}
class MyScaladslReadJournal(system: ExtendedActorSystem, config: Config)
extends akka.persistence.query.scaladsl.ReadJournal
with akka.persistence.query.scaladsl.EventsByTagQuery
with akka.persistence.query.scaladsl.EventsByPersistenceIdQuery
with akka.persistence.query.scaladsl.PersistenceIdsQuery
with akka.persistence.query.scaladsl.CurrentPersistenceIdsQuery {
private val refreshInterval: FiniteDuration =
config.getDuration("refresh-interval", MILLISECONDS).millis
/**
* You can use `NoOffset` to retrieve all events with a given tag or retrieve a
˓→subset of all
* events by specifying a `Sequence` `offset`. The `offset` corresponds to an
˓→ordered sequence number for
* the specific tag. Note that the corresponding offset of each event is
˓→provided in the
* [[akka.persistence.query.EventEnvelope]], which makes it possible to resume
˓→the
* stream at a later point from a given offset.
*
* The `offset` is exclusive, i.e. the event with the exact same sequence number
˓→will not be included
* in the returned stream. This means that you can use the offset that is
˓→returned in `EventEnvelope`
* as the `offset` parameter in a subsequent query.
*/
override def eventsByTag(
tag: String, offset: Offset = Sequence(0L)): Source[EventEnvelope, NotUsed] =
˓→offset match {
case Sequence(offsetValue) ⇒
val props = MyEventsByTagPublisher.props(tag, offsetValue, refreshInterval)
Source.actorPublisher[EventEnvelope](props)
.mapMaterializedValue(_ => NotUsed)
case NoOffset => eventsByTag(tag, Sequence(0L)) //recursive
case _ ⇒
throw new IllegalArgumentException("LevelDB does not support " + offset.
˓→getClass.getName + " offsets")
}
override def eventsByPersistenceId(
persistenceId: String, fromSequenceNr: Long = 0L,
toSequenceNr: Long = Long.MaxValue): Source[EventEnvelope, NotUsed] = {
// implement in a similar way as eventsByTag
???
}
override def persistenceIds(): Source[String, NotUsed] = {
// implement in a similar way as eventsByTag
???
}
override def currentPersistenceIds(): Source[String, NotUsed] = {
// implement in a similar way as eventsByTag
???
}
// possibility to add more plugin specific queries
4.10. Persistence Query
234
Akka Scala Documentation, Release 2.5.0-RC1
def byTagsWithMeta(tags: Set[String]): Source[RichEvent, QueryMetadata] = {
// implement in a similar way as eventsByTag
???
}
}
class MyJavadslReadJournal(scaladslReadJournal: MyScaladslReadJournal)
extends akka.persistence.query.javadsl.ReadJournal
with akka.persistence.query.javadsl.EventsByTagQuery
with akka.persistence.query.javadsl.EventsByPersistenceIdQuery
with akka.persistence.query.javadsl.PersistenceIdsQuery
with akka.persistence.query.javadsl.CurrentPersistenceIdsQuery {
override def eventsByTag(
tag: String, offset: Offset = Sequence(0L)): javadsl.Source[EventEnvelope,
˓→NotUsed] =
scaladslReadJournal.eventsByTag(tag, offset).asJava
override def eventsByPersistenceId(
persistenceId: String, fromSequenceNr: Long = 0L,
toSequenceNr: Long = Long.MaxValue): javadsl.Source[EventEnvelope, NotUsed] =
scaladslReadJournal.eventsByPersistenceId(
persistenceId, fromSequenceNr, toSequenceNr).asJava
override def persistenceIds(): javadsl.Source[String, NotUsed] =
scaladslReadJournal.persistenceIds().asJava
override def currentPersistenceIds(): javadsl.Source[String, NotUsed] =
scaladslReadJournal.currentPersistenceIds().asJava
// possibility to add more plugin specific queries
def byTagsWithMeta(
tags: java.util.Set[String]): javadsl.Source[RichEvent, QueryMetadata] = {
import scala.collection.JavaConverters._
scaladslReadJournal.byTagsWithMeta(tags.asScala.toSet).asJava
}
}
And the eventsByTag could be backed by such an Actor for example:
class MyEventsByTagPublisher(tag: String, offset: Long, refreshInterval:
˓→FiniteDuration)
extends ActorPublisher[EventEnvelope] {
private case object Continue
private val connection: java.sql.Connection = ???
private val Limit = 1000
private var currentOffset = offset
var buf = Vector.empty[EventEnvelope]
import context.dispatcher
val continueTask = context.system.scheduler.schedule(
refreshInterval, refreshInterval, self, Continue)
override def postStop(): Unit = {
continueTask.cancel()
}
def receive = {
4.10. Persistence Query
235
Akka Scala Documentation, Release 2.5.0-RC1
case _: Request | Continue =>
query()
deliverBuf()
case Cancel =>
context.stop(self)
}
object Select {
private def statement() = connection.prepareStatement(
"""
SELECT id, persistent_repr FROM journal
WHERE tag = ? AND id > ?
ORDER BY id LIMIT ?
""")
def run(tag: String, from: Long, limit: Int): Vector[(Long, Array[Byte])] = {
val s = statement()
try {
s.setString(1, tag)
s.setLong(2, from)
s.setLong(3, limit)
val rs = s.executeQuery()
val b = Vector.newBuilder[(Long, Array[Byte])]
while (rs.next())
b += (rs.getLong(1) -> rs.getBytes(2))
b.result()
} finally s.close()
}
}
def query(): Unit =
if (buf.isEmpty) {
try {
val result = Select.run(tag, currentOffset, Limit)
currentOffset = if (result.nonEmpty) result.last._1 else currentOffset
val serialization = SerializationExtension(context.system)
buf = result.map {
case (id, bytes) =>
val p = serialization.deserialize(bytes, classOf[PersistentRepr]).get
EventEnvelope(offset = Sequence(id), p.persistenceId, p.sequenceNr, p.
˓→payload)
}
} catch {
case e: Exception =>
onErrorThenStop(e)
}
}
final def deliverBuf(): Unit =
if (totalDemand > 0 && buf.nonEmpty) {
if (totalDemand <= Int.MaxValue) {
val (use, keep) = buf.splitAt(totalDemand.toInt)
buf = keep
use foreach onNext
} else {
buf foreach onNext
buf = Vector.empty
}
}
}
4.10. Persistence Query
236
Akka Scala Documentation, Release 2.5.0-RC1
The ReadJournalProvider class must have a constructor with one of these signatures:
• constructor with a ExtendedActorSystem parameter, a com.typesafe.config.Config parameter, and a String parameter for the config path
• constructor with a ExtendedActorSystem parameter, and a com.typesafe.config.Config parameter
• constructor with one ExtendedActorSystem parameter
• constructor without parameters
The plugin section of the actor system’s config will be passed in the config constructor parameter. The config path
of the plugin is passed in the String parameter.
If the underlying datastore only supports queries that are completed when they reach the end of the “result set”,
the journal has to submit new queries after a while in order to support “infinite” event streams that include events
stored after the initial query has completed. It is recommended that the plugin use a configuration property named
refresh-interval for defining such a refresh interval.
Plugin TCK
TODO, not available yet.
4.11 Persistence Query for LevelDB
This is documentation for the LevelDB implementation of the Persistence Query API. Note that implementations
for other journals may have different semantics.
4.11.1 Dependencies
Akka persistence LevelDB query implementation is bundled in the akka-persistence-query artifact.
Make sure that you have the following dependency in your project:
"com.typesafe.akka" %% "akka-persistence-query" % "2.5.0-RC1"
4.11.2 How to get the ReadJournal
The ReadJournal is retrieved via the akka.persistence.query.PersistenceQuery extension:
import akka.persistence.query.PersistenceQuery
import akka.persistence.query.journal.leveldb.scaladsl.LeveldbReadJournal
val queries = PersistenceQuery(system).readJournalFor[LeveldbReadJournal](
LeveldbReadJournal.Identifier)
4.11.3 Supported Queries
EventsByPersistenceIdQuery and CurrentEventsByPersistenceIdQuery
eventsByPersistenceId is used for retrieving events for a specific PersistentActor identified by
persistenceId.
4.11. Persistence Query for LevelDB
237
Akka Scala Documentation, Release 2.5.0-RC1
implicit val mat = ActorMaterializer()(system)
val queries = PersistenceQuery(system).readJournalFor[LeveldbReadJournal](
LeveldbReadJournal.Identifier)
val src: Source[EventEnvelope, NotUsed] =
queries.eventsByPersistenceId("some-persistence-id", 0L, Long.MaxValue)
val events: Source[Any, NotUsed] = src.map(_.event)
You can retrieve a subset of all events by specifying fromSequenceNr and toSequenceNr or use 0L and
Long.MaxValue respectively to retrieve all events. Note that the corresponding sequence number of each event
is provided in the EventEnvelope, which makes it possible to resume the stream at a later point from a given
sequence number.
The returned event stream is ordered by sequence number, i.e. the same order as the PersistentActor
persisted the events. The same prefix of stream elements (in same order) are returned for multiple executions of
the query, except for when events have been deleted.
The stream is not completed when it reaches the end of the currently stored events, but it continues to push new
events when new events are persisted. Corresponding query that is completed when it reaches the end of the
currently stored events is provided by currentEventsByPersistenceId.
The LevelDB write journal is notifying the query side as soon as events are persisted, but for efficiency
reasons the query side retrieves the events in batches that sometimes can be delayed up to the configured
refresh-interval or given RefreshInterval hint.
The stream is completed with failure if there is a failure in executing the query in the backend journal.
AllPersistenceIdsQuery and CurrentPersistenceIdsQuery
allPersistenceIds is used for retrieving all persistenceIds of all persistent actors.
implicit val mat = ActorMaterializer()(system)
val queries = PersistenceQuery(system).readJournalFor[LeveldbReadJournal](
LeveldbReadJournal.Identifier)
val src: Source[String, NotUsed] = queries.persistenceIds()
The returned event stream is unordered and you can expect different order for multiple executions of the query.
The stream is not completed when it reaches the end of the currently used persistenceIds, but it continues to push
new persistenceIds when new persistent actors are created. Corresponding query that is completed when it reaches
the end of the currently used persistenceIds is provided by currentPersistenceIds.
The LevelDB write journal is notifying the query side as soon as new persistenceIds are created and there
is no periodic polling or batching involved in this query.
The stream is completed with failure if there is a failure in executing the query in the backend journal.
EventsByTag and CurrentEventsByTag
eventsByTag is used for retrieving events that were marked with a given tag, e.g. all domain events of an
Aggregate Root type.
implicit val mat = ActorMaterializer()(system)
val queries = PersistenceQuery(system).readJournalFor[LeveldbReadJournal](
LeveldbReadJournal.Identifier)
val src: Source[EventEnvelope, NotUsed] =
queries.eventsByTag(tag = "green", offset = Sequence(0L))
4.11. Persistence Query for LevelDB
238
Akka Scala Documentation, Release 2.5.0-RC1
To tag events you create an Event Adapters that wraps the events in a akka.persistence.journal.
Tagged with the given tags.
import akka.persistence.journal.WriteEventAdapter
import akka.persistence.journal.Tagged
class MyTaggingEventAdapter extends WriteEventAdapter {
val colors = Set("green", "black", "blue")
override def toJournal(event: Any): Any = event match {
case s: String =>
var tags = colors.foldLeft(Set.empty[String]) { (acc, c) =>
if (s.contains(c)) acc + c else acc
}
if (tags.isEmpty) event
else Tagged(event, tags)
case _ => event
}
override def manifest(event: Any): String = ""
}
You can use NoOffset to retrieve all events with a given tag or retrieve a subset of all events by specifying a
Sequence offset. The offset corresponds to an ordered sequence number for the specific tag. Note that
the corresponding offset of each event is provided in the EventEnvelope, which makes it possible to resume
the stream at a later point from a given offset.
The offset is exclusive, i.e. the event with the exact same sequence number will not be included in the returned
stream. This means that you can use the offset that is returned in EventEnvelope as the offset parameter in
a subsequent query.
In addition to the offset the EventEnvelope also provides persistenceId and sequenceNr for each
event. The sequenceNr is the sequence number for the persistent actor with the persistenceId that persisted the event. The persistenceId + sequenceNr is an unique identifier for the event.
The returned event stream is ordered by the offset (tag sequence number), which corresponds to the same order as
the write journal stored the events. The same stream elements (in same order) are returned for multiple executions
of the query. Deleted events are not deleted from the tagged event stream.
Note: Events deleted using deleteMessages(toSequenceNr) are not deleted from the “tagged stream”.
The stream is not completed when it reaches the end of the currently stored events, but it continues to push new
events when new events are persisted. Corresponding query that is completed when it reaches the end of the
currently stored events is provided by currentEventsByTag.
The LevelDB write journal is notifying the query side as soon as tagged events are persisted, but for efficiency reasons the query side retrieves the events in batches that sometimes can be delayed up to the configured
refresh-interval or given RefreshInterval hint.
The stream is completed with failure if there is a failure in executing the query in the backend journal.
4.11.4 Configuration
Configuration settings can be defined in the configuration section with the absolute path corresponding to the identifier, which is "akka.persistence.query.journal.leveldb" for the default
LeveldbReadJournal.Identifier.
It can be configured with the following properties:
# Configuration for the LeveldbReadJournal
akka.persistence.query.journal.leveldb {
# Implementation class of the LevelDB ReadJournalProvider
4.11. Persistence Query for LevelDB
239
Akka Scala Documentation, Release 2.5.0-RC1
class = "akka.persistence.query.journal.leveldb.LeveldbReadJournalProvider"
# Absolute path to the write journal plugin configuration entry that this
# query journal will connect to. That must be a LeveldbJournal or
˓→SharedLeveldbJournal.
# If undefined (or "") it will connect to the default journal as specified by the
# akka.persistence.journal.plugin property.
write-plugin = ""
# The LevelDB write journal is notifying the query side as soon as things
# are persisted, but for efficiency reasons the query side retrieves the events
# in batches that sometimes can be delayed up to the configured `refresh˓→interval`.
refresh-interval = 3s
# How many events to fetch in one query (replay) and keep buffered until they
# are delivered downstreams.
max-buffer-size = 100
}
4.12 Testing Actor Systems
4.12.1 TestKit Example
Ray Roestenburg’s example code from his blog adapted to work with Akka 2.x.
import scala.util.Random
import org.scalatest.BeforeAndAfterAll
import org.scalatest.WordSpecLike
import org.scalatest.Matchers
import com.typesafe.config.ConfigFactory
import
import
import
import
import
import
import
akka.actor.Actor
akka.actor.ActorRef
akka.actor.ActorSystem
akka.actor.Props
akka.testkit.{ TestActors, DefaultTimeout, ImplicitSender, TestKit }
scala.concurrent.duration._
scala.collection.immutable
/**
* a Test to show some TestKit examples
*/
class TestKitUsageSpec
extends TestKit(ActorSystem(
"TestKitUsageSpec",
ConfigFactory.parseString(TestKitUsageSpec.config)))
with DefaultTimeout with ImplicitSender
with WordSpecLike with Matchers with BeforeAndAfterAll {
import TestKitUsageSpec._
val
val
val
val
val
val
echoRef = system.actorOf(TestActors.echoActorProps)
forwardRef = system.actorOf(Props(classOf[ForwardingActor], testActor))
filterRef = system.actorOf(Props(classOf[FilteringActor], testActor))
randomHead = Random.nextInt(6)
randomTail = Random.nextInt(10)
headList = immutable.Seq().padTo(randomHead, "0")
4.12. Testing Actor Systems
240
Akka Scala Documentation, Release 2.5.0-RC1
val tailList = immutable.Seq().padTo(randomTail, "1")
val seqRef =
system.actorOf(Props(classOf[SequencingActor], testActor, headList, tailList))
override def afterAll {
shutdown()
}
"An EchoActor" should {
"Respond with the same message it receives" in {
within(500 millis) {
echoRef ! "test"
expectMsg("test")
}
}
}
"A ForwardingActor" should {
"Forward a message it receives" in {
within(500 millis) {
forwardRef ! "test"
expectMsg("test")
}
}
}
"A FilteringActor" should {
"Filter all messages, except expected messagetypes it receives" in {
var messages = Seq[String]()
within(500 millis) {
filterRef ! "test"
expectMsg("test")
filterRef ! 1
expectNoMsg
filterRef ! "some"
filterRef ! "more"
filterRef ! 1
filterRef ! "text"
filterRef ! 1
receiveWhile(500 millis) {
case msg: String => messages = msg +: messages
}
}
messages.length should be(3)
messages.reverse should be(Seq("some", "more", "text"))
}
}
"A SequencingActor" should {
"receive an interesting message at some point " in {
within(500 millis) {
ignoreMsg {
case msg: String => msg != "something"
}
seqRef ! "something"
expectMsg("something")
ignoreMsg {
case msg: String => msg == "1"
}
expectNoMsg
ignoreNoMsg
}
}
}
}
4.12. Testing Actor Systems
241
Akka Scala Documentation, Release 2.5.0-RC1
object TestKitUsageSpec {
// Define your test specific configuration here
val config = """
akka {
loglevel = "WARNING"
}
"""
/**
* An Actor that forwards every message to a next Actor
*/
class ForwardingActor(next: ActorRef) extends Actor {
def receive = {
case msg => next ! msg
}
}
/**
* An Actor that only forwards certain messages to a next Actor
*/
class FilteringActor(next: ActorRef) extends Actor {
def receive = {
case msg: String => next ! msg
case _
=> None
}
}
/**
* An actor that sends a sequence of messages with a random head list, an
* interesting value and a random tail list. The idea is that you would
* like to test that the interesting value is received and that you cant
* be bothered with the rest
*/
class SequencingActor(next: ActorRef, head: immutable.Seq[String],
tail: immutable.Seq[String]) extends Actor {
def receive = {
case msg => {
head foreach { next ! _ }
next ! msg
tail foreach { next ! _ }
}
}
}
}
As with any piece of software, automated tests are a very important part of the development cycle. The actor
model presents a different view on how units of code are delimited and how they interact, which has an influence
on how to perform tests.
Akka comes with a dedicated module akka-testkit for supporting tests at different levels, which fall into two
clearly distinct categories:
• Testing isolated pieces of code without involving the actor model, meaning without multiple threads; this
implies completely deterministic behavior concerning the ordering of events and no concurrency concerns
and will be called Unit Testing in the following.
• Testing (multiple) encapsulated actors including multi-threaded scheduling; this implies non-deterministic
order of events but shielding from concurrency concerns by the actor model and will be called Integration
Testing in the following.
There are of course variations on the granularity of tests in both categories, where unit testing reaches down to
white-box tests and integration testing can encompass functional tests of complete actor networks. The important
4.12. Testing Actor Systems
242
Akka Scala Documentation, Release 2.5.0-RC1
distinction lies in whether concurrency concerns are part of the test or not. The tools offered are described in detail
in the following sections.
Note: Be sure to add the module akka-testkit to your dependencies.
4.12.2 Synchronous Unit Testing with TestActorRef
Testing the business logic inside Actor classes can be divided into two parts: first, each atomic operation must
work in isolation, then sequences of incoming events must be processed correctly, even in the presence of some
possible variability in the ordering of events. The former is the primary use case for single-threaded unit testing,
while the latter can only be verified in integration tests.
Normally, the ActorRef shields the underlying Actor instance from the outside, the only communications
channel is the actor’s mailbox. This restriction is an impediment to unit testing, which led to the inception of the
TestActorRef. This special type of reference is designed specifically for test purposes and allows access to
the actor in two ways: either by obtaining a reference to the underlying actor instance, or by invoking or querying
the actor’s behaviour (receive). Each one warrants its own section below.
Note: It is highly recommended to stick to traditional behavioural testing (using messaging to ask the Actor to
reply with the state you want to run assertions against), instead of using TestActorRef whenever possible.
Warning: Due to the synchronous nature of TestActorRef it will not work with some support traits that
Akka provides as they require asynchronous behaviours to function properly. Examples of traits that do not
mix well with test actor refs are PersistentActor and AtLeastOnceDelivery provided by Akka Persistence.
Obtaining a Reference to an Actor
Having access to the actual Actor object allows application of all traditional unit testing techniques on the
contained methods. Obtaining a reference is done like this:
import akka.testkit.TestActorRef
val actorRef = TestActorRef[MyActor]
val actor = actorRef.underlyingActor
Since TestActorRef is generic in the actor type it returns the underlying actor with its proper static type. From
this point on you may bring any unit testing tool to bear on your actor as usual.
Testing Finite State Machines
If your actor under test is a FSM, you may use the special TestFSMRef which offers all features of a normal
TestActorRef and in addition allows access to the internal state:
import akka.testkit.TestFSMRef
import akka.actor.FSM
import scala.concurrent.duration._
val fsm = TestFSMRef(new TestFsmActor)
val mustBeTypedProperly: TestActorRef[TestFsmActor] = fsm
assert(fsm.stateName == 1)
assert(fsm.stateData == "")
4.12. Testing Actor Systems
243
Akka Scala Documentation, Release 2.5.0-RC1
fsm ! "go" // being a TestActorRef, this runs also on the CallingThreadDispatcher
assert(fsm.stateName == 2)
assert(fsm.stateData == "go")
fsm.setState(stateName = 1)
assert(fsm.stateName == 1)
assert(fsm.isTimerActive("test") ==
fsm.setTimer("test", 12, 10 millis,
assert(fsm.isTimerActive("test") ==
fsm.cancelTimer("test")
assert(fsm.isTimerActive("test") ==
false)
true)
true)
false)
Due to a limitation in Scala’s type inference, there is only the factory method shown above, so you
will probably write code like TestFSMRef(new MyFSM) instead of the hypothetical ActorRef-inspired
TestFSMRef[MyFSM]. All methods shown above directly access the FSM state without any synchronization;
this is perfectly alright if the CallingThreadDispatcher is used and no other threads are involved, but it
may lead to surprises if you were to actually exercise timer events, because those are executed on the Scheduler
thread.
Testing the Actor’s Behavior
When the dispatcher invokes the processing behavior of an actor on a message, it actually calls apply on the
current behavior registered for the actor. This starts out with the return value of the declared receive method,
but it may also be changed using become and unbecome in response to external messages. All of this contributes to the overall actor behavior and it does not lend itself to easy testing on the Actor itself. Therefore the
TestActorRef offers a different mode of operation to complement the Actor testing: it supports all operations
also valid on normal ActorRef. Messages sent to the actor are processed synchronously on the current thread
and answers may be sent back as usual. This trick is made possible by the CallingThreadDispatcher
described below (see CallingThreadDispatcher); this dispatcher is set implicitly for any actor instantiated into a
TestActorRef.
import
import
import
import
akka.testkit.TestActorRef
scala.concurrent.duration._
scala.concurrent.Await
akka.pattern.ask
val actorRef = TestActorRef(new MyActor)
// hypothetical message stimulating a '42' answer
val future = actorRef ? Say42
val Success(result: Int) = future.value.get
result should be(42)
As the TestActorRef is a subclass of LocalActorRef with a few special extras, also aspects like supervision and restarting work properly, but beware that execution is only strictly synchronous as long as all actors
involved use the CallingThreadDispatcher. As soon as you add elements which include more sophisticated scheduling you leave the realm of unit testing as you then need to think about asynchronicity again (in most
cases the problem will be to wait until the desired effect had a chance to happen).
One more special aspect which is overridden for single-threaded tests is the receiveTimeout, as including
that would entail asynchronous queuing of ReceiveTimeout messages, violating the synchronous contract.
Note:
To summarize:
TestActorRef overwrites two fields:
it sets the dispatcher to
CallingThreadDispatcher.global and it sets the receiveTimeout to None.
4.12. Testing Actor Systems
244
Akka Scala Documentation, Release 2.5.0-RC1
The Way In-Between: Expecting Exceptions
If you want to test the actor behavior, including hotswapping, but without involving a dispatcher and without
having the TestActorRef swallow any thrown exceptions, then there is another mode available for you: just
use the receive method on TestActorRef, which will be forwarded to the underlying actor:
import akka.testkit.TestActorRef
val actorRef = TestActorRef(new Actor {
def receive = {
case "hello" => throw new IllegalArgumentException("boom")
}
})
intercept[IllegalArgumentException] { actorRef.receive("hello") }
Use Cases
You may of course mix and match both modi operandi of TestActorRef as suits your test needs:
• one common use case is setting up the actor into a specific internal state before sending the test message
• another is to verify correct internal state transitions after having sent the test message
Feel free to experiment with the possibilities, and if you find useful patterns, don’t hesitate to let the Akka forums
know about them! Who knows, common operations might even be worked into nice DSLs.
4.12.3 Asynchronous Integration Testing with TestKit
When you are reasonably sure that your actor’s business logic is correct, the next step is verifying that it works
correctly within its intended environment (if the individual actors are simple enough, possibly because they use the
FSM module, this might also be the first step). The definition of the environment depends of course very much on
the problem at hand and the level at which you intend to test, ranging for functional/integration tests to full system
tests. The minimal setup consists of the test procedure, which provides the desired stimuli, the actor under test,
and an actor receiving replies. Bigger systems replace the actor under test with a network of actors, apply stimuli
at varying injection points and arrange results to be sent from different emission points, but the basic principle
stays the same in that a single procedure drives the test.
The TestKit class contains a collection of tools which makes this common task easy.
import akka.actor.ActorSystem
import akka.testkit.{ ImplicitSender, TestActors, TestKit }
import org.scalatest.{ BeforeAndAfterAll, Matchers, WordSpecLike }
class MySpec() extends TestKit(ActorSystem("MySpec")) with ImplicitSender
with WordSpecLike with Matchers with BeforeAndAfterAll {
override def afterAll {
TestKit.shutdownActorSystem(system)
}
"An Echo actor" must {
"send back messages unchanged" in {
val echo = system.actorOf(TestActors.echoActorProps)
echo ! "hello world"
expectMsg("hello world")
}
}
}
4.12. Testing Actor Systems
245
Akka Scala Documentation, Release 2.5.0-RC1
The TestKit contains an actor named testActor which is the entry point for messages to be examined with
the various expectMsg... assertions detailed below. When mixing in the trait ImplicitSender this test
actor is implicitly used as sender reference when dispatching messages from the test procedure. The testActor
may also be passed to other actors as usual, usually subscribing it as notification listener. There is a whole set
of examination methods, e.g. receiving all consecutive messages matching certain criteria, receiving a whole
sequence of fixed messages or classes, receiving nothing for some time, etc.
The ActorSystem passed in to the constructor of TestKit is accessible via the system member. Remember to
shut down the actor system after the test is finished (also in case of failure) so that all actors—including the test
actor—are stopped.
Built-In Assertions
The above mentioned expectMsg is not the only method for formulating assertions concerning received messages. Here is the full list:
• expectMsg[T](d: Duration, msg: T): T
The given message object must be received within the specified time; the object will be returned.
• expectMsgPF[T](d: Duration)(pf: PartialFunction[Any, T]): T
Within the given time period, a message must be received and the given partial function must be defined for
that message; the result from applying the partial function to the received message is returned. The duration
may be left unspecified (empty parentheses are required in this case) to use the deadline from the innermost
enclosing within block instead.
• expectMsgClass[T](d: Duration, c: Class[T]): T
An object which is an instance of the given Class must be received within the allotted time frame; the
object will be returned. Note that this does a conformance check; if you need the class to be equal, have a
look at expectMsgAllClassOf with a single given class argument.
• expectMsgType[T: Manifest](d: Duration)
An object which is an instance of the given type (after erasure) must be received within the
allotted time frame; the object will be returned.
This method is approximately equivalent to
expectMsgClass(implicitly[ClassTag[T]].runtimeClass).
• expectMsgAnyOf[T](d: Duration, obj: T*): T
An object must be received within the given time, and it must be equal ( compared with ==) to at least one
of the passed reference objects; the received object will be returned.
• expectMsgAnyClassOf[T](d: Duration, obj: Class[_ <: T]*): T
An object must be received within the given time, and it must be an instance of at least one of the supplied
Class objects; the received object will be returned.
• expectMsgAllOf[T](d: Duration, obj: T*): Seq[T]
A number of objects matching the size of the supplied object array must be received within the given
time, and for each of the given objects there must exist at least one among the received ones which equals
(compared with ==) it. The full sequence of received objects is returned.
• expectMsgAllClassOf[T](d: Duration, c: Class[_ <: T]*): Seq[T]
A number of objects matching the size of the supplied Class array must be received within the given time,
and for each of the given classes there must exist at least one among the received objects whose class equals
(compared with ==) it (this is not a conformance check). The full sequence of received objects is returned.
• expectMsgAllConformingOf[T](d: Duration, c: Class[_ <: T]*): Seq[T]
A number of objects matching the size of the supplied Class array must be received within the given
time, and for each of the given classes there must exist at least one among the received objects which is an
instance of this class. The full sequence of received objects is returned.
4.12. Testing Actor Systems
246
Akka Scala Documentation, Release 2.5.0-RC1
• expectNoMsg(d: Duration)
No message must be received within the given time. This also fails if a message has been received before
calling this method which has not been removed from the queue using one of the other methods.
• receiveN(n: Int, d: Duration): Seq[AnyRef]
n messages must be received within the given time; the received messages are returned.
• fishForMessage(max: Duration, hint: String)(pf: PartialFunction[Any,
Boolean]): Any
Keep receiving messages as long as the time is not used up and the partial function matches and returns
false. Returns the message received for which it returned true or throws an exception, which will
include the provided hint for easier debugging.
In addition to message reception assertions there are also methods which help with message flows:
• receiveOne(d: Duration): AnyRef
Tries to receive one message for at most the given time interval and returns null in case of failure. If the
given Duration is zero, the call is non-blocking (polling mode).
• receiveWhile[T](max: Duration, idle: Duration, messages: Int)(pf: PartialFunction[An
T]): Seq[T]
Collect messages as long as
– they are matching the given partial function
– the given time interval is not used up
– the next message is received within the idle timeout
– the number of messages has not yet reached the maximum
All collected messages are returned. The maximum duration defaults to the time remaining in the innermost
enclosing within block and the idle duration defaults to infinity (thereby disabling the idle timeout feature).
The number of expected messages defaults to Int.MaxValue, which effectively disables this limit.
• awaitCond(p: => Boolean, max: Duration, interval: Duration)
Poll the given condition every interval until it returns true or the max duration is used up. The interval
defaults to 100 ms and the maximum defaults to the time remaining in the innermost enclosing within block.
• awaitAssert(a: => Any, max: Duration, interval: Duration)
Poll the given assert function every interval until it does not throw an exception or the max duration
is used up. If the timeout expires the last exception is thrown. The interval defaults to 100 ms and the
maximum defaults to the time remaining in the innermost enclosing within block.The interval defaults to
100 ms and the maximum defaults to the time remaining in the innermost enclosing within block.
• ignoreMsg(pf: PartialFunction[AnyRef, Boolean])
ignoreNoMsg
The internal testActor contains a partial function for ignoring messages: it will only enqueue messages
which do not match the function or for which the function returns false. This function can be set and reset
using the methods given above; each invocation replaces the previous function, they are not composed.
This feature is useful e.g. when testing a logging system, where you want to ignore regular messages and
are only interested in your specific ones.
Expecting Log Messages
Since an integration test does not allow to the internal processing of the participating actors, verifying expected
exceptions cannot be done directly. Instead, use the logging system for this purpose: replacing the normal event
handler with the TestEventListener and using an EventFilter allows assertions on log messages, including those which are generated by exceptions:
4.12. Testing Actor Systems
247
Akka Scala Documentation, Release 2.5.0-RC1
import akka.testkit.EventFilter
import com.typesafe.config.ConfigFactory
implicit val system = ActorSystem("testsystem", ConfigFactory.parseString("""
akka.loggers = ["akka.testkit.TestEventListener"]
"""))
try {
val actor = system.actorOf(Props.empty)
EventFilter[ActorKilledException](occurrences = 1) intercept {
actor ! Kill
}
} finally {
shutdown(system)
}
If a number of occurrences is specific—as demonstrated above—then intercept will block until that number
of matching messages have been received or the timeout configured in akka.test.filter-leeway is used
up (time starts counting after the passed-in block of code returns). In case of a timeout the test fails.
Note: Be sure to exchange the default logger with the TestEventListener in your application.conf
to enable this function:
akka.loggers = [akka.testkit.TestEventListener]
Timing Assertions
Another important part of functional testing concerns timing: certain events must not happen immediately (like a
timer), others need to happen before a deadline. Therefore, all examination methods accept an upper time limit
within the positive or negative result must be obtained. Lower time limits need to be checked external to the
examination, which is facilitated by a new construct for managing time constraints:
within([min, ]max) {
...
}
The block given to within must complete after a Duration which is between min and max, where the former
defaults to zero. The deadline calculated by adding the max parameter to the block’s start time is implicitly
available within the block to all examination methods, if you do not specify it, it is inherited from the innermost
enclosing within block.
It should be noted that if the last message-receiving assertion of the block is expectNoMsg or receiveWhile,
the final check of the within is skipped in order to avoid false positives due to wake-up latencies. This means
that while individual contained assertions still use the maximum time bound, the overall block may take arbitrarily
longer in this case.
import akka.actor.Props
import scala.concurrent.duration._
val worker = system.actorOf(Props[Worker])
within(200 millis) {
worker ! "some work"
expectMsg("some result")
expectNoMsg // will block for the rest of the 200ms
Thread.sleep(300) // will NOT make this block fail
}
4.12. Testing Actor Systems
248
Akka Scala Documentation, Release 2.5.0-RC1
Note: All times are measured using System.nanoTime, meaning that they describe wall time, not CPU time.
Ray Roestenburg has written a great article on using the TestKit: http://roestenburg.agilesquad.com/2011/02/
unit-testing-akka-actors-with-testkit_12.html. His full example is also available here.
Accounting for Slow Test Systems
The tight timeouts you use during testing on your lightning-fast notebook will invariably lead to spurious test
failures on the heavily loaded Jenkins server (or similar). To account for this situation, all maximum durations are
internally scaled by a factor taken from the Configuration, akka.test.timefactor, which defaults to 1.
You can scale other durations with the same factor by using the implicit conversion in akka.testkit package
object to add dilated function to Duration.
import scala.concurrent.duration._
import akka.testkit._
10.milliseconds.dilated
Resolving Conflicts with Implicit ActorRef
If you want the sender of messages inside your TestKit-based tests to be the testActor simply mix in
ImplicitSender into your test.
class MySpec() extends TestKit(ActorSystem("MySpec")) with ImplicitSender
with WordSpecLike with Matchers with BeforeAndAfterAll {
Using Multiple Probe Actors
When the actors under test are supposed to send various messages to different destinations, it may be difficult
distinguishing the message streams arriving at the testActor when using the TestKit as a mixin. Another
approach is to use it for creation of simple probe actors to be inserted in the message flows. To make this more
powerful and convenient, there is a concrete implementation called TestProbe. The functionality is best explained using a small example:
import
import
import
import
scala.concurrent.duration._
scala.concurrent.Future
akka.actor._
akka.testkit.TestProbe
class MyDoubleEcho extends Actor {
var dest1: ActorRef = _
var dest2: ActorRef = _
def receive = {
case (d1: ActorRef, d2: ActorRef) =>
dest1 = d1
dest2 = d2
case x =>
dest1 ! x
dest2 ! x
}
}
val probe1 = TestProbe()
val probe2 = TestProbe()
val actor = system.actorOf(Props[MyDoubleEcho])
actor ! ((probe1.ref, probe2.ref))
4.12. Testing Actor Systems
249
Akka Scala Documentation, Release 2.5.0-RC1
actor ! "hello"
probe1.expectMsg(500 millis, "hello")
probe2.expectMsg(500 millis, "hello")
Here a the system under test is simulated by MyDoubleEcho, which is supposed to mirror its input to two
outputs. Attaching two test probes enables verification of the (simplistic) behavior. Another example would
be two actors A and B which collaborate by A sending messages to B. In order to verify this message flow, a
TestProbe could be inserted as target of A, using the forwarding capabilities or auto-pilot described below to
include a real B in the test setup.
If you have many test probes, you can name them to get meaningful actor names in test logs and assertions:
val worker = TestProbe("worker")
val aggregator = TestProbe("aggregator")
worker.ref.path.name should startWith("worker")
aggregator.ref.path.name should startWith("aggregator")
Probes may also be equipped with custom assertions to make your test code even more concise and clear:
final case class Update(id: Int, value: String)
val probe = new TestProbe(system) {
def expectUpdate(x: Int) = {
expectMsgPF() {
case Update(id, _) if id == x => true
}
sender() ! "ACK"
}
}
You have complete flexibility here in mixing and matching the TestKit facilities with your own checks and
choosing an intuitive name for it. In real life your code will probably be a bit more complicated than the example
given above; just use the power!
Warning: Any message send from a TestProbe to another actor which runs on the CallingThreadDispatcher runs the risk of dead-lock, if that other actor might also send to this probe. The implementation of
TestProbe.watch and TestProbe.unwatch will also send a message to the watchee, which means
that it is dangerous to try watching e.g. TestActorRef from a TestProbe.
Watching Other Actors from Probes
A TestProbe can register itself for DeathWatch of any other actor:
val probe = TestProbe()
probe watch target
target ! PoisonPill
probe.expectTerminated(target)
Replying to Messages Received by Probes
The probes keep track of the communications channel for replies, if possible, so they can also reply:
val probe = TestProbe()
val future = probe.ref ? "hello"
probe.expectMsg(0 millis, "hello") // TestActor runs on CallingThreadDispatcher
4.12. Testing Actor Systems
250
Akka Scala Documentation, Release 2.5.0-RC1
probe.reply("world")
assert(future.isCompleted && future.value == Some(Success("world")))
Forwarding Messages Received by Probes
Given a destination actor dest which in the nominal actor network would receive a message from actor source.
If you arrange for the message to be sent to a TestProbe probe instead, you can make assertions concerning
volume and timing of the message flow while still keeping the network functioning:
class Source(target: ActorRef) extends Actor {
def receive = {
case "start" => target ! "work"
}
}
class Destination extends Actor {
def receive = {
case x => // Do something..
}
}
val probe = TestProbe()
val source = system.actorOf(Props(classOf[Source], probe.ref))
val dest = system.actorOf(Props[Destination])
source ! "start"
probe.expectMsg("work")
probe.forward(dest)
The dest actor will receive the same message invocation as if no test probe had intervened.
Auto-Pilot
Receiving messages in a queue for later inspection is nice, but in order to keep a test running and verify traces later
you can also install an AutoPilot in the participating test probes (actually in any TestKit) which is invoked
before enqueueing to the inspection queue. This code can be used to forward messages, e.g. in a chain A -->
Probe --> B, as long as a certain protocol is obeyed.
val probe = TestProbe()
probe.setAutoPilot(new TestActor.AutoPilot {
def run(sender: ActorRef, msg: Any): TestActor.AutoPilot =
msg match {
case "stop" ⇒ TestActor.NoAutoPilot
case x
⇒ testActor.tell(x, sender); TestActor.KeepRunning
}
})
The run method must return the auto-pilot for the next message, which may be KeepRunning to retain the
current one or NoAutoPilot to switch it off.
Caution about Timing Assertions
The behavior of within blocks when using test probes might be perceived as counter-intuitive: you need to
remember that the nicely scoped deadline as described above is local to each probe. Hence, probes do not react to
each other’s deadlines or to the deadline set in an enclosing TestKit instance:
4.12. Testing Actor Systems
251
Akka Scala Documentation, Release 2.5.0-RC1
val probe = TestProbe()
within(1 second) {
probe.expectMsg("hello")
}
Here, the expectMsg call will use the default timeout.
Testing parent-child relationships
The parent of an actor is always the actor that created it. At times this leads to a coupling between the two that
may not be straightforward to test. There are several approaches to improve testability of a child actor that needs
to refer to its parent:
1. when creating a child, pass an explicit reference to its parent
2. create the child with a TestProbe as parent
3. create a fabricated parent when testing
Conversely, a parent’s binding to its child can be lessened as follows:
4. when creating a parent, tell the parent how to create its child
For example, the structure of the code you want to test may follow this pattern:
class Parent extends Actor {
val child = context.actorOf(Props[Child], "child")
var ponged = false
def receive = {
case "pingit" => child ! "ping"
case "pong"
=> ponged = true
}
}
class Child extends Actor {
def receive = {
case "ping" => context.parent ! "pong"
}
}
Introduce child to its parent
The first option is to avoid use of the context.parent function and create a child with a custom parent by
passing an explicit reference to its parent instead.
class DependentChild(parent: ActorRef) extends Actor {
def receive = {
case "ping" => parent ! "pong"
}
}
Create the child using TestProbe
The TestProbe class can in fact create actors that will run with the test probe as parent. This will cause any
messages the child actor sends to context.parent to end up in the test probe.
4.12. Testing Actor Systems
252
Akka Scala Documentation, Release 2.5.0-RC1
"A TestProbe serving as parent" should {
"test its child responses" in {
val parent = TestProbe()
val child = parent.childActorOf(Props[Child])
parent.send(child, "ping")
parent.expectMsg("pong")
}
}
Using a fabricated parent
If you prefer to avoid modifying the parent or child constructor you can create a fabricated parent in your test.
This, however, does not enable you to test the parent actor in isolation.
