Akka in action

Akka in action presentation used at Xebicon 2013

Transcript

1. Akka in Action Raymond Roestenburg @RayRoestenburg rroestenburg@xebia.com github.com/RayRoestenburg
2. manning.com/roestenburg 12mp25 for 42% Off! Early Akka adopter (v0.7, 2010) Akka Committer (akka-camel in 2.x) Co-author of Akka in Action
3. “The free lunch is over” / TANSTAAFL Herb Sutter 2005 http://www.gotw.ca/publications/concurrency-ddj.htm Number of transistors Clock Speed
4. Can’t I just.. Parallelize that Sh*t™? Put that in a box! Something like this? Parallel Version 9 Hey Oracle don’t sue me, of course it’s fake
5. Nope. Parallelize that Sh*t™ Not every algorithm can be parallelized Dynamic coordination (concurrent) Amdahl’s Law Sequential fraction determines max speedup
6. Threads? Shared (mutable) state + locks Require synchronization Diminishing returns** Which primitives to use? Difficult to predict & maintain Deadlock Livelock Thread starvation Race condition *Using threads directly is meant here **Performance improvement degrades with more threads than cores.
7. “Life is too short for Blocking code”* *Mario Fusco just the other day on twitter
8. Why Async? Less latency. Ser Service A Service B Service C 4 sec 6 sec 3 sec Sync takes 13 seconds Async takes max 6 seconds Asynchronous ‘first completed response’ is easy to do
9. 1. The world is concurrent 2. Things in the world don't share data 3. Things communicate with messages 4. Things fail Joe Armstrong (creator of Erlang)
10. We need better tools to build.. Reactive* Applications * The latest need-to-know buzz word!
11. event driven scalable asynchronous resilient
12. Scale Up and Out Concurrent Distributed Many Nodes Many Threads Akka A uniform model for both up and out, embraces concurrency and failure Maximize the use of cores _and_ nodes. Concurrent (more Cores) Distributed (more Nodes)
13. Actors Futures Agents
14. Scale Up Processes (even less**) Actors (2.7 million / 1GB) Threads (4096 / 1GB) Actors are small* * Disclaimer: text is not shown to scale, actors are way smaller.. ** Especially JVM processes
15. Reactive Asynchronous Building Blocks 1. CREATE 2. SEND 3. BECOME 4. SUPERVISE*
16. Actor Actor Instance ActorRef Mailbox
17. ActorRef Actor Instance ActorRef Mailbox Points to Actor Delivers messages to Mailbox
18. Mailbox Actor Instance ActorRef Mailbox Processes Messages asynchronously* Invokes Actor Instance with message *The Mailbox runs on a Dispatcher that abstracts threading
19. Actor Instance Actor Instance ActorRef Mailbox Receives messages one at a time Your code runs here.
20. Actor Instance Crashed Actor Instance Mailbox An Actor Instance crashes instead of handling Exceptions. A Supervisor decides it’s fate. ActorRef Restart / Resume / Stop / Escalate New Actor Instance
21. ActorRef Actor Instance Mailbox Immutable. Only the ActorRef is accessible From the outside. ActorRef
22. ActorRef Actor Instance Mailbox The ActorRef is used for both local and remote actors. ActorRef
23. ActorRef Actor Instance Mailbox The ActorRef always points 1 on 1 to the ‘same’ Actor instance. (Alive or Restarted) ActorRef Actor Instance
24. ActorRef Dead Actor Instance Mailbox The ActorRef points to deadletters ActorRef when the actor is dead (terminated). ActorRef deadletters
25. msg msg Immutable messages msg Location Transparency msg Serialized messages Actor System Actor System Actor System Actor systems Event
26. Immutable messages msg msg app@node2 app@node1 app@node3 “app” clustered actor system On three nodes Clustered actor system msg msg
27. “GoTicks.com” Build a REST API PUT /events { “event” : “RHCP”, “nrOfTickets”: 300 } GET /ticket/RHCP Response: { “event”: “RHCP”, “nr”: 1 } GET /events Response: [ { “event”: “RHCP”, “nrOfTickets”: 299 } ]
28. RestApi Ticket Seller Ticket Seller BoxOffice GoTicks.com Spray & Akka “goticks” ActorSystem REST Spray-can
29. GoTicks.com REST Interface Handles HTTP calls Translates to and from JSON RestApi REST Spray-can
30. GoTicks.com BoxOffice REST Spray-can BoxOffice Creates Events Creates TicketSeller One for every Event Partition / shard per event Many sharding strategies possible
31. GoTicks.com Ticket Seller Ticket Seller REST Spray-can TicketSeller Contains list of Tickets Sells Tickets for a Event until SoldOut
32. RestApi Ticket Seller Ticket Seller BoxOffice Clustered GoTicks.com Ticket Seller Ticket Seller BoxOffice Consistent Hashing Consistent Hashing
33. Code
34. 1. CREATE
35. Inside our Main class… val system = new ActorSystem(“goticks”) Create The system ActorSystem(“goticks”)
