examples of common concurrency patterns that you can achieve with the scala.concurrent package

scalaConcurrentPatterns.scala

import scala.concurrent._
import scala.concurrent.duration._
import scala.concurrent.ExecutionContext.Implicits.global
import scala.util.Random
import java.util.{Timer, TimerTask}

object Util {
  sealed trait BaseResponse
  case class Response1(res: Int) extends BaseResponse
  case class Response2(res: String) extends BaseResponse
  
  private val timeoutTimer = {
    //in real life, use an Akka scheduler or a Netty HashWheelTimer, because they are more thread-efficient
    new Timer()
  }
  def timeoutFuture[T](timeout: Duration): Future[T] = {
    val prom = Promise[T]
    val timeoutTask = new TimerTask {
      override def run = {
        prom.failure(new TimeoutException(s"timed out after $timeout"))
      }
    }
    timeoutTimer.schedule(timeoutTask, timeout.toMillis)
    prom.future
  }
  
  private val random = new Random(System.currentTimeMillis())
  private val possibleSleepMS = List(100L, 200L, 150L, 250L)
  //creates a future that sleeps for a pseudo-random amount of time, and then executes t.
  //both actions happen on a dedicated thread (because we imported scala.concurrent.ExecutionContext.Implicits.global)
  def sleepAndThen[T](t: => T): Future[T] = {
    Future {
      val sleepLengthIdx = math.abs(random.nextInt) % possibleSleepMS.length
      val sleepLength = possibleSleepMS(sleepLengthIdx)
      Thread.sleep(sleepLength)
      t
    }
  }
}

//fan out/fan in pattern:
//execute a bunch of independent background jobs, each in their own future.
//merge them into one future that will be complete when the slowest one is.
def fanOut = {
  import Util._
  val f1 = sleepAndThen(Response1(1))
  val f2 = sleepAndThen(Response2("one"))
  val f3 = sleepAndThen(Response1(123))

  val fannedOutFutures: List[Future[BaseResponse]] = f1 :: f2 :: f3 :: Nil
  val results: Future[List[BaseResponse]] = Future.sequence(fannedOutFutures)
  results.map { resultList: List[BaseResponse] =>
    //after all the results come back, we can do stuff with them
  }
}

//timeout pattern
//execute a first request and try again if it fails
def timeout = {
  import Util._
  val request = sleepAndThen(Response1(1))
  val timeout = timeoutFuture[Response1](1.second)
  val respFuture = Future.firstCompletedOf(request :: timeout :: Nil)
  respFuture.map { resp =>
    //only executes if the request didn't time out
  }
}

//similar to timeouts - cancelling
//if you have an expensive operation, or maybe even one that runs forever,
//you can cancel it so it doesn't take up resources after we don't need it
def cancel = {
  import Util._
  
  //the variable we set to start the cancellation process
  val cancelVar = Promise[Unit]()
  //the variable that the cancelee sets to ack that it's fully done
  val cancelledVar = Promise[Unit]()
  
  //the handler for each request in our simple server
  def acceptRequest = ()
  
  //start the server
  Future {
    while(!cancelVar.isCompleted) {
      acceptRequest
    }
    cancelledVar.success(())
  }
  
  //stop the server after a while
  sleepAndThen {
    cancelVar.success(())
    while(!cancelledVar.isCompleted) {
      println("still waiting for the server to finish")
    }
  }
}

//firstCompleted
//execute the same request on multiple servers, and return the first that comes back.
//combined with timeouts and cancellation you can tightly control your call
def firstCompleted = {
  import Util._
  def makeRequest = sleepAndThen(Response1(1))
  
  val req1 = makeRequest
  val req2 = makeRequest
  val req3 = makeRequest
  
  val respFuture: Future[Response1] = Future.firstCompletedOf(req1 :: req2 :: req3 :: Nil)
  
  respFuture.map { resp =>
    //continue with the response from the fastest server
  }
}

//channels
//allow *asynchronous* bidirectional communication between two concurrently executing processes.
//scala.concurrency has a synchronous version of this
def channels = {
  import Util._
  //a class that lets you send one message. you can compose these into a queue to send multiple messages
  class OneTimeChannel[T] {
    private val sendProm = Promise[T]()
    private val recvProm = Promise[Unit]()
    
    def send(t: T): Future[Unit] = {
      sendProm.success(t)
      recvProm.future
    }
    def recv: Future[T] = {
      sendProm.future.map { t =>
        recvProm.success(())
        t
      }
    }
  }
  
  val chan = new OneTimeChannel[Int]
  //sender
  sleepAndThen {
    val sent: Future[Unit] = chan.send(123)
    sent.map { _ =>
      //do stuff after we know the value has been sent
    }
  }
  
  //receiver
  sleepAndThen {
    val receivedFuture: Future[Int] = chan.recv
    receivedFuture.map { received =>
      //do stuff with the received value
    }
  }
}

sources of inspiration

• http://talks.golang.org/2012/concurrency.slide#1 - fan-in/fan-out, first completed, timeout, channel
• http://monkey.org/~marius/funsrv.pdf - cancellation, timeout
• https://github.com/twitter/util/blob/master/util-core/src/main/scala/com/twitter/examples/Offer.scala - channels
• https://gist.github.com/andrewconner/6362491 - timeout

your timeoutFuture fails if the timeout is reached. for a timer that just returns Unit upon the timeout being reached, see Futures.alarm https://github.com/scala/scala/blob/v2.10.2/src/actors/scala/actors/Future.scala#L131