"A fabricated parent" should {
"test its child responses" in {
val proxy = TestProbe()
val parent = system.actorOf(Props(new Actor {
val child = context.actorOf(Props[Child], "child")
def receive = {
case x if sender == child => proxy.ref forward x
case x
=> child forward x
}
}))
proxy.send(parent, "ping")
proxy.expectMsg("pong")
}
}
Externalize child making from the parent
Alternatively, you can tell the parent how to create its child. There are two ways to do this: by giving it a Props
object or by giving it a function which takes care of creating the child actor:
class DependentParent(childProps: Props) extends Actor {
val child = context.actorOf(childProps, "child")
var ponged = false
def receive = {
case "pingit" => child ! "ping"
case "pong"
=> ponged = true
}
}
class GenericDependentParent(childMaker: ActorRefFactory => ActorRef) extends
˓→Actor {
val child = childMaker(context)
var ponged = false
def receive = {
case "pingit" => child ! "ping"
case "pong"
=> ponged = true
}
}
Creating the Props is straightforward and the function may look like this in your test code:
4.12. Testing Actor Systems
253
Akka Scala Documentation, Release 2.5.0-RC1
val maker = (_: ActorRefFactory) => probe.ref
val parent = system.actorOf(Props(classOf[GenericDependentParent], maker))
And like this in your application code:
val maker = (f: ActorRefFactory) => f.actorOf(Props[Child])
val parent = system.actorOf(Props(classOf[GenericDependentParent], maker))
Which of these methods is the best depends on what is most important to test. The most generic option is to create
the parent actor by passing it a function that is responsible for the Actor creation, but the fabricated parent is often
sufficient.
4.12.4 CallingThreadDispatcher
The CallingThreadDispatcher serves good purposes in unit testing, as described above, but originally
it was conceived in order to allow contiguous stack traces to be generated in case of an error. As this special
dispatcher runs everything which would normally be queued directly on the current thread, the full history of a
message’s processing chain is recorded on the call stack, so long as all intervening actors run on this dispatcher.
How to use it
Just set the dispatcher as you normally would:
import akka.testkit.CallingThreadDispatcher
val ref = system.actorOf(Props[MyActor].withDispatcher(CallingThreadDispatcher.Id))
How it works
When receiving an invocation, the CallingThreadDispatcher checks whether the receiving actor is already
active on the current thread. The simplest example for this situation is an actor which sends a message to itself.
In this case, processing cannot continue immediately as that would violate the actor model, so the invocation
is queued and will be processed when the active invocation on that actor finishes its processing; thus, it will
be processed on the calling thread, but simply after the actor finishes its previous work. In the other case, the
invocation is simply processed immediately on the current thread. Futures scheduled via this dispatcher are also
executed immediately.
This scheme makes the CallingThreadDispatcher work like a general purpose dispatcher for any actors
which never block on external events.
In the presence of multiple threads it may happen that two invocations of an actor running on this dispatcher
happen on two different threads at the same time. In this case, both will be processed directly on their respective
threads, where both compete for the actor’s lock and the loser has to wait. Thus, the actor model is left intact, but
the price is loss of concurrency due to limited scheduling. In a sense this is equivalent to traditional mutex style
concurrency.
The other remaining difficulty is correct handling of suspend and resume: when an actor is suspended, subsequent
invocations will be queued in thread-local queues (the same ones used for queuing in the normal case). The call
to resume, however, is done by one specific thread, and all other threads in the system will probably not be
executing this specific actor, which leads to the problem that the thread-local queues cannot be emptied by their
native threads. Hence, the thread calling resume will collect all currently queued invocations from all threads
into its own queue and process them.
4.12. Testing Actor Systems
254
Akka Scala Documentation, Release 2.5.0-RC1
Limitations
Warning: In case the CallingThreadDispatcher is used for top-level actors, but without going through TestActorRef, then there is a time window during which the actor is awaiting construction by the user guardian
actor. Sending messages to the actor during this time period will result in them being enqueued and then
executed on the guardian’s thread instead of the caller’s thread. To avoid this, use TestActorRef.
If an actor’s behavior blocks on a something which would normally be affected by the calling actor after having
sent the message, this will obviously dead-lock when using this dispatcher. This is a common scenario in actor
tests based on CountDownLatch for synchronization:
val latch = new CountDownLatch(1)
actor ! startWorkAfter(latch)
// actor will call latch.await() before proceeding
doSomeSetupStuff()
latch.countDown()
The example would hang indefinitely within the message processing initiated on the second line and never reach
the fourth line, which would unblock it on a normal dispatcher.
Thus, keep in mind that the CallingThreadDispatcher is not a general-purpose replacement for the normal
dispatchers. On the other hand it may be quite useful to run your actor network on it for testing, because if it runs
without dead-locking chances are very high that it will not dead-lock in production.
Warning: The above sentence is unfortunately not a strong guarantee, because your code might directly or
indirectly change its behavior when running on a different dispatcher. If you are looking for a tool to help
you debug dead-locks, the CallingThreadDispatcher may help with certain error scenarios, but keep
in mind that it has may give false negatives as well as false positives.
Thread Interruptions
If the CallingThreadDispatcher sees that the current thread has its isInterrupted() flag set when message
processing returns, it will throw an InterruptedException after finishing all its processing (i.e. all messages which need processing as described above are processed before this happens). As tell cannot throw
exceptions due to its contract, this exception will then be caught and logged, and the thread’s interrupted status
will be set again.
If during message processing an InterruptedException is thrown then it will be caught inside the CallingThreadDispatcher’s message handling loop, the thread’s interrupted flag will be set and processing continues
normally.
Note: The summary of these two paragraphs is that if the current thread is interrupted while doing work under
the CallingThreadDispatcher, then that will result in the isInterrupted flag to be true when the message
send returns and no InterruptedException will be thrown.
Benefits
To summarize, these are the features with the CallingThreadDispatcher has to offer:
• Deterministic execution of single-threaded tests while retaining nearly full actor semantics
• Full message processing history leading up to the point of failure in exception stack traces
• Exclusion of certain classes of dead-lock scenarios
4.12. Testing Actor Systems
255
Akka Scala Documentation, Release 2.5.0-RC1
4.12.5 Tracing Actor Invocations
The testing facilities described up to this point were aiming at formulating assertions about a system’s behavior.
If a test fails, it is usually your job to find the cause, fix it and verify the test again. This process is supported by
debuggers as well as logging, where the Akka toolkit offers the following options:
• Logging of exceptions thrown within Actor instances
This is always on; in contrast to the other logging mechanisms, this logs at ERROR level.
• Logging of message invocations on certain actors
This is enabled by a setting in the Configuration — namely akka.actor.debug.receive — which
enables the loggable statement to be applied to an actor’s receive function:
import akka.event.LoggingReceive
def receive = LoggingReceive {
case msg => // Do something ...
}
def otherState: Receive = LoggingReceive.withLabel("other") {
case msg => // Do something else ...
}
If the aforementioned setting is not given in the Configuration, this method will pass through the given Receive
function unmodified, meaning that there is no runtime cost unless actually enabled.
The logging feature is coupled to this specific local mark-up because enabling it uniformly on all actors is not
usually what you need, and it would lead to endless loops if it were applied to event bus logger listeners.
• Logging of special messages
Actors handle certain special messages automatically, e.g. Kill, PoisonPill, etc. Tracing of these
message invocations is enabled by the setting akka.actor.debug.autoreceive, which enables this
on all actors.
• Logging of the actor lifecycle
Actor creation, start, restart, monitor start, monitor stop and stop may be traced by enabling the setting
akka.actor.debug.lifecycle; this, too, is enabled uniformly on all actors.
All these messages are logged at DEBUG level. To summarize, you can enable full logging of actor activities using
this configuration fragment:
akka {
loglevel = "DEBUG"
actor {
debug {
receive = on
autoreceive = on
lifecycle = on
}
}
}
4.12.6 Different Testing Frameworks
Akka’s own test suite is written using ScalaTest, which also shines through in documentation examples. However,
the TestKit and its facilities do not depend on that framework, you can essentially use whichever suits your
development style best.
This section contains a collection of known gotchas with some other frameworks, which is by no means exhaustive
and does not imply endorsement or special support.
4.12. Testing Actor Systems
256
Akka Scala Documentation, Release 2.5.0-RC1
When you need it to be a trait
If for some reason it is a problem to inherit from TestKit due to it being a concrete class instead of a trait,
there’s TestKitBase:
import akka.testkit.TestKitBase
class MyTest extends TestKitBase {
implicit lazy val system = ActorSystem()
// put your test code here ...
shutdown(system)
}
The implicit lazy val system must be declared exactly like that (you can of course pass arguments to
the actor system factory as needed) because trait TestKitBase needs the system during its construction.
Warning: Use of the trait is discouraged because of potential issues with binary backwards compatibility in
the future, use at own risk.
Specs2
Some Specs2 users have contributed examples of how to work around some clashes which may arise:
• Mixing TestKit into org.specs2.mutable.Specification results in a name clash involving the
end method (which is a private variable in TestKit and an abstract method in Specification); if mixing in
TestKit first, the code may compile but might then fail at runtime. The work-around—which is actually beneficial also for the third point—is to apply the TestKit together with org.specs2.specification.
Scope.
• The Specification traits provide a Duration DSL which uses partly the same method names as scala.
concurrent.duration.Duration, resulting in ambiguous implicits if scala.concurrent.
duration._ is imported. There are two workarounds:
– either use the Specification variant of Duration and supply an implicit conversion to the Akka Duration.
This conversion is not supplied with the Akka distribution because that would mean that our JAR files
would depend on Specs2, which is not justified by this little feature.
– or mix org.specs2.time.NoTimeConversions into the Specification.
• Specifications are by default executed concurrently, which requires some care when writing the tests or
alternatively the sequential keyword.
4.12.7 Configuration
There are several configuration properties for the TestKit module, please refer to the reference configuration.
4.13 Actor DSL
Warning: Actor DSL is deprecated and will be removed in the near future. Use plain system.actorOf
or context.actorOf instead.
4.13. Actor DSL
257
Akka Scala Documentation, Release 2.5.0-RC1
4.13.1 The Actor DSL
Simple actors—for example one-off workers or even when trying things out in the REPL—can be created more
concisely using the Act trait. The supporting infrastructure is bundled in the following import:
import akka.actor.ActorDSL._
import akka.actor.ActorSystem
implicit val system = ActorSystem("demo")
This import is assumed for all code samples throughout this section. The implicit actor system serves as
ActorRefFactory for all examples below. To define a simple actor, the following is sufficient:
val a = actor(new Act {
become {
case "hello" ⇒ sender() ! "hi"
}
})
Here, actor takes the role of either system.actorOf or context.actorOf, depending on which context
it is called in: it takes an implicit ActorRefFactory, which within an actor is available in the form of the
implicit val context: ActorContext. Outside of an actor, you’ll have to either declare an implicit
ActorSystem, or you can give the factory explicitly (see further below).
The two possible ways of issuing a context.become (replacing or adding the new behavior) are offered separately to enable a clutter-free notation of nested receives:
val a = actor(new Act {
become { // this will replace the initial (empty) behavior
case "info" ⇒ sender() ! "A"
case "switch" ⇒
becomeStacked { // this will stack upon the "A" behavior
case "info"
⇒ sender() ! "B"
case "switch" ⇒ unbecome() // return to the "A" behavior
}
case "lobotomize" ⇒ unbecome() // OH NOES: Actor.emptyBehavior
}
})
Please note that calling unbecome more often than becomeStacked results in the original behavior being
installed, which in case of the Act trait is the empty behavior (the outer become just replaces it during construction).
Life-cycle management
Life-cycle hooks are also exposed as DSL elements (see Start Hook and Stop Hook), where later invocations of
the methods shown below will replace the contents of the respective hooks:
val a = actor(new Act {
whenStarting { testActor ! "started" }
whenStopping { testActor ! "stopped" }
})
The above is enough if the logical life-cycle of the actor matches the restart cycles (i.e. whenStopping is
executed before a restart and whenStarting afterwards). If that is not desired, use the following two hooks
(see Restart Hooks):
val a = actor(new Act {
become {
case "die" ⇒ throw new Exception
}
4.13. Actor DSL
258
Akka Scala Documentation, Release 2.5.0-RC1
whenFailing { case m @ (cause, msg) ⇒ testActor ! m }
whenRestarted { cause ⇒ testActor ! cause }
})
It is also possible to create nested actors, i.e. grand-children, like this:
// here we pass in the ActorRefFactory explicitly as an example
val a = actor(system, "fred")(new Act {
val b = actor("barney")(new Act {
whenStarting { context.parent ! ("hello from " + self.path) }
})
become {
case x ⇒ testActor ! x
}
})
Note: In some cases it will be necessary to explicitly pass the ActorRefFactory to the actor method (you
will notice when the compiler tells you about ambiguous implicits).
The grand-child will be supervised by the child; the supervisor strategy for this relationship can also be configured
using a DSL element (supervision directives are part of the Act trait):
superviseWith(OneForOneStrategy() {
case e: Exception if e.getMessage == "hello" ⇒ Stop
case _: Exception
⇒ Resume
})
Actor with Stash
Last but not least there is a little bit of convenience magic built-in, which detects if the runtime class of the statically
given actor subtype extends the RequiresMessageQueue trait via the Stash trait (this is a complicated way
of saying that new Act with Stash would not work because its runtime erased type is just an anonymous
subtype of Act). The purpose is to automatically use the appropriate deque-based mailbox type required by
Stash. If you want to use this magic, simply extend ActWithStash:
val a = actor(new ActWithStash {
become {
case 1 ⇒ stash()
case 2 ⇒
testActor ! 2; unstashAll(); becomeStacked {
case 1 ⇒ testActor ! 1; unbecome()
}
}
})
4.14 Typed Actors
Note: This module will be deprecated as it will be superseded by the Akka Typed project which is currently being
developed in open preview mode.
Akka Typed Actors is an implementation of the Active Objects pattern. Essentially turning method invocations
into asynchronous dispatch instead of synchronous that has been the default way since Smalltalk came out.
4.14. Typed Actors
259
Akka Scala Documentation, Release 2.5.0-RC1
Typed Actors consist of 2 “parts”, a public interface and an implementation, and if you’ve done any work in
“enterprise” Java, this will be very familiar to you. As with normal Actors you have an external API (the public
interface instance) that will delegate method calls asynchronously to a private instance of the implementation.
The advantage of Typed Actors vs. Actors is that with TypedActors you have a static contract, and don’t need to
define your own messages, the downside is that it places some limitations on what you can do and what you can’t,
i.e. you cannot use become/unbecome.
Typed Actors are implemented using JDK Proxies which provide a pretty easy-worked API to intercept method
calls.
Note: Just as with regular Akka Actors, Typed Actors process one call at a time.
4.14.1 When to use Typed Actors
Typed actors are nice for bridging between actor systems (the “inside”) and non-actor code (the “outside”), because
they allow you to write normal OO-looking code on the outside. Think of them like doors: their practicality lies
in interfacing between private sphere and the public, but you don’t want that many doors inside your house, do
you? For a longer discussion see this blog post.
A bit more background: TypedActors can easily be abused as RPC, and that is an abstraction which is wellknown to be leaky. Hence TypedActors are not what we think of first when we talk about making highly scalable
concurrent software easier to write correctly. They have their niche, use them sparingly.
4.14.2 The tools of the trade
Before we create our first Typed Actor we should first go through the tools that we have at our disposal, it’s located
in akka.actor.TypedActor.
import akka.actor.TypedActor
//Returns the Typed Actor Extension
val extension = TypedActor(system) //system is an instance of ActorSystem
//Returns whether the reference is a Typed Actor Proxy or not
TypedActor(system).isTypedActor(someReference)
//Returns the backing Akka Actor behind an external Typed Actor Proxy
TypedActor(system).getActorRefFor(someReference)
//Returns the current ActorContext,
// method only valid within methods of a TypedActor implementation
val c: ActorContext = TypedActor.context
//Returns the external proxy of the current Typed Actor,
// method only valid within methods of a TypedActor implementation
val s: Squarer = TypedActor.self[Squarer]
//Returns a contextual instance of the Typed Actor Extension
//this means that if you create other Typed Actors with this,
//they will become children to the current Typed Actor.
TypedActor(TypedActor.context)
Warning: Same as not exposing this of an Akka Actor, it’s important not to expose this of a Typed
Actor, instead you should pass the external proxy reference, which is obtained from within your Typed Actor
as TypedActor.self, this is your external identity, as the ActorRef is the external identity of an Akka
Actor.
4.14. Typed Actors
260
Akka Scala Documentation, Release 2.5.0-RC1
4.14.3 Creating Typed Actors
To create a Typed Actor you need to have one or more interfaces, and one implementation.
Our example interface:
trait Squarer {
def squareDontCare(i: Int): Unit //fire-forget
def square(i: Int): Future[Int] //non-blocking send-request-reply
def squareNowPlease(i: Int): Option[Int] //blocking send-request-reply
def squareNow(i: Int): Int //blocking send-request-reply
˓→
@throws(classOf[Exception]) //declare it or you will get an
UndeclaredThrowableException
def squareTry(i: Int): Int //blocking send-request-reply with possible exception
}
Alright, now we’ve got some methods we can call, but we need to implement those in SquarerImpl.
class SquarerImpl(val name: String) extends Squarer {
def this() = this("default")
def squareDontCare(i: Int): Unit = i * i //Nobody cares :(
def square(i: Int): Future[Int] = Future.successful(i * i)
def squareNowPlease(i: Int): Option[Int] = Some(i * i)
def squareNow(i: Int): Int = i * i
def squareTry(i: Int): Int = throw new Exception("Catch me!")
}
Excellent, now we have an interface and an implementation of that interface, and we know how to create a Typed
Actor from that, so let’s look at calling these methods.
The most trivial way of creating a Typed Actor instance of our Squarer:
val mySquarer: Squarer =
TypedActor(system).typedActorOf(TypedProps[SquarerImpl]())
First type is the type of the proxy, the second type is the type of the implementation. If you need to call a specific
constructor you do it like this:
val otherSquarer: Squarer =
TypedActor(system).typedActorOf(TypedProps(
classOf[Squarer],
new SquarerImpl("foo")), "name")
Since you supply a Props, you can specify which dispatcher to use, what the default timeout should be used and
more.
4.14.4 Method dispatch semantics
Methods returning:
• Unit will be dispatched with fire-and-forget semantics, exactly like ActorRef.tell
• scala.concurrent.Future[_] will use send-request-reply semantics, exactly like
ActorRef.ask
4.14. Typed Actors
261
Akka Scala Documentation, Release 2.5.0-RC1
• scala.Option[_] will use send-request-reply semantics, but will block to wait for an answer,
and return scala.None if no answer was produced within the timeout, or scala.Some[_] containing
the result otherwise. Any exception that was thrown during this call will be rethrown.
• Any other type of value will use send-request-reply semantics, but will block to wait for an answer, throwing java.util.concurrent.TimeoutException if there was a timeout or rethrow
any exception that was thrown during this call.
4.14.5 Messages and immutability
While Akka cannot enforce that the parameters to the methods of your Typed Actors are immutable, we strongly
recommend that parameters passed are immutable.
One-way message send
mySquarer.squareDontCare(10)
As simple as that! The method will be executed on another thread; asynchronously.
Request-reply message send
val oSquare = mySquarer.squareNowPlease(10) //Option[Int]
This will block for as long as the timeout that was set in the Props of the Typed Actor, if needed. It will return
None if a timeout occurs.
val iSquare = mySquarer.squareNow(10) //Int
This will block for as long as the timeout that was set in the Props of the Typed Actor, if needed. It will throw a
java.util.concurrent.TimeoutException if a timeout occurs.
Request-reply-with-future message send
val fSquare = mySquarer.square(10) //A Future[Int]
This call is asynchronous, and the Future returned can be used for asynchronous composition.
4.14.6 Stopping Typed Actors
Since Akka’s Typed Actors are backed by Akka Actors they must be stopped when they aren’t needed anymore.
TypedActor(system).stop(mySquarer)
This asynchronously stops the Typed Actor associated with the specified proxy ASAP.
TypedActor(system).poisonPill(otherSquarer)
This asynchronously stops the Typed Actor associated with the specified proxy after it’s done with all calls that
were made prior to this call.
4.14. Typed Actors
262
Akka Scala Documentation, Release 2.5.0-RC1
4.14.7 Typed Actor Hierarchies
Since you can obtain a contextual Typed Actor Extension by passing in an ActorContext you can create child
Typed Actors by invoking typedActorOf(..) on that:
//Inside your Typed Actor
val childSquarer: Squarer =
TypedActor(TypedActor.context).typedActorOf(TypedProps[SquarerImpl]())
//Use "childSquarer" as a Squarer
You can also create a child Typed Actor in regular Akka Actors by giving the ActorContext as an input
parameter to TypedActor.get(. . . ).
4.14.8 Supervisor Strategy
By having your Typed Actor implementation class implement TypedActor.Supervisor you can define the
strategy to use for supervising child actors, as described in Supervision and Monitoring and Fault Tolerance.
4.14.9 Lifecycle callbacks
By having your Typed Actor implementation class implement any and all of the following:
• TypedActor.PreStart
• TypedActor.PostStop
• TypedActor.PreRestart
• TypedActor.PostRestart
You can hook into the lifecycle of your Typed Actor.
4.14.10 Receive arbitrary messages
If your implementation class of your TypedActor extends akka.actor.TypedActor.Receiver, all messages that are not MethodCall instances will be passed into the onReceive-method.
This allows you to react to DeathWatch Terminated-messages and other types of messages, e.g. when interfacing with untyped actors.
4.14.11 Proxying
You can use the typedActorOf that takes a TypedProps and an ActorRef to proxy the given ActorRef as a
TypedActor. This is usable if you want to communicate remotely with TypedActors on other machines, just pass
the ActorRef to typedActorOf.
Note: The ActorRef needs to accept MethodCall messages.
4.14.12 Lookup & Remoting
Since TypedActors are backed by Akka Actors, you can use typedActorOf to proxy ActorRefs
potentially residing on remote nodes.
4.14. Typed Actors
263
Akka Scala Documentation, Release 2.5.0-RC1
val typedActor: Foo with Bar =
TypedActor(system).
typedActorOf(
TypedProps[FooBar],
actorRefToRemoteActor)
//Use "typedActor" as a FooBar
4.14.13 Supercharging
Here’s an example on how you can use traits to mix in behavior in your Typed Actors.
trait Foo {
def doFoo(times: Int): Unit = println("doFoo(" + times + ")")
}
trait Bar {
def doBar(str: String): Future[String] =
Future.successful(str.toUpperCase)
}
class FooBar extends Foo with Bar
val awesomeFooBar: Foo with Bar =
TypedActor(system).typedActorOf(TypedProps[FooBar]())
awesomeFooBar.doFoo(10)
val f = awesomeFooBar.doBar("yes")
TypedActor(system).poisonPill(awesomeFooBar)
4.14.14 Typed Router pattern
Sometimes you want to spread messages between multiple actors. The easiest way to achieve this in
Akka is to use a Router, which can implement a specific routing logic, such as smallest-mailbox or
consistent-hashing etc.
Routers are not provided directly for typed actors, but it is really easy to leverage an untyped router and use a
typed proxy in front of it. To showcase this let’s create typed actors that assign themselves some random id, so
we know that in fact, the router has sent the message to different actors:
trait HasName {
def name(): String
}
class Named extends HasName {
import scala.util.Random
private val id = Random.nextInt(1024)
def name(): String = "name-" + id
}
In order to round robin among a few instances of such actors, you can simply create a plain untyped router,
and then facade it with a TypedActor like shown in the example below. This works because typed actors of
course communicate using the same mechanisms as normal actors, and methods calls on them get transformed
into message sends of MethodCall messages.
def namedActor(): HasName = TypedActor(system).typedActorOf(TypedProps[Named]())
4.14. Typed Actors
264
Akka Scala Documentation, Release 2.5.0-RC1
// prepare routees
val routees: List[HasName] = List.fill(5) { namedActor() }
val routeePaths = routees map { r =>
TypedActor(system).getActorRefFor(r).path.toStringWithoutAddress
}
// prepare untyped router
val router: ActorRef = system.actorOf(RoundRobinGroup(routeePaths).props())
// prepare typed proxy, forwarding MethodCall messages to `router`
val typedRouter: HasName =
TypedActor(system).typedActorOf(TypedProps[Named](), actorRef = router)
println("actor
println("actor
println("actor
println("actor
was:
was:
was:
was:
4.14. Typed Actors
"
"
"
"
+
+
+
+
typedRouter.name())
typedRouter.name())
typedRouter.name())
typedRouter.name())
//
//
//
//
name-184
name-753
name-320
name-164
265
CHAPTER
FIVE
FUTURES AND AGENTS
5.1 Futures
5.1.1 Introduction
In the Scala Standard Library, a Future is a data structure used to retrieve the result of some concurrent operation.
This result can be accessed synchronously (blocking) or asynchronously (non-blocking).
5.1.2 Execution Contexts
In order to execute callbacks and operations, Futures need something called an ExecutionContext, which
is very similar to a java.util.concurrent.Executor. if you have an ActorSystem in scope, it will
use its default dispatcher as the ExecutionContext, or you can use the factory methods provided by the
ExecutionContext companion object to wrap Executors and ExecutorServices, or even create your
own.
import scala.concurrent.{ ExecutionContext, Promise }
implicit val ec = ExecutionContext.fromExecutorService(yourExecutorServiceGoesHere)
// Do stuff with your brand new shiny ExecutionContext
val f = Promise.successful("foo")
// Then shut your ExecutionContext down at some
// appropriate place in your program/application
ec.shutdown()
Within Actors
Each actor is configured to be run on a MessageDispatcher, and that dispatcher doubles as an
ExecutionContext. If the nature of the Future calls invoked by the actor matches or is compatible with
the activities of that actor (e.g. all CPU bound and no latency requirements), then it may be easiest to reuse the
dispatcher for running the Futures by importing context.dispatcher.
class A extends Actor {
import context.dispatcher
val f = Future("hello")
def receive = {
// receive omitted ...
}
}
266
Akka Scala Documentation, Release 2.5.0-RC1
5.1.3 Use With Actors
There are generally two ways of getting a reply from an Actor: the first is by a sent message (actor ! msg),
which only works if the original sender was an Actor) and the second is through a Future.
Using an Actor‘s ? method to send a message will return a Future:
import
import
import
import
scala.concurrent.Await
akka.pattern.ask
akka.util.Timeout
scala.concurrent.duration._
implicit val timeout = Timeout(5 seconds)
val future = actor ? msg // enabled by the “ask” import
val result = Await.result(future, timeout.duration).asInstanceOf[String]
This will cause the current thread to block and wait for the Actor to ‘complete’ the Future with it’s reply.
Blocking is discouraged though as it will cause performance problems. The blocking operations are located in
Await.result and Await.ready to make it easy to spot where blocking occurs. Alternatives to blocking are discussed further within this documentation. Also note that the Future returned by an Actor is a
Future[Any] since an Actor is dynamic. That is why the asInstanceOf is used in the above sample.
When using non-blocking it is better to use the mapTo method to safely try to cast a Future to an expected type:
import scala.concurrent.Future
import akka.pattern.ask
val future: Future[String] = ask(actor, msg).mapTo[String]
The mapTo method will return a new Future that contains the result if the cast was successful, or a
ClassCastException if not. Handling Exceptions will be discussed further within this documentation.
To send the result of a Future to an Actor, you can use the pipe construct:
import akka.pattern.pipe
future pipeTo actor
5.1.4 Use Directly
A common use case within Akka is to have some computation performed concurrently without needing the extra
utility of an Actor. If you find yourself creating a pool of Actors for the sole reason of performing a calculation
in parallel, there is an easier (and faster) way:
import scala.concurrent.Await
import scala.concurrent.Future
import scala.concurrent.duration._
val future = Future {
"Hello" + "World"
}
future foreach println
In the above code the block passed to Future will be executed by the default Dispatcher, with the return value
of the block used to complete the Future (in this case, the result would be the string: “HelloWorld”). Unlike
a Future that is returned from an Actor, this Future is properly typed, and we also avoid the overhead of
managing an Actor.
You can also create already completed Futures using the Future companion, which can be either successes:
val future = Future.successful("Yay!")
Or failures:
5.1. Futures
267
Akka Scala Documentation, Release 2.5.0-RC1
val otherFuture = Future.failed[String](new IllegalArgumentException("Bang!"))
It is also possible to create an empty Promise, to be filled later, and obtain the corresponding Future:
val promise = Promise[String]()
val theFuture = promise.future
promise.success("hello")
5.1.5 Functional Futures
Scala’s Future has several monadic methods that are very similar to the ones used by Scala’s collections. These
allow you to create ‘pipelines’ or ‘streams’ that the result will travel through.
Future is a Monad
The first method for working with Future functionally is map. This method takes a Function which performs
some operation on the result of the Future, and returning a new result. The return value of the map method is
another Future that will contain the new result:
val f1 = Future {
"Hello" + "World"
}
val f2 = f1 map { x =>
x.length
}
f2 foreach println
In this example we are joining two strings together within a Future. Instead of waiting for this to complete, we
apply our function that calculates the length of the string using the map method. Now we have a second Future
that will eventually contain an Int. When our original Future completes, it will also apply our function and
complete the second Future with its result. When we finally get the result, it will contain the number 10. Our
original Future still contains the string “HelloWorld” and is unaffected by the map.
The map method is fine if we are modifying a single Future, but if 2 or more Futures are involved map will
not allow you to combine them together:
val f1 = Future {
"Hello" + "World"
}
val f2 = Future.successful(3)
val f3 = f1 map { x =>
f2 map { y =>
x.length * y
}
}
f3 foreach println
f3 is a Future[Future[Int]] instead of the desired Future[Int]. Instead, the flatMap method should
be used:
val f1 = Future {
"Hello" + "World"
}
val f2 = Future.successful(3)
val f3 = f1 flatMap { x =>
f2 map { y =>
x.length * y
}
5.1. Futures
268
Akka Scala Documentation, Release 2.5.0-RC1
}
f3 foreach println
Composing futures using nested combinators it can sometimes become quite complicated and hard to read, in
these cases using Scala’s ‘for comprehensions’ usually yields more readable code. See next section for examples.
If you need to do conditional propagation, you can use filter:
val future1 = Future.successful(4)
val future2 = future1.filter(_ % 2 == 0)
future2 foreach println
val failedFilter = future1.filter(_ % 2 == 1).recover {
// When filter fails, it will have a java.util.NoSuchElementException
case m: NoSuchElementException => 0
}
failedFilter foreach println
For Comprehensions
Since Future has a map, filter and flatMap method it can be easily used in a ‘for comprehension’:
val f = for {
a <- Future(10 / 2) // 10 / 2 = 5
b <- Future(a + 1) // 5 + 1 = 6
c <- Future(a - 1) // 5 - 1 = 4
if c > 3 // Future.filter
} yield b * c // 6 * 4 = 24
// Note that the execution of futures a, b, and c
// are not done in parallel.
f foreach println
Something to keep in mind when doing this is even though it looks like parts of the above example can run in
parallel, each step of the for comprehension is run sequentially. This will happen on separate threads for each
step but there isn’t much benefit over running the calculations all within a single Future. The real benefit comes
when the Futures are created first, and then combining them together.
Composing Futures
The example for comprehension above is an example of composing Futures. A common use case for this is
combining the replies of several Actors into a single calculation without resorting to calling Await.result
or Await.ready to block for each result. First an example of using Await.result:
val f1 = ask(actor1, msg1)
val f2 = ask(actor2, msg2)
val a = Await.result(f1, 3 seconds).asInstanceOf[Int]
val b = Await.result(f2, 3 seconds).asInstanceOf[Int]
val f3 = ask(actor3, (a + b))
val result = Await.result(f3, 3 seconds).asInstanceOf[Int]
5.1. Futures
269
Akka Scala Documentation, Release 2.5.0-RC1
Warning: Await.result and Await.ready are provided for exceptional situations where you must
block, a good rule of thumb is to only use them if you know why you must block. For all other cases, use
asynchronous composition as described below.
Here we wait for the results from the first 2 Actors before sending that result to the third Actor. We called
Await.result 3 times, which caused our little program to block 3 times before getting our final result. Now
compare that to this example:
val f1 = ask(actor1, msg1)
val f2 = ask(actor2, msg2)
val f3 = for {
a <- f1.mapTo[Int]
b <- f2.mapTo[Int]
c <- ask(actor3, (a + b)).mapTo[Int]
} yield c
f3 foreach println
Here we have 2 actors processing a single message each. Once the 2 results are available (note that we don’t block
to get these results!), they are being added together and sent to a third Actor, which replies with a string, which
we assign to ‘result’.
This is fine when dealing with a known amount of Actors, but can grow unwieldy if we have more than a
handful. The sequence and traverse helper methods can make it easier to handle more complex use
cases. Both of these methods are ways of turning, for a subclass T of Traversable, T[Future[A]] into
a Future[T[A]]. For example:
// oddActor returns odd numbers sequentially from 1 as a List[Future[Int]]
val listOfFutures = List.fill(100)(akka.pattern.ask(oddActor, GetNext).mapTo[Int])
// now we have a Future[List[Int]]
val futureList = Future.sequence(listOfFutures)
// Find the sum of the odd numbers
val oddSum = futureList.map(_.sum)
oddSum foreach println
To better explain what happened in the example, Future.sequence is taking the List[Future[Int]]
and turning it into a Future[List[Int]]. We can then use map to work with the List[Int] directly, and
we find the sum of the List.
The traverse method is similar to sequence, but it takes a T[A] and a function A => Future[B] to
return a Future[T[B]], where T is again a subclass of Traversable. For example, to use traverse to sum
the first 100 odd numbers:
val futureList = Future.traverse((1 to 100).toList)(x => Future(x * 2 - 1))
val oddSum = futureList.map(_.sum)
oddSum foreach println
This is the same result as this example:
val futureList = Future.sequence((1 to 100).toList.map(x => Future(x * 2 - 1)))
val oddSum = futureList.map(_.sum)
oddSum foreach println
But it may be faster to use traverse as it doesn’t have to create an intermediate List[Future[Int]].
Then there’s a method that’s called fold that takes a start-value, a sequence of Futures and a function from the
type of the start-value and the type of the futures and returns something with the same type as the start-value, and
5.1. Futures
270
Akka Scala Documentation, Release 2.5.0-RC1
then applies the function to all elements in the sequence of futures, asynchronously, the execution will start when
the last of the Futures is completed.
// Create a sequence of Futures
val futures = for (i <- 1 to 1000) yield Future(i * 2)
val futureSum = Future.fold(futures)(0)(_ + _)
futureSum foreach println
That’s all it takes!
If the sequence passed to fold is empty, it will return the start-value, in the case above, that will be 0. In some
cases you don’t have a start-value and you’re able to use the value of the first completing Future in the sequence
as the start-value, you can use reduce, it works like this:
// Create a sequence of Futures
val futures = for (i <- 1 to 1000) yield Future(i * 2)
val futureSum = Future.reduce(futures)(_ + _)
futureSum foreach println
Same as with fold, the execution will be done asynchronously when the last of the Future is completed, you
can also parallelize it by chunking your futures into sub-sequences and reduce them, and then reduce the reduced
results again.
5.1.6 Callbacks
Sometimes you just want to listen to a Future being completed, and react to that not by creating a new Future,
but by side-effecting. For this Scala supports onComplete, onSuccess and onFailure, of which the last
two are specializations of the first.
future onSuccess {
case "bar"
=> println("Got my bar alright!")
case x: String => println("Got some random string: " + x)
}
future onFailure {
case ise: IllegalStateException if ise.getMessage == "OHNOES" =>
//OHNOES! We are in deep trouble, do something!
case e: Exception =>
//Do something else
}
future onComplete {
case Success(result) => doSomethingOnSuccess(result)
case Failure(failure) => doSomethingOnFailure(failure)
}
5.1.7 Define Ordering
Since callbacks are executed in any order and potentially in parallel, it can be tricky at the times when you need
sequential ordering of operations. But there’s a solution and it’s name is andThen. It creates a new Future
with the specified callback, a Future that will have the same result as the Future it’s called on, which allows
for ordering like in the following sample:
val result = Future { loadPage(url) } andThen {
case Failure(exception) => log(exception)
} andThen {
case _ => watchSomeTV()
}
result foreach println
5.1. Futures
271
Akka Scala Documentation, Release 2.5.0-RC1
5.1.8 Auxiliary Methods
Future fallbackTo combines 2 Futures into a new Future, and will hold the successful value of the second
Future if the first Future fails.
val future4 = future1 fallbackTo future2 fallbackTo future3
future4 foreach println
You can also combine two Futures into a new Future that will hold a tuple of the two Futures successful results,
using the zip operation.
val future3 = future1 zip future2 map { case (a, b) => a + " " + b }
future3 foreach println
5.1.9 Exceptions
Since the result of a Future is created concurrently to the rest of the program, exceptions must be handled
differently. It doesn’t matter if an Actor or the dispatcher is completing the Future, if an Exception is
caught the Future will contain it instead of a valid result. If a Future does contain an Exception, calling
Await.result will cause it to be thrown again so it can be handled properly.
It is also possible to handle an Exception by returning a different result. This is done with the recover
method. For example:
val future = akka.pattern.ask(actor, msg1) recover {
case e: ArithmeticException => 0
}
future foreach println
In this example, if the actor replied with a akka.actor.Status.Failure containing the
ArithmeticException, our Future would have a result of 0. The recover method works very
similarly to the standard try/catch blocks, so multiple Exceptions can be handled in this manner, and if an
Exception is not handled this way it will behave as if we hadn’t used the recover method.
You can also use the recoverWith method, which has the same relationship to recover as flatMap has to
map, and is use like this:
val future = akka.pattern.ask(actor, msg1) recoverWith {
case e: ArithmeticException => Future.successful(0)
case foo: IllegalArgumentException =>
Future.failed[Int](new IllegalStateException("All br0ken!"))
}
future foreach println
5.1.10 After
akka.pattern.after makes it easy to complete a Future with a value or exception after a timeout.
// TODO after is unfortunately shadowed by ScalaTest, fix as part of #3759
// import akka.pattern.after
val delayed = akka.pattern.after(200 millis, using = system.scheduler)(Future.
˓→failed(
new IllegalStateException("OHNOES")))
val future = Future { Thread.sleep(1000); "foo" }
val result = Future firstCompletedOf Seq(future, delayed)
5.1. Futures
272
Akka Scala Documentation, Release 2.5.0-RC1
5.2 Agents
Agents in Akka are inspired by agents in Clojure.
Warning: Deprecation warning - Agents have been deprecated and are scheduled for removal in the next
major version. We have found that their leaky abstraction (they do not work over the network) make them
inferior to pure Actors, and in face of the soon inclusion of Akka Typed we see little value in maintaining the
current Agents.
Agents provide asynchronous change of individual locations. Agents are bound to a single storage location for
their lifetime, and only allow mutation of that location (to a new state) to occur as a result of an action. Update
actions are functions that are asynchronously applied to the Agent’s state and whose return value becomes the
Agent’s new state. The state of an Agent should be immutable.
While updates to Agents are asynchronous, the state of an Agent is always immediately available for reading by
any thread (using get or apply) without any messages.
Agents are reactive.
The update actions of all Agents get interleaved amongst threads in an
ExecutionContext. At any point in time, at most one send action for each Agent is being executed. Actions
dispatched to an agent from another thread will occur in the order they were sent, potentially interleaved with
actions dispatched to the same agent from other threads.
Note: Agents are local to the node on which they are created. This implies that you should generally not include
them in messages that may be passed to remote Actors or as constructor parameters for remote Actors; those
remote Actors will not be able to read or update the Agent.
5.2.1 Creating Agents
Agents are created by invoking Agent(value) passing in the Agent’s initial value and providing an implicit
ExecutionContext to be used for it, for these examples we’re going to use the default global one, but YMMV:
import scala.concurrent.ExecutionContext.Implicits.global
import akka.agent.Agent
val agent = Agent(5)
5.2.2 Reading an Agent’s value
Agents can be dereferenced (you can get an Agent’s value) by invoking the Agent with parentheses like this:
val result = agent()
Or by using the get method:
val result = agent.get
Reading an Agent’s current value does not involve any message passing and happens immediately. So while
updates to an Agent are asynchronous, reading the state of an Agent is synchronous.
5.2.3 Updating Agents (send & alter)
You update an Agent by sending a function that transforms the current value or by sending just a new value. The
Agent will apply the new value or function atomically and asynchronously. The update is done in a fire-forget
manner and you are only guaranteed that it will be applied. There is no guarantee of when the update will be
5.2. Agents
273
Akka Scala Documentation, Release 2.5.0-RC1
applied but dispatches to an Agent from a single thread will occur in order. You apply a value or a function by
invoking the send function.
// send a value, enqueues this change
// of the value of the Agent
agent send 7
// send a function, enqueues this change
// to the value of the Agent
agent send (_ + 1)
agent send (_ * 2)
You can also dispatch a function to update the internal state but on its own thread. This does not use the reactive
thread pool and can be used for long-running or blocking operations. You do this with the sendOff method.
Dispatches using either sendOff or send will still be executed in order.
// the ExecutionContext you want to run the function on
implicit val ec = someExecutionContext()
// sendOff a function
agent sendOff longRunningOrBlockingFunction
All send methods also have a corresponding alter method that returns a Future. See Futures for more
information on Futures.
// alter a value
val f1: Future[Int] = agent alter 7
// alter a function
val f2: Future[Int] = agent alter (_ + 1)
val f3: Future[Int] = agent alter (_ * 2)
// the ExecutionContext you want to run the function on
implicit val ec = someExecutionContext()
// alterOff a function
val f4: Future[Int] = agent alterOff longRunningOrBlockingFunction
5.2.4 Awaiting an Agent’s value
You can also get a Future to the Agents value, that will be completed after the currently queued updates have
completed:
val future = agent.future
See Futures for more information on Futures.
5.2.5 Monadic usage
Agents are also monadic, allowing you to compose operations using for-comprehensions. In monadic usage, new
Agents are created leaving the original Agents untouched. So the old values (Agents) are still available as-is. They
are so-called ‘persistent’.
