Марк Ткаченко "Virtual actors: что, как и зачем"

Поговорим о модели виртуальных акторов в контексте первой их реализации — Microsoft Orleans — что привело к созданию, какие проблемы классической модели решались, какие инструменты имеются на сегодня.

Transcript

1. VIRTUAL ACTORS The What, Why, and How Mark Tkachenko AtlasDelivery.io Moscow, 2019-12-14
2. We use Akka: why? 2 • CQRS + ES • Scalable parallelism • Easier state managment
3. We use Akka: problems 3 • Learning curve • Boilerplate • Language support
4. https://www.karlrupp.net/2018/02/42-years-of-microprocessor-trend-data/
5. Problem: Concurrency 5 • Shared state • Resource management • Communications • Performance
6. Multithreading 6
7. Problem: Scalability 7 • State storage scalability • Resilience • Error handling
8. 3-tier 8
9. 3-tier with cache 9
10. Distributed DB 10
11. Stateful Application 11
12. Actor model: 1973 12
13. Universal primitive of concurrent computation • Maintain its own private state • Send messages • Receive messages • Create more actors 13 Actor
14. Actor hierarchy 14
15. • In-memory • Sequential message processing • No shared state • Asynchronous messaging • Automatic thread management 15 Advantages
16. • Erlang • Facebook WhatsApp (Erlang) • RabbitMQ (Erlang) • CouchDB (Erlang) • LinkedIn (JVM Akka) • Walmart (JVM Akka) • Blizzard (JVM Akka) 16 Usages
17. 17 Classic model problems • Explicit actor lifetime management • Boilerplate • Learning curve
18. 18 C# and Akka.NET
19. 19 C# and Akka.NET
20. 20 C# and Akka.NET
21. 21 F# and Akka.NET (with Akkling)
22. 22 F# and Akka.NET (with Akkling)
23. 23 Virtual actors: 2009
24. 24 Key concepts • Perpetual existence • Automatic instantiation • Location transparency • Automatic scale out
25. 25 Virtual vs Classic Actor Implicit lifetime management Manual create and shutdown Implicit messaging (async method calls) Explicit messaging Automatic Actor placement Manual Actor placement No hierarchy Hierarchy At least once message delivery At most once message delivery You don’t know physical location Address reflects location
27. 27 Orleans Grain Lifecycle
28. 28 Orleans Grain into the wild
29. 29 Actor-based Middle Tier
30. 30 Silo Failure
31. 31 Silo Failure
32. 32 Silo Failure
33. 33 Silo Failure
34. 34 Silo Failure
35. 35 Silo Failure
36. 36 Silo Failure
37. 37 Error handling
38. 38 Error handling
39. 39 Error handling
40. 40 Error handling
41. 41 Error handling
42. 42 Error handling
43. 43 Error handling
44. 44 Grain Versioning
45. 45 Orleans history • Started in Microsoft Research - 2009 • In production – 2011 • Halo 4 – 2012 • Public preview – 2014 • GitHub (MIT) - 2015
46. 46 Other implementations
47. 47 Links • A Universal Modular ACTOR Formalism for Artificial Intelligence • Orleans: Distributed Virtual Actors for Programmability and Scalability • Microsoft Orleans • How to build real-world applications with Orleans • Distributed Transactions are dead
48. 48 Contacts @KiloOscarTango @KiloOscarTango github.com/Rizzen