Concurrency, Actor model and Akka

The talk intends to convey the common problems of building concurrent systems and how a concept called Actor Model alleviates some of these problems and helps us build saner concurrent systems. We also see some real-world applications and use a popular actor framework - Akka - to realize the concept of Actor Model in code in concurrent as well as a distributed setup. We will look deeper into Akka actors, their pros and cons and finally some best practices.
Whole talk at: https://www.youtube.com/watch?v=X4U8hlr6jzA

Transcript

1. CONCURRENCY - ACTOR MODEL - AKKA
2. AYUSHI SHARMA Software Engineer, Avalara R&D Worked on: u Data pipeline tool (MDA) u Classification Ops u Search u HS Code Classification
3. INDEX u Why even talk about it – what are the issues? u What is the Actor Model ? u Components of Actor Model u How does it solve the addressed issues? u Akka Actor model u Akka Actor model primitives
4. WHY TALK ABOUT IT? u Encapsulation u Concurrency u Parallelism u Resource management u Failure/ error handling (ext. Fault Tolerance) BANK ACCT 1 ACCT 2 ACCT 3 ACCT 4 W T D W T W T D D
5. Encapsulation Fault tolerance
6. DEMO1
7. ENTER ACTOR MODEL
8. Threads share state, actors share nothing.
9. PRIMITIVES OF ACTOR MODEL u Actor u Message passing u Execution environment
10. u Actor is the fundamental unit of computation which has: • Address • State • Behaviour • Create more actors • Send messages. • Designate what to do with the next message (update it’s local state for next message) • Mailbox Isolated mutable state + Asynchronous messaging
11. 2. MESSAGE PASSING u Communication between actors through messages u Asynchronous u Immutable messages u Best efforts delivery u At-most once u Order not preserved (among actors)
12. 3. EXECUTION ENVIRONMENT u Engine that runs all actors. u Like execution contexts in respective languages.
13. BANK ACTOR BANK MAILBOX ACCT1 ACCT MAILBOX ACCT2 ACCT MAILBOX ACCT3 ACCT MAILBOX D W W D W D T T D2 W1 W2 D1 D3 W3
14. Rewind: How does actor model solve the challenges?
15. u Encapsulation via message passing. u Eliminates race condition u Resource management u Simple to reason u Fault tolerance (let it crash, self heal & responsive) u Exploit potential of modern CPUs u Distribution
16. Threading model Actor model Models****** Behaviour and state Communication Communication Method invocation Message passing State Shared Isolated Concurrency Threads Actors Synchronization Locks, mutexes, semaphores - Control flow Mostly Synchronous Completely asynchronous
18. AKKA ACTOR
19. ACTOR SYSTEM & LIFECYCLE u Collaborating ensemble of actors arranged in a hierarchy. u Initial actors(guardians): • Root • System • User u User actors = sub-tasks u Resource cleanup u Death watch
20. DEMO1
21. CONS OF ACTOR MODEL u Serialization/Deserialization cost involved in message passing. u Overflowing mailbox u Mandates applications manage message guarantee and ordering u Not trivial to test u Code cleanliness
22. ACTOR BEST PRACTICES 1. Event driven and non-blocking 2. Passing immutable messages 3. Do not send behavior within messages 4. Create user actors sparingly
23. DEMO2: Distributed MapReduce using Akka