Example of monadic usage:
import scala.concurrent.ExecutionContext.Implicits.global
val agent1 = Agent(3)
val agent2 = Agent(5)
// uses foreach
for (value <- agent1)
5.2. Agents
274
Akka Scala Documentation, Release 2.5.0-RC1
println(value)
// uses map
val agent3 = for (value <- agent1) yield value + 1
// or using map directly
val agent4 = agent1 map (_ + 1)
// uses flatMap
val agent5 = for {
value1 <- agent1
value2 <- agent2
} yield value1 + value2
5.2.6 Configuration
There are several configuration properties for the agents module, please refer to the reference configuration.
5.2.7 Deprecated Transactional Agents
Agents participating in enclosing STM transaction is a deprecated feature in 2.3.
If an Agent is used within an enclosing transaction, then it will participate in that transaction. If you send to an
Agent within a transaction then the dispatch to the Agent will be held until that transaction commits, and discarded
if the transaction is aborted. Here’s an example:
import
import
import
import
scala.concurrent.ExecutionContext.Implicits.global
akka.agent.Agent
scala.concurrent.duration._
scala.concurrent.stm._
def transfer(from:
atomic { txn =>
if (from.get <
else {
from send (_
to send (_ +
true
}
}
}
Agent[Int], to: Agent[Int], amount: Int): Boolean = {
amount) false
- amount)
amount)
val from = Agent(100)
val to = Agent(20)
val ok = transfer(from, to, 50)
val fromValue = from.future // -> 50
val toValue = to.future // -> 70
5.2. Agents
275
CHAPTER
SIX
NETWORKING
6.1 Cluster Specification
Note: This document describes the design concepts of the clustering.
6.1.1 Intro
Akka Cluster provides a fault-tolerant decentralized peer-to-peer based cluster membership service with no single
point of failure or single point of bottleneck. It does this using gossip protocols and an automatic failure detector.
6.1.2 Terms
node A logical member of a cluster. There could be multiple nodes on a physical machine. Defined by a hostname:port:uid tuple.
cluster A set of nodes joined together through the membership service.
leader A single node in the cluster that acts as the leader. Managing cluster convergence and membership state
transitions.
6.1.3 Membership
A cluster is made up of a set of member nodes. The identifier for each node is a hostname:port:uid tuple.
An Akka application can be distributed over a cluster with each node hosting some part of the application. Cluster
membership and the actors running on that node of the application are decoupled. A node could be a member of
a cluster without hosting any actors. Joining a cluster is initiated by issuing a Join command to one of the nodes
in the cluster to join.
The node identifier internally also contains a UID that uniquely identifies this actor system instance at that
hostname:port. Akka uses the UID to be able to reliably trigger remote death watch. This means that
the same actor system can never join a cluster again once it’s been removed from that cluster. To re-join an actor
system with the same hostname:port to a cluster you have to stop the actor system and start a new one with
the same hostname:port which will then receive a different UID.
The cluster membership state is a specialized CRDT, which means that it has a monotonic merge function. When
concurrent changes occur on different nodes the updates can always be merged and converge to the same end
result.
Gossip
The cluster membership used in Akka is based on Amazon’s Dynamo system and particularly the approach taken
in Basho’s‘ Riak distributed database. Cluster membership is communicated using a Gossip Protocol, where the
276
Akka Scala Documentation, Release 2.5.0-RC1
current state of the cluster is gossiped randomly through the cluster, with preference to members that have not
seen the latest version.
Vector Clocks
Vector clocks are a type of data structure and algorithm for generating a partial ordering of events in a distributed
system and detecting causality violations.
We use vector clocks to reconcile and merge differences in cluster state during gossiping. A vector clock is a set
of (node, counter) pairs. Each update to the cluster state has an accompanying update to the vector clock.
Gossip Convergence
Information about the cluster converges locally at a node at certain points in time. This is when a node can
prove that the cluster state he is observing has been observed by all other nodes in the cluster. Convergence is
implemented by passing a set of nodes that have seen current state version during gossip. This information is
referred to as the seen set in the gossip overview. When all nodes are included in the seen set there is convergence.
Gossip convergence cannot occur while any nodes are unreachable. The nodes need to become reachable
again, or moved to the down and removed states (see the Membership Lifecycle section below). This only blocks
the leader from performing its cluster membership management and does not influence the application running on
top of the cluster. For example this means that during a network partition it is not possible to add more nodes to
the cluster. The nodes can join, but they will not be moved to the up state until the partition has healed or the
unreachable nodes have been downed.
Failure Detector
The failure detector is responsible for trying to detect if a node is unreachable from the rest of the cluster. For
this we are using an implementation of The Phi Accrual Failure Detector by Hayashibara et al.
An accrual failure detector decouple monitoring and interpretation. That makes them applicable to a wider area
of scenarios and more adequate to build generic failure detection services. The idea is that it is keeping a history
of failure statistics, calculated from heartbeats received from other nodes, and is trying to do educated guesses by
taking multiple factors, and how they accumulate over time, into account in order to come up with a better guess
if a specific node is up or down. Rather than just answering “yes” or “no” to the question “is the node down?” it
returns a phi value representing the likelihood that the node is down.
The threshold that is the basis for the calculation is configurable by the user. A low threshold is prone
to generate many wrong suspicions but ensures a quick detection in the event of a real crash. Conversely, a high
threshold generates fewer mistakes but needs more time to detect actual crashes. The default threshold is
8 and is appropriate for most situations. However in cloud environments, such as Amazon EC2, the value could
be increased to 12 in order to account for network issues that sometimes occur on such platforms.
In a cluster each node is monitored by a few (default maximum 5) other nodes, and when any of these detects the
node as unreachable that information will spread to the rest of the cluster through the gossip. In other words,
only one node needs to mark a node unreachable to have the rest of the cluster mark that node unreachable.
The nodes to monitor are picked out of neighbors in a hashed ordered node ring. This is to increase the likelihood
to monitor across racks and data centers, but the order is the same on all nodes, which ensures full coverage.
Heartbeats are sent out every second and every heartbeat is performed in a request/reply handshake with the replies
used as input to the failure detector.
The failure detector will also detect if the node becomes reachable again. When all nodes that monitored the
unreachable node detects it as reachable again the cluster, after gossip dissemination, will consider it as
reachable.
If system messages cannot be delivered to a node it will be quarantined and then it cannot come back from
unreachable. This can happen if the there are too many unacknowledged system messages (e.g. watch,
Terminated, remote actor deployment, failures of actors supervised by remote parent). Then the node needs to be
6.1. Cluster Specification
277
Akka Scala Documentation, Release 2.5.0-RC1
moved to the down or removed states (see the Membership Lifecycle section below) and the actor system must
be restarted before it can join the cluster again.
Leader
After gossip convergence a leader for the cluster can be determined. There is no leader election process,
the leader can always be recognised deterministically by any node whenever there is gossip convergence. The
leader is just a role, any node can be the leader and it can change between convergence rounds. The leader is
simply the first node in sorted order that is able to take the leadership role, where the preferred member states for a
leader are up and leaving (see the Membership Lifecycle section below for more information about member
states).
The role of the leader is to shift members in and out of the cluster, changing joining members to the up
state or exiting members to the removed state. Currently leader actions are only triggered by receiving a
new cluster state with gossip convergence.
The leader also has the power, if configured so, to “auto-down” a node that according to the Failure Detector
is considered unreachable. This means setting the unreachable node status to down automatically after a
configured time of unreachability.
Seed Nodes
The seed nodes are configured contact points for new nodes joining the cluster. When a new node is started it
sends a message to all seed nodes and then sends a join command to the seed node that answers first.
The seed nodes configuration value does not have any influence on the running cluster itself, it is only relevant for
new nodes joining the cluster as it helps them to find contact points to send the join command to; a new member
can send this command to any current member of the cluster, not only to the seed nodes.
Gossip Protocol
A variation of push-pull gossip is used to reduce the amount of gossip information sent around the cluster. In
push-pull gossip a digest is sent representing current versions but not actual values; the recipient of the gossip
can then send back any values for which it has newer versions and also request values for which it has outdated
versions. Akka uses a single shared state with a vector clock for versioning, so the variant of push-pull gossip
used in Akka makes use of this version to only push the actual state as needed.
Periodically, the default is every 1 second, each node chooses another random node to initiate a round of gossip
with. If less than ½ of the nodes resides in the seen set (have seen the new state) then the cluster gossips 3 times
instead of once every second. This adjusted gossip interval is a way to speed up the convergence process in the
early dissemination phase after a state change.
The choice of node to gossip with is random but it is biased to towards nodes that might not have seen the
current state version. During each round of gossip exchange when no convergence it uses a probability of 0.8
(configurable) to gossip to a node not part of the seen set, i.e. that probably has an older version of the state.
Otherwise gossip to any random live node.
This biased selection is a way to speed up the convergence process in the late dissemination phase after a state
change.
For clusters larger than 400 nodes (configurable, and suggested by empirical evidence) the 0.8 probability is
gradually reduced to avoid overwhelming single stragglers with too many concurrent gossip requests. The gossip
receiver also has a mechanism to protect itself from too many simultaneous gossip messages by dropping messages
that have been enqueued in the mailbox for too long of a time.
While the cluster is in a converged state the gossiper only sends a small gossip status message containing the
gossip version to the chosen node. As soon as there is a change to the cluster (meaning non-convergence) then it
goes back to biased gossip again.
The recipient of the gossip state or the gossip status can use the gossip version (vector clock) to determine whether:
6.1. Cluster Specification
278
Akka Scala Documentation, Release 2.5.0-RC1
1. it has a newer version of the gossip state, in which case it sends that back to the gossiper
2. it has an outdated version of the state, in which case the recipient requests the current state from the gossiper
by sending back its version of the gossip state
3. it has conflicting gossip versions, in which case the different versions are merged and sent back
If the recipient and the gossip have the same version then the gossip state is not sent or requested.
The periodic nature of the gossip has a nice batching effect of state changes, e.g. joining several nodes quickly
after each other to one node will result in only one state change to be spread to other members in the cluster.
The gossip messages are serialized with protobuf and also gzipped to reduce payload size.
Membership Lifecycle
A node begins in the joining state. Once all nodes have seen that the new node is joining (through gossip
convergence) the leader will set the member state to up.
If a node is leaving the cluster in a safe, expected manner then it switches to the leaving state. Once the leader
sees the convergence on the node in the leaving state, the leader will then move it to exiting. Once all
nodes have seen the exiting state (convergence) the leader will remove the node from the cluster, marking it as
removed.
If a node is unreachable then gossip convergence is not possible and therefore any leader actions are also
not possible (for instance, allowing a node to become a part of the cluster). To be able to move forward the state
of the unreachable nodes must be changed. It must become reachable again or marked as down. If the
node is to join the cluster again the actor system must be restarted and go through the joining process again. The
cluster can, through the leader, also auto-down a node after a configured time of unreachability. If new incarnation
of unreachable node tries to rejoin the cluster old incarnation will be marked as down and new incarnation can
rejoin the cluster without manual intervention.
Note: If you have auto-down enabled and the failure detector triggers, you can over time end up with a lot of
single node clusters if you don’t put measures in place to shut down nodes that have become unreachable.
This follows from the fact that the unreachable node will likely see the rest of the cluster as unreachable,
become its own leader and form its own cluster.
As mentioned before, if a node is unreachable then gossip convergence is not possible and therefore any
leader actions are also not possible. By enabling akka.cluster.allow-weakly-up-members (enabled by default) it is possible to let new joining nodes be promoted while convergence is not yet reached. These
Joining nodes will be promoted as WeaklyUp. Once gossip convergence is reached, the leader will move
WeaklyUp members to Up.
Note that members on the other side of a network partition have no knowledge about the existence of the new
members. You should for example not count WeaklyUp members in quorum decisions.
6.1. Cluster Specification
279
Akka Scala Documentation, Release 2.5.0-RC1
State Diagram for the Member States (akka.cluster.allow-weakly-up-members=off)
State Diagram for the Member States (akka.cluster.allow-weakly-up-members=on)
6.1. Cluster Specification
280
Akka Scala Documentation, Release 2.5.0-RC1
Member States
• joining transient state when joining a cluster
• weakly up transient
state
while
network
allow-weakly-up-members=on)
split
(only
if
akka.cluster.
• up normal operating state
• leaving / exiting states during graceful removal
• down marked as down (no longer part of cluster decisions)
• removed tombstone state (no longer a member)
User Actions
• join join a single node to a cluster - can be explicit or automatic on startup if a node to join have been
specified in the configuration
• leave tell a node to leave the cluster gracefully
• down mark a node as down
Leader Actions
The leader has the following duties:
• shifting members in and out of the cluster
– joining -> up
– exiting -> removed
Failure Detection and Unreachability
• fd* the failure detector of one of the monitoring nodes has triggered causing the monitored node to be
marked as unreachable
• unreachable* unreachable is not a real member states but more of a flag in addition to the state signaling
that the cluster is unable to talk to this node, after being unreachable the failure detector may detect it
as reachable again and thereby remove the flag
6.2 Cluster Usage
For introduction to the Akka Cluster concepts please see Cluster Specification.
6.2.1 Preparing Your Project for Clustering
The Akka cluster is a separate jar file. Make sure that you have the following dependency in your project:
"com.typesafe.akka" %% "akka-cluster" % "2.5.0-RC1"
6.2. Cluster Usage
281
Akka Scala Documentation, Release 2.5.0-RC1
6.2.2 A Simple Cluster Example
The following configuration enables the Cluster extension to be used. It joins the cluster and an actor subscribes
to cluster membership events and logs them.
The application.conf configuration looks like this:
akka {
actor {
provider = "cluster"
}
remote {
log-remote-lifecycle-events = off
netty.tcp {
hostname = "127.0.0.1"
port = 0
}
}
cluster {
seed-nodes = [
"akka.tcp://ClusterSystem@127.0.0.1:2551",
"akka.tcp://ClusterSystem@127.0.0.1:2552"]
# auto downing is NOT safe for production deployments.
# you may want to use it during development, read more about it in the docs.
#
# auto-down-unreachable-after = 10s
}
}
# Disable legacy metrics in akka-cluster.
akka.cluster.metrics.enabled=off
# Enable metrics extension in akka-cluster-metrics.
akka.extensions=["akka.cluster.metrics.ClusterMetricsExtension"]
# Sigar native library extract location during tests.
# Note: use per-jvm-instance folder when running multiple jvm on one host.
akka.cluster.metrics.native-library-extract-folder=${user.dir}/target/native
To enable cluster capabilities in your Akka project you should, at a minimum, add the Remoting settings, but with
cluster. The akka.cluster.seed-nodes should normally also be added to your application.conf
file.
Note: If you are running Akka in a Docker container or the nodes for some other reason have separate internal
and external ip addresses you must configure remoting according to Akka behind NAT or in a Docker container
The seed nodes are configured contact points for initial, automatic, join of the cluster.
Note that if you are going to start the nodes on different machines you need to specify the ip-addresses or host
names of the machines in application.conf instead of 127.0.0.1
An actor that uses the cluster extension may look like this:
package scala.docs.cluster
import
import
import
import
akka.cluster.Cluster
akka.cluster.ClusterEvent._
akka.actor.ActorLogging
akka.actor.Actor
6.2. Cluster Usage
282
Akka Scala Documentation, Release 2.5.0-RC1
class SimpleClusterListener extends Actor with ActorLogging {
val cluster = Cluster(context.system)
// subscribe to cluster changes, re-subscribe when restart
override def preStart(): Unit = {
//#subscribe
cluster.subscribe(self, initialStateMode = InitialStateAsEvents,
classOf[MemberEvent], classOf[UnreachableMember])
//#subscribe
}
override def postStop(): Unit = cluster.unsubscribe(self)
def receive = {
case MemberUp(member) =>
log.info("Member is Up: {}", member.address)
case UnreachableMember(member) =>
log.info("Member detected as unreachable: {}", member)
case MemberRemoved(member, previousStatus) =>
log.info(
"Member is Removed: {} after {}",
member.address, previousStatus)
case _: MemberEvent => // ignore
}
}
The actor registers itself as subscriber of certain cluster events. It receives events corresponding to the current
state of the cluster when the subscription starts and then it receives events for changes that happen in the cluster.
The easiest way to run this example yourself is to download Lightbend Activator and open the tutorial named
Akka Cluster Samples with Scala. It contains instructions of how to run the SimpleClusterApp.
6.2.3 Joining to Seed Nodes
You may decide if joining to the cluster should be done manually or automatically to configured initial contact
points, so-called seed nodes. When a new node is started it sends a message to all seed nodes and then sends join
command to the one that answers first. If no one of the seed nodes replied (might not be started yet) it retries this
procedure until successful or shutdown.
You define the seed nodes in the Configuration file (application.conf):
akka.cluster.seed-nodes = [
"akka.tcp://ClusterSystem@host1:2552",
"akka.tcp://ClusterSystem@host2:2552"]
This can also be defined as Java system properties when starting the JVM using the following syntax:
-Dakka.cluster.seed-nodes.0=akka.tcp://ClusterSystem@host1:2552
-Dakka.cluster.seed-nodes.1=akka.tcp://ClusterSystem@host2:2552
The seed nodes can be started in any order and it is not necessary to have all seed nodes running, but the node
configured as the first element in the seed-nodes configuration list must be started when initially starting a
cluster, otherwise the other seed-nodes will not become initialized and no other node can join the cluster. The
reason for the special first seed node is to avoid forming separated islands when starting from an empty cluster. It
is quickest to start all configured seed nodes at the same time (order doesn’t matter), otherwise it can take up to
the configured seed-node-timeout until the nodes can join.
Once more than two seed nodes have been started it is no problem to shut down the first seed node. If the first
seed node is restarted, it will first try to join the other seed nodes in the existing cluster.
If you don’t configure seed nodes you need to join the cluster programmatically or manually.
6.2. Cluster Usage
283
Akka Scala Documentation, Release 2.5.0-RC1
Manual joining can be performed by using JMX or HTTP. Joining programmatically can be performed with
Cluster(system).join. Unsuccessful join attempts are automatically retried after the time period defined in
configuration property retry-unsuccessful-join-after. Retries can be disabled by setting the property
to off.
You can join to any node in the cluster. It does not have to be configured as a seed node. Note that you can only
join to an existing cluster member, which means that for bootstrapping some node must join itself,and then the
following nodes could join them to make up a cluster.
You may also use Cluster(system).joinSeedNodes to join programmatically, which is attractive
when dynamically discovering other nodes at startup by using some external tool or API. When using
joinSeedNodes you should not include the node itself except for the node that is supposed to be the first
seed node, and that should be placed first in parameter to joinSeedNodes.
Unsuccessful attempts to contact seed nodes are automatically retried after the time period defined in configuration
property seed-node-timeout. Unsuccessful attempt to join a specific seed node is automatically retried after
the configured retry-unsuccessful-join-after. Retrying means that it tries to contact all seed nodes
and then joins the node that answers first. The first node in the list of seed nodes will join itself if it cannot contact
any of the other seed nodes within the configured seed-node-timeout.
An actor system can only join a cluster once. Additional attempts will be ignored. When it has successfully joined
it must be restarted to be able to join another cluster or to join the same cluster again.It can use the same host name
and port after the restart, when it come up as new incarnation of existing member in the cluster, trying to join in,
then the existing one will be removed from the cluster and then it will be allowed to join.
Note: The name of the ActorSystem must be the same for all members of a cluster. The name is given when
you start the ActorSystem.
6.2.4 Downing
When a member is considered by the failure detector to be unreachable the leader is not allowed to perform its
duties, such as changing status of new joining members to ‘Up’. The node must first become reachable again, or
the status of the unreachable member must be changed to ‘Down’. Changing status to ‘Down’ can be performed
automatically or manually. By default it must be done manually, using JMX or HTTP.
It can also be performed programmatically with Cluster(system).down(address).
A pre-packaged solution for the downing problem is provided by Split Brain Resolver, which is part of the Lightbend Reactive Platform. If you don’t use RP, you should anyway carefully read the documentation of the Split
Brain Resolver and make sure that the solution you are using handles the concerns described there.
Auto-downing (DO NOT USE)
There is an automatic downing feature that you should not use in production. For testing purpose you can enable
it with configuration:
akka.cluster.auto-down-unreachable-after = 120s
This means that the cluster leader member will change the unreachable node status to down automatically
after the configured time of unreachability.
This is a naïve approach to remove unreachable nodes from the cluster membership. It works great for crashes
and short transient network partitions, but not for long network partitions. Both sides of the network partition will
see the other side as unreachable and after a while remove it from its cluster membership. Since this happens on
both sides the result is that two separate disconnected clusters have been created. This can also happen because of
long GC pauses or system overload.
6.2. Cluster Usage
284
Akka Scala Documentation, Release 2.5.0-RC1
Warning: We recommend against using the auto-down feature of Akka Cluster in production. This is crucial
for correct behavior if you use Cluster Singleton or Cluster Sharding, especially together with Akka Persistence. For Akka Persistence with Cluster Sharding it can result in corrupt data in case of network partitions.
6.2.5 Leaving
There are two ways to remove a member from the cluster.
You can just stop the actor system (or the JVM process). It will be detected as unreachable and removed after the
automatic or manual downing as described above.
A more graceful exit can be performed if you tell the cluster that a node shall leave. This can be performed using
JMX or HTTP. It can also be performed programmatically with:
val cluster = Cluster(system)
cluster.leave(cluster.selfAddress)
Note that this command can be issued to any member in the cluster, not necessarily the one that is leaving.
The Coordinated Shutdown will automatically run when the cluster node sees itself as Exiting, i.e. leaving from
another node will trigger the shutdown process on the leaving node. Tasks for graceful leaving of cluster including
graceful shutdown of Cluster Singletons and Cluster Sharding are added automatically when Akka Cluster is used,
i.e. running the shutdown process will also trigger the graceful leaving if it’s not already in progress.
Normally this is handled automatically, but in case of network failures during this process it might still be necessary
to set the node’s status to Down in order to complete the removal.
6.2.6 WeaklyUp Members
If a node is unreachable then gossip convergence is not possible and therefore any leader actions are also
not possible. However, we still might want new nodes to join the cluster in this scenario.
Joining members will be promoted to WeaklyUp and become part of the cluster if convergence can’t be
reached. Once gossip convergence is reached, the leader will move WeaklyUp members to Up.
This feature is enabled by default, but it can be disabled with configuration option:
akka.cluster.allow-weakly-up-members = off
You can subscribe to the WeaklyUp membership event to make use of the members that are in this state, but you
should be aware of that members on the other side of a network partition have no knowledge about the existence
of the new members. You should for example not count WeaklyUp members in quorum decisions.
6.2.7 Subscribe to Cluster Events
You can subscribe to change notifications of the cluster membership by using Cluster(system).
subscribe.
package scala.docs.cluster
import
import
import
import
akka.cluster.Cluster
akka.cluster.ClusterEvent._
akka.actor.ActorLogging
akka.actor.Actor
class SimpleClusterListener2 extends Actor with ActorLogging {
val cluster = Cluster(context.system)
6.2. Cluster Usage
285
Akka Scala Documentation, Release 2.5.0-RC1
// subscribe to cluster changes, re-subscribe when restart
override def preStart(): Unit = {
cluster.subscribe(self, classOf[MemberEvent], classOf[UnreachableMember])
}
override def postStop(): Unit = cluster.unsubscribe(self)
def receive = {
case state: CurrentClusterState =>
log.info("Current members: {}", state.members.mkString(", "))
case MemberUp(member) =>
log.info("Member is Up: {}", member.address)
case UnreachableMember(member) =>
log.info("Member detected as unreachable: {}", member)
case MemberRemoved(member, previousStatus) =>
log.info(
"Member is Removed: {} after {}",
member.address, previousStatus)
case _: MemberEvent => // ignore
}
}
A snapshot of the full state, akka.cluster.ClusterEvent.CurrentClusterState, is sent to the
subscriber as the first message, followed by events for incremental updates.
Note that you may receive an empty CurrentClusterState, containing no members, if you start the subscription before the initial join procedure has completed. This is expected behavior. When the node has been
accepted in the cluster you will receive MemberUp for that node, and other nodes.
If you find it inconvenient to handle the CurrentClusterState you can use ClusterEvent.
InitialStateAsEvents as parameter to subscribe.
That means that instead of receiving
CurrentClusterState as the first message you will receive the events corresponding to the current state
to mimic what you would have seen if you were listening to the events when they occurred in the past. Note that
those initial events only correspond to the current state and it is not the full history of all changes that actually has
occurred in the cluster.
cluster.subscribe(self, initialStateMode = InitialStateAsEvents,
classOf[MemberEvent], classOf[UnreachableMember])
The events to track the life-cycle of members are:
• ClusterEvent.MemberJoined - A new member has joined the cluster and its status has been changed
to Joining.
• ClusterEvent.MemberUp - A new member has joined the cluster and its status has been changed to
Up.
• ClusterEvent.MemberExited - A member is leaving the cluster and its status has been changed to
Exiting. Note that the node might already have been shutdown when this event is published on another
node.
• ClusterEvent.MemberRemoved - Member completely removed from the cluster.
• ClusterEvent.UnreachableMember - A member is considered as unreachable, detected by the
failure detector of at least one other node.
• ClusterEvent.ReachableMember - A member is considered as reachable again, after having been
unreachable. All nodes that previously detected it as unreachable has detected it as reachable again.
There are more types of change events, consult the API documentation of classes that extends akka.cluster.
ClusterEvent.ClusterDomainEvent for details about the events.
Instead of subscribing to cluster events it can sometimes be convenient to only get the full membership state with
Cluster(system).state. Note that this state is not necessarily in sync with the events published to a cluster
subscription.
6.2. Cluster Usage
286
Akka Scala Documentation, Release 2.5.0-RC1
Worker Dial-in Example
Let’s take a look at an example that illustrates how workers, here named backend, can detect and register to new
master nodes, here named frontend.
The example application provides a service to transform text. When some text is sent to one of the frontend
services, it will be delegated to one of the backend workers, which performs the transformation job, and sends the
result back to the original client. New backend nodes, as well as new frontend nodes, can be added or removed to
the cluster dynamically.
Messages:
final case class TransformationJob(text: String)
final case class TransformationResult(text: String)
final case class JobFailed(reason: String, job: TransformationJob)
case object BackendRegistration
The backend worker that performs the transformation job:
class TransformationBackend extends Actor {
val cluster = Cluster(context.system)
// subscribe to cluster changes, MemberUp
// re-subscribe when restart
override def preStart(): Unit = cluster.subscribe(self, classOf[MemberUp])
override def postStop(): Unit = cluster.unsubscribe(self)
def receive = {
case TransformationJob(text) => sender() ! TransformationResult(text.
˓→toUpperCase)
case state: CurrentClusterState =>
state.members.filter(_.status == MemberStatus.Up) foreach register
case MemberUp(m) => register(m)
}
def register(member: Member): Unit =
if (member.hasRole("frontend"))
context.actorSelection(RootActorPath(member.address) / "user" / "frontend") !
BackendRegistration
}
Note that the TransformationBackend actor subscribes to cluster events to detect new, potential, frontend
nodes, and send them a registration message so that they know that they can use the backend worker.
The frontend that receives user jobs and delegates to one of the registered backend workers:
class TransformationFrontend extends Actor {
var backends = IndexedSeq.empty[ActorRef]
var jobCounter = 0
def receive = {
case job: TransformationJob if backends.isEmpty =>
sender() ! JobFailed("Service unavailable, try again later", job)
case job: TransformationJob =>
jobCounter += 1
backends(jobCounter % backends.size) forward job
case BackendRegistration if !backends.contains(sender()) =>
context watch sender()
backends = backends :+ sender()
6.2. Cluster Usage
287
Akka Scala Documentation, Release 2.5.0-RC1
case Terminated(a) =>
backends = backends.filterNot(_ == a)
}
}
Note that the TransformationFrontend actor watch the registered backend to be able to remove it from
its list of available backend workers. Death watch uses the cluster failure detector for nodes in the cluster, i.e.
it detects network failures and JVM crashes, in addition to graceful termination of watched actor. Death watch
generates the Terminated message to the watching actor when the unreachable cluster node has been downed
and removed.
The Lightbend Activator tutorial named Akka Cluster Samples with Scala. contains the full source code and
instructions of how to run the Worker Dial-in Example.
6.2.8 Node Roles
Not all nodes of a cluster need to perform the same function: there might be one sub-set which runs the web
front-end, one which runs the data access layer and one for the number-crunching. Deployment of actors—for
example by cluster-aware routers—can take node roles into account to achieve this distribution of responsibilities.
The roles of a node is defined in the configuration property named akka.cluster.roles and it is typically
defined in the start script as a system property or environment variable.
The roles of the nodes is part of the membership information in MemberEvent that you can subscribe to.
6.2.9 How To Startup when Cluster Size Reached
A common use case is to start actors after the cluster has been initialized, members have joined, and the cluster
has reached a certain size.
With a configuration option you can define required number of members before the leader changes member status
of ‘Joining’ members to ‘Up’.:
akka.cluster.min-nr-of-members = 3
In a similar way you can define required number of members of a certain role before the leader changes member
status of ‘Joining’ members to ‘Up’.:
akka.cluster.role {
frontend.min-nr-of-members = 1
backend.min-nr-of-members = 2
}
You can start the actors in a registerOnMemberUp callback, which will be invoked when the current member
status is changed to ‘Up’, i.e. the cluster has at least the defined number of members.
Cluster(system) registerOnMemberUp {
system.actorOf(
Props(classOf[FactorialFrontend], upToN, true),
name = "factorialFrontend")
}
This callback can be used for other things than starting actors.
6.2.10 How To Cleanup when Member is Removed
You can do some clean up in a registerOnMemberRemoved callback, which will be invoked when the current
member status is changed to ‘Removed’ or the cluster have been shutdown.
6.2. Cluster Usage
288
Akka Scala Documentation, Release 2.5.0-RC1
An alternative is to register tasks to the Coordinated Shutdown.
Note: Register a OnMemberRemoved callback on a cluster that have been shutdown, the callback will be invoked
immediately on the caller thread, otherwise it will be invoked later when the current member status changed to
‘Removed’. You may want to install some cleanup handling after the cluster was started up, but the cluster might
already be shutting down when you installing, and depending on the race is not healthy.
6.2.11 Cluster Singleton
For some use cases it is convenient and sometimes also mandatory to ensure that you have exactly one actor of a
certain type running somewhere in the cluster.
This can be implemented by subscribing to member events, but there are several corner cases to consider. Therefore, this specific use case is made easily accessible by the Cluster Singleton.
6.2.12 Cluster Sharding
Distributes actors across several nodes in the cluster and supports interaction with the actors using their logical
identifier, but without having to care about their physical location in the cluster.
See Cluster Sharding
6.2.13 Distributed Publish Subscribe
Publish-subscribe messaging between actors in the cluster, and point-to-point messaging using the logical path of
the actors, i.e. the sender does not have to know on which node the destination actor is running.
See Distributed Publish Subscribe in Cluster.
6.2.14 Cluster Client
Communication from an actor system that is not part of the cluster to actors running somewhere in the cluster.
The client does not have to know on which node the destination actor is running.
See Cluster Client.
6.2.15 Distributed Data
Akka Distributed Data is useful when you need to share data between nodes in an Akka Cluster. The data is
accessed with an actor providing a key-value store like API.
See Distributed Data.
6.2.16 Failure Detector
In a cluster each node is monitored by a few (default maximum 5) other nodes, and when any of these detects the
node as unreachable that information will spread to the rest of the cluster through the gossip. In other words,
only one node needs to mark a node unreachable to have the rest of the cluster mark that node unreachable.
The failure detector will also detect if the node becomes reachable again. When all nodes that monitored the
unreachable node detects it as reachable again the cluster, after gossip dissemination, will consider it as
reachable.
If system messages cannot be delivered to a node it will be quarantined and then it cannot come back from
unreachable. This can happen if the there are too many unacknowledged system messages (e.g. watch,
6.2. Cluster Usage
289
Akka Scala Documentation, Release 2.5.0-RC1
Terminated, remote actor deployment, failures of actors supervised by remote parent). Then the node needs to be
moved to the down or removed states and the actor system of the quarantined node must be restarted before it
can join the cluster again.
The nodes in the cluster monitor each other by sending heartbeats to detect if a node is unreachable from the rest
of the cluster. The heartbeat arrival times is interpreted by an implementation of The Phi Accrual Failure Detector.
The suspicion level of failure is given by a value called phi. The basic idea of the phi failure detector is to express
the value of phi on a scale that is dynamically adjusted to reflect current network conditions.
The value of phi is calculated as:
phi = -log10(1 - F(timeSinceLastHeartbeat))
where F is the cumulative distribution function of a normal distribution with mean and standard deviation estimated
from historical heartbeat inter-arrival times.
In the Configuration you can adjust the akka.cluster.failure-detector.threshold to define when
a phi value is considered to be a failure.
A low threshold is prone to generate many false positives but ensures a quick detection in the event of a real
crash. Conversely, a high threshold generates fewer mistakes but needs more time to detect actual crashes.
The default threshold is 8 and is appropriate for most situations. However in cloud environments, such as
Amazon EC2, the value could be increased to 12 in order to account for network issues that sometimes occur on
such platforms.
The following chart illustrates how phi increase with increasing time since the previous heartbeat.
Phi is calculated from the mean and standard deviation of historical inter arrival times. The previous chart is an
example for standard deviation of 200 ms. If the heartbeats arrive with less deviation the curve becomes steeper,
i.e. it is possible to determine failure more quickly. The curve looks like this for a standard deviation of 100 ms.
6.2. Cluster Usage
290
Akka Scala Documentation, Release 2.5.0-RC1
To be able to survive sudden abnormalities, such as garbage collection pauses and transient network failures the failure detector is configured with a margin, akka.cluster.failure-detector.
acceptable-heartbeat-pause. You may want to adjust the Configuration of this depending on you
environment. This is how the curve looks like for acceptable-heartbeat-pause configured to 3 seconds.
Death watch uses the cluster failure detector for nodes in the cluster, i.e. it detects network failures and JVM
crashes, in addition to graceful termination of watched actor. Death watch generates the Terminated message
to the watching actor when the unreachable cluster node has been downed and removed.
6.2. Cluster Usage
291
Akka Scala Documentation, Release 2.5.0-RC1
If you encounter suspicious false positives when the system is under load you should define a separate dispatcher
for the cluster actors as described in Cluster Dispatcher.
6.2.17 Cluster Aware Routers
All routers can be made aware of member nodes in the cluster, i.e. deploying new routees or looking up routees
on nodes in the cluster. When a node becomes unreachable or leaves the cluster the routees of that node are
automatically unregistered from the router. When new nodes join the cluster, additional routees are added to the
router, according to the configuration. Routees are also added when a node becomes reachable again, after having
been unreachable.
Cluster aware routers make use of members with status WeaklyUp if that feature is enabled.
There are two distinct types of routers.
• Group - router that sends messages to the specified path using actor selection The routees can be shared
among routers running on different nodes in the cluster. One example of a use case for this type of router is
a service running on some backend nodes in the cluster and used by routers running on front-end nodes in
the cluster.
• Pool - router that creates routees as child actors and deploys them on remote nodes. Each router will
have its own routee instances. For example, if you start a router on 3 nodes in a 10-node cluster, you will
have 30 routees in total if the router is configured to use one instance per node. The routees created by the
different routers will not be shared among the routers. One example of a use case for this type of router is
a single master that coordinates jobs and delegates the actual work to routees running on other nodes in the
cluster.
Router with Group of Routees
When using a Group you must start the routee actors on the cluster member nodes. That is not done by the router.
The configuration for a group looks like this::
akka.actor.deployment {
/statsService/workerRouter {
router = consistent-hashing-group
routees.paths = ["/user/statsWorker"]
cluster {
enabled = on
allow-local-routees = on
use-role = compute
}
}
}
Note: The routee actors should be started as early as possible when starting the actor system, because the router
will try to use them as soon as the member status is changed to ‘Up’.
The actor paths without address information that are defined in routees.paths are used for selecting the
actors to which the messages will be forwarded to by the router. Messages will be forwarded to the routees using
ActorSelection, so the same delivery semantics should be expected. It is possible to limit the lookup of routees to
member nodes tagged with a certain role by specifying use-role.
max-total-nr-of-instances defines total number of routees in the cluster.
By default
max-total-nr-of-instances is set to a high value (10000) that will result in new routees added to the
router when nodes join the cluster. Set it to a lower value if you want to limit total number of routees.
The same type of router could also have been defined in code:
6.2. Cluster Usage
292
Akka Scala Documentation, Release 2.5.0-RC1
import akka.cluster.routing.{ ClusterRouterGroup, ClusterRouterGroupSettings }
import akka.routing.ConsistentHashingGroup
val workerRouter = context.actorOf(
ClusterRouterGroup(ConsistentHashingGroup(Nil), ClusterRouterGroupSettings(
totalInstances = 100, routeesPaths = List("/user/statsWorker"),
allowLocalRoutees = true, useRole = Some("compute"))).props(),
name = "workerRouter2")
See Configuration section for further descriptions of the settings.
Router Example with Group of Routees
Let’s take a look at how to use a cluster aware router with a group of routees, i.e. router sending to the paths of
the routees.
The example application provides a service to calculate statistics for a text. When some text is sent to the service
it splits it into words, and delegates the task to count number of characters in each word to a separate worker, a
routee of a router. The character count for each word is sent back to an aggregator that calculates the average
number of characters per word when all results have been collected.
Messages:
final case class StatsJob(text: String)
final case class StatsResult(meanWordLength: Double)
final case class JobFailed(reason: String)
The worker that counts number of characters in each word:
class StatsWorker extends Actor {
var cache = Map.empty[String, Int]
def receive = {
case word: String ⇒
val length = cache.get(word) match {
case Some(x) ⇒ x
case None ⇒
val x = word.length
cache += (word → x)
x
}
sender() ! length
}
}
The service that receives text from users and splits it up into words, delegates to workers and aggregates:
class StatsService extends Actor {
// This router is used both with lookup and deploy of routees. If you
// have a router with only lookup of routees you can use Props.empty
// instead of Props[StatsWorker.class].
val workerRouter = context.actorOf(
FromConfig.props(Props[StatsWorker]),
name = "workerRouter")
def receive = {
case StatsJob(text) if text != "" ⇒
val words = text.split(" ")
val replyTo = sender() // important to not close over sender()
// create actor that collects replies from workers
val aggregator = context.actorOf(Props(
classOf[StatsAggregator], words.size, replyTo))
6.2. Cluster Usage
293
Akka Scala Documentation, Release 2.5.0-RC1
words foreach { word ⇒
workerRouter.tell(
ConsistentHashableEnvelope(word, word), aggregator)
}
}
}
class StatsAggregator(expectedResults: Int, replyTo: ActorRef) extends Actor {
var results = IndexedSeq.empty[Int]
context.setReceiveTimeout(3.seconds)
def receive = {
case wordCount: Int ⇒
results = results :+ wordCount
if (results.size == expectedResults) {
val meanWordLength = results.sum.toDouble / results.size
replyTo ! StatsResult(meanWordLength)
context.stop(self)
}
case ReceiveTimeout ⇒
replyTo ! JobFailed("Service unavailable, try again later")
context.stop(self)
}
}
Note, nothing cluster specific so far, just plain actors.
All nodes start StatsService and StatsWorker actors. Remember, routees are the workers in this case.
The router is configured with routees.paths::
akka.actor.deployment {
/statsService/workerRouter {
router = consistent-hashing-group
routees.paths = ["/user/statsWorker"]
cluster {
enabled = on
allow-local-routees = on
use-role = compute
}
}
}
This means that user requests can be sent to StatsService on any node and it will use StatsWorker on all
nodes.
The Lightbend Activator tutorial named Akka Cluster Samples with Scala. contains the full source code and
instructions of how to run the Router Example with Group of Routees.
Router with Pool of Remote Deployed Routees
When using a Pool with routees created and deployed on the cluster member nodes the configuration for a router
looks like this::
akka.actor.deployment {
/statsService/singleton/workerRouter {
router = consistent-hashing-pool
cluster {
enabled = on
max-nr-of-instances-per-node = 3
allow-local-routees = on
use-role = compute
}
6.2. Cluster Usage
294
Akka Scala Documentation, Release 2.5.0-RC1
}
}
It is possible to limit the deployment of routees to member nodes tagged with a certain role by specifying
use-role.
max-total-nr-of-instances defines total number of routees in the cluster, but the number
of routees per node, max-nr-of-instances-per-node, will not be exceeded.
By default
max-total-nr-of-instances is set to a high value (10000) that will result in new routees added to the
router when nodes join the cluster. Set it to a lower value if you want to limit total number of routees.
The same type of router could also have been defined in code:
import akka.cluster.routing.{ ClusterRouterPool, ClusterRouterPoolSettings }
import akka.routing.ConsistentHashingPool
val workerRouter = context.actorOf(
ClusterRouterPool(ConsistentHashingPool(0), ClusterRouterPoolSettings(
totalInstances = 100, maxInstancesPerNode = 3,
allowLocalRoutees = false, useRole = None)).props(Props[StatsWorker]),
name = "workerRouter3")
See Configuration section for further descriptions of the settings.
Router Example with Pool of Remote Deployed Routees
Let’s take a look at how to use a cluster aware router on single master node that creates and deploys workers. To keep track of a single master we use the Cluster Singleton in the cluster-tools module. The
ClusterSingletonManager is started on each node.:
system.actorOf(
ClusterSingletonManager.props(
singletonProps = Props[StatsService],
terminationMessage = PoisonPill,
settings = ClusterSingletonManagerSettings(system).withRole("compute")),
name = "statsService")
We also need an actor on each node that keeps track of where current single master exists and delegates jobs to
the StatsService. That is provided by the ClusterSingletonProxy.:
system.actorOf(
ClusterSingletonProxy.props(
singletonManagerPath = "/user/statsService",
settings = ClusterSingletonProxySettings(system).withRole("compute")),
name = "statsServiceProxy")
The ClusterSingletonProxy receives text from users and delegates to the current StatsService, the
single master. It listens to cluster events to lookup the StatsService on the oldest node.
