Scale up and out with Akka.NET

Transcript

1. Scale up and out with Akka.NET Roger Johansson akka.net Akka.NET Founder Twitter: @rogeralsing Github: rogeralsing Mail: rogeralsing@gmail.com Github.com/rogeralsing/presentations
2. Actor Model Three axioms: Send – an actor can send messages to other actors Create – an actor can create new actors Become – an actor can decide how to handle it’s next message ”An island of consistency in a sea of concurrency” ”Shared nothing”, ”Black box” ”Location transparent”, ”Distributable by design”
3. OOP Actor Model Behavior Stateful Synchronous method calls Behavior Stateful Asynchronous message passing OOP vs. Actor Model Why Actors?? The synchronous model has worked nicely for 60 years!?
4. Moore’s Law We can no longer build faster processors Instead, we stack them next to eachother and call them ”cores” 200 300 400 500 1000 1800 2530 3200 3600 2200 2930 3000 3200 3330 3330 3150 3200 3150 3150 3150 1 1 1 1 1 1 1 1 1 2 2 4 4 8 8 16 16 32 32 64 0 10 20 30 40 50 60 70 0 500 1000 1500 2000 2500 3000 3500 4000 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 MHZ and Cores per year Mhz Cores
5. CPU
6. Core Core Core Core
7. Code Invocation The invocation will run on a single core
8. Code Invocation Buying a larger CPU will not change this
9. Code Invocation Multithreading Multithreaded systems let us utilize more than one core
10. Multithreading Multithreaded systems let us utilize more than one core Code Request
11. Scale out When one CPU is no longer enough
12. Classic .NET technologies Elasticity: Scale out: WCF Web API MSMQ Scale up: Parallel Linq TPL – async await Threads OMG-WTF-Difficult
13. akka.net Scale up: Akka.Actor Scale out: Akka.Remote Elasticity: Akka.Cluster
14. Scale up and out are essentially the same That is; we want to execute code ”somewhere”, on a core, on a machine, in a cluster. Why should we have to resort to different technologies to accomplish the same thing?
15. Scale up and out are essentially the same That is; we want to execute code ”somewhere”, on a core, on a machine, in a cluster. Why should we have to resort to different technologies to accomplish the same thing?
16. Event-driven thread ActorRef ActorRef Actor State Supervision Children Mailbox Behavior Transport Location Transparency, Distributed by default Shared Nothing, Black box An island of consistency in a sea of concurrency
17. Thread Pool Multiplex Scheduling Actor1 Actor1 Actor2 Actor2 Actor3 Actor4 Actor4 Actor1 Cheaper than threads, no context switching Actor3 Actor4 Actor2 Actor1 Actor3 Actor2 Time
18. Actor Model Products basered on the actor model: Facebook WhatsApp (Erlang) RabbitMQ (Erlang) CouchDB (Erlang) LinkedIn.com (JVM Akka) Walmart.com (JVM Akka) Azure Service Fabric alternative backend (Akka.NET)
19. Akka.Actor
20. Demo – Create your first actor ActorSystem ReceiveActor Props ActorRef
21. Akka.Remote
23. System 1 System 2 System 3
24. Demo – Build a chat RemoteActorRefProvider ActorSelection
25. Demo – Remote Deployment Deployment configuration RemoteDaemon
26. Akka.Routing
27. A router delegates messages to other ”routee” actors Group routers – delegate messages to your existing actors Pool routers – delegate messages to a dedicated pool of actors Routers
28. BroadcastRouter 1 2 1 2 1 2 1 2 Router Routee2 Routee3 Routee1 Notifies all ”routees”
29. RoundRobinRouter 1 2 1 2 3 3 4 4 Router Routee1 Routee2 Routee3 Scale up! .. Or down!
30. RoundRobinRouter Router Routee1 Routee2 Routee3 Routee1 Routee2 Routee3 Routee1 Routee2 Routee3 Scale up! Scale out! Remote1 Remote2 Remote3 Router Router Router
31. ConsistentHashRouter M Y M Y X A X A Router Routee1 Routee2 Routee3 Affinity between hash index and routee
32. ConsistentHashRouter Id: 789 Payment … Id: 123 Stop … Id:456 Create … Id: 123 Create … Router
33. ConsistentHashRouter Id: 123 Create … Id: 123 Stop … Id:456 Create … Id: 789 Payment … Routee1 Routee2 Routee3 Prevents race conditions Router Id: 789 Payment … Id: 123 Stop … Id:456 Create … Id: 123 Create …
34. Concurrency akka.net Coordinate access to shared resources ”Hacking Starbucks for unlimited coffee”
35. Routee1 Routee2 Routee3 ? ? ? ? Sender ! Similar to broadcast, forwards the first reply to the original sender ScatterGatherFirstCompletedRouter Router
36. Demo – Use routers RoundRobin Deployment configuration
37. Demo – Scale up and out RoundRobin Remote Deployment Executing actors on a GUI Thread
38. Error handling in Java, C# and C
39. Isolate failures in bulkheads Fault tolerance – fail gracefully akka.net
40. Component Consumer 1 Consumer 2
41. Component Website Businesslogic
42. Service Request Application error Supervisor Manage failures Response Validation error Client
43. Supervision Every actor is being supervised by another actor that dictates how failures should be handled A supervisor can decide to: • Force a restart • Terminate • Ignore and resume • Escalate up to it’s own supervisor
44. /user /system / Error Kernel User actor Hierarchy System actor Hierarchy ”System Guardian” ”Guardian” ”Root Guardian”
45. /c2 /b2 /b1 /a1 /a2 /b3 /c3 /c4 /c5 /user /system / Error Kernel /c1 OneForOne supervisor AllForOne supervisor /b1 /b3
47. Become Become or hotswap means that an actor can decide how it should process the next incoming message. In Akka.NET this is handled by the .Become and .Unbecome methods. Very useful for creating state machines
48. Private void Alive(){ …//other code if (this.hitpoints <= 0) this.Become(Dead); … } private void Dead(){ Receive<Resurrect>(_ => { this.hitpoints = this.maxHitpoints; this.Become(Alive); }); }
49. Other features • Persistence • Cluster • Cluster Sharding • Dependency Injection • Finite State Machines • F# API • Reactive Streams
50. Before Akka.NET 200-300 simulated Excel workbooks. Aprox. 7000 Excel celles. Excel compatible formulas. Single user per worksheet.
51. With Akka.NET 800 000 Excel cells. Multiuser, multiple users editing worksheets – google docs style. Two devs rewrote the entire system in two weeks.
52. Before Akka.NET Legacy system, written around 2000. One customer per server, SPOF. Huge amount of synchronization/locking code.
53. With Akka.NET 85% smaller codebase, code focuses on business logic, Akka.NET handles infrastructure. High Availability - Failover with Akka.Cluster.
54. Q & A
55. FAQ Akka/NET vs. Project Orleans http://bit.ly/25KAuzC Akka/NET vs. Message Queues, e.g. NServiceBus http://stackoverflow.com/a/13018198 OT, but how did you get the strange arrows and symbols in the code? https://github.com/tonsky/FiraCode
56. Scale up and out with Akka.NET Roger Johansson akka.net Akka.NET Co-Founder Twitter: @rogeralsing Github: rogeralsing Mail: roger.johansson@betsson.com Github.com/rogeralsing/presentations