Download to read offline
![EventLoop in Actor System
asynchronous synchronous
proactive Work-balancing Work-distribution
reactive Work-stealing Work-requesting
SCHEDULER = (STEALING ^ REQUESTING) [+DISTRIBUTION] [+BALANCING]](https://image.slidesharecdn.com/random-170216090627/85/slide-51-320.jpg)
Uploaded bycorehard_by
модели акторов в с++ миф или реальность
The document discusses the actor model in C++, including its fundamental concepts, internal mechanisms, and lifecycle management within thread pools. It examines implementations like CAF and SObjectizer, along with examples of message handling and processing strategies. Additionally, it outlines potential applications of the actor model in various programming contexts and provides references for further exploration.
More Related Content
Similar to модели акторов в с++ миф или реальность
модели акторов в с++ миф или реальность
- 1. Модели акторов вС++ миф или реальность Боргардт Александр Александрович https://github.com/smart-cloud
- 2. What is inthe report: 1. The general concept of the actor model; 2. Consider the example of a solved problem with Productions; 3. The internal device of the actor; 4. A little about the life cycle of threadpool; 5. About Different strategies work threadpool.
- 3.
- 4.
- 5.
- 6. Intro Project Architecture SmartProxy Smart Proxy Smart Proxy Smart Proxy Nginx Nginx Nginx Service Service Service ServiceService Client Client Client
- 7. Smart Proxy Other Service Service Statistic Shard1 Service User Service Statistic Shard N Queue db CACHE IntroPart of the project Architecture
- 8. Intro The componentsthat hold the system EventLoop db EventLoop queue EventLoop network ThreadPool 1 2 3 4 PipeLine PipeLine PipeLine PipeLine PipeLine
- 9.
- 10. ● Single Thread-> Nodejs, python3.4+ ● Actor model -> Erlang, Elixir, jvm family, C# ● Communicating sequential processes -> go ● Software transactional memory -> clojure ● etc ... What models of concurrent programming are?
- 11. Actor Model whatkind of animal?
- 12. Formally, the descriptionof the actor model It has an ID on which the actor can be identified by other actors; May interact with other actors only by sending messages with the identifiers of the actors addresses; Implements its behavior depending on the incoming messages; May create one or more new actors; Can change its state; Can terminate its execution. https://en.wikipedia.org/wiki/Actor_model
- 13. Actor Model Actor C++ OPP Smalltalk Class ≈≠ What looks like an actor?
- 14. What are theimplementation? C++ Actor Framework: CAF https://www.actor-framework.org/ CAF -> КАФ SObjectizer https://github.com/eao197/so-5-5 SObjectizer -> собжектайзер
- 15.
- 16. One of themain ideas of the actor model Actor Actor Actor Message Message Message
- 17. One of themain ideas of the actor model Actorping {}; request response pong {}; does_not_understand{};Actorother_method{};
- 18.
- 19. Supervisor Supervisord Case : restoredafter the fall of the actor Time Supervisord Supervisord
- 20. Supervisor Case : balancingmessaging Supervisord MSG Time
- 21.
- 22.
- 23. Description of aproblem Productions Counters Json User service Case 1: We obtain the unique identifier and increase the counter by one Case 2: Return a list of all the stored hash from uid and the counter service
- 24. Description of possibletypes of messages. ActorCounters put { string }; get_all {}; request response get_all { vector<string,std::size_t> }; put {}; does_not_understand{}; ActorCounters ActorCountersother_method{};
- 25. Scheme of theinitial solution Dispatcher Counters gen-hash Broker IO Client
- 26. Driving the initialdecision to supervisord Supervisord Counters Supervisord gen-hash Broker IO Counters gen-hash Client Client Client
- 27.
- 28. General Schematic Environment Actor Actor ActorActor cooperation ThreadPool
- 29.
- 30.
- 31. Data locality andIsolated HashMap Message(“update”,{1,1,2}) Actor
- 32. Data locality andIsolated HashMap HashMap HashMap Message(“get”) Message(“merge”, )
- 33.
- 34. Data locality andIsolated Shared State HashMap HashMap Message(“get”) Message(“merge”, hm*) or Message(“merge”, hm&)
- 35.
- 36.
- 37. System + Un-boundQueue ≈ Key word:Queueing theory Un-bound Queue in Your systems
- 38. Un-bound Queue inYour systems ... Size ? Size ?
- 39.
- 40. EventLoop in Actorblocking approach void run(args ..., args ...,) { message *msg_ptr = nullptr; while (has_next_message()) { msg_ptr = next_message(); if (msg_ptr != nullptr) { *************************************** auto response = life.invoke(msg_ptr); *************************************** } else { *************************************** } }
- 41. EventLoop in ActorPseudo non-blocking auto run(args ..., size_t max_throughput, args ...,) -> event_type { message *msg_ptr = nullptr; for (size_t handled_msgs = 0; handled_msgs < max_throughput;) { msg_ptr = next_message(); if (msg_ptr != nullptr) { ************************************ auto response = life.invoke(request); ************************************ ++handled_msgs; ************************************ } else { return executable_result::awaiting; } } return executable_result::resume; }
- 42. EventLoop in ActorPseudo non-blocking auto run(args ..., size_tmax_throughput, args ...,) -> event_type { message *msg_ptr = nullptr; for (size_t handled_msgs = 0; handled_msgs < max_throughput;) { msg_ptr = next_message(); if (msg_ptr != nullptr) { ************************************ auto response = life.invoke(request); ************************************ ++handled_msgs; ************************************ } else { return executable_result::awaiting; } } return executable_result::resume; }
- 43. struct magick_method final: public abstract_action { *********************************** response *operator()(request *msg) override final { *********************************** a lot of data processing *********************************** return nullptr; } }; EventLoop in Actor Pseudo non-blocking
- 44. EventLoop in RunOnce autorun_once(args ..., args ...,) { return run(args ..., 1, args ...,) } How to run once?
- 45. General Schematic Environment Actor Actor ActorActor cooperation ThreadPool
- 46. EventLoop in ActorSystem Job Shared Queue Worker Worker Worker Strategy: shared-work
- 47. EventLoop in ActorSystem Strategy: shared-work
- 48. EventLoop in ActorSystem
- 49. EventLoop in ActorSystem asynchronous synchronous proactive Work-balancing Work-distribution reactive Work-stealing Work-requesting And what are the tasks scheduling strategy?
- 50. EventLoop in ActorSystem asynchronous synchronous proactive Work-balancing Work-distribution reactive Work-stealing Work-requesting And what are the tasks scheduling strategy?
- 51. EventLoop in ActorSystem asynchronous synchronous proactive Work-balancing Work-distribution reactive Work-stealing Work-requesting SCHEDULER = (STEALING ^ REQUESTING) [+DISTRIBUTION] [+BALANCING]
- 52. Reality Possible applications: ● back-endgame dev; ● micro-service; ● service; ● macro-service; ● distribution system; ● etc … .
- 53.
- 54.
- 55. What do youwant to change? abstarct_actor local_actor template<ParentActor=local_actor, Supervisor=base_supervisor, async=true > scheduler_actor: public ParentActor, public Supervisor, public executable{}; base_supervisor vs fake_supervisor Pseudo non-blocking aka async vs blocking Bonus slide
- 56.