All nodes start ClusterSingletonProxy and the ClusterSingletonManager. The router is now
configured like this::
akka.actor.deployment {
/statsService/singleton/workerRouter {
router = consistent-hashing-pool
cluster {
enabled = on
max-nr-of-instances-per-node = 3
allow-local-routees = on
use-role = compute
}
6.2. Cluster Usage
295
Akka Scala Documentation, Release 2.5.0-RC1
}
}
The Lightbend Activator tutorial named Akka Cluster Samples with Scala. contains the full source code and
instructions of how to run the Router Example with Pool of Remote Deployed Routees.
6.2.18 Cluster Metrics
The member nodes of the cluster can collect system health metrics and publish that to other cluster nodes and to
the registered subscribers on the system event bus with the help of Cluster Metrics Extension.
6.2.19 How to Test
Multi Node Testing is useful for testing cluster applications.
Set up your project according to the instructions in Multi Node Testing and Multi JVM Testing, i.e. add the
sbt-multi-jvm plugin and the dependency to akka-multi-node-testkit.
First, as described in Multi Node Testing, we need some scaffolding to configure the MultiNodeSpec. Define
the participating roles and their Configuration in an object extending MultiNodeConfig:
import akka.remote.testkit.MultiNodeConfig
import com.typesafe.config.ConfigFactory
object StatsSampleSpecConfig extends MultiNodeConfig {
// register the named roles (nodes) of the test
val first = role("first")
val second = role("second")
val third = role("third")
def nodeList = Seq(first, second, third)
// Extract individual sigar library for every node.
nodeList foreach { role ⇒
nodeConfig(role) {
ConfigFactory.parseString(s"""
# Enable metrics extension in akka-cluster-metrics.
akka.extensions=["akka.cluster.metrics.ClusterMetricsExtension"]
# Sigar native library extract location during tests.
akka.cluster.metrics.native-library-extract-folder=target/native/${role.name}
""")
}
}
// this configuration will be used for all nodes
// note that no fixed host names and ports are used
commonConfig(ConfigFactory.parseString("""
akka.actor.provider = cluster
akka.remote.log-remote-lifecycle-events = off
akka.cluster.roles = [compute]
// router lookup config ...
"""))
}
Define one concrete test class for each role/node. These will be instantiated on the different nodes (JVMs). They
can be implemented differently, but often they are the same and extend an abstract test class, as illustrated here.
// need one concrete test class per node
class StatsSampleSpecMultiJvmNode1 extends StatsSampleSpec
6.2. Cluster Usage
296
Akka Scala Documentation, Release 2.5.0-RC1
class StatsSampleSpecMultiJvmNode2 extends StatsSampleSpec
class StatsSampleSpecMultiJvmNode3 extends StatsSampleSpec
Note the naming convention of these classes. The name of the classes must end with MultiJvmNode1,
MultiJvmNode2 and so on. It is possible to define another suffix to be used by the sbt-multi-jvm, but
the default should be fine in most cases.
Then the abstract MultiNodeSpec, which takes the MultiNodeConfig as constructor parameter.
import akka.remote.testkit.MultiNodeSpec
import akka.testkit.ImplicitSender
import org.scalatest.{ BeforeAndAfterAll, Matchers, WordSpecLike }
abstract class StatsSampleSpec extends MultiNodeSpec(StatsSampleSpecConfig)
with WordSpecLike with Matchers with BeforeAndAfterAll
with ImplicitSender {
import StatsSampleSpecConfig._
override def initialParticipants = roles.size
override def beforeAll() = multiNodeSpecBeforeAll()
override def afterAll() = multiNodeSpecAfterAll()
Most of this can of course be extracted to a separate trait to avoid repeating this in all your tests.
Typically you begin your test by starting up the cluster and let the members join, and create some actors. That can
be done like this:
"illustrate how to startup cluster" in within(15 seconds) {
Cluster(system).subscribe(testActor, classOf[MemberUp])
expectMsgClass(classOf[CurrentClusterState])
val firstAddress = node(first).address
val secondAddress = node(second).address
val thirdAddress = node(third).address
Cluster(system) join firstAddress
system.actorOf(Props[StatsWorker], "statsWorker")
system.actorOf(Props[StatsService], "statsService")
receiveN(3).collect { case MemberUp(m) ⇒ m.address }.toSet should be(
Set(firstAddress, secondAddress, thirdAddress))
Cluster(system).unsubscribe(testActor)
testConductor.enter("all-up")
}
From the test you interact with the cluster using the Cluster extension, e.g. join.
Cluster(system) join firstAddress
Notice how the testActor from testkit is added as subscriber to cluster changes and then waiting for certain events,
such as in this case all members becoming ‘Up’.
The above code was running for all roles (JVMs). runOn is a convenient utility to declare that a certain block of
code should only run for a specific role.
"show usage of the statsService from one node" in within(15 seconds) {
runOn(second) {
6.2. Cluster Usage
297
Akka Scala Documentation, Release 2.5.0-RC1
assertServiceOk()
}
testConductor.enter("done-2")
}
def assertServiceOk(): Unit = {
val service = system.actorSelection(node(third) / "user" / "statsService")
// eventually the service should be ok,
// first attempts might fail because worker actors not started yet
awaitAssert {
service ! StatsJob("this is the text that will be analyzed")
expectMsgType[StatsResult](1.second).meanWordLength should be(
3.875 +- 0.001)
}
}
Once again we take advantage of the facilities in testkit to verify expected behavior. Here using testActor as
sender (via ImplicitSender) and verifying the reply with expectMsgPF.
In the above code you can see node(third), which is useful facility to get the root actor reference of the actor
system for a specific role. This can also be used to grab the akka.actor.Address of that node.
val firstAddress = node(first).address
val secondAddress = node(second).address
val thirdAddress = node(third).address
6.2.20 Management
HTTP
Information and management of the cluster is available with a HTTP API. See documentation of akka/akkacluster-management.
JMX
Information and management of the cluster is available as JMX MBeans with the root name akka.Cluster.
The JMX information can be displayed with an ordinary JMX console such as JConsole or JVisualVM.
From JMX you can:
• see what members that are part of the cluster
• see status of this node
• see roles of each member
• join this node to another node in cluster
• mark any node in the cluster as down
• tell any node in the cluster to leave
Member nodes are identified
name>@<hostname>:<port>.
6.2. Cluster Usage
by
their
address,
in
format
akka.<protocol>://<actor-system-
298
Akka Scala Documentation, Release 2.5.0-RC1
Command Line
Warning: Deprecation warning - The command line script has been deprecated and is scheduled for removal
in the next major version. Use the HTTP API with curl or similar instead.
The cluster can be managed with the script akka-cluster provided in the Akka github repository here: http://
github.com/akka/akka/tree/v2.5.0-RC1/akka-cluster/jmx-client. Place the script and the jmxsh-R5.jar library
in the same directory.
Run it without parameters to see instructions about how to use the script:
Usage: ./akka-cluster <node-hostname> <jmx-port> <command> ...
Supported commands are:
join <node-url>
leave <node-url>
down <node-url>
member-status
members
unreachable
cluster-status
-
Sends request a JOIN node with the specified URL
Sends a request for node with URL to LEAVE the cluster
Sends a request for marking node with URL as DOWN
Asks the member node for its current status
Asks the cluster for addresses of current members
Asks the cluster for addresses of unreachable members
Asks the cluster for its current status (member ring,
unavailable nodes, meta data etc.)
leader - Asks the cluster who the current leader is
is-singleton - Checks if the cluster is a singleton cluster (single
node cluster)
is-available - Checks if the member node is available
Where the <node-url> should be on the format of
'akka.<protocol>://<actor-system-name>@<hostname>:<port>'
Examples: ./akka-cluster localhost 9999 is-available
./akka-cluster localhost 9999 join akka.tcp://MySystem@darkstar:2552
./akka-cluster localhost 9999 cluster-status
To be able to use the script you must enable remote monitoring and management when starting the JVMs of the
cluster nodes, as described in Monitoring and Management Using JMX Technology. Make sure you understand
the security implications of enabling remote monitoring and management.
6.2.21 Configuration
There are several configuration properties for the cluster. We refer to the reference configuration for more information.
Cluster Info Logging
You can silence the logging of cluster events at info level with configuration property:
akka.cluster.log-info = off
Cluster Dispatcher
Under the hood the cluster extension is implemented with actors and it can be necessary to create a bulkhead
for those actors to avoid disturbance from other actors. Especially the heartbeating actors that is used for failure
detection can generate false positives if they are not given a chance to run at regular intervals. For this purpose
you can define a separate dispatcher to be used for the cluster actors:
6.2. Cluster Usage
299
Akka Scala Documentation, Release 2.5.0-RC1
akka.cluster.use-dispatcher = cluster-dispatcher
cluster-dispatcher {
type = "Dispatcher"
executor = "fork-join-executor"
fork-join-executor {
parallelism-min = 2
parallelism-max = 4
}
}
Note: Normally it should not be necessary to configure a separate dispatcher for the Cluster. The defaultdispatcher should be sufficient for performing the Cluster tasks, i.e. akka.cluster.use-dispatcher
should not be changed. If you have Cluster related problems when using the default-dispatcher that is typically
an indication that you are running blocking or CPU intensive actors/tasks on the default-dispatcher. Use dedicated dispatchers for such actors/tasks instead of running them on the default-dispatcher, because that may starve
system internal tasks. Related config properties: akka.cluster.use-dispatcher = akka.cluster.
cluster-dispatcher. Corresponding default values: akka.cluster.use-dispatcher =.
6.3 Cluster Singleton
For some use cases it is convenient and sometimes also mandatory to ensure that you have exactly one actor of a
certain type running somewhere in the cluster.
Some examples:
• single point of responsibility for certain cluster-wide consistent decisions, or coordination of actions across
the cluster system
• single entry point to an external system
• single master, many workers
• centralized naming service, or routing logic
Using a singleton should not be the first design choice. It has several drawbacks, such as single-point of bottleneck.
Single-point of failure is also a relevant concern, but for some cases this feature takes care of that by making sure
that another singleton instance will eventually be started.
The
cluster
singleton
pattern
is
implemented
by
akka.cluster.singleton.
ClusterSingletonManager. It manages one singleton actor instance among all cluster nodes or a
group of nodes tagged with a specific role. ClusterSingletonManager is an actor that is supposed to
be started on all nodes, or all nodes with specified role, in the cluster. The actual singleton actor is started
by the ClusterSingletonManager on the oldest node by creating a child actor from supplied Props.
ClusterSingletonManager makes sure that at most one singleton instance is running at any point in time.
The singleton actor is always running on the oldest member with specified role. The oldest member is determined
by akka.cluster.Member#isOlderThan. This can change when removing that member from the cluster.
Be aware that there is a short time period when there is no active singleton during the hand-over process.
The cluster failure detector will notice when oldest node becomes unreachable due to things like JVM crash, hard
shut down, or network failure. Then a new oldest node will take over and a new singleton actor is created. For
these failure scenarios there will not be a graceful hand-over, but more than one active singletons is prevented by
all reasonable means. Some corner cases are eventually resolved by configurable timeouts.
You can access the singleton actor by using the provided akka.cluster.singleton.
ClusterSingletonProxy, which will route all messages to the current instance of the singleton. The
proxy will keep track of the oldest node in the cluster and resolve the singleton’s ActorRef by explicitly
sending the singleton’s actorSelection the akka.actor.Identify message and waiting for it to reply.
This is performed periodically if the singleton doesn’t reply within a certain (configurable) time. Given the
6.3. Cluster Singleton
300
Akka Scala Documentation, Release 2.5.0-RC1
implementation, there might be periods of time during which the ActorRef is unavailable, e.g., when a node
leaves the cluster. In these cases, the proxy will buffer the messages sent to the singleton and then deliver them
when the singleton is finally available. If the buffer is full the ClusterSingletonProxy will drop old
messages when new messages are sent via the proxy. The size of the buffer is configurable and it can be disabled
by using a buffer size of 0.
It’s worth noting that messages can always be lost because of the distributed nature of these actors. As always,
additional logic should be implemented in the singleton (acknowledgement) and in the client (retry) actors to
ensure at-least-once message delivery.
The singleton instance will not run on members with status WeaklyUp.
6.3.1 Potential problems to be aware of
This pattern may seem to be very tempting to use at first, but it has several drawbacks, some of them are listed
below:
• the cluster singleton may quickly become a performance bottleneck,
• you can not rely on the cluster singleton to be non-stop available — e.g. when the node on which the
singleton has been running dies, it will take a few seconds for this to be noticed and the singleton be
migrated to another node,
• in the case of a network partition appearing in a Cluster that is using Automatic Downing (see Auto Downing
docs for Downing), it may happen that the isolated clusters each decide to spin up their own singleton,
meaning that there might be multiple singletons running in the system, yet the Clusters have no way of
finding out about them (because of the partition).
Especially the last point is something you should be aware of — in general when using the Cluster Singleton
pattern you should take care of downing nodes yourself and not rely on the timing based auto-down feature.
Warning: Don’t use Cluster Singleton together with Automatic Downing, since it allows the cluster to
split up into two separate clusters, which in turn will result in multiple Singletons being started, one in each
separate cluster!
6.3.2 An Example
Assume that we need one single entry point to an external system. An actor that receives messages from a JMS
queue with the strict requirement that only one JMS consumer must exist to be make sure that the messages are
processed in order. That is perhaps not how one would like to design things, but a typical real-world scenario
when integrating with external systems.
On each node in the cluster you need to start the ClusterSingletonManager and supply the Props of the
singleton actor, in this case the JMS queue consumer.
system.actorOf(
ClusterSingletonManager.props(
singletonProps = Props(classOf[Consumer], queue, testActor),
terminationMessage = End,
settings = ClusterSingletonManagerSettings(system).withRole("worker")),
name = "consumer")
Here we limit the singleton to nodes tagged with the "worker" role, but all nodes, independent of role, can be
used by not specifying withRole.
Here we use an application specific terminationMessage to be able to close the resources before actually
stopping the singleton actor. Note that PoisonPill is a perfectly fine terminationMessage if you only
need to stop the actor.
Here is how the singleton actor handles the terminationMessage in this example.
6.3. Cluster Singleton
301
Akka Scala Documentation, Release 2.5.0-RC1
case End ⇒
queue ! UnregisterConsumer
case UnregistrationOk ⇒
stoppedBeforeUnregistration = false
context stop self
case Ping ⇒
sender() ! Pong
With the names given above, access to the singleton can be obtained from any cluster node using a properly
configured proxy.
system.actorOf(
ClusterSingletonProxy.props(
singletonManagerPath = "/user/consumer",
settings = ClusterSingletonProxySettings(system).withRole("worker")),
name = "consumerProxy")
A more comprehensive sample is available in the Lightbend Activator tutorial named Distributed workers with
Akka and Scala!.
6.3.3 Dependencies
To use the Cluster Singleton you must add the following dependency in your project.
sbt:
"com.typesafe.akka" %% "akka-cluster-tools" % "2.5.0-RC1"
maven:
<dependency>
<groupId>com.typesafe.akka</groupId>
<artifactId>akka-cluster-tools_2.11</artifactId>
<version>2.5.0-RC1</version>
</dependency>
6.3.4 Configuration
The following configuration properties are read by the ClusterSingletonManagerSettings
when created with a ActorSystem parameter.
It is also possible to amend the
ClusterSingletonManagerSettings or create it from another config section with the same layout as
below. ClusterSingletonManagerSettings is a parameter to the ClusterSingletonManager.
props factory method, i.e. each singleton can be configured with different settings if needed.
akka.cluster.singleton {
# The actor name of the child singleton actor.
singleton-name = "singleton"
# Singleton among the nodes tagged with specified role.
# If the role is not specified it's a singleton among all nodes in the cluster.
role = ""
# When a node is becoming oldest it sends hand-over request to previous oldest,
# that might be leaving the cluster. This is retried with this interval until
# the previous oldest confirms that the hand over has started or the previous
# oldest member is removed from the cluster (+ akka.cluster.down-removal-margin).
hand-over-retry-interval = 1s
# The number of retries are derived from hand-over-retry-interval and
6.3. Cluster Singleton
302
Akka Scala Documentation, Release 2.5.0-RC1
˓→
# akka.cluster.down-removal-margin (or ClusterSingletonManagerSettings.
removalMargin),
# but it will never be less than this property.
min-number-of-hand-over-retries = 10
}
The following configuration properties are read by the ClusterSingletonProxySettings when created
with a ActorSystem parameter. It is also possible to amend the ClusterSingletonProxySettings or
create it from another config section with the same layout as below. ClusterSingletonProxySettings
is a parameter to the ClusterSingletonProxy.props factory method, i.e. each singleton proxy can be
configured with different settings if needed.
akka.cluster.singleton-proxy {
# The actor name of the singleton actor that is started by the
˓→ClusterSingletonManager
singleton-name = ${akka.cluster.singleton.singleton-name}
# The role of the cluster nodes where the singleton can be deployed.
# If the role is not specified then any node will do.
role = ""
# Interval at which the proxy will try to resolve the singleton instance.
singleton-identification-interval = 1s
# If the location of the singleton is unknown the proxy will buffer this
# number of messages and deliver them when the singleton is identified.
# When the buffer is full old messages will be dropped when new messages are
# sent via the proxy.
# Use 0 to disable buffering, i.e. messages will be dropped immediately if
# the location of the singleton is unknown.
# Maximum allowed buffer size is 10000.
buffer-size = 1000
}
6.4 Distributed Publish Subscribe in Cluster
How do I send a message to an actor without knowing which node it is running on?
How do I send messages to all actors in the cluster that have registered interest in a named topic?
This pattern provides a mediator actor, akka.cluster.pubsub.DistributedPubSubMediator, that
manages a registry of actor references and replicates the entries to peer actors among all cluster nodes or a group
of nodes tagged with a specific role.
The DistributedPubSubMediator actor is supposed to be started on all nodes, or all nodes with specified
role, in the cluster. The mediator can be started with the DistributedPubSub extension or as an ordinary
actor.
The registry is eventually consistent, i.e. changes are not immediately visible at other nodes, but typically they
will be fully replicated to all other nodes after a few seconds. Changes are only performed in the own part of the
registry and those changes are versioned. Deltas are disseminated in a scalable way to other nodes with a gossip
protocol.
Cluster members with status WeaklyUp, will participate in Distributed Publish Subscribe, i.e. subscribers on
nodes with WeaklyUp status will receive published messages if the publisher and subscriber are on same side of
a network partition.
You can send messages via the mediator on any node to registered actors on any other node.
There a two different modes of message delivery, explained in the sections Publish and Send below.
6.4. Distributed Publish Subscribe in Cluster
303
Akka Scala Documentation, Release 2.5.0-RC1
A more comprehensive sample is available in the Lightbend Activator tutorial named Akka Clustered PubSub with
Scala!.
6.4.1 Publish
This is the true pub/sub mode. A typical usage of this mode is a chat room in an instant messaging application.
Actors are registered to a named topic. This enables many subscribers on each node. The message will be delivered
to all subscribers of the topic.
For efficiency the message is sent over the wire only once per node (that has a matching topic), and then delivered
to all subscribers of the local topic representation. (See more in )
You register actors to the local mediator with DistributedPubSubMediator.Subscribe. Successful Subscribe and Unsubscribe is acknowledged with DistributedPubSubMediator.
SubscribeAck and DistributedPubSubMediator.UnsubscribeAck replies. The acknowledgment
means that the subscription is registered, but it can still take some time until it is replicated to other nodes.
You publish messages by sending DistributedPubSubMediator.Publish message to the local mediator.
Actors are automatically removed from the registry when they are terminated, or you can explicitly remove entries
with DistributedPubSubMediator.Unsubscribe.
An example of a subscriber actor:
class Subscriber extends Actor with ActorLogging {
import DistributedPubSubMediator.{ Subscribe, SubscribeAck }
val mediator = DistributedPubSub(context.system).mediator
// subscribe to the topic named "content"
mediator ! Subscribe("content", self)
def receive = {
case s: String ⇒
log.info("Got {}", s)
case SubscribeAck(Subscribe("content", None, `self`)) ⇒
log.info("subscribing")
}
}
Subscriber actors can be started on several nodes in the cluster, and all will receive messages published to the
“content” topic.
runOn(first) {
system.actorOf(Props[Subscriber], "subscriber1")
}
runOn(second) {
system.actorOf(Props[Subscriber], "subscriber2")
system.actorOf(Props[Subscriber], "subscriber3")
}
A simple actor that publishes to this “content” topic:
class Publisher extends Actor {
import DistributedPubSubMediator.Publish
// activate the extension
val mediator = DistributedPubSub(context.system).mediator
def receive = {
case in: String ⇒
val out = in.toUpperCase
mediator ! Publish("content", out)
}
}
6.4. Distributed Publish Subscribe in Cluster
304
Akka Scala Documentation, Release 2.5.0-RC1
It can publish messages to the topic from anywhere in the cluster:
runOn(third) {
val publisher = system.actorOf(Props[Publisher], "publisher")
later()
// after a while the subscriptions are replicated
publisher ! "hello"
}
Topic Groups
Actors may also be subscribed to a named topic with a group id. If subscribing with a group id, each message
published to a topic with the sendOneMessageToEachGroup flag set to true is delivered via the supplied
RoutingLogic (default random) to one actor within each subscribing group.
If all the subscribed actors have the same group id, then this works just like Send and each message is only
delivered to one subscriber.
If all the subscribed actors have different group names, then this works like normal Publish and each message
is broadcasted to all subscribers.
Note:
Note that if the group id is used it is part of the topic identifier. Messages published with
sendOneMessageToEachGroup=false will not be delivered to subscribers that subscribed with a group
id. Messages published with sendOneMessageToEachGroup=true will not be delivered to subscribers
that subscribed without a group id.
6.4.2 Send
This is a point-to-point mode where each message is delivered to one destination, but you still do not have to
know where the destination is located. A typical usage of this mode is private chat to one other user in an instant
messaging application. It can also be used for distributing tasks to registered workers, like a cluster aware router
where the routees dynamically can register themselves.
The message will be delivered to one recipient with a matching path, if any such exists in the registry. If several
entries match the path because it has been registered on several nodes the message will be sent via the supplied
RoutingLogic (default random) to one destination. The sender() of the message can specify that local affinity
is preferred, i.e. the message is sent to an actor in the same local actor system as the used mediator actor, if any
such exists, otherwise route to any other matching entry.
You register actors to the local mediator with DistributedPubSubMediator.Put. The ActorRef in
Put must belong to the same local actor system as the mediator. The path without address information is the key
to which you send messages. On each node there can only be one actor for a given path, since the path is unique
within one local actor system.
You send messages by sending DistributedPubSubMediator.Send message to the local mediator with
the path (without address information) of the destination actors.
Actors are automatically removed from the registry when they are terminated, or you can explicitly remove entries
with DistributedPubSubMediator.Remove.
An example of a destination actor:
class Destination extends Actor with ActorLogging {
import DistributedPubSubMediator.Put
val mediator = DistributedPubSub(context.system).mediator
// register to the path
mediator ! Put(self)
def receive = {
6.4. Distributed Publish Subscribe in Cluster
305
Akka Scala Documentation, Release 2.5.0-RC1
case s: String ⇒
log.info("Got {}", s)
}
}
Destination actors can be started on several nodes in the cluster, and all will receive messages sent to the path
(without address information).
runOn(first) {
system.actorOf(Props[Destination], "destination")
}
runOn(second) {
system.actorOf(Props[Destination], "destination")
}
A simple actor that sends to the path:
class Sender extends Actor {
import DistributedPubSubMediator.Send
// activate the extension
val mediator = DistributedPubSub(context.system).mediator
def receive = {
case in: String ⇒
val out = in.toUpperCase
mediator ! Send(path = "/user/destination", msg = out, localAffinity = true)
}
}
It can send messages to the path from anywhere in the cluster:
runOn(third) {
val sender = system.actorOf(Props[Sender], "sender")
later()
// after a while the destinations are replicated
sender ! "hello"
}
It is also possible to broadcast messages to the actors that have been registered with Put.
Send
DistributedPubSubMediator.SendToAll message to the local mediator and the wrapped message will
then be delivered to all recipients with a matching path. Actors with the same path, without address information,
can be registered on different nodes. On each node there can only be one such actor, since the path is unique
within one local actor system.
Typical usage of this mode is to broadcast messages to all replicas with the same path, e.g. 3 actors on different nodes that all perform the same actions, for redundancy. You can also optionally specify a property
(allButSelf) deciding if the message should be sent to a matching path on the self node or not.
6.4.3 DistributedPubSub Extension
In the example above the mediator is started and accessed with the akka.cluster.pubsub.
DistributedPubSub extension. That is convenient and perfectly fine in most cases, but it can be good to
know that it is possible to start the mediator actor as an ordinary actor and you can have several different mediators at the same time to be able to divide a large number of actors/topics to different mediators. For example you
might want to use different cluster roles for different mediators.
The DistributedPubSub extension can be configured with the following properties:
# Settings for the DistributedPubSub extension
akka.cluster.pub-sub {
# Actor name of the mediator actor, /system/distributedPubSubMediator
6.4. Distributed Publish Subscribe in Cluster
306
Akka Scala Documentation, Release 2.5.0-RC1
name = distributedPubSubMediator
# Start the mediator on members tagged with this role.
# All members are used if undefined or empty.
role = ""
# The routing logic to use for 'Send'
# Possible values: random, round-robin, broadcast
routing-logic = random
# How often the DistributedPubSubMediator should send out gossip information
gossip-interval = 1s
# Removed entries are pruned after this duration
removed-time-to-live = 120s
˓→
# Maximum number of elements to transfer in one message when synchronizing the
registries.
# Next chunk will be transferred in next round of gossip.
max-delta-elements = 3000
# The id of the dispatcher to use for DistributedPubSubMediator actors.
# If not specified default dispatcher is used.
# If specified you need to define the settings of the actual dispatcher.
use-dispatcher = ""
}
It is recommended to load the extension when the actor system is started by defining it in akka.extensions
configuration property. Otherwise it will be activated when first used and then it takes a while for it to be populated.
akka.extensions = ["akka.cluster.pubsub.DistributedPubSub"]
6.4.4 Delivery Guarantee
As in Message Delivery Reliability of Akka, message delivery guarantee in distributed pub sub modes is at-mostonce delivery. In other words, messages can be lost over the wire.
If you are looking for at-least-once delivery guarantee, we recommend Kafka Akka Streams integration.
6.4.5 Dependencies
To use Distributed Publish Subscribe you must add the following dependency in your project.
sbt:
"com.typesafe.akka" %% "akka-cluster-tools" % "2.5.0-RC1"
maven:
<dependency>
<groupId>com.typesafe.akka</groupId>
<artifactId>akka-cluster-tools_2.11</artifactId>
<version>2.5.0-RC1</version>
</dependency>
6.4. Distributed Publish Subscribe in Cluster
307
Akka Scala Documentation, Release 2.5.0-RC1
6.5 Cluster Client
An actor system that is not part of the cluster can communicate with actors somewhere in the cluster via this
ClusterClient. The client can of course be part of another cluster. It only needs to know the location of one
(or more) nodes to use as initial contact points. It will establish a connection to a ClusterReceptionist
somewhere in the cluster. It will monitor the connection to the receptionist and establish a new connection if
the link goes down. When looking for a new receptionist it uses fresh contact points retrieved from previous
establishment, or periodically refreshed contacts, i.e. not necessarily the initial contact points.
Note: ClusterClient should not be used when sending messages to actors that run within the same cluster. Similar functionality as the ClusterClient is provided in a more efficient way by Distributed Publish
Subscribe in Cluster for actors that belong to the same cluster.
Also, note it’s necessary to change akka.actor.provider from local to remote or cluster when
using the cluster client.
The receptionist is supposed to be started on all nodes, or all nodes with specified role, in the cluster. The
receptionist can be started with the ClusterClientReceptionist extension or as an ordinary actor.
You can send messages via the ClusterClient to any actor in the cluster that is registered in the DistributedPubSubMediator used by the ClusterReceptionist.
The
ClusterClientReceptionist provides methods for registration of actors that should be reachable
from the client. Messages are wrapped in ClusterClient.Send, ClusterClient.SendToAll or
ClusterClient.Publish.
Both the ClusterClient and the ClusterClientReceptionist emit events that can be subscribed to.
The ClusterClient sends out notifications in relation to having received a list of contact points from the
ClusterClientReceptionist. One use of this list might be for the client to record its contact points. A
client that is restarted could then use this information to supersede any previously configured contact points.
The ClusterClientReceptionist sends out notifications in relation to having received contact from a
ClusterClient. This notification enables the server containing the receptionist to become aware of what
clients are connected.
1. ClusterClient.Send
The message will be delivered to one recipient with a matching path, if any such exists. If several entries match
the path the message will be delivered to one random destination. The sender() of the message can specify that
local affinity is preferred, i.e. the message is sent to an actor in the same local actor system as the used receptionist
actor, if any such exists, otherwise random to any other matching entry.
2. ClusterClient.SendToAll
The message will be delivered to all recipients with a matching path.
3. ClusterClient.Publish
The message will be delivered to all recipients Actors that have been registered as subscribers to the named topic.
Response messages from the destination actor are tunneled via the receptionist to avoid inbound connections from
other cluster nodes to the client, i.e. the sender(), as seen by the destination actor, is not the client itself. The
sender() of the response messages, as seen by the client, is deadLetters since the client should normally
send subsequent messages via the ClusterClient. It is possible to pass the original sender inside the reply
messages if the client is supposed to communicate directly to the actor in the cluster.
While establishing a connection to a receptionist the ClusterClient will buffer messages and send them
when the connection is established. If the buffer is full the ClusterClient will drop old messages when new
messages are sent via the client. The size of the buffer is configurable and it can be disabled by using a buffer size
of 0.
It’s worth noting that messages can always be lost because of the distributed nature of these actors. As always,
additional logic should be implemented in the destination (acknowledgement) and in the client (retry) actors to
ensure at-least-once message delivery.
6.5. Cluster Client
308
Akka Scala Documentation, Release 2.5.0-RC1
6.5.1 An Example
On the cluster nodes first start the receptionist. Note, it is recommended to load the extension when the actor
system is started by defining it in the akka.extensions configuration property:
akka.extensions = ["akka.cluster.client.ClusterClientReceptionist"]
Next, register the actors that should be available for the client.
runOn(host1) {
val serviceA = system.actorOf(Props[Service], "serviceA")
ClusterClientReceptionist(system).registerService(serviceA)
}
runOn(host2, host3) {
val serviceB = system.actorOf(Props[Service], "serviceB")
ClusterClientReceptionist(system).registerService(serviceB)
}
On the client you create the ClusterClient actor and use it as a gateway for sending messages to the actors
identified by their path (without address information) somewhere in the cluster.
runOn(client) {
val c = system.actorOf(ClusterClient.props(
ClusterClientSettings(system).withInitialContacts(initialContacts)), "client")
c ! ClusterClient.Send("/user/serviceA", "hello", localAffinity = true)
c ! ClusterClient.SendToAll("/user/serviceB", "hi")
}
The initialContacts parameter is a Set[ActorPath], which can be created like this:
val initialContacts = Set(
ActorPath.fromString("akka.tcp://OtherSys@host1:2552/system/receptionist"),
ActorPath.fromString("akka.tcp://OtherSys@host2:2552/system/receptionist"))
val settings = ClusterClientSettings(system)
.withInitialContacts(initialContacts)
You will probably define the address information of the initial contact points in configuration or system property.
See also Configuration.
A more comprehensive sample is available in the Lightbend Activator tutorial named Distributed workers with
Akka and Scala!.
6.5.2 ClusterClientReceptionist Extension
In the example above the receptionist is started and accessed with the akka.cluster.client.
ClusterClientReceptionist extension. That is convenient and perfectly fine in most cases, but it can
be good to know that it is possible to start the akka.cluster.client.ClusterReceptionist actor
as an ordinary actor and you can have several different receptionists at the same time, serving different types of
clients.
Note that the ClusterClientReceptionist uses the DistributedPubSub extension, which is described in Distributed Publish Subscribe in Cluster.
It is recommended to load the extension when the actor system is started by defining it in the akka.extensions
configuration property:
akka.extensions = ["akka.cluster.client.ClusterClientReceptionist"]
6.5. Cluster Client
309
Akka Scala Documentation, Release 2.5.0-RC1
6.5.3 Events
As mentioned earlier, both the ClusterClient and ClusterClientReceptionist emit events that can
be subscribed to. The following code snippet declares an actor that will receive notifications on contact points
(addresses to the available receptionists), as they become available. The code illustrates subscribing to the events
and receiving the ClusterClient initial state.
class ClientListener(targetClient: ActorRef) extends Actor {
override def preStart(): Unit =
targetClient ! SubscribeContactPoints
def receive: Receive =
receiveWithContactPoints(Set.empty)
def receiveWithContactPoints(contactPoints: Set[ActorPath]): Receive = {
case ContactPoints(cps) ⇒
context.become(receiveWithContactPoints(cps))
// Now do something with the up-to-date "cps"
case ContactPointAdded(cp) ⇒
context.become(receiveWithContactPoints(contactPoints + cp))
// Now do something with an up-to-date "contactPoints + cp"
case ContactPointRemoved(cp) ⇒
context.become(receiveWithContactPoints(contactPoints - cp))
// Now do something with an up-to-date "contactPoints - cp"
}
}
Similarly we can have an actor that behaves in a similar fashion for learning what cluster clients contact a
ClusterClientReceptionist:
class ReceptionistListener(targetReceptionist: ActorRef) extends Actor {
override def preStart(): Unit =
targetReceptionist ! SubscribeClusterClients
def receive: Receive =
receiveWithClusterClients(Set.empty)
def receiveWithClusterClients(clusterClients: Set[ActorRef]): Receive = {
case ClusterClients(cs) ⇒
context.become(receiveWithClusterClients(cs))
// Now do something with the up-to-date "c"
case ClusterClientUp(c) ⇒
context.become(receiveWithClusterClients(clusterClients + c))
// Now do something with an up-to-date "clusterClients + c"
case ClusterClientUnreachable(c) ⇒
context.become(receiveWithClusterClients(clusterClients - c))
// Now do something with an up-to-date "clusterClients - c"
}
}
6.5.4 Dependencies
To use the Cluster Client you must add the following dependency in your project.
sbt:
"com.typesafe.akka" %% "akka-cluster-tools" % "2.5.0-RC1"
maven:
6.5. Cluster Client
310
Akka Scala Documentation, Release 2.5.0-RC1
<dependency>
<groupId>com.typesafe.akka</groupId>
<artifactId>akka-cluster-tools_2.11</artifactId>
<version>2.5.0-RC1</version>
</dependency>
6.5.5 Configuration
The ClusterClientReceptionist extension (or ClusterReceptionistSettings) can be configured with the following properties:
# Settings for the ClusterClientReceptionist extension
akka.cluster.client.receptionist {
# Actor name of the ClusterReceptionist actor, /system/receptionist
name = receptionist
# Start the receptionist on members tagged with this role.
# All members are used if undefined or empty.
role = ""
# The receptionist will send this number of contact points to the client
number-of-contacts = 3
# The actor that tunnel response messages to the client will be stopped
# after this time of inactivity.
response-tunnel-receive-timeout = 30s
# The id of the dispatcher to use for ClusterReceptionist actors.
# If not specified default dispatcher is used.
# If specified you need to define the settings of the actual dispatcher.
use-dispatcher = ""
# How often failure detection heartbeat messages should be received for
# each ClusterClient
heartbeat-interval = 2s
# Number of potentially lost/delayed heartbeats that will be
# accepted before considering it to be an anomaly.
# The ClusterReceptionist is using the akka.remote.DeadlineFailureDetector, which
# will trigger if there are no heartbeats within the duration
# heartbeat-interval + acceptable-heartbeat-pause, i.e. 15 seconds with
# the default settings.
acceptable-heartbeat-pause = 13s
# Failure detection checking interval for checking all ClusterClients
failure-detection-interval = 2s
}
The following configuration properties are read by the ClusterClientSettings when created with a
ActorSystem parameter. It is also possible to amend the ClusterClientSettings or create it from
another config section with the same layout as below. ClusterClientSettings is a parameter to the
ClusterClient.props factory method, i.e. each client can be configured with different settings if needed.
# Settings for the ClusterClient
akka.cluster.client {
# Actor paths of the ClusterReceptionist actors on the servers (cluster nodes)
# that the client will try to contact initially. It is mandatory to specify
# at least one initial contact.
# Comma separated full actor paths defined by a string on the form of
# "akka.tcp://system@hostname:port/system/receptionist"
6.5. Cluster Client
311
Akka Scala Documentation, Release 2.5.0-RC1
initial-contacts = []
# Interval at which the client retries to establish contact with one of
# ClusterReceptionist on the servers (cluster nodes)
establishing-get-contacts-interval = 3s
# Interval at which the client will ask the ClusterReceptionist for
# new contact points to be used for next reconnect.
refresh-contacts-interval = 60s
# How often failure detection heartbeat messages should be sent
heartbeat-interval = 2s
# Number of potentially lost/delayed heartbeats that will be
# accepted before considering it to be an anomaly.
# The ClusterClient is using the akka.remote.DeadlineFailureDetector, which
# will trigger if there are no heartbeats within the duration
# heartbeat-interval + acceptable-heartbeat-pause, i.e. 15 seconds with
# the default settings.
acceptable-heartbeat-pause = 13s
# If connection to the receptionist is not established the client will buffer
# this number of messages and deliver them the connection is established.
# When the buffer is full old messages will be dropped when new messages are sent
# via the client. Use 0 to disable buffering, i.e. messages will be dropped
# immediately if the location of the singleton is unknown.
# Maximum allowed buffer size is 10000.
buffer-size = 1000
# If connection to the receiptionist is lost and the client has not been
# able to acquire a new connection for this long the client will stop itself.
# This duration makes it possible to watch the cluster client and react on a
˓→more permanent
# loss of connection with the cluster, for example by accessing some kind of
# service registry for an updated set of initial contacts to start a new cluster
˓→client with.
# If this is not wanted it can be set to "off" to disable the timeout and retry
# forever.
reconnect-timeout = off
}
6.5.6 Failure handling
When the cluster client is started it must be provided with a list of initial contacts which are cluster nodes where receptionists are running. It will then repeatedly (with an interval configurable by
establishing-get-contacts-interval) try to contact those until it gets in contact with one of them.
While running, the list of contacts are continuously updated with data from the receptionists (again, with an interval configurable with refresh-contacts-interval), so that if there are more receptionists in the cluster
than the initial contacts provided to the client the client will learn about them.
While the client is running it will detect failures in its connection to the receptionist by heartbeats if more than a
configurable amount of heartbeats are missed the client will try to reconnect to its known set of contacts to find a
receptionist it can access.
6.5.7 When the cluster cannot be reached at all
It is possible to make the cluster client stop entirely if it cannot find a receptionist it can talk to within a configurable
interval. This is configured with the reconnect-timeout, which defaults to off. This can be useful when
initial contacts are provided from some kind of service registry, cluster node addresses are entirely dynamic and
6.5. Cluster Client
312
Akka Scala Documentation, Release 2.5.0-RC1
the entire cluster might shut down or crash, be restarted on new addresses. Since the client will be stopped in that
case a monitoring actor can watch it and upon Terminate a new set of initial contacts can be fetched and a new
cluster client started.
6.6 Cluster Sharding
Cluster sharding is useful when you need to distribute actors across several nodes in the cluster and want to be
able to interact with them using their logical identifier, but without having to care about their physical location in
the cluster, which might also change over time.
It could for example be actors representing Aggregate Roots in Domain-Driven Design terminology. Here we
call these actors “entities”. These actors typically have persistent (durable) state, but this feature is not limited to
actors with persistent state.
Cluster sharding is typically used when you have many stateful actors that together consume more resources (e.g.
memory) than fit on one machine. If you only have a few stateful actors it might be easier to run them on a Cluster
Singleton node.
In this context sharding means that actors with an identifier, so called entities, can be automatically distributed
across multiple nodes in the cluster. Each entity actor runs only at one place, and messages can be sent to the
entity without requiring the sender to know the location of the destination actor. This is achieved by sending the
messages via a ShardRegion actor provided by this extension, which knows how to route the message with the
entity id to the final destination.
Cluster sharding will not be active on members with status WeaklyUp if that feature is enabled.
Warning: Don’t use Cluster Sharding together with Automatic Downing, since it allows the cluster to
split up into two separate clusters, which in turn will result in multiple shards and entities being started, one in
each separate cluster! See automatic-vs-manual-downing-java.
6.6.1 An Example
This is how an entity actor may look like:
case object Increment
case object Decrement
final case class Get(counterId: Long)
final case class EntityEnvelope(id: Long, payload: Any)
case object Stop
final case class CounterChanged(delta: Int)
class Counter extends PersistentActor {
import ShardRegion.Passivate
context.setReceiveTimeout(120.seconds)
// self.path.name is the entity identifier (utf-8 URL-encoded)
override def persistenceId: String = "Counter-" + self.path.name
var count = 0
def updateState(event: CounterChanged): Unit =
count += event.delta
override def receiveRecover: Receive = {
case evt: CounterChanged ⇒ updateState(evt)
}
6.6. Cluster Sharding
313
Akka Scala Documentation, Release 2.5.0-RC1
override def receiveCommand: Receive = {
case Increment
⇒ persist(CounterChanged(+1))(updateState)
case Decrement
⇒ persist(CounterChanged(-1))(updateState)
case Get(_)
⇒ sender() ! count
case ReceiveTimeout ⇒ context.parent ! Passivate(stopMessage = Stop)
case Stop
⇒ context.stop(self)
}
}
The above actor uses event sourcing and the support provided in PersistentActor to store its state. It does
not have to be a persistent actor, but in case of failure or migration of entities between nodes it must be able to
recover its state if it is valuable.