36. Inside Main… val system = new ActorSystem(“goticks”) val api = system.actorOf( Props[RestInterface], “httpInterface”) … CREATE RestInterface ActorSystem(“goticks”) Top Level Actor ActorSystem(“goticks”) REST Interface
37. Inside Main… val system = new ActorSystem(“goticks”) val api = system.actorOf( Props[RestInterface], “httpInterface”) … CREATE RestInterface ActorSystem(“goticks”) Factory/Creator/ Immutable Configuration object ActorSystem(“goticks”) REST Interface
38. Inside Main… val system = new ActorSystem(“goticks”) val api = system.actorOf( Props[RestInterface], “httpInterface”) … CREATE RestInterface ActorSystem(“goticks”) ActorRef to the Actor ActorSystem(“goticks”) REST Interface
39. ActorSystem(“goticks”) ActorSystem(“goticks”) Inside RestInterface.. val master = context.actorOf( Props[BoxOffice],”boxOffice” ) 1. CREATE BoxOffice REST Interface BoxOffice Child of REST Interface Actor
40. ActorSystem(“goticks”) ActorSystem(“goticks”) Inside RestInterface.. val master = context.actorOf( Props[BoxOffice],”boxOffice” ) 1. CREATE BoxOffice REST Interface BoxOffice An Actor has a name, part of an ActorPath
41. Inside BoxOffice… val child = context.actorOf( Props[TicketSeller], eventName ) 1. CREATE TicketSeller ActorSystem(“goticks”) ActorSystem(“goticks”) REST Interface BoxOffice TicketSeller Child of BoxOffice Actor
42. Inside BoxOffice… val child = context.actorOf( Props[TicketSeller], eventName ) 1. CREATE TicketSeller ActorSystem(“goticks”) ActorSystem(“goticks”) REST Interface BoxOffice TicketSeller Ticket sellers are named by event
43. 2. SEND/RECEIVE
44. Receiving messages class BoxOffice extends Actor { def receive = { case TicketRequest(name) => … case GetEvents => … case event: Event => … } } Pattern Matching On Messages
45. Receiving messages class TicketSeller extends Actor { def receive = soldOut def soldOut:Receive = { case Tickets(newTickets) => .. //store new tickets become(selling) } def selling:Receive = { case BuyTicket(name) => … } } Partial Function 3. BECOME: You can change th behavior at every msg
46. Create Event ActorSystem(“goticks”) ActorSystem(“goticks”) REST Interface BoxOffice Event(“RHCP”, 250) { event: “RHCP” nrOfTickets: 250} Inside RestInterface… (receives JSON, unmarshall to event) boxOffice ! event
47. Create Event ActorSystem(“goticks”) ActorSystem(“goticks”) REST Interface BoxOffice Event(“RHCP”, 250) { event: “RHCP” nrOfTickets: 250} Inside RestInterface… (receives JSON, unmarshall to event) boxOffice ! event Bang Operator ‘Tell’ ‘One-way’
48. ActorSystem(“goticks”) ActorSystem(“goticks”) REST Interface BoxOffice Event(“RHCP”, 250) { event: “RHCP” nrOfTickets: 250} Inside RestInterface… boxOffice ! event Receiver can respond to ‘sender’ Send Event To BoxOffice
49. ActorSystem(“goticks”) ActorSystem(“goticks”) REST Interface BoxOffice EventCreated Inside BoxOffice… sender ! EventCreated Respond with EventCreated, results in OK HTTP OK
50. ActorSystem(“goticks”) ActorSystem(“goticks”) REST Interface BoxOffice { event: “RHCP” nrOfTickets: 250} BuyTicket TicketSeller Inside BoxOffice… ticketSeller forward BuyTicket Get a Ticket Forward keeps sender (RestInterface) TicketRequest(“RHCP”)
51. ActorSystem(“goticks”) ActorSystem(“goticks”) REST Interface BoxOffice { event: “RHCP”, nr :1} TicketSeller Inside TicketSeller… sender ! ticket Respond to original sender BuyTicket Ticket(“RHCP”, 1)
52. CODE
53. Questions? @RayRoestenburg rroestenburg@xebia.com www.xebia.com github.com/RayRoestenburg/xebicon
54. Backup slides
55. “Async sending subscription data” Send Thread Consumer Consumer Consumer Thread Thread Thread Service A
56. Send Thread Consumer Consumer Consumer Thread Thread Thread Service A Thread 2 fails
57. Send Thread Consumer Consumer Consumer Thread Thread Thread Service A Thread 2 fails Handle every possible error inside Thread 2, never fail? Not completely under your control.
58. Send Thread Consumer Consumer Consumer Thread Thread Thread Service A Thread 2 fails Restart / Stop? Monitor? Handle errors across all subscriptions?
59. 4. Supervise* Untangle behavior and errors ‘Just Reboot’ Let It Crash ™ Error Kernel
60. 4. Supervise* Strategies One for One (Poison Pill /Seppuku) All for One
61. 4. Supervise* val supervisorStrategy = OneForOneStrategy(maxNrOfRetries = 5, withinTimeRange = 1 minute) { case _: Exception => Restart }
62. 4. Supervise* val supervisorStrategy = OneForOneStrategy(maxNrOfRetries = 5, withinTimeRange = 1 minute) { case _: Exception => Restart } Restart, Resume, Escalate, Stop,