如何并行执行多个任务?

时间:2016-08-02 05:18:13

标签: scala concurrency parallel-processing

我参加了课程Parallel Programming,它显示了并行界面:

def parallel[A, B](taskA: => A, taskB: => B): (A, B) = {
  val ta = taskA
  val tb = task {taskB}
  (ta, tb.join())
}

以下是错误的:

def parallel[A, B](taskA: => A, taskB: => B): (A, B) = {
  val ta = taskB
  val tb = task {taskB}.join()
  (ta, tb)
}

https://gist.github.com/ChenZhongPu/fe389d30626626294306264a148bd2aa

更多地查看界面

它还向我们展示了执行四项任务的正确方法:

def parallel[A, B, C, D](taskA: => A, taskB: => B, taskC: => C, taskD: => D): (A, B, C, D) = {
    val ta = task { taskA }
    val tb = task { taskB }
    val tc = task { taskC }
    val td = taskD
    (ta.join(), tb.join(), tc.join(), td)
}

我的问题:如果我不知道推进的任务数量(任务列表),我该如何正确地为每项任务打电话join

tasks.map(_.join()) // wrong

修改

类似的讨论也发生在Discuss this week's module: Parallel Programming

3 个答案:

答案 0 :(得分:2)

使用framework课程

中的Parallel Programming

您可以实现以下方法:

def parallel[A](tasks: (() => A)*): Seq[A] = {
  if (tasks.isEmpty) Nil
  else {
    val pendingTasks = tasks.tail.map(t => task { t() })
    tasks.head() +: pendingTasks.map(_.join())
  }
}

(请注意,您can't have variable number of by-name arguments - 尽管此can change

然后像这样使用它:

object ParallelUsage {
  def main(args: Array[String]) {
    val start = System.currentTimeMillis()

    // Use a list of tasks:
    val tasks = List(longTask _, longTask _, longTask _, longTask _)
    val results = parallel(tasks: _*)
    println(results)

    // or pass any number of individual tasks directly:
    println(parallel(longTask, longTask, longTask))
    println(parallel(longTask, longTask))
    println(parallel(longTask))
    println(parallel())

    println(s"Done in ${ System.currentTimeMillis() - start } ms")
  }

  def longTask() = {
    println("starting longTask execution")
    Thread.sleep(1000)
    42 + Math.random
  }
}

使用Scala's parallel collections

你不能比这简单:

val tasks = Vector(longTask _, longTask _, longTask _)
val results = tasks.par.map(_()).seq

答案 1 :(得分:0)

Future.sequence启发并作弊。您需要一个Task实施,这也是一个Monad,可以使这个设计工作。

  /** Transforms a `TraversableOnce[Task[A]]` into a `Task[TraversableOnce[A]]`.
   *  Useful for reducing many `Task`s into a single `Task`.
   */
  def parallel[
    A,
    M[X] <: TraversableOnce[X]
  ](in: M[Task[A]])(
    implicit cbf: CanBuildFrom[M[Task[A]], A, M[A]],
    executor: ExecutionContext
  ): Task[M[A]] = {
    in.foldLeft(Task.point(cbf(in))) {
      (fr, fa) => for (r <- fr; a <- fa) yield (r += a)
    }.map(_.result())(executor)
  }

这可以为大多数Scala集合并行执行操作,唯一的条件是Task定义mapflatMap,无论哪个实现都是,因为您可以抽象特定的使用implicit builder构造的集合类型,它是Scala库的内部构件。

答案 2 :(得分:0)

寻找构建parallel()的实用方法,我发现它可以从Future构建。任何使用现代Javascript Promises的人都会熟悉这种范例:

import scala.concurrent.{Await,Future}
import scala.concurrent.duration.Duration
import scala.concurrent.ExecutionContext.Implicits.global

def parallel[A, B](taskA: =>A, taskB: =>B): (A,B) = {
  val fB:Future[B] = Future { taskB }
  val a:A = taskA
  val b:B = Await.result(fB, Duration.Inf)
  (a,b)
}

这会将taskB旋转到它自己的线程并在主线程中执行taskA。我们会taskA并在必要时永远等待fB完成。请注意,我没有使用此设置测试异常,它可能会停滞或行为异常。