Note how the persistenceId is defined. The name of the actor is the entity identifier (utf-8 URL-encoded).
You may define it another way, but it must be unique.
When using the sharding extension you are first, typically at system startup on each node in the cluster, supposed
to register the supported entity types with the ClusterSharding.start method. ClusterSharding.
start gives you the reference which you can pass along.
val counterRegion: ActorRef = ClusterSharding(system).start(
typeName = "Counter",
entityProps = Props[Counter],
settings = ClusterShardingSettings(system),
extractEntityId = extractEntityId,
extractShardId = extractShardId)
The extractEntityId and extractShardId are two application specific functions to extract the entity
identifier and the shard identifier from incoming messages.
val extractEntityId: ShardRegion.ExtractEntityId = {
case EntityEnvelope(id, payload) ⇒ (id.toString, payload)
case msg @ Get(id)
⇒ (id.toString, msg)
}
val numberOfShards = 100
val extractShardId: ShardRegion.ExtractShardId = {
case EntityEnvelope(id, _) ⇒ (id % numberOfShards).toString
case Get(id)
⇒ (id % numberOfShards).toString
}
This example illustrates two different ways to define the entity identifier in the messages:
• The Get message includes the identifier itself.
• The EntityEnvelope holds the identifier, and the actual message that is sent to the entity actor is
wrapped in the envelope.
Note how these two messages types are handled in the extractEntityId function shown above. The message
sent to the entity actor is the second part of the tuple return by the extractEntityId and that makes it possible
to unwrap envelopes if needed.
A shard is a group of entities that will be managed together. The grouping is defined by the extractShardId
function shown above. For a specific entity identifier the shard identifier must always be the same.
Creating a good sharding algorithm is an interesting challenge in itself. Try to produce a uniform distribution, i.e.
same amount of entities in each shard. As a rule of thumb, the number of shards should be a factor ten greater than
the planned maximum number of cluster nodes. Less shards than number of nodes will result in that some nodes
will not host any shards. Too many shards will result in less efficient management of the shards, e.g. rebalancing
overhead, and increased latency because the coordinator is involved in the routing of the first message for each
shard. The sharding algorithm must be the same on all nodes in a running cluster. It can be changed after stopping
all nodes in the cluster.
6.6. Cluster Sharding
314
Akka Scala Documentation, Release 2.5.0-RC1
A simple sharding algorithm that works fine in most cases is to take the absolute value of the hashCode
of the entity identifier modulo number of shards. As a convenience this is provided by the ShardRegion.
HashCodeMessageExtractor.
Messages to the entities are always sent via the local ShardRegion. The ShardRegion actor reference for a named entity type is returned by ClusterSharding.start and it can also be retrieved with
ClusterSharding.shardRegion. The ShardRegion will lookup the location of the shard for the entity
if it does not already know its location. It will delegate the message to the right node and it will create the entity
actor on demand, i.e. when the first message for a specific entity is delivered.
val counterRegion: ActorRef = ClusterSharding(system).shardRegion("Counter")
counterRegion ! Get(123)
expectMsg(0)
counterRegion ! EntityEnvelope(123, Increment)
counterRegion ! Get(123)
expectMsg(1)
A more comprehensive sample is available in the Lightbend Activator tutorial named Akka Cluster Sharding with
Scala!.
6.6.2 How it works
The ShardRegion actor is started on each node in the cluster, or group of nodes tagged with a specific role.
The ShardRegion is created with two application specific functions to extract the entity identifier and the shard
identifier from incoming messages. A shard is a group of entities that will be managed together. For the first
message in a specific shard the ShardRegion request the location of the shard from a central coordinator, the
ShardCoordinator.
The ShardCoordinator decides which ShardRegion shall own the Shard and informs that
ShardRegion. The region will confirm this request and create the Shard supervisor as a child actor. The
individual Entities will then be created when needed by the Shard actor. Incoming messages thus travel via
the ShardRegion and the Shard to the target Entity.
If the shard home is another ShardRegion instance messages will be forwarded to that ShardRegion instance instead. While resolving the location of a shard incoming messages for that shard are buffered and later
delivered when the shard home is known. Subsequent messages to the resolved shard can be delivered to the target
destination immediately without involving the ShardCoordinator.
Scenario 1:
1. Incoming message M1 to ShardRegion instance R1.
2. M1 is mapped to shard S1. R1 doesn’t know about S1, so it asks the coordinator C for the location of S1.
3. C answers that the home of S1 is R1.
4. R1 creates child actor for the entity E1 and sends buffered messages for S1 to E1 child
5. All incoming messages for S1 which arrive at R1 can be handled by R1 without C. It creates entity children
as needed, and forwards messages to them.
Scenario 2:
1. Incoming message M2 to R1.
2. M2 is mapped to S2. R1 doesn’t know about S2, so it asks C for the location of S2.
3. C answers that the home of S2 is R2.
4. R1 sends buffered messages for S2 to R2
5. All incoming messages for S2 which arrive at R1 can be handled by R1 without C. It forwards messages to
R2.
6.6. Cluster Sharding
315
Akka Scala Documentation, Release 2.5.0-RC1
6. R2 receives message for S2, ask C, which answers that the home of S2 is R2, and we are in Scenario 1 (but
for R2).
To make sure that at most one instance of a specific entity actor is running somewhere in the cluster it is important
that all nodes have the same view of where the shards are located. Therefore the shard allocation decisions are
taken by the central ShardCoordinator, which is running as a cluster singleton, i.e. one instance on the oldest
member among all cluster nodes or a group of nodes tagged with a specific role.
The logic that decides where a shard is to be located is defined in a pluggable shard allocation strategy. The
default implementation ShardCoordinator.LeastShardAllocationStrategy allocates new shards
to the ShardRegion with least number of previously allocated shards. This strategy can be replaced by an
application specific implementation.
To be able to use newly added members in the cluster the coordinator facilitates rebalancing of shards, i.e. migrate
entities from one node to another. In the rebalance process the coordinator first notifies all ShardRegion actors
that a handoff for a shard has started. That means they will start buffering incoming messages for that shard, in
the same way as if the shard location is unknown. During the rebalance process the coordinator will not answer
any requests for the location of shards that are being rebalanced, i.e. local buffering will continue until the handoff
is completed. The ShardRegion responsible for the rebalanced shard will stop all entities in that shard by
sending the specified handOffStopMessage (default PoisonPill) to them. When all entities have been
terminated the ShardRegion owning the entities will acknowledge the handoff as completed to the coordinator.
Thereafter the coordinator will reply to requests for the location of the shard and thereby allocate a new home for
the shard and then buffered messages in the ShardRegion actors are delivered to the new location. This means
that the state of the entities are not transferred or migrated. If the state of the entities are of importance it should
be persistent (durable), e.g. with Persistence, so that it can be recovered at the new location.
The logic that decides which shards to rebalance is defined in a pluggable shard allocation strategy. The default implementation ShardCoordinator.LeastShardAllocationStrategy picks shards for handoff from the ShardRegion with most number of previously allocated shards. They will then be allocated to the
ShardRegion with least number of previously allocated shards, i.e. new members in the cluster. There is a
configurable threshold of how large the difference must be to begin the rebalancing. This strategy can be replaced
by an application specific implementation.
The state of shard locations in the ShardCoordinator is persistent (durable) with Distributed Data or Persistence to survive failures. When a crashed or unreachable coordinator node has been removed (via down) from the
cluster a new ShardCoordinator singleton actor will take over and the state is recovered. During such a failure period shards with known location are still available, while messages for new (unknown) shards are buffered
until the new ShardCoordinator becomes available.
As long as a sender uses the same ShardRegion actor to deliver messages to an entity actor the order of the
messages is preserved. As long as the buffer limit is not reached messages are delivered on a best effort basis, with
at-most once delivery semantics, in the same way as ordinary message sending. Reliable end-to-end messaging,
with at-least-once semantics can be added by using AtLeastOnceDelivery in Persistence.
Some additional latency is introduced for messages targeted to new or previously unused shards due to the roundtrip to the coordinator. Rebalancing of shards may also add latency. This should be considered when designing
the application specific shard resolution, e.g. to avoid too fine grained shards.
6.6.3 Distributed Data vs. Persistence Mode
The state of the coordinator and the state of Remembering Entities of the shards are persistent (durable) to survive
failures. Distributed Data or Persistence can be used for the storage. Distributed Data is used by default.
The functionality when using the two modes is the same. If your sharded entities are not using Akka Persistence
themselves it is more convenient to use the Distributed Data mode, since then you don’t have to setup and operate
a separate data store (e.g. Cassandra) for persistence. Aside from that, there are no major reasons for using one
mode over the the other.
It’s important to use the same mode on all nodes in the cluster, i.e. it’s not possible to perform a rolling upgrade
to change this setting.
6.6. Cluster Sharding
316
Akka Scala Documentation, Release 2.5.0-RC1
Distributed Data Mode
This mode is enabled with configuration (enabled by default):
akka.cluster.sharding.state-store-mode = ddata
The state of the ShardCoordinator will be replicated inside a cluster by the Distributed Data module with
WriteMajority/ReadMajority consistency. The state of the coordinator is not durable, it’s not stored to
disk. When all nodes in the cluster have been stopped the state is lost and not needed any more.
The state of Remembering Entities is also durable, i.e. it is stored to disk. The stored entities are started also after
a complete cluster restart.
Cluster Sharding is using its own Distributed Data Replicator per node role. In this way you can use a subset
of all nodes for some entity types and another subset for other entity types. Each such replicator has a name that
contains the node role and therefore the role configuration must be the same on all nodes in the cluster, i.e. you
can’t change the roles when performing a rolling upgrade.
The settings for Distributed Data is configured in the the section akka.cluster.sharding.
distributed-data. It’s not possible to have different distributed-data settings for different sharding
entity types.
Persistence Mode
This mode is enabled with configuration:
akka.cluster.sharding.state-store-mode = persistence
Since it is running in a cluster Persistence must be configured with a distributed journal.
6.6.4 Startup after minimum number of members
It’s good to use Cluster Sharding with the Cluster setting akka.cluster.min-nr-of-members or akka.
cluster.role.<role-name>.min-nr-of-members. That will defer the allocation of the shards until
at least that number of regions have been started and registered to the coordinator. This avoids that many shards
are allocated to the first region that registers and only later are rebalanced to other nodes.
See How To Startup when Cluster Size Reached for more information about min-nr-of-members.
6.6.5 Proxy Only Mode
The ShardRegion actor can also be started in proxy only mode, i.e. it will not host any entities itself, but knows
how to delegate messages to the right location. A ShardRegion is started in proxy only mode with the method
ClusterSharding.startProxy method.
6.6.6 Passivation
If the state of the entities are persistent you may stop entities that are not used to reduce memory consumption.
This is done by the application specific implementation of the entity actors for example by defining receive timeout
(context.setReceiveTimeout). If a message is already enqueued to the entity when it stops itself the
enqueued message in the mailbox will be dropped. To support graceful passivation without losing such messages
the entity actor can send ShardRegion.Passivate to its parent Shard. The specified wrapped message in
Passivate will be sent back to the entity, which is then supposed to stop itself. Incoming messages will be
buffered by the Shard between reception of Passivate and termination of the entity. Such buffered messages
are thereafter delivered to a new incarnation of the entity.
6.6. Cluster Sharding
317
Akka Scala Documentation, Release 2.5.0-RC1
6.6.7 Remembering Entities
The list of entities in each Shard can be made persistent (durable) by setting the rememberEntities flag
to true in ClusterShardingSettings when calling ClusterSharding.start. When configured to
remember entities, whenever a Shard is rebalanced onto another node or recovers after a crash it will recreate all
the entities which were previously running in that Shard. To permanently stop entities, a Passivate message
must be sent to the parent of the entity actor, otherwise the entity will be automatically restarted after the entity
restart backoff specified in the configuration.
When Distributed Data mode is used the identifiers of the entities are stored in Durable Storage of Distributed
Data. You may want to change the configuration of the akka.cluster.sharding.distributed-data.durable.lmdb.dir‘,
since the default directory contains the remote port of the actor system. If using a dynamically assigned port (0) it
will be different each time and the previously stored data will not be loaded.
When rememberEntities is set to false, a Shard will not automatically restart any entities after a rebalance
or recovering from a crash. Entities will only be started once the first message for that entity has been received in
the Shard. Entities will not be restarted if they stop without using a Passivate.
Note that the state of the entities themselves will not be restored unless they have been made persistent, e.g. with
Persistence.
The performance cost of rememberEntities is rather high when starting/stopping entities and when shards
are rebalanced. This cost increases with number of entities per shard and we currently don’t recommend using it
with more than 10000 entities per shard.
6.6.8 Supervision
If you need to use another supervisorStrategy for the entity actors than the default (restarting) strategy you
need to create an intermediate parent actor that defines the supervisorStrategy to the child entity actor.
class CounterSupervisor extends Actor {
val counter = context.actorOf(Props[Counter], "theCounter")
override val supervisorStrategy = OneForOneStrategy() {
case _: IllegalArgumentException
⇒ SupervisorStrategy.Resume
case _: ActorInitializationException ⇒ SupervisorStrategy.Stop
case _: DeathPactException
⇒ SupervisorStrategy.Stop
case _: Exception
⇒ SupervisorStrategy.Restart
}
def receive = {
case msg ⇒ counter forward msg
}
}
You start such a supervisor in the same way as if it was the entity actor.
ClusterSharding(system).start(
typeName = "SupervisedCounter",
entityProps = Props[CounterSupervisor],
settings = ClusterShardingSettings(system),
extractEntityId = extractEntityId,
extractShardId = extractShardId)
Note that stopped entities will be started again when a new message is targeted to the entity.
6.6.9 Graceful Shutdown
You can send the message ShardRegion.GracefulShutdown message to the ShardRegion actor to
handoff all shards that are hosted by that ShardRegion and then the ShardRegion actor will be stopped.
6.6. Cluster Sharding
318
Akka Scala Documentation, Release 2.5.0-RC1
You can watch the ShardRegion actor to know when it is completed. During this period other regions will
buffer messages for those shards in the same way as when a rebalance is triggered by the coordinator. When the
shards have been stopped the coordinator will allocate these shards elsewhere.
This is performed automatically by the Coordinated Shutdown and is therefore part of the graceful leaving process
of a cluster member.
6.6.10 Removal of Internal Cluster Sharding Data
The Cluster Sharding coordinator stores the locations of the shards using Akka Persistence. This data can safely
be removed when restarting the whole Akka Cluster. Note that this is not application data.
There is a utility program akka.cluster.sharding.RemoveInternalClusterShardingData that
removes this data.
Warning: Never use this program while there are running Akka Cluster nodes that are using Cluster Sharding.
Stop all Cluster nodes before using this program.
It can be needed to remove the data if the Cluster Sharding coordinator cannot startup because of corrupt data,
which may happen if accidentally two clusters were running at the same time, e.g. caused by using auto-down and
there was a network partition.
Warning: Don’t use Cluster Sharding together with Automatic Downing, since it allows the cluster to
split up into two separate clusters, which in turn will result in multiple shards and entities being started, one in
each separate cluster! See Downing.
Use this program as a standalone Java main program:
java -classpath <jar files, including akka-cluster-sharding>
akka.cluster.sharding.RemoveInternalClusterShardingData
-2.3 entityType1 entityType2 entityType3
The program is included in the akka-cluster-sharding jar file. It is easiest to run it with same classpath
and configuration as your ordinary application. It can be run from sbt or maven in similar way.
Specify the entity type names (same as you use in the start method of ClusterSharding) as program
arguments.
If you specify -2.3 as the first program argument it will also try to remove data that was stored by Cluster
Sharding in Akka 2.3.x using different persistenceId.
6.6.11 Dependencies
To use the Cluster Sharding you must add the following dependency in your project.
sbt:
"com.typesafe.akka" %% "akka-cluster-sharding" % "2.5.0-RC1"
maven:
<dependency>
<groupId>com.typesafe.akka</groupId>
<artifactId>akka-cluster-sharding_2.11</artifactId>
<version>2.5.0-RC1</version>
</dependency>
6.6. Cluster Sharding
319
Akka Scala Documentation, Release 2.5.0-RC1
6.6.12 Configuration
The ClusterSharding extension can be configured with the following properties. These configuration properties are read by the ClusterShardingSettings when created with a ActorSystem parameter. It is also
possible to amend the ClusterShardingSettings or create it from another config section with the same layout as below. ClusterShardingSettings is a parameter to the start method of the ClusterSharding
extension, i.e. each each entity type can be configured with different settings if needed.
# Settings for the ClusterShardingExtension
akka.cluster.sharding {
˓→
# The extension creates a top level actor with this name in top level system
scope,
# e.g. '/system/sharding'
guardian-name = sharding
# Specifies that entities runs on cluster nodes with a specific role.
# If the role is not specified (or empty) all nodes in the cluster are used.
role = ""
˓→
# When this is set to 'on' the active entity actors will automatically be
restarted
# upon Shard restart. i.e. if the Shard is started on a different ShardRegion
# due to rebalance or crash.
remember-entities = off
# If the coordinator can't store state changes it will be stopped
# and started again after this duration, with an exponential back-off
# of up to 5 times this duration.
coordinator-failure-backoff = 5 s
# The ShardRegion retries registration and shard location requests to the
# ShardCoordinator with this interval if it does not reply.
retry-interval = 2 s
# Maximum number of messages that are buffered by a ShardRegion actor.
buffer-size = 100000
# Timeout of the shard rebalancing process.
handoff-timeout = 60 s
# Time given to a region to acknowledge it's hosting a shard.
shard-start-timeout = 10 s
# If the shard is remembering entities and can't store state changes
# will be stopped and then started again after this duration. Any messages
# sent to an affected entity may be lost in this process.
shard-failure-backoff = 10 s
# If the shard is remembering entities and an entity stops itself without
# using passivate. The entity will be restarted after this duration or when
# the next message for it is received, which ever occurs first.
entity-restart-backoff = 10 s
# Rebalance check is performed periodically with this interval.
rebalance-interval = 10 s
# Absolute path to the journal plugin configuration entity that is to be
# used for the internal persistence of ClusterSharding. If not defined
# the default journal plugin is used. Note that this is not related to
# persistence used by the entity actors.
# Only used when state-store-mode=persistence
journal-plugin-id = ""
6.6. Cluster Sharding
320
Akka Scala Documentation, Release 2.5.0-RC1
# Absolute path to the snapshot plugin configuration entity that is to be
# used for the internal persistence of ClusterSharding. If not defined
# the default snapshot plugin is used. Note that this is not related to
# persistence used by the entity actors.
# Only used when state-store-mode=persistence
snapshot-plugin-id = ""
# Defines how the coordinator stores its state. Same is also used by the
# shards for rememberEntities.
# Valid values are "ddata" or "persistence".
state-store-mode = "ddata"
# The shard saves persistent snapshots after this number of persistent
# events. Snapshots are used to reduce recovery times.
# Only used when state-store-mode=persistence
snapshot-after = 1000
# The shard deletes persistent events (messages and snapshots) after doing
snapshot
# keeping this number of old persistent batches.
# Batch is of size `snapshot-after`.
# When set to 0 after snapshot is successfully done all messages with equal or
˓→lower sequence number will be deleted.
# Default value of 2 leaves last maximum 2*`snapshot-after` messages and 3
˓→snapshots (2 old ones + fresh snapshot)
keep-nr-of-batches = 2
˓→
# Setting for the default shard allocation strategy
least-shard-allocation-strategy {
# Threshold of how large the difference between most and least number of
# allocated shards must be to begin the rebalancing.
rebalance-threshold = 10
# The number of ongoing rebalancing processes is limited to this number.
max-simultaneous-rebalance = 3
}
˓→
˓→
# Timeout of waiting the initial distributed state (an initial state will be
queried again if the timeout happened)
# Only used when state-store-mode=ddata
waiting-for-state-timeout = 5 s
# Timeout of waiting for update the distributed state (update will be retried if
the timeout happened)
# Only used when state-store-mode=ddata
updating-state-timeout = 5 s
# The shard uses this strategy to determines how to recover the underlying
entity actors. The strategy is only used
# by the persistent shard when rebalancing or restarting. The value can either
˓→be "all" or "constant". The "all"
# strategy start all the underlying entity actors at the same time. The constant
˓→strategy will start the underlying
# entity actors at a fix rate. The default strategy "all".
entity-recovery-strategy = "all"
˓→
# Default settings for the constant rate entity recovery strategy
entity-recovery-constant-rate-strategy {
# Sets the frequency at which a batch of entity actors is started.
frequency = 100 ms
# Sets the number of entity actors to be restart at a particular interval
number-of-entities = 5
6.6. Cluster Sharding
321
Akka Scala Documentation, Release 2.5.0-RC1
}
# Settings for the coordinator singleton. Same layout as akka.cluster.singleton.
# The "role" of the singleton configuration is not used. The singleton role will
# be the same as "akka.cluster.sharding.role".
coordinator-singleton = ${akka.cluster.singleton}
# Settings for the Distributed Data replicator.
# Same layout as akka.cluster.distributed-data.
# The "role" of the distributed-data configuration is not used. The distributed˓→data
# role will be the same as "akka.cluster.sharding.role".
# Note that there is one Replicator per role and it's not possible
# to have different distributed-data settings for different sharding entity
˓→types.
# Only used when state-store-mode=ddata
distributed-data = ${akka.cluster.distributed-data}
distributed-data {
# minCap parameter to MajorityWrite and MajorityRead consistency level.
majority-min-cap = 5
durable.keys = ["shard-*"]
# When using many entities with "remember entities" the Gossip message
# can become to large if including to many in same message. Limit to
# the same number as the number of ORSet per shard.
akka.cluster.sharding.distributed-data.max-delta-elements = 5
}
# The id of the dispatcher to use for ClusterSharding actors.
# If not specified default dispatcher is used.
# If specified you need to define the settings of the actual dispatcher.
# This dispatcher for the entity actors is defined by the user provided
# Props, i.e. this dispatcher is not used for the entity actors.
use-dispatcher = ""
}
Custom shard allocation strategy can be defined in an optional parameter to ClusterSharding.start. See
the API documentation of ShardAllocationStrategy for details of how to implement a custom shard
allocation strategy.
6.6.13 Inspecting cluster sharding state
Two requests to inspect the cluster state are available:
ShardRegion.GetShardRegionState
which
will
return
a
ShardRegion.
CurrentShardRegionState that contains the identifiers of the shards running in a Region and what
entities are alive for each of them.
ShardRegion.GetClusterShardingStats which will query all the regions in the cluster and return a
ShardRegion.ClusterShardingStats containing the identifiers of the shards running in each region
and a count of entities that are alive in each shard.
The purpose of these messages is testing and monitoring, they are not provided to give access to directly sending
messages to the individual entities.
6.6. Cluster Sharding
322
Akka Scala Documentation, Release 2.5.0-RC1
6.7 Cluster Metrics Extension
6.7.1 Introduction
The member nodes of the cluster can collect system health metrics and publish that to other cluster nodes and to
the registered subscribers on the system event bus with the help of Cluster Metrics Extension.
Cluster metrics information is primarily used for load-balancing routers, and can also be used to implement advanced metrics-based node life cycles, such as “Node Let-it-crash” when CPU steal time becomes excessive.
Cluster Metrics Extension is a separate Akka module delivered in akka-cluster-metrics jar.
To enable usage of the extension you need to add the following dependency to your project:
"com.typesafe.akka" % "akka-cluster-metrics_2.11" % "2.5.0-RC1"
and add the following configuration stanza to your application.conf
akka.extensions = [ "akka.cluster.metrics.ClusterMetricsExtension" ]
Cluster members with status WeaklyUp, if that feature is enabled, will participate in Cluster Metrics collection and
dissemination.
6.7.2 Metrics Collector
Metrics collection is delegated to an implementation of akka.cluster.metrics.MetricsCollector.
Different collector implementations provide different subsets of metrics published to the cluster. Certain message
routing and let-it-crash functions may not work when Sigar is not provisioned.
Cluster metrics extension comes with two built-in collector implementations:
1. akka.cluster.metrics.SigarMetricsCollector, which requires Sigar provisioning, and is
more rich/precise
2. akka.cluster.metrics.JmxMetricsCollector, which is used as fall back, and is less
rich/precise
You can also plug-in your own metrics collector implementation.
By default, metrics extension will use collector provider fall back and will try to load them in this order:
1. configured user-provided collector
2. built-in akka.cluster.metrics.SigarMetricsCollector
3. and finally akka.cluster.metrics.JmxMetricsCollector
6.7.3 Metrics Events
Metrics extension periodically publishes current snapshot of the cluster metrics to the node system event bus.
The publication interval is controlled by the akka.cluster.metrics.collector.sample-interval
setting.
The payload of the akka.cluster.metrics.ClusterMetricsChanged event will contain latest metrics of the node as well as other cluster member nodes metrics gossip which was received during the collector
sample interval.
You can subscribe your metrics listener actors to these events in order to implement custom node lifecycle
ClusterMetricsExtension(system).subscribe(metricsListenerActor)
6.7. Cluster Metrics Extension
323
Akka Scala Documentation, Release 2.5.0-RC1
6.7.4 Hyperic Sigar Provisioning
Both user-provided and built-in metrics collectors can optionally use Hyperic Sigar for a wider and more accurate
range of metrics compared to what can be retrieved from ordinary JMX MBeans.
Sigar is using a native o/s library, and requires library provisioning, i.e. deployment, extraction and loading of the
o/s native library into JVM at runtime.
User can provision Sigar classes and native library in one of the following ways:
1. Use Kamon sigar-loader as a project dependency for the user project. Metrics extension will extract and
load sigar library on demand with help of Kamon sigar provisioner.
2. Use Kamon sigar-loader as java agent: java -javaagent:/path/to/sigar-loader.jar. Kamon sigar loader agent will extract and load sigar library during JVM start.
3. Place sigar.jar on the classpath and Sigar native library for the o/s on the java.library.path.
User is required to manage both project dependency and library deployment manually.
Warning: When using Kamon sigar-loader and running multiple instances of the same application on the
same host, you have to make sure that sigar library is extracted to a unique per instance directory. You can control the extract directory with the akka.cluster.metrics.native-library-extract-folder
configuration setting.
To enable usage of Sigar you can add the following dependency to the user project
"io.kamon" % "sigar-loader" % "1.6.6-rev002"
You can download Kamon sigar-loader from Maven Central
6.7.5 Adaptive Load Balancing
The AdaptiveLoadBalancingPool / AdaptiveLoadBalancingGroup performs load balancing of
messages to cluster nodes based on the cluster metrics data. It uses random selection of routees with probabilities derived from the remaining capacity of the corresponding node. It can be configured to use a specific
MetricsSelector to produce the probabilities, a.k.a. weights:
• heap / HeapMetricsSelector - Used and max JVM heap memory. Weights based on remaining heap
capacity; (max - used) / max
• load / SystemLoadAverageMetricsSelector - System load average for the past 1 minute, corresponding value can be found in top of Linux systems. The system is possibly nearing a bottleneck if the
system load average is nearing number of cpus/cores. Weights based on remaining load capacity; 1 - (load
/ processors)
• cpu / CpuMetricsSelector - CPU utilization in percentage, sum of User + Sys + Nice + Wait. Weights
based on remaining cpu capacity; 1 - utilization
• mix / MixMetricsSelector - Combines heap, cpu and load. Weights based on mean of remaining
capacity of the combined selectors.
• Any custom implementation of akka.cluster.metrics.MetricsSelector
The collected metrics values are smoothed with exponential weighted moving average. In the Configuration you
can adjust how quickly past data is decayed compared to new data.
Let’s take a look at this router in action. What can be more demanding than calculating factorials?
The backend worker that performs the factorial calculation:
class FactorialBackend extends Actor with ActorLogging {
import context.dispatcher
6.7. Cluster Metrics Extension
324
Akka Scala Documentation, Release 2.5.0-RC1
def receive = {
case (n: Int) =>
Future(factorial(n)) map { result => (n, result) } pipeTo sender()
}
def factorial(n: Int): BigInt = {
@tailrec def factorialAcc(acc: BigInt, n: Int): BigInt = {
if (n <= 1) acc
else factorialAcc(acc * n, n - 1)
}
factorialAcc(BigInt(1), n)
}
}
The frontend that receives user jobs and delegates to the backends via the router:
class FactorialFrontend(upToN: Int, repeat: Boolean) extends Actor with
˓→ActorLogging {
val backend = context.actorOf(
FromConfig.props(),
name = "factorialBackendRouter")
override def preStart(): Unit = {
sendJobs()
if (repeat) {
context.setReceiveTimeout(10.seconds)
}
}
def receive = {
case (n: Int, factorial: BigInt) =>
if (n == upToN) {
log.debug("{}! = {}", n, factorial)
if (repeat) sendJobs()
else context.stop(self)
}
case ReceiveTimeout =>
log.info("Timeout")
sendJobs()
}
def sendJobs(): Unit = {
log.info("Starting batch of factorials up to [{}]", upToN)
1 to upToN foreach { backend ! _ }
}
}
As you can see, the router is defined in the same way as other routers, and in this case it is configured as follows:
akka.actor.deployment {
/factorialFrontend/factorialBackendRouter = {
# Router type provided by metrics extension.
router = cluster-metrics-adaptive-group
# Router parameter specific for metrics extension.
# metrics-selector = heap
# metrics-selector = load
# metrics-selector = cpu
metrics-selector = mix
#
routees.paths = ["/user/factorialBackend"]
6.7. Cluster Metrics Extension
325
Akka Scala Documentation, Release 2.5.0-RC1
cluster {
enabled = on
use-role = backend
allow-local-routees = off
}
}
}
It is only router type and the metrics-selector parameter that is specific to this router, other things work
in the same way as other routers.
The same type of router could also have been defined in code:
import
import
import
import
akka.cluster.routing.ClusterRouterGroup
akka.cluster.routing.ClusterRouterGroupSettings
akka.cluster.metrics.AdaptiveLoadBalancingGroup
akka.cluster.metrics.HeapMetricsSelector
val backend = context.actorOf(
ClusterRouterGroup(
AdaptiveLoadBalancingGroup(HeapMetricsSelector),
ClusterRouterGroupSettings(
totalInstances = 100, routeesPaths = List("/user/factorialBackend"),
allowLocalRoutees = true, useRole = Some("backend"))).props(),
name = "factorialBackendRouter2")
import
import
import
import
akka.cluster.routing.ClusterRouterPool
akka.cluster.routing.ClusterRouterPoolSettings
akka.cluster.metrics.AdaptiveLoadBalancingPool
akka.cluster.metrics.SystemLoadAverageMetricsSelector
val backend = context.actorOf(
ClusterRouterPool(AdaptiveLoadBalancingPool(
SystemLoadAverageMetricsSelector), ClusterRouterPoolSettings(
totalInstances = 100, maxInstancesPerNode = 3,
allowLocalRoutees = false, useRole = Some("backend"))).
˓→props(Props[FactorialBackend]),
name = "factorialBackendRouter3")
The Lightbend Activator tutorial named Akka Cluster Samples with Scala. contains the full source code and
instructions of how to run the Adaptive Load Balancing sample.
6.7.6 Subscribe to Metrics Events
It is possible to subscribe to the metrics events directly to implement other functionality.
import
import
import
import
import
import
import
import
import
import
akka.actor.ActorLogging
akka.actor.Actor
akka.cluster.Cluster
akka.cluster.metrics.ClusterMetricsEvent
akka.cluster.metrics.ClusterMetricsChanged
akka.cluster.ClusterEvent.CurrentClusterState
akka.cluster.metrics.NodeMetrics
akka.cluster.metrics.StandardMetrics.HeapMemory
akka.cluster.metrics.StandardMetrics.Cpu
akka.cluster.metrics.ClusterMetricsExtension
class MetricsListener extends Actor with ActorLogging {
val selfAddress = Cluster(context.system).selfAddress
val extension = ClusterMetricsExtension(context.system)
6.7. Cluster Metrics Extension
326
Akka Scala Documentation, Release 2.5.0-RC1
// Subscribe unto ClusterMetricsEvent events.
override def preStart(): Unit = extension.subscribe(self)
// Unsubscribe from ClusterMetricsEvent events.
override def postStop(): Unit = extension.unsubscribe(self)
def receive = {
case ClusterMetricsChanged(clusterMetrics) =>
clusterMetrics.filter(_.address == selfAddress) foreach { nodeMetrics =>
logHeap(nodeMetrics)
logCpu(nodeMetrics)
}
case state: CurrentClusterState => // Ignore.
}
def logHeap(nodeMetrics: NodeMetrics): Unit = nodeMetrics match {
case HeapMemory(address, timestamp, used, committed, max) =>
log.info("Used heap: {} MB", used.doubleValue / 1024 / 1024)
case _ => // No heap info.
}
def logCpu(nodeMetrics: NodeMetrics): Unit = nodeMetrics match {
case Cpu(address, timestamp, Some(systemLoadAverage), cpuCombined, cpuStolen,
˓→processors) =>
log.info("Load: {} ({} processors)", systemLoadAverage, processors)
case _ => // No cpu info.
}
}
6.7.7 Custom Metrics Collector
Metrics collection is delegated to the implementation of akka.cluster.metrics.MetricsCollector
You can plug-in your own metrics collector instead of built-in akka.cluster.metrics.
SigarMetricsCollector or akka.cluster.metrics.JmxMetricsCollector.
Look at those two implementations for inspiration.
Custom metrics collector implementation class must be specified in the akka.cluster.metrics.
collector.provider configuration property.
6.7.8 Configuration
The Cluster metrics extension can be configured with the following properties:
##############################################
# Akka Cluster Metrics Reference Config File #
##############################################
# This is the reference config file that contains all the default settings.
# Make your edits in your application.conf in order to override these settings.
# Sigar provisioning:
#
# User can provision sigar classes and native library in one of the following
˓→ways:
#
# 1) Use https://github.com/kamon-io/sigar-loader Kamon sigar-loader as a project
˓→dependency for the user project.
# Metrics extension will extract and load sigar library on demand with help of
˓→Kamon sigar provisioner.
6.7. Cluster Metrics Extension
327
Akka Scala Documentation, Release 2.5.0-RC1
#
#
2) Use https://github.com/kamon-io/sigar-loader Kamon sigar-loader as java
agent: `java -javaagent:/path/to/sigar-loader.jar`
# Kamon sigar loader agent will extract and load sigar library during JVM start.
#
# 3) Place `sigar.jar` on the `classpath` and sigar native library for the o/s on
˓→the `java.library.path`
# User is required to manage both project dependency and library deployment
˓→manually.
˓→
# Cluster metrics extension.
# Provides periodic statistics collection and publication throughout the cluster.
akka.cluster.metrics {
# Full path of dispatcher configuration key.
# Use "" for default key `akka.actor.default-dispatcher`.
dispatcher = ""
# How long should any actor wait before starting the periodic tasks.
periodic-tasks-initial-delay = 1s
# Sigar native library extract location.
# Use per-application-instance scoped location, such as program working
˓→directory.
native-library-extract-folder = ${user.dir}"/native"
# Metrics supervisor actor.
supervisor {
# Actor name. Example name space: /system/cluster-metrics
name = "cluster-metrics"
# Supervision strategy.
strategy {
#
# FQCN of class providing `akka.actor.SupervisorStrategy`.
# Must have a constructor with signature `<init>(com.typesafe.config.
˓→Config)`.
# Default metrics strategy provider is a configurable extension of
˓→`OneForOneStrategy`.
provider = "akka.cluster.metrics.ClusterMetricsStrategy"
#
# Configuration of the default strategy provider.
# Replace with custom settings when overriding the provider.
configuration = {
# Log restart attempts.
loggingEnabled = true
# Child actor restart-on-failure window.
withinTimeRange = 3s
# Maximum number of restart attempts before child actor is stopped.
maxNrOfRetries = 3
}
}
}
# Metrics collector actor.
collector {
# Enable or disable metrics collector for load-balancing nodes.
# Metrics collection can also be controlled at runtime by sending control
˓→messages
# to /system/cluster-metrics actor: `akka.cluster.metrics.
˓→{CollectionStartMessage,CollectionStopMessage}`
enabled = on
# FQCN of the metrics collector implementation.
# It must implement `akka.cluster.metrics.MetricsCollector` and
# have public constructor with akka.actor.ActorSystem parameter.
# Will try to load in the following order of priority:
# 1) configured custom collector 2) internal `SigarMetricsCollector` 3)
˓→internal `JmxMetricsCollector`
provider = ""
6.7. Cluster Metrics Extension
328
Akka Scala Documentation, Release 2.5.0-RC1
# Try all 3 available collector providers, or else fail on the configured
custom collector provider.
fallback = true
# How often metrics are sampled on a node.
# Shorter interval will collect the metrics more often.
# Also controls frequency of the metrics publication to the node system event
˓→bus.
sample-interval = 3s
# How often a node publishes metrics information to the other nodes in the
˓→cluster.
# Shorter interval will publish the metrics gossip more often.
gossip-interval = 3s
# How quickly the exponential weighting of past data is decayed compared to
# new data. Set lower to increase the bias toward newer values.
# The relevance of each data sample is halved for every passing half-life
# duration, i.e. after 4 times the half-life, a data sample’s relevance is
# reduced to 6% of its original relevance. The initial relevance of a data
# sample is given by 1 - 0.5 ^ (collect-interval / half-life).
# See http://en.wikipedia.org/wiki/Moving_average#Exponential_moving_average
moving-average-half-life = 12s
}
}
˓→
# Cluster metrics extension serializers and routers.
akka.actor {
# Protobuf serializer for remote cluster metrics messages.
serializers {
akka-cluster-metrics = "akka.cluster.metrics.protobuf.MessageSerializer"
}
# Interface binding for remote cluster metrics messages.
serialization-bindings {
"akka.cluster.metrics.ClusterMetricsMessage" = akka-cluster-metrics
"akka.cluster.metrics.AdaptiveLoadBalancingPool" = akka-cluster-metrics
"akka.cluster.metrics.MixMetricsSelector" = akka-cluster-metrics
"akka.cluster.metrics.CpuMetricsSelector$" = akka-cluster-metrics
"akka.cluster.metrics.HeapMetricsSelector$" = akka-cluster-metrics
"akka.cluster.metrics.SystemLoadAverageMetricsSelector$" = akka-cluster-metrics
}
# Globally unique metrics extension serializer identifier.
serialization-identifiers {
"akka.cluster.metrics.protobuf.MessageSerializer" = 10
}
# Provide routing of messages based on cluster metrics.
router.type-mapping {
cluster-metrics-adaptive-pool = "akka.cluster.metrics.
˓→AdaptiveLoadBalancingPool"
cluster-metrics-adaptive-group = "akka.cluster.metrics.
˓→AdaptiveLoadBalancingGroup"
}
}
6.8 Distributed Data
Akka Distributed Data is useful when you need to share data between nodes in an Akka Cluster. The data is
accessed with an actor providing a key-value store like API. The keys are unique identifiers with type information
of the data values. The values are Conflict Free Replicated Data Types (CRDTs).
All data entries are spread to all nodes, or nodes with a certain role, in the cluster via direct replication and gossip
based dissemination. You have fine grained control of the consistency level for reads and writes.
The nature CRDTs makes it possible to perform updates from any node without coordination. Concurrent updates
6.8. Distributed Data
329
Akka Scala Documentation, Release 2.5.0-RC1
from different nodes will automatically be resolved by the monotonic merge function, which all data types must
provide. The state changes always converge. Several useful data types for counters, sets, maps and registers are
provided and you can also implement your own custom data types.
It is eventually consistent and geared toward providing high read and write availability (partition tolerance), with
low latency. Note that in an eventually consistent system a read may return an out-of-date value.
6.8.1 Using the Replicator
The akka.cluster.ddata.Replicator actor provides the API for interacting with the data. The
Replicator actor must be started on each node in the cluster, or group of nodes tagged with a specific role. It
communicates with other Replicator instances with the same path (without address) that are running on other
nodes . For convenience it can be used with the akka.cluster.ddata.DistributedData extension but
it can also be started as an ordinary actor using the Replicator.props. If it is started as an ordinary actor it
is important that it is given the same name, started on same path, on all nodes.
Cluster members with status WeaklyUp, will participate in Distributed Data. This means that the data will be
replicated to the WeaklyUp nodes with the background gossip protocol. Note that it will not participate in any
actions where the consistency mode is to read/write from all nodes or the majority of nodes. The WeaklyUp node
is not counted as part of the cluster. So 3 nodes + 5 WeaklyUp is essentially a 3 node cluster as far as consistent
actions are concerned.
Below is an example of an actor that schedules tick messages to itself and for each tick adds or removes elements
from a ORSet (observed-remove set). It also subscribes to changes of this.
import
import
import
import
import
import
import
import
import
java.util.concurrent.ThreadLocalRandom
akka.actor.Actor
akka.actor.ActorLogging
akka.cluster.Cluster
akka.cluster.ddata.DistributedData
akka.cluster.ddata.ORSet
akka.cluster.ddata.ORSetKey
akka.cluster.ddata.Replicator
akka.cluster.ddata.Replicator._
object DataBot {
private case object Tick
}
class DataBot extends Actor with ActorLogging {
import DataBot._
val replicator = DistributedData(context.system).replicator
implicit val node = Cluster(context.system)
import context.dispatcher
val tickTask = context.system.scheduler.schedule(5.seconds, 5.seconds, self,
˓→Tick)
val DataKey = ORSetKey[String]("key")
replicator ! Subscribe(DataKey, self)
def receive = {
case Tick =>
val s = ThreadLocalRandom.current().nextInt(97, 123).toChar.toString
if (ThreadLocalRandom.current().nextBoolean()) {
// add
log.info("Adding: {}", s)
replicator ! Update(DataKey, ORSet.empty[String], WriteLocal)(_ + s)
} else {
// remove
6.8. Distributed Data
330
Akka Scala Documentation, Release 2.5.0-RC1
log.info("Removing: {}", s)
replicator ! Update(DataKey, ORSet.empty[String], WriteLocal)(_ - s)
}
case _: UpdateResponse[_] => // ignore
case c @ Changed(DataKey) =>
val data = c.get(DataKey)
log.info("Current elements: {}", data.elements)
}
override def postStop(): Unit = tickTask.cancel()
}
Update
To modify and replicate a data value you send a Replicator.Update message to the local Replicator.
The current data value for the key of the Update is passed as parameter to the modify function of the Update.
The function is supposed to return the new value of the data, which will then be replicated according to the given
consistency level.
The modify function is called by the Replicator actor and must therefore be a pure function that only uses
the data parameter and stable fields from enclosing scope. It must for example not access sender() reference
of an enclosing actor.
Update is intended to only be sent from an actor running in same local ActorSystem as the
Replicator, because the modify function is typically not serializable.
You supply a write consistency level which has the following meaning:
• WriteLocal the value will immediately only be written to the local replica, and later disseminated with
gossip
• WriteTo(n) the value will immediately be written to at least n replicas, including the local replica
• WriteMajority the value will immediately be written to a majority of replicas, i.e. at least N/2 + 1
replicas, where N is the number of nodes in the cluster (or cluster role group)
• WriteAll the value will immediately be written to all nodes in the cluster (or all nodes in the cluster role
group)
When you specify to write to n out of x nodes, the update will first replicate to n nodes. If there are not
enough Acks after 1/5th of the timeout, the update will be replicated to n other nodes. If there are less than
n nodes left all of the remaining nodes are used. Reachable nodes are prefered over unreachable nodes.
Note that WriteMajority has a minCap parameter that is useful to specify to achieve better safety for small
clusters.
implicit val node = Cluster(system)
val replicator = DistributedData(system).replicator
val
val
val
val
Counter1Key = PNCounterKey("counter1")
Set1Key = GSetKey[String]("set1")
Set2Key = ORSetKey[String]("set2")
ActiveFlagKey = FlagKey("active")
replicator ! Update(Counter1Key, PNCounter(), WriteLocal)(_ + 1)
val writeTo3 = WriteTo(n = 3, timeout = 1.second)
replicator ! Update(Set1Key, GSet.empty[String], writeTo3)(_ + "hello")
val writeMajority = WriteMajority(timeout = 5.seconds)
6.8. Distributed Data
331
Akka Scala Documentation, Release 2.5.0-RC1
replicator ! Update(Set2Key, ORSet.empty[String], writeMajority)(_ + "hello")
val writeAll = WriteAll(timeout = 5.seconds)
replicator ! Update(ActiveFlagKey, Flag.empty, writeAll)(_.switchOn)
As reply of the Update a Replicator.UpdateSuccess is sent to the sender of the Update if the value
was successfully replicated according to the supplied consistency level within the supplied timeout. Otherwise a
Replicator.UpdateFailure subclass is sent back. Note that a Replicator.UpdateTimeout reply
does not mean that the update completely failed or was rolled back. It may still have been replicated to some
nodes, and will eventually be replicated to all nodes with the gossip protocol.
case UpdateSuccess(Counter1Key, req) => // ok
case UpdateSuccess(Set1Key, req) => // ok
case UpdateTimeout(Set1Key, req) =>
// write to 3 nodes failed within 1.second
You will always see your own writes. For example if you send two Update messages changing the value of
the same key, the modify function of the second message will see the change that was performed by the first
Update message.
In the Update message you can pass an optional request context, which the Replicator does not care about,
but is included in the reply messages. This is a convenient way to pass contextual information (e.g. original
sender) without having to use ask or maintain local correlation data structures.
implicit val node = Cluster(system)
val replicator = DistributedData(system).replicator
val writeTwo = WriteTo(n = 2, timeout = 3.second)
val Counter1Key = PNCounterKey("counter1")
def receive: Receive = {
case "increment" =>
// incoming command to increase the counter
val upd = Update(Counter1Key, PNCounter(), writeTwo, request =
˓→Some(sender()))(_ + 1)
replicator ! upd
case UpdateSuccess(Counter1Key, Some(replyTo: ActorRef)) =>
replyTo ! "ack"
case UpdateTimeout(Counter1Key, Some(replyTo: ActorRef)) =>
replyTo ! "nack"
}
Get
To retrieve the current value of a data you send Replicator.Get message to the Replicator. You supply
a consistency level which has the following meaning:
• ReadLocal the value will only be read from the local replica
• ReadFrom(n) the value will be read and merged from n replicas, including the local replica
• ReadMajority the value will be read and merged from a majority of replicas, i.e. at least N/2 + 1 replicas,
where N is the number of nodes in the cluster (or cluster role group)
• ReadAll the value will be read and merged from all nodes in the cluster (or all nodes in the cluster role
group)
Note that ReadMajority has a minCap parameter that is useful to specify to achieve better safety for small
clusters.
6.8. Distributed Data
332
Akka Scala Documentation, Release 2.5.0-RC1
val
val
val
val
val
replicator = DistributedData(system).replicator
Counter1Key = PNCounterKey("counter1")
Set1Key = GSetKey[String]("set1")
Set2Key = ORSetKey[String]("set2")
ActiveFlagKey = FlagKey("active")
replicator ! Get(Counter1Key, ReadLocal)
val readFrom3 = ReadFrom(n = 3, timeout = 1.second)
replicator ! Get(Set1Key, readFrom3)
val readMajority = ReadMajority(timeout = 5.seconds)
replicator ! Get(Set2Key, readMajority)
val readAll = ReadAll(timeout = 5.seconds)
replicator ! Get(ActiveFlagKey, readAll)
As reply of the Get a Replicator.GetSuccess is sent to the sender of the Get if the value was successfully
retrieved according to the supplied consistency level within the supplied timeout. Otherwise a Replicator.
GetFailure is sent. If the key does not exist the reply will be Replicator.NotFound.
case g @ GetSuccess(Counter1Key, req) =>
val value = g.get(Counter1Key).value
case NotFound(Counter1Key, req) => // key counter1 does not exist
case g @ GetSuccess(Set1Key, req) =>
val elements = g.get(Set1Key).elements
case GetFailure(Set1Key, req) =>
// read from 3 nodes failed within 1.second
case NotFound(Set1Key, req)
=> // key set1 does not exist
You will always read your own writes. For example if you send a Update message followed by a Get of the
same key the Get will retrieve the change that was performed by the preceding Update message. However,
the order of the reply messages are not defined, i.e. in the previous example you may receive the GetSuccess
before the UpdateSuccess.
In the Get message you can pass an optional request context in the same way as for the Update message,
described above. For example the original sender can be passed and replied to after receiving and transforming
GetSuccess.
implicit val node = Cluster(system)
val replicator = DistributedData(system).replicator
val readTwo = ReadFrom(n = 2, timeout = 3.second)
val Counter1Key = PNCounterKey("counter1")
def receive: Receive = {
case "get-count" =>
// incoming request to retrieve current value of the counter
replicator ! Get(Counter1Key, readTwo, request = Some(sender()))
case g @ GetSuccess(Counter1Key, Some(replyTo: ActorRef)) =>
val value = g.get(Counter1Key).value.longValue
replyTo ! value
case GetFailure(Counter1Key, Some(replyTo: ActorRef)) =>
replyTo ! -1L
case NotFound(Counter1Key, Some(replyTo: ActorRef)) =>
replyTo ! 0L
}
6.8. Distributed Data
333
Akka Scala Documentation, Release 2.5.0-RC1
Consistency
The consistency level that is supplied in the Update and Get specifies per request how many replicas that must
respond successfully to a write and read request.
For low latency reads you use ReadLocal with the risk of retrieving stale data, i.e. updates from other nodes
might not be visible yet.
When using WriteLocal the update is only written to the local replica and then disseminated in the background
with the gossip protocol, which can take few seconds to spread to all nodes.
WriteAll and ReadAll is the strongest consistency level, but also the slowest and with lowest availability. For
example, it is enough that one node is unavailable for a Get request and you will not receive the value.
If consistency is important, you can ensure that a read always reflects the most recent write by using the following
formula:
(nodes_written + nodes_read) > N
where N is the total number of nodes in the cluster, or the number of nodes with the role that is used for the
Replicator.
For example, in a 7 node cluster this these consistency properties are achieved by writing to 4 nodes and reading
from 4 nodes, or writing to 5 nodes and reading from 3 nodes.
By combining WriteMajority and ReadMajority levels a read always reflects the most recent write. The
Replicator writes and reads to a majority of replicas, i.e. N / 2 + 1. For example, in a 5 node cluster it writes
to 3 nodes and reads from 3 nodes. In a 6 node cluster it writes to 4 nodes and reads from 4 nodes.
You can define a minimum number of nodes for WriteMajority and ReadMajority, this will minimize the risk of reading steal data. Minimum cap is provided by minCap property of WriteMajority and
ReadMajority and defines the required majority. If the minCap is higher then N / 2 + 1 the minCap will be
used.
For example if the minCap is 5 the WriteMajority and ReadMajority for cluster of 3 nodes will be 3, for
cluster of 6 nodes will be 5 and for cluster of 12 nodes will be 7(N / 2 + 1).
For small clusters (<7) the risk of membership changes between a WriteMajority and ReadMajority is rather
high and then the nice properties of combining majority write and reads are not guaranteed. Therefore the
ReadMajority and WriteMajority have a minCap parameter that is useful to specify to achieve better
safety for small clusters. It means that if the cluster size is smaller than the majority size it will use the minCap
number of nodes but at most the total size of the cluster.
Here is an example of using WriteMajority and ReadMajority:
private val timeout = 3.seconds
private val readMajority = ReadMajority(timeout)
private val writeMajority = WriteMajority(timeout)
def receiveGetCart: Receive = {
case GetCart =>
replicator ! Get(DataKey, readMajority, Some(sender()))
case g @ GetSuccess(DataKey, Some(replyTo: ActorRef)) =>
val data = g.get(DataKey)
val cart = Cart(data.entries.values.toSet)
replyTo ! cart
case NotFound(DataKey, Some(replyTo: ActorRef)) =>
replyTo ! Cart(Set.empty)
case GetFailure(DataKey, Some(replyTo: ActorRef)) =>
// ReadMajority failure, try again with local read
replicator ! Get(DataKey, ReadLocal, Some(replyTo))
}
6.8. Distributed Data
334
Akka Scala Documentation, Release 2.5.0-RC1
def receiveAddItem: Receive = {
case cmd @ AddItem(item) =>
val update = Update(DataKey, LWWMap.empty[String, LineItem], writeMajority,
˓→Some(cmd)) {
cart => updateCart(cart, item)
}
replicator ! update
}
In some rare cases, when performing an Update it is needed to first try to fetch latest data from other nodes.
That can be done by first sending a Get with ReadMajority and then continue with the Update when the
GetSuccess, GetFailure or NotFound reply is received. This might be needed when you need to base a
decision on latest information or when removing entries from ORSet or ORMap. If an entry is added to an ORSet
or ORMap from one node and removed from another node the entry will only be removed if the added entry is
visible on the node where the removal is performed (hence the name observed-removed set).
The following example illustrates how to do that:
def receiveRemoveItem: Receive = {
case cmd @ RemoveItem(productId) =>
// Try to fetch latest from a majority of nodes first, since ORMap
// remove must have seen the item to be able to remove it.
replicator ! Get(DataKey, readMajority, Some(cmd))
case GetSuccess(DataKey, Some(RemoveItem(productId))) =>
replicator ! Update(DataKey, LWWMap(), writeMajority, None) {
_ - productId
}
case GetFailure(DataKey, Some(RemoveItem(productId))) =>
// ReadMajority failed, fall back to best effort local value
replicator ! Update(DataKey, LWWMap(), writeMajority, None) {
_ - productId
}
case NotFound(DataKey, Some(RemoveItem(productId))) =>
// nothing to remove
}
Warning: Caveat: Even if you use WriteMajority and ReadMajority there is small risk that you
may read stale data if the cluster membership has changed between the Update and the Get. For example,
in cluster of 5 nodes when you Update and that change is written to 3 nodes: n1, n2, n3. Then 2 more nodes
are added and a Get request is reading from 4 nodes, which happens to be n4, n5, n6, n7, i.e. the value on n1,
n2, n3 is not seen in the response of the Get request.
Subscribe
You may also register interest in change notifications by sending Replicator.Subscribe message to
the Replicator. It will send Replicator.Changed messages to the registered subscriber when
the data for the subscribed key is updated. Subscribers will be notified periodically with the configured notify-subscribers-interval, and it is also possible to send an explicit Replicator.
FlushChanges message to the Replicator to notify the subscribers immediately.
The subscriber is automatically removed if the subscriber is terminated. A subscriber can also be deregistered
with the Replicator.Unsubscribe message.
6.8. Distributed Data
335
Akka Scala Documentation, Release 2.5.0-RC1
val replicator = DistributedData(system).replicator
val Counter1Key = PNCounterKey("counter1")
// subscribe to changes of the Counter1Key value
replicator ! Subscribe(Counter1Key, self)
var currentValue = BigInt(0)
def receive: Receive = {
case c @ Changed(Counter1Key) =>
currentValue = c.get(Counter1Key).value
case "get-count" =>
// incoming request to retrieve current value of the counter
sender() ! currentValue
}
Delete
A data entry can be deleted by sending a Replicator.Delete message to the local local Replicator.
As reply of the Delete a Replicator.DeleteSuccess is sent to the sender of the Delete if the value
was successfully deleted according to the supplied consistency level within the supplied timeout. Otherwise a
Replicator.ReplicationDeleteFailure is sent. Note that ReplicationDeleteFailure does
not mean that the delete completely failed or was rolled back. It may still have been replicated to some nodes, and
may eventually be replicated to all nodes.
A deleted key cannot be reused again, but it is still recommended to delete unused data entries because that reduces
the replication overhead when new nodes join the cluster. Subsequent Delete, Update and Get requests will
be replied with Replicator.DataDeleted. Subscribers will receive Replicator.Deleted.
In the Delete message you can pass an optional request context in the same way as for the Update message,
described above. For example the original sender can be passed and replied to after receiving and transforming
DeleteSuccess.
val replicator = DistributedData(system).replicator
val Counter1Key = PNCounterKey("counter1")
val Set2Key = ORSetKey[String]("set2")
replicator ! Delete(Counter1Key, WriteLocal)
val writeMajority = WriteMajority(timeout = 5.seconds)
replicator ! Delete(Set2Key, writeMajority)
Warning: As deleted keys continue to be included in the stored data on each node as well as in gossip
messages, a continuous series of updates and deletes of top-level entities will result in growing memory usage
until an ActorSystem runs out of memory. To use Akka Distributed Data where frequent adds and removes are
required, you should use a fixed number of top-level data types that support both updates and removals, for
example ORMap or ORSet.
delta-CRDT
Delta State Replicated Data Types are supported. delta-CRDT is a way to reduce the need for sending the full state
for updates. For example adding element 'c' and 'd' to set {'a', 'b'} would result in sending the delta
{'c', 'd'} and merge that with the state on the receiving side, resulting in set {'a', 'b', 'c', 'd'}.
The protocol for replicating the deltas supports causal consistency if the data type is marked with
RequiresCausalDeliveryOfDeltas. Otherwise it is only eventually consistent. Without causal consistency it means that if elements 'c' and 'd' are added in two separate Update operations these deltas may
occasionally be propagated to nodes in different order than the causal order of the updates. For this example it
6.8. Distributed Data
336
Akka Scala Documentation, Release 2.5.0-RC1
can result in that set {'a', 'b', 'd'} can be seen before element ‘c’ is seen. Eventually it will be {'a',
'b', 'c', 'd'}.
Note that the full state is occasionally also replicated for delta-CRDTs, for example when new nodes are added to
the cluster or when deltas could not be propagated because of network partitions or similar problems.
The the delta propagation can be disabled with configuration property:
akka.cluster.distributed-data.delta-crdt.enabled=off
6.8.2 Data Types
The data types must be convergent (stateful) CRDTs and implement the ReplicatedData trait, i.e. they
provide a monotonic merge function and the state changes always converge.
You can use your own custom ReplicatedData or DeltaReplicatedData types, and several types are
provided by this package, such as:
• Counters: GCounter, PNCounter
• Sets: GSet, ORSet
• Maps: ORMap, ORMultiMap, LWWMap, PNCounterMap
• Registers: LWWRegister, Flag
Counters
GCounter is a “grow only counter”. It only supports increments, no decrements.
It works in a similar way as a vector clock. It keeps track of one counter per node and the total value is the sum of
these counters. The merge is implemented by taking the maximum count for each node.
If you need both increments and decrements you can use the PNCounter (positive/negative counter).
It is tracking the increments (P) separate from the decrements (N). Both P and N are represented as two internal
GCounter. Merge is handled by merging the internal P and N counters. The value of the counter is the value of
the P counter minus the value of the N counter.
implicit val node = Cluster(system)
val c0 = PNCounter.empty
val c1 = c0 + 1
val c2 = c1 + 7
val c3: PNCounter = c2 - 2
println(c3.value) // 6
GCounter and PNCounter have support for delta-CRDT and don’t need causal delivery of deltas.
Several related counters can be managed in a map with the PNCounterMap data type. When the counters are
placed in a PNCounterMap as opposed to placing them as separate top level values they are guaranteed to be
replicated together as one unit, which is sometimes necessary for related data.
implicit val node = Cluster(system)
val m0 = PNCounterMap.empty[String]
val m1 = m0.increment("a", 7)
val m2 = m1.decrement("a", 2)
val m3 = m2.increment("b", 1)
println(m3.get("a")) // 5
m3.entries.foreach { case (key, value) => println(s"$key -> $value") }
6.8. Distributed Data
337
Akka Scala Documentation, Release 2.5.0-RC1
Sets
If you only need to add elements to a set and not remove elements the GSet (grow-only set) is the data type to
use. The elements can be any type of values that can be serialized. Merge is simply the union of the two sets.
val s0 = GSet.empty[String]
val s1 = s0 + "a"
val s2 = s1 + "b" + "c"
if (s2.contains("a"))
println(s2.elements) // a, b, c
GSet has support for delta-CRDT and it doesn’t require causal delivery of deltas.
If you need add and remove operations you should use the ORSet (observed-remove set). Elements can be added
and removed any number of times. If an element is concurrently added and removed, the add will win. You cannot
remove an element that you have not seen.
The ORSet has a version vector that is incremented when an element is added to the set. The version for the node
that added the element is also tracked for each element in a so called “birth dot”. The version vector and the dots
are used by the merge function to track causality of the operations and resolve concurrent updates.
implicit val node = Cluster(system)
val s0 = ORSet.empty[String]
val s1 = s0 + "a"
val s2 = s1 + "b"
val s3 = s2 - "a"
println(s3.elements) // b
ORSet has support for delta-CRDT and it requires causal delivery of deltas.
Maps
ORMap (observed-remove map) is a map with keys of Any type and the values are ReplicatedData types
themselves. It supports add, remove and delete any number of times for a map entry.
If an entry is concurrently added and removed, the add will win. You cannot remove an entry that you have not
seen. This is the same semantics as for the ORSet.
If an entry is concurrently updated to different values the values will be merged, hence the requirement that the
values must be ReplicatedData types.
It is rather inconvenient to use the ORMap directly since it does not expose specific types of the values. The
ORMap is intended as a low level tool for building more specific maps, such as the following specialized maps.
ORMultiMap (observed-remove multi-map) is a multi-map implementation that wraps an ORMap with an
ORSet for the map’s value.
PNCounterMap (positive negative counter map) is a map of named counters (where the name can be of any
type). It is a specialized ORMap with PNCounter values.
LWWMap (last writer wins map) is a specialized ORMap with LWWRegister (last writer wins register) values.
implicit val node = Cluster(system)
val m0 = ORMultiMap.empty[String, Int]
val m1 = m0 + ("a" -> Set(1, 2, 3))
val m2 = m1.addBinding("a", 4)
val m3 = m2.removeBinding("a", 2)
val m4 = m3.addBinding("b", 1)
println(m4.entries)
When a data entry is changed the full state of that entry is replicated to other nodes, i.e. when you update a map
the whole map is replicated. Therefore, instead of using one ORMap with 1000 elements it is more efficient to split
that up in 10 top level ORMap entries with 100 elements each. Top level entries are replicated individually, which
6.8. Distributed Data
338
Akka Scala Documentation, Release 2.5.0-RC1
has the trade-off that different entries may not be replicated at the same time and you may see inconsistencies
between related entries. Separate top level entries cannot be updated atomically together.
Note that LWWRegister and therefore LWWMap relies on synchronized clocks and should only be used when the
choice of value is not important for concurrent updates occurring within the clock skew. Read more in the below
section about LWWRegister.
Flags and Registers
Flag is a data type for a boolean value that is initialized to false and can be switched to true. Thereafter it
cannot be changed. true wins over false in merge.
val f0 = Flag.empty
val f1 = f0.switchOn
println(f1.enabled)
LWWRegister (last writer wins register) can hold any (serializable) value.
Merge of a LWWRegister takes the register with highest timestamp. Note that this relies on synchronized
clocks. LWWRegister should only be used when the choice of value is not important for concurrent updates
occurring within the clock skew.
Merge takes the register updated by the node with lowest address (UniqueAddress is ordered) if the timestamps
are exactly the same.
implicit val node = Cluster(system)
val r1 = LWWRegister("Hello")
val r2 = r1.withValue("Hi")
println(s"${r1.value} by ${r1.updatedBy} at ${r1.timestamp}")
Instead of using timestamps based on System.currentTimeMillis() time it is possible to use a timestamp
value based on something else, for example an increasing version number from a database record that is used for
optimistic concurrency control.
case class Record(version: Int, name: String, address: String)
implicit val node = Cluster(system)
implicit val recordClock = new LWWRegister.Clock[Record] {
override def apply(currentTimestamp: Long, value: Record): Long =
value.version
}
val record1 = Record(version = 1, "Alice", "Union Square")
val r1 = LWWRegister(record1)
val record2 = Record(version = 2, "Alice", "Madison Square")
val r2 = LWWRegister(record2)
val r3 = r1.merge(r2)
println(r3.value)
For first-write-wins semantics you can use the LWWRegister#reverseClock instead of the
LWWRegister#defaultClock.
The defaultClock is using max value of System.currentTimeMillis() and currentTimestamp
+ 1. This means that the timestamp is increased for changes on the same node that occurs within the same
millisecond. It also means that it is safe to use the LWWRegister without synchronized clocks when there is
only one active writer, e.g. a Cluster Singleton. Such a single writer should then first read current value with
ReadMajority (or more) before changing and writing the value with WriteMajority (or more).
6.8. Distributed Data
339
Akka Scala Documentation, Release 2.5.0-RC1
Custom Data Type
You can rather easily implement your own data types. The only requirement is that it implements the merge
function of the ReplicatedData trait.
A nice property of stateful CRDTs is that they typically compose nicely, i.e. you can combine several smaller
data types to build richer data structures. For example, the PNCounter is composed of two internal GCounter
instances to keep track of increments and decrements separately.
Here is s simple implementation of a custom TwoPhaseSet that is using two internal GSet types to keep track
of addition and removals. A TwoPhaseSet is a set where an element may be added and removed, but never
added again thereafter.
case class TwoPhaseSet(
adds:
GSet[String] = GSet.empty,
removals: GSet[String] = GSet.empty)
extends ReplicatedData {
type T = TwoPhaseSet
def add(element: String): TwoPhaseSet =
copy(adds = adds.add(element))
def remove(element: String): TwoPhaseSet =
copy(removals = removals.add(element))
def elements: Set[String] = adds.elements diff removals.elements
override def merge(that: TwoPhaseSet): TwoPhaseSet =
copy(
adds = this.adds.merge(that.adds),
removals = this.removals.merge(that.removals))
}
Data types should be immutable, i.e. “modifying” methods should return a new instance.
Implement the additional methods of DeltaReplicatedData if it has support for delta-CRDT replication.
Serialization
The data types must be serializable with an Akka Serializer. It is highly recommended that you implement efficient serialization with Protobuf or similar for your custom data types. The built in data
types are marked with ReplicatedDataSerialization and serialized with akka.cluster.ddata.
protobuf.ReplicatedDataSerializer.
Serialization of the data types are used in remote messages and also for creating message digests (SHA-1) to
detect changes. Therefore it is important that the serialization is efficient and produce the same bytes for the same
content. For example sets and maps should be sorted deterministically in the serialization.
This is a protobuf representation of the above TwoPhaseSet:
option java_package = "docs.ddata.protobuf.msg";
option optimize_for = SPEED;
message TwoPhaseSet {
repeated string adds = 1;
repeated string removals = 2;
}
The serializer for the TwoPhaseSet:
import java.util.ArrayList
import java.util.Collections
6.8. Distributed Data
340
Akka Scala Documentation, Release 2.5.0-RC1
import
import
import
import
import
import
import
scala.collection.JavaConverters._
akka.actor.ExtendedActorSystem
akka.cluster.ddata.GSet
akka.cluster.ddata.protobuf.SerializationSupport
akka.serialization.Serializer
docs.ddata.TwoPhaseSet
docs.ddata.protobuf.msg.TwoPhaseSetMessages
class TwoPhaseSetSerializer(val system: ExtendedActorSystem)
extends Serializer with SerializationSupport {
override def includeManifest: Boolean = false
override def identifier = 99999
override def toBinary(obj: AnyRef): Array[Byte] = obj match {
case m: TwoPhaseSet => twoPhaseSetToProto(m).toByteArray
case _ => throw new IllegalArgumentException(
s"Can't serialize object of type ${obj.getClass}")
}
override def fromBinary(bytes: Array[Byte], clazz: Option[Class[_]]): AnyRef = {
twoPhaseSetFromBinary(bytes)
}
˓→
def twoPhaseSetToProto(twoPhaseSet: TwoPhaseSet): TwoPhaseSetMessages.
TwoPhaseSet = {
val b = TwoPhaseSetMessages.TwoPhaseSet.newBuilder()
// using java collections and sorting for performance (avoid conversions)
val adds = new ArrayList[String]
twoPhaseSet.adds.elements.foreach(adds.add)
if (!adds.isEmpty) {
Collections.sort(adds)
b.addAllAdds(adds)
}
val removals = new ArrayList[String]
twoPhaseSet.removals.elements.foreach(removals.add)
if (!removals.isEmpty) {
Collections.sort(removals)
b.addAllRemovals(removals)
}
b.build()
}
def twoPhaseSetFromBinary(bytes: Array[Byte]): TwoPhaseSet = {
val msg = TwoPhaseSetMessages.TwoPhaseSet.parseFrom(bytes)
val addsSet = msg.getAddsList.iterator.asScala.toSet
val removalsSet = msg.getRemovalsList.iterator.asScala.toSet
val adds = addsSet.foldLeft(GSet.empty[String])((acc, el) => acc.add(el))
val removals = removalsSet.foldLeft(GSet.empty[String])((acc, el) => acc.
˓→add(el))
// GSet will accumulate deltas when adding elements,
// but those are not of interest in the result of the deserialization
TwoPhaseSet(adds.resetDelta, removals.resetDelta)
}
}
Note that the elements of the sets are sorted so the SHA-1 digests are the same for the same elements.
You register the serializer in configuration:
akka.actor {
serializers {
6.8. Distributed Data
341
Akka Scala Documentation, Release 2.5.0-RC1
two-phase-set = "docs.ddata.protobuf.TwoPhaseSetSerializer"
}
serialization-bindings {
"docs.ddata.TwoPhaseSet" = two-phase-set
}
}
Using compression can sometimes be a good idea to reduce the data size. Gzip compression is provided by the
akka.cluster.ddata.protobuf.SerializationSupport trait:
override def toBinary(obj: AnyRef): Array[Byte] = obj match {
case m: TwoPhaseSet => compress(twoPhaseSetToProto(m))
case _ => throw new IllegalArgumentException(
s"Can't serialize object of type ${obj.getClass}")
}
override def fromBinary(bytes: Array[Byte], clazz: Option[Class[_]]): AnyRef = {
twoPhaseSetFromBinary(decompress(bytes))
}
The two embedded GSet can be serialized as illustrated above, but in general when composing new data types
from the existing built in types it is better to make use of the existing serializer for those types. This can be done
by declaring those as bytes fields in protobuf:
message TwoPhaseSet2 {
optional bytes adds = 1;
optional bytes removals = 2;
}
and use the methods otherMessageToProto and otherMessageFromBinary that are provided by the
SerializationSupport trait to serialize and deserialize the GSet instances. This works with any type that
has a registered Akka serializer. This is how such an serializer would look like for the TwoPhaseSet:
import
import
import
import
import
import
import
akka.actor.ExtendedActorSystem
akka.cluster.ddata.GSet
akka.cluster.ddata.protobuf.ReplicatedDataSerializer
akka.cluster.ddata.protobuf.SerializationSupport
akka.serialization.Serializer
docs.ddata.TwoPhaseSet
docs.ddata.protobuf.msg.TwoPhaseSetMessages
class TwoPhaseSetSerializer2(val system: ExtendedActorSystem)
extends Serializer with SerializationSupport {
override def includeManifest: Boolean = false
override def identifier = 99999
val replicatedDataSerializer = new ReplicatedDataSerializer(system)
override def toBinary(obj: AnyRef): Array[Byte] = obj match {
case m: TwoPhaseSet => twoPhaseSetToProto(m).toByteArray
case _ => throw new IllegalArgumentException(
s"Can't serialize object of type ${obj.getClass}")
}
override def fromBinary(bytes: Array[Byte], clazz: Option[Class[_]]): AnyRef = {
twoPhaseSetFromBinary(bytes)
}
˓→
def twoPhaseSetToProto(twoPhaseSet: TwoPhaseSet): TwoPhaseSetMessages.
TwoPhaseSet2 = {
6.8. Distributed Data
342
Akka Scala Documentation, Release 2.5.0-RC1
val b = TwoPhaseSetMessages.TwoPhaseSet2.newBuilder()
if (!twoPhaseSet.adds.isEmpty)
b.setAdds(otherMessageToProto(twoPhaseSet.adds).toByteString())
if (!twoPhaseSet.removals.isEmpty)
b.setRemovals(otherMessageToProto(twoPhaseSet.removals).toByteString())
b.build()
}
def twoPhaseSetFromBinary(bytes: Array[Byte]): TwoPhaseSet = {
val msg = TwoPhaseSetMessages.TwoPhaseSet2.parseFrom(bytes)
val adds =
if (msg.hasAdds)
otherMessageFromBinary(msg.getAdds.toByteArray).asInstanceOf[GSet[String]]
else
GSet.empty[String]
val removals =
if (msg.hasRemovals)
otherMessageFromBinary(msg.getRemovals.toByteArray).
˓→asInstanceOf[GSet[String]]
else
GSet.empty[String]
TwoPhaseSet(adds, removals)
}
}
Durable Storage
By default the data is only kept in memory. It is redundant since it is replicated to other nodes in the cluster, but if
you stop all nodes the data is lost, unless you have saved it elsewhere.
Entries can be configured to be durable, i.e. stored on local disk on each node. The stored data will be loaded next
time the replicator is started, i.e. when actor system is restarted. This means data will survive as long as at least
one node from the old cluster takes part in a new cluster. The keys of the durable entries are configured with:
akka.cluster.distributed-data.durable.keys = ["a", "b", "durable*"]
Prefix matching is supported by using * at the end of a key.
All entries can be made durable by specifying:
akka.cluster.distributed-data.durable.keys = ["*"]
LMDB is the default storage implementation. It is possible to replace that with another implementation by implementing the actor protocol described in akka.cluster.ddata.DurableStore and defining the akka.
cluster.distributed-data.durable.store-actor-class property for the new implementation.
The location of the files for the data is configured with:
# Directory of LMDB file. There are two options:
# 1. A relative or absolute path to a directory that ends with 'ddata'
#
the full name of the directory will contain name of the ActorSystem
#
and its remote port.
# 2. Otherwise the path is used as is, as a relative or absolute path to
#
a directory.
akka.cluster.distributed-data.durable.lmdb.dir = "ddata"
When running in production you may want to configure the directory to a specific path (alt 2), since the default
directory contains the remote port of the actor system to make the name unique. If using a dynamically assigned
port (0) it will be different each time and the previously stored data will not be loaded.
Making the data durable has of course a performance cost. By default, each update is flushed to disk before the
UpdateSuccess reply is sent. For better performance, but with the risk of losing the last writes if the JVM
6.8. Distributed Data
343
Akka Scala Documentation, Release 2.5.0-RC1
crashes, you can enable write behind mode. Changes are then accumulated during a time period before it is written
to LMDB and flushed to disk. Enabling write behind is especially efficient when performing many writes to the
same key, because it is only the last value for each key that will be serialized and stored. The risk of losing writes
if the JVM crashes is small since the data is typically replicated to other nodes immediately according to the given
WriteConsistency.
akka.cluster.distributed-data.lmdb.write-behind-interval = 200 ms
Note that you should be prepared to receive WriteFailure as reply to an Update of a durable entry if the
data could not be stored for some reason. When enabling write-behind-interval such errors will only be
logged and UpdateSuccess will still be the reply to the Update.
There is one important caveat when it comes pruning of CRDT Garbage for durable data. If and old data entry
that was never pruned is injected and merged with existing data after that the pruning markers have been removed
the value will not be correct. The time-to-live of the markers is defined by configuration akka.cluster.
distributed-data.durable.remove-pruning-marker-after and is in the magnitude of days.
This would be possible if a node with durable data didn’t participate in the pruning (e.g. it was shutdown) and
later started after this time. A node with durable data should not be stopped for longer time than this duration and
if it is joining again after this duration its data should first be manually removed (from the lmdb directory).
CRDT Garbage
One thing that can be problematic with CRDTs is that some data types accumulate history (garbage). For example
a GCounter keeps track of one counter per node. If a GCounter has been updated from one node it will
associate the identifier of that node forever. That can become a problem for long running systems with many
cluster nodes being added and removed. To solve this problem the Replicator performs pruning of data
associated with nodes that have been removed from the cluster. Data types that need pruning have to implement
the RemovedNodePruning trait. See the API documentation of the Replicator for details.
6.8.3 Samples
Several interesting samples are included and described in the Lightbend Activator tutorial named Akka Distributed
Data Samples with Scala.
• Low Latency Voting Service
• Highly Available Shopping Cart
• Distributed Service Registry
• Replicated Cache
• Replicated Metrics
6.8.4 Limitations
There are some limitations that you should be aware of.
CRDTs cannot be used for all types of problems, and eventual consistency does not fit all domains. Sometimes
you need strong consistency.
It is not intended for Big Data. The number of top level entries should not exceed 100000. When a new node is
added to the cluster all these entries are transferred (gossiped) to the new node. The entries are split up in chunks
and all existing nodes collaborate in the gossip, but it will take a while (tens of seconds) to transfer all entries and
this means that you cannot have too many top level entries. The current recommended limit is 100000. We will
be able to improve this if needed, but the design is still not intended for billions of entries.
All data is held in memory, which is another reason why it is not intended for Big Data.
When a data entry is changed the full state of that entry may be replicated to other nodes if it doesn’t support
delta-CRDT. The full state is also replicated for delta-CRDTs, for example when new nodes are added to the
6.8. Distributed Data
344
Akka Scala Documentation, Release 2.5.0-RC1
cluster or when deltas could not be propagated because of network partitions or similar problems. This means that
you cannot have too large data entries, because then the remote message size will be too large.
6.8.5 Learn More about CRDTs
• The Final Causal Frontier talk by Sean Cribbs
• Eventually Consistent Data Structures talk by Sean Cribbs
• Strong Eventual Consistency and Conflict-free Replicated Data Types talk by Mark Shapiro
• A comprehensive study of Convergent and Commutative Replicated Data Types paper by Mark Shapiro et.
al.
6.8.6 Dependencies
To use Distributed Data you must add the following dependency in your project.
sbt:
"com.typesafe.akka" %% "akka-distributed-data" % "2.5.0-RC1"
maven:
<dependency>
<groupId>com.typesafe.akka</groupId>
<artifactId>akka-distributed-data_2.11</artifactId>
<version>2.5.0-RC1</version>
</dependency>
6.8.7 Configuration
The DistributedData extension can be configured with the following properties:
# Settings for the DistributedData extension
akka.cluster.distributed-data {
# Actor name of the Replicator actor, /system/ddataReplicator
name = ddataReplicator
# Replicas are running on members tagged with this role.
# All members are used if undefined or empty.
role = ""
# How often the Replicator should send out gossip information
gossip-interval = 2 s
# How often the subscribers will be notified of changes, if any
notify-subscribers-interval = 500 ms
# Maximum number of entries to transfer in one gossip message when synchronizing
# the replicas. Next chunk will be transferred in next round of gossip.
max-delta-elements = 1000
# The id of the dispatcher to use for Replicator actors. If not specified
# default dispatcher is used.
# If specified you need to define the settings of the actual dispatcher.
use-dispatcher = ""
# How often the Replicator checks for pruning of data associated with
# removed cluster nodes. If this is set to 'off' the pruning feature will
6.8. Distributed Data
345
Akka Scala Documentation, Release 2.5.0-RC1
# be completely disabled.
pruning-interval = 120 s
# How long time it takes to spread the data to all other replica nodes.
# This is used when initiating and completing the pruning process of data
˓→associated
# with removed cluster nodes. The time measurement is stopped when any replica is
# unreachable, but it's still recommended to configure this with certain margin.
# It should be in the magnitude of minutes even though typical dissemination time
# is shorter (grows logarithmic with number of nodes). There is no advantage of
# setting this too low. Setting it to large value will delay the pruning process.
max-pruning-dissemination = 300 s
# The markers of that pruning has been performed for a removed node are kept for
this
# time and thereafter removed. If and old data entry that was never pruned is
˓→somehow
# injected and merged with existing data after this time the value will not be
˓→correct.
# This would be possible (although unlikely) in the case of a long network
˓→partition.
# It should be in the magnitude of hours. For durable data it is configured by
# 'akka.cluster.distributed-data.durable.pruning-marker-time-to-live'.
pruning-marker-time-to-live = 6 h
˓→
# Serialized Write and Read messages are cached when they are sent to
# several nodes. If no further activity they are removed from the cache
# after this duration.
serializer-cache-time-to-live = 10s
# Settings for delta-CRDT
delta-crdt {
# enable or disable delta-CRDT replication
enabled = on
}
durable {
# List of keys that are durable. Prefix matching is supported by using * at the
# end of a key.
keys = []
# The markers of that pruning has been performed for a removed node are kept
for this
# time and thereafter removed. If and old data entry that was never pruned is
# injected and merged with existing data after this time the value will not be
˓→correct.
# This would be possible if replica with durable data didn't participate in
˓→the pruning
# (e.g. it was shutdown) and later started after this time. A durable replica
˓→should not
# be stopped for longer time than this duration and if it is joining again
˓→after this
# duration its data should first be manually removed (from the lmdb directory).
# It should be in the magnitude of days. Note that there is a corresponding
˓→setting
# for non-durable data: 'akka.cluster.distributed-data.pruning-marker-time-to˓→live'.
pruning-marker-time-to-live = 10 d
˓→
#
#
#
#
Fully qualified class name of the durable store actor. It must be a subclass
of akka.actor.Actor and handle the protocol defined in
akka.cluster.ddata.DurableStore. The class must have a constructor with
com.typesafe.config.Config parameter.
6.8. Distributed Data
346
Akka Scala Documentation, Release 2.5.0-RC1
store-actor-class = akka.cluster.ddata.LmdbDurableStore
use-dispatcher = akka.cluster.distributed-data.durable.pinned-store
pinned-store {
executor = thread-pool-executor
type = PinnedDispatcher
}
# Config for the LmdbDurableStore
lmdb {
# Directory of LMDB file. There are two options:
# 1. A relative or absolute path to a directory that ends with 'ddata'
#
the full name of the directory will contain name of the ActorSystem
#
and its remote port.
# 2. Otherwise the path is used as is, as a relative or absolute path to
#
a directory.
#
# When running in production you may want to configure this to a specific
# path (alt 2), since the default directory contains the remote port of the
# actor system to make the name unique. If using a dynamically assigned
# port (0) it will be different each time and the previously stored data
# will not be loaded.
dir = "ddata"
# Size in bytes of the memory mapped file.
map-size = 100 MiB
# Accumulate changes before storing improves performance with the
# risk of losing the last writes if the JVM crashes.
# The interval is by default set to 'off' to write each update immediately.
# Enabling write behind by specifying a duration, e.g. 200ms, is especially
# efficient when performing many writes to the same key, because it is only
# the last value for each key that will be serialized and stored.
# write-behind-interval = 200 ms
write-behind-interval = off
}
}
}
6.9 Remoting
For an introduction of remoting capabilities of Akka please see Location Transparency.
Note: As explained in that chapter Akka remoting is designed for communication in a peer-to-peer fashion and
it has limitations for client-server setups. In particular Akka Remoting does not work transparently with Network
Address Translation, Load Balancers, or in Docker containers. For symmetric communication in these situations
network and/or Akka configuration will have to be changed as described in Akka behind NAT or in a Docker
container.
6.9.1 Preparing your ActorSystem for Remoting
The Akka remoting is a separate jar file. Make sure that you have the following dependency in your project:
6.9. Remoting
347
Akka Scala Documentation, Release 2.5.0-RC1
"com.typesafe.akka" %% "akka-remote" % "2.5.0-RC1"
To enable remote capabilities in your Akka project you should, at a minimum, add the following changes to your
application.conf file:
akka {
actor {
provider = remote
}
remote {
enabled-transports = ["akka.remote.netty.tcp"]
netty.tcp {
hostname = "127.0.0.1"
port = 2552
}
}
}
As you can see in the example above there are four things you need to add to get started:
• Change provider from local to remote
• Add host name - the machine you want to run the actor system on; this host name is exactly what is passed
to remote systems in order to identify this system and consequently used for connecting back to this system
if need be, hence set it to a reachable IP address or resolvable name in case you want to communicate across
the network.
• Add port number - the port the actor system should listen on, set to 0 to have it chosen automatically
Note: The port number needs to be unique for each actor system on the same machine even if the actor systems have different names. This is because each actor system has its own networking subsystem listening for
connections and handling messages as not to interfere with other actor systems.
The example above only illustrates the bare minimum of properties you have to add to enable remoting. All
settings are described in Remote Configuration.
6.9.2 Types of Remote Interaction
Akka has two ways of using remoting:
• Lookup : used to look up an actor on a remote node with actorSelection(path)
• Creation : used to create an actor on a remote node with actorOf(Props(...), actorName)
In the next sections the two alternatives are described in detail.
6.9.3 Looking up Remote Actors
actorSelection(path) will obtain an ActorSelection to an Actor on a remote node, e.g.:
val selection =
context.actorSelection("akka.tcp://actorSystemName@10.0.0.1:2552/user/actorName")
As you can see from the example above the following pattern is used to find an actor on a remote node:
akka.<protocol>://<actor system>@<hostname>:<port>/<actor path>
Once you obtained a selection to the actor you can interact with it in the same way you would with a local actor,
e.g.:
6.9. Remoting
348
Akka Scala Documentation, Release 2.5.0-RC1
selection ! "Pretty awesome feature"
To acquire an ActorRef for an ActorSelection you need to send a message to the selection and use the
sender reference of the reply from the actor. There is a built-in Identify message that all Actors will understand and automatically reply to with a ActorIdentity message containing the ActorRef. This can also
be done with the resolveOne method of the ActorSelection, which returns a Future of the matching
ActorRef.
Note: For more details on how actor addresses and paths are formed and used, please refer to Actor References,
Paths and Addresses.
Note: Message sends to actors that are actually in the sending actor system do not get delivered via the remote
actor ref provider. They’re delivered directly, by the local actor ref provider.
Aside from providing better performance, this also means that if the hostname you configure remoting to listen
as cannot actually be resolved from within the very same actor system, such messages will (perhaps counterintuitively) be delivered just fine.
6.9.4 Creating Actors Remotely
If you want to use the creation functionality in Akka remoting you have to further amend the application.
conf file in the following way (only showing deployment section):
akka {
actor {
deployment {
/sampleActor {
remote = "akka.tcp://sampleActorSystem@127.0.0.1:2553"
}
}
}
}
The configuration above instructs Akka to react when an actor with path /sampleActor is created, i.e. using
system.actorOf(Props(...), "sampleActor"). This specific actor will not be directly instantiated,
but instead the remote daemon of the remote system will be asked to create the actor, which in this sample
corresponds to sampleActorSystem@127.0.0.1:2553.
Once you have configured the properties above you would do the following in code:
val actor = system.actorOf(Props[SampleActor], "sampleActor")
actor ! "Pretty slick"
The actor class SampleActor has to be available to the runtimes using it, i.e. the classloader of the actor systems
has to have a JAR containing the class.
Note: In order to ensure serializability of Props when passing constructor arguments to the actor being created,
do not make the factory an inner class: this will inherently capture a reference to its enclosing object, which in
most cases is not serializable. It is best to create a factory method in the companion object of the actor’s class.
Serializability of all Props can be tested by setting the configuration item akka.actor.
serialize-creators=on. Only Props whose deploy has LocalScope are exempt from this
check.
6.9. Remoting
349
Akka Scala Documentation, Release 2.5.0-RC1
Note: You can use asterisks as wildcard matches for the actor paths, so you could specify: /*/sampleActor
and that would match all sampleActor on that level in the hierarchy. You can also use wildcard in the last
position to match all actors at a certain level: /someParent/*. Non-wildcard matches always have higher
priority to match than wildcards, so: /foo/bar is considered more specific than /foo/* and only the highest
priority match is used. Please note that it cannot be used to partially match section, like this: /foo*/bar,
/f*o/bar etc.
Programmatic Remote Deployment
To allow dynamically deployed systems, it is also possible to include deployment configuration in the Props
which are used to create an actor: this information is the equivalent of a deployment section from the configuration
file, and if both are given, the external configuration takes precedence.
With these imports:
import akka.actor.{ Props, Deploy, Address, AddressFromURIString }
import akka.remote.RemoteScope
and a remote address like this:
val one = AddressFromURIString("akka.tcp://sys@host:1234")
val two = Address("akka.tcp", "sys", "host", 1234) // this gives the same
you can advise the system to create a child on that remote node like so:
val ref = system.actorOf(Props[SampleActor].
withDeploy(Deploy(scope = RemoteScope(address))))
Remote deployment whitelist
As remote deployment can potentially be abused by both users and even attackers a whitelist feature is available
to guard the ActorSystem from deploying unexpected actors. Please note that remote deployment is not remote
code loading, the Actors class to be deployed onto a remote system needs to be present on that remote system.
This still however may pose a security risk, and one may want to restrict remote deployment to only a specific set
of known actors by enabling the whitelist feature.
To enable remote deployment whitelisting set the akka.remote.deployment.enable-whitelist value
to on. The list of allowed classes has to be configured on the “remote” system, in other words on the system onto
which others will be attempting to remote deploy Actors. That system, locally, knows best which Actors it should
or should not allow others to remote deploy onto it. The full settings section may for example look like this:
akka.remote.deployment {
enable-whitelist = on
whitelist = [
"NOT_ON_CLASSPATH", # verify we don't throw if a class not on classpath is
˓→listed here
"akka.remote.RemoteDeploymentWhitelistSpec.EchoWhitelisted"
]
}
Actor classes not included in the whitelist will not be allowed to be remote deployed onto this system.
6.9. Remoting
350
Akka Scala Documentation, Release 2.5.0-RC1
6.9.5 Lifecycle and Failure Recovery Model
Each link with a remote system can be in one of the four states as illustrated above. Before any communication
happens with a remote system at a given Address the state of the association is Idle. The first time a message
is attempted to be sent to the remote system or an inbound connection is accepted the state of the link transitions
to Active denoting that the two systems has messages to send or receive and no failures were encountered so
far. When a communication failure happens and the connection is lost between the two systems the link becomes
Gated.
In this state the system will not attempt to connect to the remote host and all outbound messages will
be dropped. The time while the link is in the Gated state is controlled by the setting akka.remote.
retry-gate-closed-for: after this time elapses the link state transitions to Idle again. Gate is onesided in the sense that whenever a successful inbound connection is accepted from a remote system during Gate
it automatically transitions to Active and communication resumes immediately.
In the face of communication failures that are unrecoverable because the state of the participating systems are
inconsistent, the remote system becomes Quarantined. Unlike Gate, quarantining is permanent and lasts
until one of the systems is restarted. After a restart communication can be resumed again and the link can become
Active again.
6.9. Remoting
351
Akka Scala Documentation, Release 2.5.0-RC1
6.9.6 Watching Remote Actors
Watching a remote actor is not different than watching a local actor, as described in Lifecycle Monitoring aka
DeathWatch.
Failure Detector
Under the hood remote death watch uses heartbeat messages and a failure detector to generate Terminated
message from network failures and JVM crashes, in addition to graceful termination of watched actor.
The heartbeat arrival times is interpreted by an implementation of The Phi Accrual Failure Detector.
The suspicion level of failure is given by a value called phi. The basic idea of the phi failure detector is to express
the value of phi on a scale that is dynamically adjusted to reflect current network conditions.
The value of phi is calculated as:
phi = -log10(1 - F(timeSinceLastHeartbeat))
where F is the cumulative distribution function of a normal distribution with mean and standard deviation estimated
from historical heartbeat inter-arrival times.
In the Remote Configuration you can adjust the akka.remote.watch-failure-detector.threshold
to define when a phi value is considered to be a failure.
A low threshold is prone to generate many false positives but ensures a quick detection in the event of a real
crash. Conversely, a high threshold generates fewer mistakes but needs more time to detect actual crashes.
The default threshold is 10 and is appropriate for most situations. However in cloud environments, such as
Amazon EC2, the value could be increased to 12 in order to account for network issues that sometimes occur on
such platforms.
The following chart illustrates how phi increase with increasing time since the previous heartbeat.
Phi is calculated from the mean and standard deviation of historical inter arrival times. The previous chart is an
example for standard deviation of 200 ms. If the heartbeats arrive with less deviation the curve becomes steeper,
i.e. it is possible to determine failure more quickly. The curve looks like this for a standard deviation of 100 ms.
6.9. Remoting
352
Akka Scala Documentation, Release 2.5.0-RC1
To be able to survive sudden abnormalities, such as garbage collection pauses and transient network failures the failure detector is configured with a margin, akka.remote.watch-failure-detector.
acceptable-heartbeat-pause. You may want to adjust the Remote Configuration of this depending
on you environment. This is how the curve looks like for acceptable-heartbeat-pause configured to 3
seconds.
6.9. Remoting
353
Akka Scala Documentation, Release 2.5.0-RC1
6.9.7 Serialization
When using remoting for actors you must ensure that the props and messages used for those actors are serializable. Failing to do so will cause the system to behave in an unintended way.
For more information please see Serialization.
Disabling the Java Serializer
Since the 2.4.11 release of Akka it is possible to entirely disable the default Java Serialization mechanism.
Please note that new remoting implementation (codename Artery) does not use Java serialization for internal
messages by default. For compatibility reasons, the current remoting still uses Java serialization for some classes,
however you can disable it in this remoting implementation as well by following the steps below.
The first step is to enable some additional serializers that replace previous Java serialization of some internal
messages. This is recommended also when you can’t disable Java serialization completely. Those serializers are
enabled with this configuration:
akka.actor {
# Set this to on to enable serialization-bindings define in
# additional-serialization-bindings. Those are by default not included
# for backwards compatibility reasons. They are enabled by default if
# akka.remote.artery.enabled=on.
enable-additional-serialization-bindings = on
}
The reason these are not enabled by default is wire-level compatibility between any 2.4.x Actor Systems. If you
roll out a new cluster, all on the same Akka version that can enable these serializers it is recommended to enable
this setting. When using Remoting (codename Artery) these serializers are enabled by default.
Warning: Please note that when enabling the additional-serialization-bindings when using the old remoting,
you must do so on all nodes participating in a cluster, otherwise the mis-aligned serialization configurations
will cause deserialization errors on the receiving nodes.
Java serialization is known to be slow and prone to attacks of various kinds - it never was designed for high
throughput messaging after all. However, it is very convenient to use, thus it remained the default serialization
mechanism that Akka used to serialize user messages as well as some of its internal messages in previous versions.
Since the release of Artery, Akka internals do not rely on Java serialization anymore (one exception being java.
lang.Throwable).
Note: When using the new remoting implementation (codename Artery), Akka does not use Java Serialization
for any of its internal messages. It is highly encouraged to disable java serialization, so please plan to do so at the
earliest possibility you have in your project.
One may think that network bandwidth and latency limit the performance of remote messaging, but serialization
is a more typical bottleneck.
For user messages, the default serializer, implemented using Java serialization, remains available and enabled. We
do however recommend to disable it entirely and utilise a proper serialization library instead in order effectively
utilise the improved performance and ability for rolling deployments using Artery. Libraries that we recommend
to use include, but are not limited to, Kryo by using the akka-kryo-serialization library or Google Protocol Buffers
if you want more control over the schema evolution of your messages.
In order to completely disable Java Serialization in your Actor system you need to add the following configuration
to your application.conf:
akka.actor.allow-java-serialization = off
6.9. Remoting
354
Akka Scala Documentation, Release 2.5.0-RC1
This will completely disable the use of akka.serialization.JavaSerialization by the Akka Serialization extension, instead DisabledJavaSerializer will be inserted which will fail explicitly if attempts
to use java serialization are made.
It will also enable the above mentioned enable-additional-serialization-bindings.
The log messages emitted by such serializer SHOULD be be treated as potential attacks which the serializer
prevented, as they MAY indicate an external operator attempting to send malicious messages intending to use java
serialization as attack vector. The attempts are logged with the SECURITY marker.
Please note that this option does not stop you from manually invoking java serialization.
Please note that this means that you will have to configure different serializers which will able to handle all of
your remote messages. Please refer to the Serialization documentation as well as ByteBuffer based serialization
to learn how to do this.
6.9.8 Routers with Remote Destinations
It is absolutely feasible to combine remoting with Routing.
A pool of remote deployed routees can be configured as:
akka.actor.deployment {
/parent/remotePool {
router = round-robin-pool
nr-of-instances = 10
target.nodes = ["akka.tcp://app@10.0.0.2:2552", "akka.tcp://app@10.0.0.3:2552"]
}
}
This configuration setting will clone the actor defined in the Props of the remotePool 10 times and deploy it
evenly distributed across the two given target nodes.
A group of remote actors can be configured as:
akka.actor.deployment {
/parent/remoteGroup {
router = round-robin-group
routees.paths = [
"akka.tcp://app@10.0.0.1:2552/user/workers/w1",
"akka.tcp://app@10.0.0.2:2552/user/workers/w1",
"akka.tcp://app@10.0.0.3:2552/user/workers/w1"]
}
}
This configuration setting will send messages to the defined remote actor paths. It requires that you create the
destination actors on the remote nodes with matching paths. That is not done by the router.
6.9.9 Remoting Sample
There is a more extensive remote example that comes with Lightbend Activator. The tutorial named Akka Remote
Samples with Scala demonstrates both remote deployment and look-up of remote actors.
Remote Events
It is possible to listen to events that occur in Akka Remote, and to subscribe/unsubscribe to these events you
simply register as listener to the below described types in on the ActorSystem.eventStream.
6.9. Remoting
355
Akka Scala Documentation, Release 2.5.0-RC1
Note: To subscribe to any remote event, subscribe to RemotingLifecycleEvent. To subscribe to events
related only to the lifecycle of associations, subscribe to akka.remote.AssociationEvent.
Note: The use of term “Association” instead of “Connection” reflects that the remoting subsystem may use connectionless transports, but an association similar to transport layer connections is maintained between endpoints
by the Akka protocol.
By default an event listener is registered which logs all of the events described below. This default was chosen to
help setting up a system, but it is quite common to switch this logging off once that phase of the project is finished.
Note: In order to switch off the logging, set akka.remote.log-remote-lifecycle-events = off
in your application.conf.
To be notified when an association is over (“disconnected”) listen to DisassociatedEvent which holds the
direction of the association (inbound or outbound) and the addresses of the involved parties.
To be notified when an association is successfully established (“connected”) listen to AssociatedEvent which
holds the direction of the association (inbound or outbound) and the addresses of the involved parties.
To intercept errors directly related to associations, listen to AssociationErrorEvent which holds the direction of the association (inbound or outbound), the addresses of the involved parties and the Throwable cause.
To be notified when the remoting subsystem is ready to accept associations, listen to RemotingListenEvent
which contains the addresses the remoting listens on.
To be notified when the current system is quarantined by the remote system, listen to
ThisActorSystemQuarantinedEvent, which includes the addresses of local and remote ActorSystems.
To be notified when the remoting subsystem has been shut down, listen to RemotingShutdownEvent.
To intercept generic remoting related errors, listen to RemotingErrorEvent which holds the Throwable
cause.
6.9.10 Remote Security
An ActorSystem should not be exposed via Akka Remote over plain TCP to an untrusted network (e.g. internet). It should be protected by network security, such as a firewall. If that is not considered as enough protection
TLS with mutual authentication should be enabled.
It is also security best-practice to disable the Java serializer because of its multiple known attack surfaces.
Configuring SSL/TLS for Akka Remoting
SSL can be used as the remote transport by adding akka.remote.netty.ssl to the enabled-transport
configuration section. An example of setting up the default Netty based SSL driver as default:
akka {
remote {
enabled-transports = [akka.remote.netty.ssl]
}
}
Next the actual SSL/TLS parameters have to be configured:
6.9. Remoting
356
Akka Scala Documentation, Release 2.5.0-RC1
akka {
remote {
netty.ssl.security {
key-store = "/example/path/to/mykeystore.jks"
trust-store = "/example/path/to/mytruststore.jks"
key-store-password = "changeme"
key-password = "changeme"
trust-store-password = "changeme"
protocol = "TLSv1.2"
enabled-algorithms = [TLS_DHE_RSA_WITH_AES_128_GCM_SHA256]
random-number-generator = "AES128CounterSecureRNG"
}
}
}
According to RFC 7525 the recommended algorithms to use with TLS 1.2 (as of writing this document) are:
• TLS_DHE_RSA_WITH_AES_128_GCM_SHA256
• TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256
• TLS_DHE_RSA_WITH_AES_256_GCM_SHA384
• TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384
You should always check the latest information about security and algorithm recommendations though before you
configure your system.
Creating and working with keystores and certificates is well documented in the Generating X.509 Certificates
section of Lightbend’s SSL-Config library.
Since an Akka remoting is inherently peer-to-peer both the key-store as well as trust-store need to be configured
on each remoting node participating in the cluster.
The official Java Secure Socket Extension documentation as well as the Oracle documentation on creating KeyStore and TrustStores are both great resources to research when setting up security on the JVM. Please consult
those resources when troubleshooting and configuring SSL.
Since Akka 2.5.0 mutual authentication between TLS peers is enabled by default.
Mutual authentication means that the the passive side (the TLS server side) of a connection will also request and
verify a certificate from the connecting peer. Without this mode only the client side is requesting and verifying
certificates. While Akka is a peer-to-peer technology, each connection between nodes starts out from one side (the
“client”) towards the other (the “server”).
Note that if TLS is enabled with mutual authentication there is still a risk that an attacker can gain access to a valid
certificate by compromising any node with certificates issued by the same internal PKI tree.
See also a description of the settings in the Remote Configuration section.
Note: When using SHA1PRNG on Linux it’s recommended specify -Djava.security.egd=file:/
dev/urandom as argument to the JVM to prevent blocking. It is NOT as secure because it reuses the seed.
Untrusted Mode
As soon as an actor system can connect to another remotely, it may in principle send any possible message to any
actor contained within that remote system. One example may be sending a PoisonPill to the system guardian,
shutting that system down. This is not always desired, and it can be disabled with the following setting:
6.9. Remoting
357
Akka Scala Documentation, Release 2.5.0-RC1
akka.remote.untrusted-mode = on
This disallows sending of system messages (actor life-cycle commands, DeathWatch, etc.) and any message
extending PossiblyHarmful to the system on which this flag is set. Should a client send them nonetheless
they are dropped and logged (at DEBUG level in order to reduce the possibilities for a denial of service attack).
PossiblyHarmful covers the predefined messages like PoisonPill and Kill, but it can also be added as
a marker trait to user-defined messages.
Warning: Untrusted mode does not give full protection against attacks by itself. It makes it slightly harder
to perform malicious or unintended actions but it should be complemented with disabled Java serializer. Additional protection can be achieved when running in an untrusted network by network security (e.g. firewalls)
and/or enabling TLS with mutual authentication.
Messages sent with actor selection are by default discarded in untrusted mode, but permission to receive actor
selection messages can be granted to specific actors defined in configuration:
akka.remote.trusted-selection-paths = ["/user/receptionist", "/user/namingService"]
The actual message must still not be of type PossiblyHarmful.
In summary, the following operations are ignored by a system configured in untrusted mode when incoming via
the remoting layer:
• remote deployment (which also means no remote supervision)
• remote DeathWatch
• system.stop(), PoisonPill, Kill
• sending any message which extends from the PossiblyHarmful marker interface, which includes
Terminated
• messages sent with actor selection, unless destination defined in trusted-selection-paths.
Note: Enabling the untrusted mode does not remove the capability of the client to freely choose the target of
its message sends, which means that messages not prohibited by the above rules can be sent to any actor in the
remote system. It is good practice for a client-facing system to only contain a well-defined set of entry point actors,
which then forward requests (possibly after performing validation) to another actor system containing the actual
worker actors. If messaging between these two server-side systems is done using local ActorRef (they can be
exchanged safely between actor systems within the same JVM), you can restrict the messages on this interface by
marking them PossiblyHarmful so that a client cannot forge them.
6.9.11 Remote Configuration
There are lots of configuration properties that are related to remoting in Akka. We refer to the reference configuration for more information.
Note: Setting properties like the listening IP and port number programmatically is best done by using something
like the following:
ConfigFactory.parseString("akka.remote.netty.tcp.hostname=\"1.2.3.4\"")
.withFallback(ConfigFactory.load());
6.9. Remoting
358
Akka Scala Documentation, Release 2.5.0-RC1
Akka behind NAT or in a Docker container
In setups involving Network Address Translation (NAT), Load Balancers or Docker containers the hostname and
port pair that Akka binds to will be different than the “logical” host name and port pair that is used to connect to
the system from the outside. This requires special configuration that sets both the logical and the bind pairs for
remoting.
akka {
remote {
netty.tcp {
hostname = my.domain.com
port = 8000
# external (logical) hostname
# external (logical) port
bind-hostname = local.address # internal (bind) hostname
bind-port = 2552
# internal (bind) port
}
}
}
6.10 Remoting (codename Artery)
Note: This page describes the may change remoting subsystem, codenamed Artery that will eventually replace
the old remoting implementation. For the current stable remoting system please refer to Remoting.
Remoting enables Actor systems on different hosts or JVMs to communicate with each other. By enabling remoting the system will start listening on a provided network address and also gains the ability to connect to other
systems through the network. From the application’s perspective there is no API difference between local or remote systems, ActorRef instances that point to remote systems look exactly the same as local ones: they can
be sent messages to, watched, etc. Every ActorRef contains hostname and port information and can be passed
around even on the network. This means that on a network every ActorRef is a unique identifier of an actor on
that network.
Remoting is not a server-client technology. All systems using remoting can contact any other system on the
network if they possess an ActorRef pointing to those system. This means that every system that is remoting
enabled acts as a “server” to which arbitrary systems on the same network can connect to.
6.10.1 What is new in Artery
Artery is a reimplementation of the old remoting module aimed at improving performance and stability. It is
mostly backwards compatible with the old implementation and it is a drop-in replacement in many cases. Main
features of Artery compared to the previous implementation:
• Based on Aeron (UDP) instead of TCP
• Focused on high-throughput, low-latency communication
• Isolation of internal control messages from user messages improving stability and reducing false failure
detection in case of heavy traffic by using a dedicated subchannel.
• Mostly allocation-free operation
• Support for a separate subchannel for large messages to avoid interference with smaller messages
• Compression of actor paths on the wire to reduce overhead for smaller messages
• Support for faster serialization/deserialization using ByteBuffers directly
• Built-in Flight-Recorder to help debugging implementation issues without polluting users logs with implementaiton specific events
6.10. Remoting (codename Artery)
359
Akka Scala Documentation, Release 2.5.0-RC1
• Providing protocol stability across major Akka versions to support rolling updates of large-scale systems
The main incompatible change from the previous implementation that the protocol field of the string representation
of an ActorRef is always akka instead of the previously used akka.tcp or akka.ssl.tcp. Configuration properties
are also different.
6.10.2 Preparing your ActorSystem for Remoting
The Akka remoting is a separate jar file. Make sure that you have the following dependency in your project:
"com.typesafe.akka" %% "akka-remote" % "2.5.0-RC1"
To enable remote capabilities in your Akka project you should, at a minimum, add the following changes to your
application.conf file:
akka {
actor {
provider = remote
}
remote {
artery {
enabled = on
canonical.hostname = "127.0.0.1"
canonical.port = 25520
}
}
}
As you can see in the example above there are four things you need to add to get started:
• Change provider from local to remote
• Enable Artery to use it as the remoting implementation
• Add host name - the machine you want to run the actor system on; this host name is exactly what is passed
to remote systems in order to identify this system and consequently used for connecting back to this system
if need be, hence set it to a reachable IP address or resolvable name in case you want to communicate across
the network.
• Add port number - the port the actor system should listen on, set to 0 to have it chosen automatically
Note: The port number needs to be unique for each actor system on the same machine even if the actor systems have different names. This is because each actor system has its own networking subsystem listening for
connections and handling messages as not to interfere with other actor systems.
The example above only illustrates the bare minimum of properties you have to add to enable remoting. All
settings are described in Remote Configuration.
Note: Aeron requires 64bit JVM to work reliably.
Canonical address
In order to remoting to work properly, where each system can send messages to any other system on the same
network (for example a system forwards a message to a third system, and the third replies directly to the sender
system) it is essential for every system to have a unique, globally reachable address and port. This address is
part of the unique name of the system and will be used by other systems to open a connection to it and send
messages. This means that if a host has multiple names (different DNS records pointing to the same IP address)
then only one of these can be canonical. If a message arrives to a system but it contains a different hostname
6.10. Remoting (codename Artery)
360
Akka Scala Documentation, Release 2.5.0-RC1
than the expected canonical name then the message will be dropped. If multiple names for a system would be
allowed, then equality checks among ActorRef instances would no longer to be trusted and this would violate
the fundamental assumption that an actor has a globally unique reference on a given network. As a consequence,
this also means that localhost addresses (e.g. 127.0.0.1) cannot be used in general (apart from local development)
since they are not unique addresses in a real network.
In cases, where Network Address Translation (NAT) is used or other network bridging is involved, it is important
to configure the system so that it understands that there is a difference between his externally visible, canonical
address and between the host-port pair that is used to listen for connections. See Akka behind NAT or in a Docker
container for details.
6.10.3 Acquiring references to remote actors
In order to communicate with an actor, it is necessary to have its ActorRef. In the local case it is usually the
creator of the actor (the caller of actorOf()) is who gets the ActorRef for an actor that it can then send to
other actors. In other words:
• An Actor can get a remote Actor’s reference simply by receiving a message from it (as it’s available as
sender() then), or inside of a remote message (e.g. PleaseReply(message: String, remoteActorRef: ActorRef))
Alternatively, an actor can look up another located at a known path using ActorSelection. These methods
are available even in remoting enabled systems:
• Remote Lookup : used to look up an actor on a remote node with actorSelection(path)
• Remote Creation :
actorName)
used to create an actor on a remote node with actorOf(Props(...),
In the next sections the two alternatives are described in detail.
Looking up Remote Actors
actorSelection(path) will obtain an ActorSelection to an Actor on a remote node, e.g.:
val selection =
context.actorSelection("akka://actorSystemName@10.0.0.1:25520/user/actorName")
As you can see from the example above the following pattern is used to find an actor on a remote node:
akka://<actor system>@<hostname>:<port>/<actor path>
Note: Unlike with earlier remoting, the protocol field is always akka as pluggable transports are no longer
supported.
Once you obtained a selection to the actor you can interact with it in the same way you would with a local actor,
e.g.:
selection ! "Pretty awesome feature"
To acquire an ActorRef for an ActorSelection you need to send a message to the selection and use the
sender reference of the reply from the actor. There is a built-in Identify message that all Actors will understand and automatically reply to with a ActorIdentity message containing the ActorRef. This can also
be done with the resolveOne method of the ActorSelection, which returns a Future of the matching
ActorRef.
For more details on how actor addresses and paths are formed and used, please refer to Actor References, Paths
and Addresses.
6.10. Remoting (codename Artery)
361
Akka Scala Documentation, Release 2.5.0-RC1
Note: Message sends to actors that are actually in the sending actor system do not get delivered via the remote
actor ref provider. They’re delivered directly, by the local actor ref provider.
Aside from providing better performance, this also means that if the hostname you configure remoting to listen
as cannot actually be resolved from within the very same actor system, such messages will (perhaps counterintuitively) be delivered just fine.
Creating Actors Remotely
If you want to use the creation functionality in Akka remoting you have to further amend the application.
conf file in the following way (only showing deployment section):
akka {
actor {
deployment {
/sampleActor {
remote = "akka://sampleActorSystem@127.0.0.1:2553"
}
}
}
}
The configuration above instructs Akka to react when an actor with path /sampleActor is created, i.e. using
system.actorOf(Props(...), "sampleActor"). This specific actor will not be directly instantiated,
but instead the remote daemon of the remote system will be asked to create the actor, which in this sample
corresponds to sampleActorSystem@127.0.0.1:2553.
Once you have configured the properties above you would do the following in code:
val actor = system.actorOf(Props[SampleActor], "sampleActor")
actor ! "Pretty slick"
The actor class SampleActor has to be available to the runtimes using it, i.e. the classloader of the actor systems
has to have a JAR containing the class.
Note: In order to ensure serializability of Props when passing constructor arguments to the actor being created,
do not make the factory an inner class: this will inherently capture a reference to its enclosing object, which in
most cases is not serializable. It is best to create a factory method in the companion object of the actor’s class.
Serializability of all Props can be tested by setting the configuration item akka.actor.
serialize-creators=on. Only Props whose deploy has LocalScope are exempt from this
check.
You can use asterisks as wildcard matches for the actor paths, so you could specify: /*/sampleActor and
that would match all sampleActor on that level in the hierarchy. You can also use wildcard in the last position
to match all actors at a certain level: /someParent/*. Non-wildcard matches always have higher priority to
match than wildcards, so: /foo/bar is considered more specific than /foo/* and only the highest priority
match is used. Please note that it cannot be used to partially match section, like this: /foo*/bar, /f*o/bar
etc.
Programmatic Remote Deployment
To allow dynamically deployed systems, it is also possible to include deployment configuration in the Props
which are used to create an actor: this information is the equivalent of a deployment section from the configuration
file, and if both are given, the external configuration takes precedence.
With these imports:
6.10. Remoting (codename Artery)
362
Akka Scala Documentation, Release 2.5.0-RC1
import akka.actor.{ Props, Deploy, Address, AddressFromURIString }
import akka.remote.RemoteScope
and a remote address like this:
val one = AddressFromURIString("akka://sys@host:1234")
val two = Address("akka", "sys", "host", 1234) // this gives the same
you can advise the system to create a child on that remote node like so:
val ref = system.actorOf(Props[SampleActor].
withDeploy(Deploy(scope = RemoteScope(address))))
Remote deployment whitelist
As remote deployment can potentially be abused by both users and even attackers a whitelist feature is available
to guard the ActorSystem from deploying unexpected actors. Please note that remote deployment is not remote
code loading, the Actors class to be deployed onto a remote system needs to be present on that remote system.
This still however may pose a security risk, and one may want to restrict remote deployment to only a specific set
of known actors by enabling the whitelist feature.
To enable remote deployment whitelisting set the akka.remote.deployment.enable-whitelist value
to on. The list of allowed classes has to be configured on the “remote” system, in other words on the system onto
which others will be attempting to remote deploy Actors. That system, locally, knows best which Actors it should
or should not allow others to remote deploy onto it. The full settings section may for example look like this:
akka.remote.deployment {
enable-whitelist = on
whitelist = [
"NOT_ON_CLASSPATH", # verify we don't throw if a class not on classpath is
˓→listed here
"akka.remote.RemoteDeploymentWhitelistSpec.EchoWhitelisted"
]
}
Actor classes not included in the whitelist will not be allowed to be remote deployed onto this system.
6.10.4 Remote Security
An ActorSystem should not be exposed via Akka Remote (Artery) over plain Aeron/UDP to an untrusted
network (e.g. internet). It should be protected by network security, such as a firewall. There is currently no support
for encryption with Artery so if network security is not considered as enough protection the classic remoting with
TLS and mutual authentication should be used.
Best practice is that Akka remoting nodes should only be accessible from the adjacent network.
It is also security best practice to disable the Java serializer because of its multiple known attack surfaces.
Untrusted Mode
As soon as an actor system can connect to another remotely, it may in principle send any possible message to any
actor contained within that remote system. One example may be sending a PoisonPill to the system guardian,
shutting that system down. This is not always desired, and it can be disabled with the following setting:
akka.remote.artery.untrusted-mode = on
6.10. Remoting (codename Artery)
363
Akka Scala Documentation, Release 2.5.0-RC1
This disallows sending of system messages (actor life-cycle commands, DeathWatch, etc.) and any message
extending PossiblyHarmful to the system on which this flag is set. Should a client send them nonetheless
they are dropped and logged (at DEBUG level in order to reduce the possibilities for a denial of service attack).
PossiblyHarmful covers the predefined messages like PoisonPill and Kill, but it can also be added as
a marker trait to user-defined messages.
Warning: Untrusted mode does not give full protection against attacks by itself. It makes it slightly harder to
perform malicious or unintended actions but it should be complemented with disabled Java serializer. Additional protection can be achieved when running in an untrusted network by network security (e.g. firewalls).
Messages sent with actor selection are by default discarded in untrusted mode, but permission to receive actor
selection messages can be granted to specific actors defined in configuration:
akka.remote.artery..trusted-selection-paths = ["/user/receptionist", "/user/
˓→namingService"]
The actual message must still not be of type PossiblyHarmful.
In summary, the following operations are ignored by a system configured in untrusted mode when incoming via
the remoting layer:
• remote deployment (which also means no remote supervision)
• remote DeathWatch
• system.stop(), PoisonPill, Kill
• sending any message which extends from the PossiblyHarmful marker interface, which includes
Terminated
• messages sent with actor selection, unless destination defined in trusted-selection-paths.
Note: Enabling the untrusted mode does not remove the capability of the client to freely choose the target of
its message sends, which means that messages not prohibited by the above rules can be sent to any actor in the
remote system. It is good practice for a client-facing system to only contain a well-defined set of entry point actors,
which then forward requests (possibly after performing validation) to another actor system containing the actual
worker actors. If messaging between these two server-side systems is done using local ActorRef (they can be
exchanged safely between actor systems within the same JVM), you can restrict the messages on this interface by
marking them PossiblyHarmful so that a client cannot forge them.
6.10.5 Quarantine
Akka remoting is using Aeron as underlying message transport. Aeron is using UDP and adds among other
things reliable delivery and session semantics, very similar to TCP. This means that the order of the messages are
preserved, which is needed for the Actor message ordering guarantees. Under normal circumstances all messages
will be delivered but there are cases when messages may not be delivered to the destination:
• during a network partition and the Aeron session is broken, this automatically recovered once the partition
is over
• when sending too many messages without flow control and thereby filling up the outbound send queue
(outbound-message-queue-size config)
• if serialization or deserialization of a message fails (only that message will be dropped)
• if an unexpected exception occurs in the remoting infrastructure
In short, Actor message delivery is “at-most-once” as described in Message Delivery Reliability
Some messages in Akka are called system messages and those cannot be dropped because that would result in
an inconsistent state between the systems. Such messages are used for essentially two features; remote death
6.10. Remoting (codename Artery)
364
Akka Scala Documentation, Release 2.5.0-RC1
watch and remote deployment. These messages are delivered by Akka remoting with “exactly-once” guarantee by
confirming each message and resending unconfirmed messages. If a system message anyway cannot be delivered
the association with the destination system is irrecoverable failed, and Terminated is signaled for all watched
actors on the remote system. It is placed in a so called quarantined state. Quarantine usually does not happen if
remote watch or remote deployment is not used.
Each ActorSystem instance has an unique identifier (UID), which is important for differentiating between
incarnations of a system when it is restarted with the same hostname and port. It is the specific incarnation (UID)
that is quarantined. The only way to recover from this state is to restart one of the actor systems.
Messages that are sent to and received from a quarantined system will be dropped. However, it is possible to send
messages with actorSelection to the address of a quarantined system, which is useful to probe if the system
has been restarted.
An association will be quarantined when:
• Cluster node is removed from the cluster membership.
• Remote failure detector triggers, i.e. remote watch is used. This is different when Akka Cluster is used.
The unreachable observation by the cluster failure detector can go back to reachable if the network partition
heals. A cluster member is not quarantined when the failure detector triggers.
• Overflow of the system message delivery buffer, e.g. because of too many watch requests at the same time
(system-message-buffer-size config).
• Unexpected exception occurs in the control subchannel of the remoting infrastructure.
The UID of the ActorSystem is exchanged in a two-way handshake when the first message is sent to a destination. The handshake will be retried until the other system replies and no other messages will pass through until the
handshake is completed. If the handshake cannot be established within a timeout (handshake-timeout config) the association is stopped (freeing up resources). Queued messages will be dropped if the handshake cannot
be established. It will not be quarantined, because the UID is unknown. New handshake attempt will start when
next message is sent to the destination.
Handshake requests are actually also sent periodically to be able to establish a working connection when the
destination system has been restarted.
Watching Remote Actors
Watching a remote actor is API wise not different than watching a local actor, as described in Lifecycle Monitoring
aka DeathWatch. However, it is important to note, that unlike in the local case, remoting has to handle when a
remote actor does not terminate in a graceful way sending a system message to notify the watcher actor about the
event, but instead being hosted on a system which stopped abruptly (crashed). These situations are handled by the
built-in failure detector.
Failure Detector
Under the hood remote death watch uses heartbeat messages and a failure detector to generate Terminated
message from network failures and JVM crashes, in addition to graceful termination of watched actor.
The heartbeat arrival times is interpreted by an implementation of The Phi Accrual Failure Detector.
The suspicion level of failure is given by a value called phi. The basic idea of the phi failure detector is to express
the value of phi on a scale that is dynamically adjusted to reflect current network conditions.
The value of phi is calculated as:
phi = -log10(1 - F(timeSinceLastHeartbeat))
where F is the cumulative distribution function of a normal distribution with mean and standard deviation estimated
from historical heartbeat inter-arrival times.
In the Remote Configuration you can adjust the akka.remote.watch-failure-detector.threshold
to define when a phi value is considered to be a failure.
6.10. Remoting (codename Artery)
365
Akka Scala Documentation, Release 2.5.0-RC1
A low threshold is prone to generate many false positives but ensures a quick detection in the event of a real
crash. Conversely, a high threshold generates fewer mistakes but needs more time to detect actual crashes.
The default threshold is 10 and is appropriate for most situations. However in cloud environments, such as
Amazon EC2, the value could be increased to 12 in order to account for network issues that sometimes occur on
such platforms.
The following chart illustrates how phi increase with increasing time since the previous heartbeat.
Phi is calculated from the mean and standard deviation of historical inter arrival times. The previous chart is an
example for standard deviation of 200 ms. If the heartbeats arrive with less deviation the curve becomes steeper,
i.e. it is possible to determine failure more quickly. The curve looks like this for a standard deviation of 100 ms.
6.10. Remoting (codename Artery)
366
Akka Scala Documentation, Release 2.5.0-RC1
To be able to survive sudden abnormalities, such as garbage collection pauses and transient network failures the failure detector is configured with a margin, akka.remote.watch-failure-detector.
acceptable-heartbeat-pause. You may want to adjust the Remote Configuration of this depending
on you environment. This is how the curve looks like for acceptable-heartbeat-pause configured to 3
seconds.
6.10. Remoting (codename Artery)
367
Akka Scala Documentation, Release 2.5.0-RC1
6.10.6 Serialization
When using remoting for actors you must ensure that the props and messages used for those actors are serializable. Failing to do so will cause the system to behave in an unintended way.
For more information please see Serialization.
ByteBuffer based serialization
Artery introduces a new serialization mechanism which allows the ByteBufferSerializer to directly write
into a shared java.nio.ByteBuffer instead of being forced to allocate and return an Array[Byte] for
each serialized message. For high-throughput messaging this API change can yield significant performance benefits, so we recommend changing your serializers to use this new mechanism.
This new API also plays well with new versions of Google Protocol Buffers and other serialization libraries, which
gained the ability to serialize directly into and from ByteBuffers.
As the new feature only changes how bytes are read and written, and the rest of the serialization infrastructure
remained the same, we recommend reading the Serialization documentation first.
Implementing an akka.serialization.ByteBufferSerializer works the same way as any other serializer,
trait ByteBufferSerializer {
/**
* Serializes the given object into the `ByteBuffer`.
*/
def toBinary(o: AnyRef, buf: ByteBuffer): Unit
/**
* Produces an object from a `ByteBuffer`, with an optional type-hint;
* the class should be loaded using ActorSystem.dynamicAccess.
*/
def fromBinary(buf: ByteBuffer, manifest: String): AnyRef
}
Implementing a serializer for Artery is therefore as simple as implementing this interface, and binding the serializer
as usual (which is explained in Serialization).
Implementations should typically extend SerializerWithStringManifest and in addition to the
ByteBuffer based toBinary and fromBinary methods also implement the array based toBinary and
fromBinary methods. The array based methods will be used when ByteBuffer is not used, e.g. in Akka
Persistence.
Note that the array based methods can be implemented by delegation like this:
import java.nio.ByteBuffer
import akka.serialization.ByteBufferSerializer
import akka.serialization.SerializerWithStringManifest
˓→
class ExampleByteBufSerializer extends SerializerWithStringManifest with
ByteBufferSerializer {
override def identifier: Int = 1337
override def manifest(o: AnyRef): String = "naive-toStringImpl"
// Implement this method for compatibility with `SerializerWithStringManifest`.
override def toBinary(o: AnyRef): Array[Byte] = {
// in production code, aquire this from a BufferPool
val buf = ByteBuffer.allocate(256)
toBinary(o, buf)
6.10. Remoting (codename Artery)
368
Akka Scala Documentation, Release 2.5.0-RC1
buf.flip()
val bytes = new Array[Byte](buf.remaining)
buf.get(bytes)
bytes
}
// Implement this method for compatibility with `SerializerWithStringManifest`.
override def fromBinary(bytes: Array[Byte], manifest: String): AnyRef =
fromBinary(ByteBuffer.wrap(bytes), manifest)
// Actual implementation in the ByteBuffer versions of to/fromBinary:
override def toBinary(o: AnyRef, buf: ByteBuffer): Unit = ??? // implement
˓→actual logic here
override def fromBinary(buf: ByteBuffer, manifest: String): AnyRef = ??? //
˓→implement actual logic here
}
Disabling the Java Serializer
It is possible to completely disable Java Serialization for the entire Actor system.
Java serialization is known to be slow and prone to attacks of various kinds - it never was designed for high
throughput messaging after all. However, it is very convenient to use, thus it remained the default serialization
mechanism that Akka used to serialize user messages as well as some of its internal messages in previous versions.
Since the release of Artery, Akka internals do not rely on Java serialization anymore (exceptions to that being
java.lang.Throwable and “remote deployment”).
Note: Akka does not use Java Serialization for any of its internal messages. It is highly encouraged to disable
java serialization, so please plan to do so at the earliest possibility you have in your project.
One may think that network bandwidth and latency limit the performance of remote messaging, but serialization
is a more typical bottleneck.
For user messages, the default serializer, implemented using Java serialization, remains available and enabled. We
do however recommend to disable it entirely and utilise a proper serialization library instead in order effectively
utilise the improved performance and ability for rolling deployments using Artery. Libraries that we recommend
to use include, but are not limited to, Kryo by using the akka-kryo-serialization library or Google Protocol Buffers
if you want more control over the schema evolution of your messages.
In order to completely disable Java Serialization in your Actor system you need to add the following configuration
to your application.conf:
akka.actor.allow-java-serialization = off
This will completely disable the use of akka.serialization.JavaSerialization by the Akka Serialization extension, instead DisabledJavaSerializer will be inserted which will fail explicitly if attempts
to use java serialization are made.
It will also enable the above mentioned enable-additional-serialization-bindings.
The log messages emitted by such serializer SHOULD be be treated as potential attacks which the serializer
prevented, as they MAY indicate an external operator attempting to send malicious messages intending to use java
serialization as attack vector. The attempts are logged with the SECURITY marker.
Please note that this option does not stop you from manually invoking java serialization.
Please note that this means that you will have to configure different serializers which will able to handle all of
your remote messages. Please refer to the Serialization documentation as well as ByteBuffer based serialization
to learn how to do this.
6.10. Remoting (codename Artery)
369
Akka Scala Documentation, Release 2.5.0-RC1
6.10.7 Routers with Remote Destinations
It is absolutely feasible to combine remoting with Routing.
A pool of remote deployed routees can be configured as:
akka.actor.deployment {
/parent/remotePool {
router = round-robin-pool
nr-of-instances = 10
target.nodes = ["tcp://app@10.0.0.2:2552", "akka://app@10.0.0.3:2552"]
}
}
This configuration setting will clone the actor defined in the Props of the remotePool 10 times and deploy it
evenly distributed across the two given target nodes.
A group of remote actors can be configured as:
akka.actor.deployment {
/parent/remoteGroup2 {
router = round-robin-group
routees.paths = [
"akka://app@10.0.0.1:2552/user/workers/w1",
"akka://app@10.0.0.2:2552/user/workers/w1",
"akka://app@10.0.0.3:2552/user/workers/w1"]
}
}
This configuration setting will send messages to the defined remote actor paths. It requires that you create the
destination actors on the remote nodes with matching paths. That is not done by the router.
6.10.8 Remoting Sample
There is a more extensive remote example that comes with Lightbend Activator. The tutorial named Akka Remote
Samples with Scala demonstrates both remote deployment and look-up of remote actors.
6.10.9 Performance tuning
Dedicated subchannel for large messages
All the communication between user defined remote actors are isolated from the channel of Akka internal messages so a large user message cannot block an urgent system message. While this provides good isolation for Akka
services, all user communications by default happen through a shared network connection (an Aeron stream).
When some actors send large messages this can cause other messages to suffer higher latency as they need to wait
until the full message has been transported on the shared channel (and hence, shared bottleneck). In these cases it
is usually helpful to separate actors that have different QoS requirements: large messages vs. low latency.
Akka remoting provides a dedicated channel for large messages if configured. Since actor message ordering must
not be violated the channel is actually dedicated for actors instead of messages, to ensure all of the messages arrive
in send order. It is possible to assign actors on given paths to use this dedicated channel by using path patterns
that have to be specified in the actor system’s configuration on both the sending and the receiving side:
akka.remote.artery.large-message-destinations = [
"/user/largeMessageActor",
"/user/largeMessagesGroup/*",
"/user/anotherGroup/*/largeMesssages",
"/user/thirdGroup/**",
]
6.10. Remoting (codename Artery)
370
Akka Scala Documentation, Release 2.5.0-RC1
This means that all messages sent to the following actors will pass through the dedicated, large messages channel:
• /user/largeMessageActor
• /user/largeMessageActorGroup/actor1
• /user/largeMessageActorGroup/actor2
• /user/anotherGroup/actor1/largeMessages
• /user/anotherGroup/actor2/largeMessages
• /user/thirdGroup/actor3/
• /user/thirdGroup/actor4/actor5
Messages destined for actors not matching any of these patterns are sent using the default channel as before.
External, shared Aeron media driver
The Aeron transport is running in a so called media driver. By default, Akka starts the media driver embedded in
the same JVM process as application. This is convenient and simplifies operational concerns by only having one
process to start and monitor.
The media driver may use rather much CPU resources. If you run more than one Akka application JVM on the
same machine it can therefore be wise to share the media driver by running it as a separate process.
The media driver has also different resource usage characteristics than a normal application and it can therefore
be more efficient and stable to run the media driver as a separate process.
Given that Aeron jar files are in the classpath the standalone media driver can be started with:
java io.aeron.driver.MediaDriver
The needed classpath:
Agrona-0.5.4.jar:aeron-driver-1.0.1.jar:aeron-client-1.0.1.jar
You find those jar files on maven central, or you can create a package with your preferred build tool.
You can pass Aeron properties as command line -D system properties:
-Daeron.dir=/dev/shm/aeron
You can also define Aeron properties in a file:
java io.aeron.driver.MediaDriver config/aeron.properties
An example of such a properties file:
aeron.mtu.length=16384
aeron.socket.so_sndbuf=2097152
aeron.socket.so_rcvbuf=2097152
aeron.rcv.buffer.length=16384
aeron.rcv.initial.window.length=2097152
agrona.disable.bounds.checks=true
aeron.threading.mode=SHARED_NETWORK
# low latency settings
#aeron.threading.mode=DEDICATED
#aeron.sender.idle.strategy=org.agrona.concurrent.BusySpinIdleStrategy
#aeron.receiver.idle.strategy=org.agrona.concurrent.BusySpinIdleStrategy
# use same director in akka.remote.artery.advanced.aeron-dir config
6.10. Remoting (codename Artery)
371
Akka Scala Documentation, Release 2.5.0-RC1
# of the Akka application
aeron.dir=/dev/shm/aeron
Read more about the media driver in the Aeron documentation.
To use the external media driver from the Akka application you need to define the following two configuration
properties:
akka.remote.artery.advanced {
embedded-media-driver = off
aeron-dir = /dev/shm/aeron
}
The aeron-dir must match the directory you started the media driver with, i.e. the aeron.dir property.
Several Akka applications can then be configured to use the same media driver by pointing to the same directory.
Note that if the media driver process is stopped the Akka applications that are using it will also be stopped.
Aeron Tuning
See Aeron documentation about Performance Testing.
Fine-tuning CPU usage latency tradeoff
Artery has been designed for low latency and as a result it can be CPU hungry when the system is mostly idle.
This is not always desirable. It is possible to tune the tradeoff between CPU usage and latency with the following
configuration:
# Values can be from 1 to 10, where 10 strongly prefers low latency # and 1 strongly prefers less CPU
usage akka.remote.artery.advanced.idle-cpu-level = 1
By setting this value to a lower number, it tells Akka to do longer “sleeping” periods on its thread dedicated for
spin-waiting and hence reducing CPU load when there is no immediate task to execute at the cost of a longer
reaction time to an event when it actually happens. It is worth to be noted though that during a continuously
high-throughput period this setting makes not much difference as the thread mostly has tasks to execute. This also
means that under high throughput (but below maximum capacity) the system might have less latency than at low
message rates.
6.10.10 Internal Event Log for Debugging (Flight Recorder)
Note: In this version (2.5.0-RC1) the flight-recorder is disabled by default because there is no automatic file name
and path calculation implemented to make it possible to reuse the same file for every restart of the same actor
system without clashing with files produced by other systems (possibly running on the same machine). Currently,
you have to set the path and file names yourself to avoid creating an unbounded number of files and enable flight
recorder manually by adding akka.remote.artery.advanced.flight-recorder.enabled=on to your configuration file.
This a limitation of the current version and will not be necessary in the future.
Emitting event information (logs) from internals is always a tradeoff. The events that are usable for the Akka
developers are usually too low level to be of any use for users and usually need to be fine-grained enough to
provide enough information to be able to debug issues in the internal implementation. This usually means that
these logs are hidden behind special flags and emitted at low log levels to not clutter the log output of the user
system. Unfortunately this means that during production or integration testing these flags are usually off and events
are not available when an actual failure happens - leaving maintainers in the dark about details of the event. To
solve this contradiction, remoting has an internal, high-performance event store for debug events which is always
on. This log and the events that it contains are highly specialized and not directly exposed to users, their primary
purpose is to help the maintainers of Akka to identify and solve issues discovered during daily usage. When you
6.10. Remoting (codename Artery)
372
Akka Scala Documentation, Release 2.5.0-RC1
encounter production issues involving remoting, you can include the flight recorder log file in your bug report to
give us more insight into the nature of the failure.
There are various important features of this event log:
• Flight Recorder produces a fixed size file completely encapsulating log rotation. This means that this file
will never grow in size and will not cause any unexpected disk space shortage in production.
• This file is crash resistant, i.e. its contents can be recovered even if the JVM hosting the ActorSystem
crashes unexpectedly.
• Very low overhead, specialized, binary logging that has no significant overhead and can be safely left enabled for production systems.
The location of the file can be controlled via the akka.remote.artery.advanced.flight-recoder.destination setting
(see akka-remote (artery) for details). By default, a file with the .afr extension is produced in the temporary
directory of the operating system. In cases where the flight recorder casuses issues, it can be disabled by adding
the setting akka.remote.artery.advanced.flight-recorder.enabled=off, although this is not recommended.
6.10.11 Remote Configuration
There are lots of configuration properties that are related to remoting in Akka. We refer to the reference configuration for more information.
Note: Setting properties like the listening IP and port number programmatically is best done by using something
like the following:
ConfigFactory.parseString("akka.remote.artery.canonical.hostname=\"1.2.3.4\"")
.withFallback(ConfigFactory.load());
Akka behind NAT or in a Docker container
In setups involving Network Address Translation (NAT), Load Balancers or Docker containers the hostname and
port pair that Akka binds to will be different than the “logical” host name and port pair that is used to connect to
the system from the outside. This requires special configuration that sets both the logical and the bind pairs for
remoting.
akka {
remote {
artery {
canonical.hostname = my.domain.com
canonical.port = 8000
# external (logical) hostname
# external (logical) port
bind.hostname = local.address # internal (bind) hostname
bind.port = 25520
# internal (bind) port
}
}
}
6.11 Serialization
Akka has a built-in Extension for serialization, and it is both possible to use the built-in serializers and to write
your own.
The serialization mechanism is both used by Akka internally to serialize messages, and available for ad-hoc
serialization of whatever you might need it for.
6.11. Serialization
373
Akka Scala Documentation, Release 2.5.0-RC1
6.11.1 Usage
Configuration
For Akka to know which Serializer to use for what, you need edit your Configuration, in
the “akka.actor.serializers”-section you bind names to implementations of the akka.serialization.
Serializer you wish to use, like this:
akka {
actor {
serializers {
java = "akka.serialization.JavaSerializer"
proto = "akka.remote.serialization.ProtobufSerializer"
myown = "docs.serialization.MyOwnSerializer"
}
}
}
After you’ve bound names to different implementations of Serializer you need to wire which classes should
be serialized using which Serializer, this is done in the “akka.actor.serialization-bindings”-section:
akka {
actor {
serializers {
java = "akka.serialization.JavaSerializer"
proto = "akka.remote.serialization.ProtobufSerializer"
myown = "docs.serialization.MyOwnSerializer"
}
serialization-bindings {
"java.lang.String" = java
"docs.serialization.Customer" = java
"com.google.protobuf.Message" = proto
"docs.serialization.MyOwnSerializable" = myown
"java.lang.Boolean" = myown
}
}
}
You only need to specify the name of an interface or abstract base class of the messages. In case of ambiguity,
i.e. the message implements several of the configured classes, the most specific configured class will be used, i.e.
the one of which all other candidates are superclasses. If this condition cannot be met, because e.g. java.io.
Serializable and MyOwnSerializable both apply and neither is a subtype of the other, a warning will
be issued
Note: If your messages are contained inside of a Scala object, then in order to reference those messages, you will
need use the fully qualified Java class name. For a message named Message contained inside the object named
Wrapper you would need to reference it as Wrapper$Message instead of Wrapper.Message.
Akka provides serializers for java.io.Serializable and protobuf com.google.protobuf.
GeneratedMessage by default (the latter only if depending on the akka-remote module), so normally you
don’t need to add configuration for that; since com.google.protobuf.GeneratedMessage implements
java.io.Serializable, protobuf messages will always be serialized using the protobuf protocol unless
specifically overridden. In order to disable a default serializer, map its marker type to “none”:
akka.actor.serialization-bindings {
"java.io.Serializable" = none
}
6.11. Serialization
374
Akka Scala Documentation, Release 2.5.0-RC1
Verification
If you want to verify that your messages are serializable you can enable the following config option:
akka {
actor {
serialize-messages = on
}
}
Warning: We only recommend using the config option turned on when you’re running tests. It is completely
pointless to have it turned on in other scenarios.
If you want to verify that your Props are serializable you can enable the following config option:
akka {
actor {
serialize-creators = on
}
}
Warning: We only recommend using the config option turned on when you’re running tests. It is completely
pointless to have it turned on in other scenarios.
Programmatic
If you want to programmatically serialize/deserialize using Akka Serialization, here’s some examples:
import akka.actor.{ ActorRef, ActorSystem }
import akka.serialization._
import com.typesafe.config.ConfigFactory
val system = ActorSystem("example")
// Get the Serialization Extension
val serialization = SerializationExtension(system)
// Have something to serialize
val original = "woohoo"
// Find the Serializer for it
val serializer = serialization.findSerializerFor(original)
// Turn it into bytes
val bytes = serializer.toBinary(original)
// Turn it back into an object
val back = serializer.fromBinary(bytes, manifest = None)
// Voilá!
back should be(original)
For more information, have a look at the ScalaDoc for akka.serialization._
6.11. Serialization
375
Akka Scala Documentation, Release 2.5.0-RC1
6.11.2 Customization
So, lets say that you want to create your own Serializer, you saw the docs.serialization.
MyOwnSerializer in the config example above?
Creating new Serializers
First you need to create a class definition of your Serializer like so:
import akka.actor.{ ActorRef, ActorSystem }
import akka.serialization._
import com.typesafe.config.ConfigFactory
class MyOwnSerializer extends Serializer {
// This is whether "fromBinary" requires a "clazz" or not
def includeManifest: Boolean = true
// Pick a unique identifier for your Serializer,
// you've got a couple of billions to choose from,
// 0 - 40 is reserved by Akka itself
def identifier = 1234567
// "toBinary" serializes the given object to an Array of Bytes
def toBinary(obj: AnyRef): Array[Byte] = {
// Put the code that serializes the object here
// ... ...
}
// "fromBinary" deserializes the given array,
// using the type hint (if any, see "includeManifest" above)
def fromBinary(
bytes: Array[Byte],
clazz: Option[Class[_]]): AnyRef = {
// Put your code that deserializes here
// ... ...
}
}
The manifest is a type hint so that the same serializer can be used for different classes. The manifest parameter
in fromBinary is the class of the object that was serialized. In fromBinary you can match on the class and
deserialize the bytes to different objects.
Then you only need to fill in the blanks, bind it to a name in your Configuration and then list which classes that
should be serialized using it.
Serializer with String Manifest
The Serializer illustrated above supports a class based manifest (type hint). For serialization of data that need
to evolve over time the SerializerWithStringManifest is recommended instead of Serializer because the manifest (type hint) is a String instead of a Class. That means that the class can be moved/removed
and the serializer can still deserialize old data by matching on the String. This is especially useful for Persistence.
The manifest string can also encode a version number that can be used in fromBinary to deserialize in different
ways to migrate old data to new domain objects.
If the data was originally serialized with Serializer and in a later version of the system you change
to SerializerWithStringManifest the manifest string will be the full class name if you used
includeManifest=true, otherwise it will be the empty string.
6.11. Serialization
376
Akka Scala Documentation, Release 2.5.0-RC1
This is how a SerializerWithStringManifest looks like:
class MyOwnSerializer2 extends SerializerWithStringManifest {
val CustomerManifest = "customer"
val UserManifest = "user"
val UTF_8 = StandardCharsets.UTF_8.name()
// Pick a unique identifier for your Serializer,
// you've got a couple of billions to choose from,
// 0 - 40 is reserved by Akka itself
def identifier = 1234567
// The manifest (type hint) that will be provided in the fromBinary method
// Use `""` if manifest is not needed.
def manifest(obj: AnyRef): String =
obj match {
case _: Customer => CustomerManifest
case _: User
=> UserManifest
}
// "toBinary" serializes the given object to an Array of Bytes
def toBinary(obj: AnyRef): Array[Byte] = {
// Put the real code that serializes the object here
obj match {
case Customer(name) => name.getBytes(UTF_8)
case User(name)
=> name.getBytes(UTF_8)
}
}
// "fromBinary" deserializes the given array,
// using the type hint
def fromBinary(bytes: Array[Byte], manifest: String): AnyRef = {
// Put the real code that deserializes here
manifest match {
case CustomerManifest =>
Customer(new String(bytes, UTF_8))
case UserManifest =>
User(new String(bytes, UTF_8))
}
}
}
You must also bind it to a name in your Configuration and then list which classes that should be serialized using
it.
It’s recommended to throw java.io.NotSerializableException in fromBinary if the manifest is
unknown. This makes it possible to introduce new message types and send them to nodes that don’t know about
them. This is typically needed when performing rolling upgrades, i.e. running a cluster with mixed versions for
while. NotSerializableException is treated as a transient problem in the TCP based remoting layer. The
problem will be logged and message is dropped. Other exceptions will tear down the TCP connection because it
can be an indication of corrupt bytes from the underlying transport.
Serializing ActorRefs
All ActorRefs are serializable using JavaSerializer, but in case you are writing your own serializer, you might want
to know how to serialize and deserialize them properly. In the general case, the local address to be used depends
on the type of remote address which shall be the recipient of the serialized information. Use Serialization.
serializedActorPath(actorRef) like this:
import akka.actor.{ ActorRef, ActorSystem }
import akka.serialization._
6.11. Serialization
377
Akka Scala Documentation, Release 2.5.0-RC1
import com.typesafe.config.ConfigFactory
// Serialize
// (beneath toBinary)
val identifier: String = Serialization.serializedActorPath(theActorRef)
// Then just serialize the identifier however you like
// Deserialize
// (beneath fromBinary)
val deserializedActorRef = extendedSystem.provider.resolveActorRef(identifier)
// Then just use the ActorRef
This assumes that serialization happens in the context of sending a message through the remote transport. There
are other uses of serialization, though, e.g. storing actor references outside of an actor application (database, etc.).
In this case, it is important to keep in mind that the address part of an actor’s path determines how that actor is
communicated with. Storing a local actor path might be the right choice if the retrieval happens in the same logical
context, but it is not enough when deserializing it on a different network host: for that it would need to include
the system’s remote transport address. An actor system is not limited to having just one remote transport per
se, which makes this question a bit more interesting. To find out the appropriate address to use when sending to
remoteAddr you can use ActorRefProvider.getExternalAddressFor(remoteAddr) like this:
object ExternalAddress extends ExtensionId[ExternalAddressExt] with
˓→ExtensionIdProvider {
override def lookup() = ExternalAddress
override def createExtension(system: ExtendedActorSystem): ExternalAddressExt =
new ExternalAddressExt(system)
override def get(system: ActorSystem): ExternalAddressExt = super.get(system)
}
class ExternalAddressExt(system: ExtendedActorSystem) extends Extension {
def addressFor(remoteAddr: Address): Address =
system.provider.getExternalAddressFor(remoteAddr) getOrElse
(throw new UnsupportedOperationException("cannot send to " + remoteAddr))
}
def serializeTo(ref: ActorRef, remote: Address): String =
ref.path.toSerializationFormatWithAddress(ExternalAddress(extendedSystem).
addressFor(remote))
Note: ActorPath.toSerializationFormatWithAddress differs from toString if the address
does not already have host and port components, i.e. it only inserts address information for local addresses.
toSerializationFormatWithAddress also adds the unique id of the actor, which will change when the
actor is stopped and then created again with the same name. Sending messages to a reference pointing the old
actor will not be delivered to the new actor. If you don’t want this behavior, e.g. in case of long term storage of
the reference, you can use toStringWithAddress, which doesn’t include the unique id.
This requires that you know at least which type of address will be supported by the system which will deserialize
the resulting actor reference; if you have no concrete address handy you can create a dummy one for the right
protocol using Address(protocol, "", "", 0) (assuming that the actual transport used is as lenient as
Akka’s RemoteActorRefProvider).
There is also a default remote address which is the one used by cluster support (and typical systems have just this
one); you can get it like this:
object ExternalAddress extends ExtensionId[ExternalAddressExt] with
˓→ExtensionIdProvider {
6.11. Serialization
378
Akka Scala Documentation, Release 2.5.0-RC1
override def lookup() = ExternalAddress
override def createExtension(system: ExtendedActorSystem): ExternalAddressExt =
new ExternalAddressExt(system)
override def get(system: ActorSystem): ExternalAddressExt = super.get(system)
}
class ExternalAddressExt(system: ExtendedActorSystem) extends Extension {
def addressForAkka: Address = system.provider.getDefaultAddress
}
def serializeAkkaDefault(ref: ActorRef): String =
ref.path.toSerializationFormatWithAddress(ExternalAddress(theActorSystem).
addressForAkka)
Deep serialization of Actors
The recommended approach to do deep serialization of internal actor state is to use Akka Persistence.
6.11.3 A Word About Java Serialization
When using Java serialization without employing the JavaSerializer for the task, you must make sure to
supply a valid ExtendedActorSystem in the dynamic variable JavaSerializer.currentSystem.
This is used when reading in the representation of an ActorRef for turning the string representation into a real
reference. DynamicVariable is a thread-local variable, so be sure to have it set while deserializing anything
which might contain actor references.
6.11.4 Serialization compatibility
It is not safe to mix major Scala versions when using the Java serialization as Scala does not guarantee compatibility and this could lead to very surprising errors.
If using the Akka Protobuf serializers (implicitly with akka.actor.allow-java-serialization =
off or explicitly with enable-additional-serialization-bindings = true) for the internal
Akka messages those will not require the same major Scala version however you must also ensure the serializers used for your own types does not introduce the same incompatibility as Java serialization does.
6.11.5 External Akka Serializers
Akka-protostuff by Roman Levenstein
Akka-quickser by Roman Levenstein
Akka-kryo by Roman Levenstein
Twitter Chill Scala extensions for Kryo (based on Akka Version 2.3.x but due to backwards compatibility of the
Serializer Interface this extension also works with 2.4.x)
6.12 I/O
6.12.1 Introduction
The akka.io package has been developed in collaboration between the Akka and spray.io teams. Its design
combines experiences from the spray-io module with improvements that were jointly developed for more
6.12. I/O
379
Akka Scala Documentation, Release 2.5.0-RC1
general consumption as an actor-based service.
The guiding design goal for this I/O implementation was to reach extreme scalability, make no compromises
in providing an API correctly matching the underlying transport mechanism and to be fully event-driven, nonblocking and asynchronous. The API is meant to be a solid foundation for the implementation of network protocols
and building higher abstractions; it is not meant to be a full-service high-level NIO wrapper for end users.
6.12.2 Terminology, Concepts
The I/O API is completely actor based, meaning that all operations are implemented with message passing instead
of direct method calls. Every I/O driver (TCP, UDP) has a special actor, called a manager that serves as an entry
point for the API. I/O is broken into several drivers. The manager for a particular driver is accessible through the
IO entry point. For example the following code looks up the TCP manager and returns its ActorRef:
import akka.io.{ IO, Tcp }
import context.system // implicitly used by IO(Tcp)
val manager = IO(Tcp)
The manager receives I/O command messages and instantiates worker actors in response. The worker actors
present themselves to the API user in the reply to the command that was sent. For example after a Connect
command sent to the TCP manager the manager creates an actor representing the TCP connection. All operations
related to the given TCP connections can be invoked by sending messages to the connection actor which announces
itself by sending a Connected message.
DeathWatch and Resource Management
I/O worker actors receive commands and also send out events. They usually need a user-side counterpart actor
listening for these events (such events could be inbound connections, incoming bytes or acknowledgements for
writes). These worker actors watch their listener counterparts. If the listener stops then the worker will automatically release any resources that it holds. This design makes the API more robust against resource leaks.
Thanks to the completely actor based approach of the I/O API the opposite direction works as well: a user actor
responsible for handling a connection can watch the connection actor to be notified if it unexpectedly terminates.
Write models (Ack, Nack)
I/O devices have a maximum throughput which limits the frequency and size of writes. When an application tries
to push more data than a device can handle, the driver has to buffer bytes until the device is able to write them.
With buffering it is possible to handle short bursts of intensive writes — but no buffer is infinite. “Flow control”
is needed to avoid overwhelming device buffers.
Akka supports two types of flow control:
• Ack-based, where the driver notifies the writer when writes have succeeded.
• Nack-based, where the driver notifies the writer when writes have failed.
Each of these models is available in both the TCP and the UDP implementations of Akka I/O.
Individual writes can be acknowledged by providing an ack object in the write message (Write in the case of
TCP and Send for UDP). When the write is complete the worker will send the ack object to the writing actor. This
can be used to implement ack-based flow control; sending new data only when old data has been acknowledged.
If a write (or any other command) fails, the driver notifies the actor that sent the command with a special message
(CommandFailed in the case of UDP and TCP). This message will also notify the writer of a failed write,
serving as a nack for that write. Please note, that in a nack-based flow-control setting the writer has to be prepared
for the fact that the failed write might not be the most recent write it sent. For example, the failure notification for
a write W1 might arrive after additional write commands W2 and W3 have been sent. If the writer wants to resend
any nacked messages it may need to keep a buffer of pending messages.
6.12. I/O
380
Akka Scala Documentation, Release 2.5.0-RC1
Warning: An acknowledged write does not mean acknowledged delivery or storage; receiving an ack for a
write simply signals that the I/O driver has successfully processed the write. The Ack/Nack protocol described
here is a means of flow control not error handling. In other words, data may still be lost, even if every write is
acknowledged.
ByteString
To maintain isolation, actors should communicate with immutable objects only. ByteString is an immutable
container for bytes. It is used by Akka’s I/O system as an efficient, immutable alternative the traditional byte
containers used for I/O on the JVM, such as Array[Byte] and ByteBuffer.
ByteString is a rope-like data structure that is immutable and provides fast concatenation and slicing operations (perfect for I/O). When two ByteStrings are concatenated together they are both stored within the
resulting ByteString instead of copying both to a new Array. Operations such as drop and take return
ByteStrings that still reference the original Array, but just change the offset and length that is visible. Great
care has also been taken to make sure that the internal Array cannot be modified. Whenever a potentially unsafe
Array is used to create a new ByteString a defensive copy is created. If you require a ByteString that only
blocks as much memory as necessary for it’s content, use the compact method to get a CompactByteString
instance. If the ByteString represented only a slice of the original array, this will result in copying all bytes in
that slice.
ByteString inherits all methods from IndexedSeq, and it also has some new ones. For more information,
look up the akka.util.ByteString class and it’s companion object in the ScalaDoc.
ByteString also comes with its own optimized builder and iterator classes ByteStringBuilder and
ByteIterator which provide extra features in addition to those of normal builders and iterators.
Compatibility with java.io
A ByteStringBuilder can be wrapped in a java.io.OutputStream via the asOutputStream
method. Likewise, ByteIterator can be wrapped in a java.io.InputStream via asInputStream.
Using these, akka.io applications can integrate legacy code based on java.io streams.
6.12.3 Architecture in-depth
For further details on the design and internal architecture see I/O Layer Design.
6.13 Using TCP
The code snippets through-out this section assume the following imports:
import
import
import
import
akka.actor.{ Actor, ActorRef, Props }
akka.io.{ IO, Tcp }
akka.util.ByteString
java.net.InetSocketAddress
All of the Akka I/O APIs are accessed through manager objects. When using an I/O API, the first step is to acquire
a reference to the appropriate manager. The code below shows how to acquire a reference to the Tcp manager.
import akka.io.{ IO, Tcp }
import context.system // implicitly used by IO(Tcp)
val manager = IO(Tcp)
6.13. Using TCP
381
Akka Scala Documentation, Release 2.5.0-RC1
The manager is an actor that handles the underlying low level I/O resources (selectors, channels) and instantiates
workers for specific tasks, such as listening to incoming connections.
6.13.1 Connecting
object Client {
def props(remote: InetSocketAddress, replies: ActorRef) =
Props(classOf[Client], remote, replies)
}
class Client(remote: InetSocketAddress, listener: ActorRef) extends Actor {
import Tcp._
import context.system
IO(Tcp) ! Connect(remote)
def receive = {
case CommandFailed(_: Connect) =>
listener ! "connect failed"
context stop self
case c @ Connected(remote, local) =>
listener ! c
val connection = sender()
connection ! Register(self)
context become {
case data: ByteString =>
connection ! Write(data)
case CommandFailed(w: Write) =>
// O/S buffer was full
listener ! "write failed"
case Received(data) =>
listener ! data
case "close" =>
connection ! Close
case _: ConnectionClosed =>
listener ! "connection closed"
context stop self
}
}
}
The first step of connecting to a remote address is sending a Connect message to the TCP manager; in addition
to the simplest form shown above there is also the possibility to specify a local InetSocketAddress to bind
to and a list of socket options to apply.
Note: The SO_NODELAY (TCP_NODELAY on Windows) socket option defaults to true in Akka, independently of the OS default settings. This setting disables Nagle’s algorithm, considerably improving latency for
most applications. This setting could be overridden by passing SO.TcpNoDelay(false) in the list of socket
options of the Connect message.
The TCP manager will then reply either with a CommandFailed or it will spawn an internal actor representing
the new connection. This new actor will then send a Connected message to the original sender of the Connect
message.
In order to activate the new connection a Register message must be sent to the connection actor, informing
that one about who shall receive data from the socket. Before this step is done the connection cannot be used, and
there is an internal timeout after which the connection actor will shut itself down if no Register message is
received.
6.13. Using TCP
382
Akka Scala Documentation, Release 2.5.0-RC1
The connection actor watches the registered handler and closes the connection when that one terminates, thereby
cleaning up all internal resources associated with that connection.
The actor in the example above uses become to switch from unconnected to connected operation, demonstrating
the commands and events which are observed in that state. For a discussion on CommandFailed see Throttling
Reads and Writes below. ConnectionClosed is a trait, which marks the different connection close events.
The last line handles all connection close events in the same way. It is possible to listen for more fine-grained
connection close events, see Closing Connections below.
6.13.2 Accepting connections
class Server extends Actor {
import Tcp._
import context.system
IO(Tcp) ! Bind(self, new InetSocketAddress("localhost", 0))
def receive = {
case b @ Bound(localAddress) =>
// do some logging or setup ...
case CommandFailed(_: Bind) => context stop self
case c @ Connected(remote, local) =>
val handler = context.actorOf(Props[SimplisticHandler])
val connection = sender()
connection ! Register(handler)
}
}
To create a TCP server and listen for inbound connections, a Bind command has to be sent to the TCP manager.
This will instruct the TCP manager to listen for TCP connections on a particular InetSocketAddress; the
port may be specified as 0 in order to bind to a random port.
The actor sending the Bind message will receive a Bound message signaling that the server is ready to accept
incoming connections; this message also contains the InetSocketAddress to which the socket was actually
bound (i.e. resolved IP address and correct port number).
From this point forward the process of handling connections is the same as for outgoing connections. The example
demonstrates that handling the reads from a certain connection can be delegated to another actor by naming it as
the handler when sending the Register message. Writes can be sent from any actor in the system to the
connection actor (i.e. the actor which sent the Connected message). The simplistic handler is defined as:
class SimplisticHandler extends Actor {
import Tcp._
def receive = {
case Received(data) => sender() ! Write(data)
case PeerClosed
=> context stop self
}
}
For a more complete sample which also takes into account the possibility of failures when sending please see
Throttling Reads and Writes below.
The only difference to outgoing connections is that the internal actor managing the listen port—the sender of the
Bound message—watches the actor which was named as the recipient for Connected messages in the Bind
message. When that actor terminates the listen port will be closed and all resources associated with it will be
released; existing connections will not be terminated at this point.
6.13. Using TCP
383
Akka Scala Documentation, Release 2.5.0-RC1
6.13.3 Closing connections
A connection can be closed by sending one of the commands Close, ConfirmedClose or Abort to the
connection actor.
Close will close the connection by sending a FIN message, but without waiting for confirmation from the remote
endpoint. Pending writes will be flushed. If the close is successful, the listener will be notified with Closed.
ConfirmedClose will close the sending direction of the connection by sending a FIN message, but data will
continue to be received until the remote endpoint closes the connection, too. Pending writes will be flushed. If the
close is successful, the listener will be notified with ConfirmedClosed.
Abort will immediately terminate the connection by sending a RST message to the remote endpoint. Pending
writes will be not flushed. If the close is successful, the listener will be notified with Aborted.
PeerClosed will be sent to the listener if the connection has been closed by the remote endpoint. Per default,
the connection will then automatically be closed from this endpoint as well. To support half-closed connections
set the keepOpenOnPeerClosed member of the Register message to true in which case the connection
stays open until it receives one of the above close commands.
ErrorClosed will be sent to the listener whenever an error happened that forced the connection to be closed.
All close notifications are sub-types of ConnectionClosed so listeners who do not need fine-grained close
events may handle all close events in the same way.
6.13.4 Writing to a connection
Once a connection has been established data can be sent to it from any actor in the form of a Tcp.
WriteCommand. Tcp.WriteCommand is an abstract class with three concrete implementations:
Tcp.Write The simplest WriteCommand implementation which wraps a ByteString instance and an “ack”
event. A ByteString (as explained in this section) models one or more chunks of immutable in-memory
data with a maximum (total) size of 2 GB (2^31 bytes).
Tcp.WriteFile If you want to send “raw” data from a file you can do so efficiently with the Tcp.WriteFile
command. This allows you do designate a (contiguous) chunk of on-disk bytes for sending across the
connection without the need to first load them into the JVM memory. As such Tcp.WriteFile can
“hold” more than 2GB of data and an “ack” event if required.
Tcp.CompoundWrite Sometimes you might want to group (or interleave) several Tcp.Write and/or Tcp.
WriteFile commands into one atomic write command which gets written to the connection in one go.
The Tcp.CompoundWrite allows you to do just that and offers three benefits:
1. As explained in the following section the TCP connection actor can only handle one single write
command at a time. By combining several writes into one CompoundWrite you can have them be
sent across the connection with minimum overhead and without the need to spoon feed them to the
connection actor via an ACK-based message protocol.
2. Because a WriteCommand is atomic you can be sure that no other actor can “inject” other writes
into your series of writes if you combine them into one single CompoundWrite. In scenarios where
several actors write to the same connection this can be an important feature which can be somewhat
hard to achieve otherwise.
3. The “sub writes” of a CompoundWrite are regular Write or WriteFile commands that themselves can request “ack” events. These ACKs are sent out as soon as the respective “sub write” has
been completed. This allows you to attach more than one ACK to a Write or WriteFile (by combining it with an empty write that itself requests an ACK) or to have the connection actor acknowledge
the progress of transmitting the CompoundWrite by sending out intermediate ACKs at arbitrary
points.
6.13. Using TCP
384
Akka Scala Documentation, Release 2.5.0-RC1
6.13.5 Throttling Reads and Writes
The basic model of the TCP connection actor is that it has no internal buffering (i.e. it can only process one write
at a time, meaning it can buffer one write until it has been passed on to the O/S kernel in full). Congestion needs
to be handled at the user level, for both writes and reads.
For back-pressuring writes there are three modes of operation
• ACK-based: every Write command carries an arbitrary object, and if this object is not Tcp.NoAck then
it will be returned to the sender of the Write upon successfully writing all contained data to the socket. If
no other write is initiated before having received this acknowledgement then no failures can happen due to
buffer overrun.
• NACK-based: every write which arrives while a previous write is not yet completed will be replied to with
a CommandFailed message containing the failed write. Just relying on this mechanism requires the
implemented protocol to tolerate skipping writes (e.g. if each write is a valid message on its own and it
is not required that all are delivered). This mode is enabled by setting the useResumeWriting flag to
false within the Register message during connection activation.
• NACK-based with write suspending: this mode is very similar to the NACK-based one, but once a single
write has failed no further writes will succeed until a ResumeWriting message is received. This message
will be answered with a WritingResumed message once the last accepted write has completed. If the
actor driving the connection implements buffering and resends the NACK’ed messages after having awaited
the WritingResumed signal then every message is delivered exactly once to the network socket.
These write back-pressure models (with the exception of the second which is rather specialised) are demonstrated
in complete examples below. The full and contiguous source is available on GitHub.
For back-pressuring reads there are two modes of operation
• Push-reading: in this mode the connection actor sends the registered reader actor incoming data as soon as
available as Received events. Whenever the reader actor wants to signal back-pressure to the remote TCP
endpoint it can send a SuspendReading message to the connection actor to indicate that it wants to suspend the reception of new data. No Received events will arrive until a corresponding ResumeReading
is sent indicating that the receiver actor is ready again.
• Pull-reading: after sending a Received event the connection actor automatically suspends accepting data
from the socket until the reader actor signals with a ResumeReading message that it is ready to process
more input data. Hence new data is “pulled” from the connection by sending ResumeReading messages.
Note: It should be obvious that all these flow control schemes only work between one writer/reader and one
connection actor; as soon as multiple actors send write commands to a single connection no consistent result can
be achieved.
6.13.6 ACK-Based Write Back-Pressure
For proper function of the following example it is important to configure the connection to remain half-open when
the remote side closed its writing end: this allows the example EchoHandler to write all outstanding data back
to the client before fully closing the connection. This is enabled using a flag upon connection activation (observe
the Register message):
case Connected(remote, local) =>
log.info("received connection from {}", remote)
val handler = context.actorOf(Props(handlerClass, sender(), remote))
sender() ! Register(handler, keepOpenOnPeerClosed = true)
With this preparation let us dive into the handler itself:
// storage omitted ...
class SimpleEchoHandler(connection: ActorRef, remote: InetSocketAddress)
6.13. Using TCP
385
Akka Scala Documentation, Release 2.5.0-RC1
extends Actor with ActorLogging {
import Tcp._
// sign death pact: this actor terminates when connection breaks
context watch connection
case object Ack extends Event
def receive = {
case Received(data) =>
buffer(data)
connection ! Write(data, Ack)
context.become({
case Received(data) => buffer(data)
case Ack
=> acknowledge()
case PeerClosed
=> closing = true
}, discardOld = false)
case PeerClosed => context stop self
}
// storage omitted ...
}
The principle is simple: when having written a chunk always wait for the Ack to come back before sending the
next chunk. While waiting we switch behavior such that new incoming data are buffered. The helper functions
used are a bit lengthy but not complicated:
private def buffer(data: ByteString): Unit = {
storage :+= data
stored += data.size
if (stored > maxStored) {
log.warning(s"drop connection to [$remote] (buffer overrun)")
context stop self
} else if (stored > highWatermark) {
log.debug(s"suspending reading")
connection ! SuspendReading
suspended = true
}
}
private def acknowledge(): Unit = {
require(storage.nonEmpty, "storage was empty")
val size = storage(0).size
stored -= size
transferred += size
storage = storage drop 1
if (suspended && stored < lowWatermark) {
log.debug("resuming reading")
connection ! ResumeReading
suspended = false
}
if (storage.isEmpty) {
if (closing) context stop self
6.13. Using TCP
386
Akka Scala Documentation, Release 2.5.0-RC1
else context.unbecome()
} else connection ! Write(storage(0), Ack)
}
The most interesting part is probably the last: an Ack removes the oldest data chunk from the buffer, and if that
was the last chunk then we either close the connection (if the peer closed its half already) or return to the idle
behavior; otherwise we just send the next buffered chunk and stay waiting for the next Ack.
Back-pressure can be propagated also across the reading side back to the writer on the other end of the connection
by sending the SuspendReading command to the connection actor. This will lead to no data being read from
the socket anymore (although this does happen after a delay because it takes some time until the connection actor
processes this command, hence appropriate head-room in the buffer should be present), which in turn will lead
to the O/S kernel buffer filling up on our end, then the TCP window mechanism will stop the remote side from
writing, filling up its write buffer, until finally the writer on the other side cannot push any data into the socket
anymore. This is how end-to-end back-pressure is realized across a TCP connection.
6.13.7 NACK-Based Write Back-Pressure with Suspending
object EchoHandler {
fi