我有这种情况需要使用迭代器,对于调用函数f(item)的每个项目并返回Future[Unit]
。
但是,我需要让每个f(item)
调用按顺序执行,它们不能并行运行。
for(item <- it)
f(item)
不起作用,因为这会并行启动所有调用。
我该怎么做,以便按顺序进行?
答案 0 :(得分:40)
如果您不介意非常本地化的var
,您可以按如下方式序列化异步处理(每个f(item)
)(flatMap
进行序列化):
val fSerialized = {
var fAccum = Future{()}
for(item <- it) {
println(s"Processing ${item}")
fAccum = fAccum flatMap { _ => f(item) }
}
fAccum
}
fSerialized.onComplete{case resTry => println("All Done.")}
一般情况下,避免Await
操作 - 它们会阻止(有点像异步,消耗资源和设计草率,会死锁)
酷技巧1:
您可以通过通常的嫌疑人Futures
将flatmap
链接在一起 - 它会序列化异步操作。它有什么不能做的吗? ; - )
def f1 = Future { // some background running logic here...}
def f2 = Future { // other background running logic here...}
val fSerialized: Future[Unit] = f1 flatMap(res1 => f2)
fSerialized.onComplete{case resTry => println("Both Done: Success=" + resTry.isSuccess)}
以上都没有 - 主线程在几十纳秒内直接运行。在所有情况下都使用Futures来执行并行线程并跟踪异步状态/结果和链逻辑。
fSerialized
表示链接在一起的两个不同异步操作的组合。一旦评估了val,它就会立即启动f1
(异步运行)。 f1
像任何Future
一样运行 - 当它最终完成时,它会调用它的onComplete
回调块。这是很酷的位 - flatMap
将它的参数安装为f1
onComplete回调块 - 所以f2
在f1
完成后立即启动,没有阻塞,轮询或浪费资源用法。完成f2
后,fSerialized
即完成 - 因此它会运行fSerialized.onComplete
回调块 - 打印“两个完成”。
不仅如此,您还可以使用整洁的非意大利面条代码尽可能多地链接平面地图
f1 flatmap(res1 => f2) flatMap(res2 => f3) flatMap(res3 => f4) ...
如果你是通过Future.onComplete来做的,你必须将连续的操作嵌入到嵌套的onComplete图层中:
f1.onComplete{case res1Try =>
f2
f2.onComplete{case res2Try =>
f3
f3.onComplete{case res3Try =>
f4
f4.onComplete{ ...
}
}
}
}
不太好。
测试证明:
import scala.concurrent.Future
import scala.concurrent.ExecutionContext.Implicits.global
import scala.concurrent.blocking
import scala.concurrent.duration._
def f(item: Int): Future[Unit] = Future{
print("Waiting " + item + " seconds ...")
Console.flush
blocking{Thread.sleep((item seconds).toMillis)}
println("Done")
}
val fSerial = f(4) flatMap(res1 => f(16)) flatMap(res2 => f(2)) flatMap(res3 => f(8))
fSerial.onComplete{case resTry => println("!!!! That's a wrap !!!! Success=" + resTry.isSuccess)}
酷技巧2:
这样的理解:
for {a <- aExpr; b <- bExpr; c <- cExpr; d <- dExpr} yield eExpr
只不过是语法糖:
aExpr.flatMap{a => bExpr.flatMap{b => cExpr.flatMap{c => dExpr.map{d => eExpr} } } }
这是一个flatMaps链,后面是最终地图。
这意味着
f1 flatmap(res1 => f2) flatMap(res2 => f3) flatMap(res3 => f4) map(res4 => "Did It!")
与
相同for {res1 <- f1; res2 <- f2; res3 <- f3; res4 <- f4} yield "Did It!"
测试证明(继之前的测试之后):
val fSerial = for {res1 <- f(4); res2 <- f(16); res3 <- f(2); res4 <- f(8)} yield "Did It!"
fSerial.onComplete{case resTry => println("!!!! That's a wrap !!!! Success=" + resTry.isSuccess)}
不那么酷的技巧3:
不幸的是你无法混合迭代器和放大器。期货在同样的理解中。编译错误:
val fSerial = {for {nextItem <- itemIterable; nextRes <- f(nextItem)} yield "Did It"}.last
嵌套fors会带来挑战。以下不是序列化,而是并行运行异步块(嵌套的理解不会将后续的Futures与flatMap / Map链接,而是链为Iterable.flatMap {item =&gt; f(item)} - 不一样!)
val fSerial = {for {nextItem <- itemIterable} yield
for {nextRes <- f(nextItem)} yield "Did It"}.last
同样使用foldLeft / foldRight和flatMap并不像你期望的那样工作 - 似乎是一个错误/限制;所有异步块都是并行处理的(因此Iterator.foldLeft/Right
与Future.flatMap
不合作):
import scala.concurrent.Future
import scala.concurrent.ExecutionContext.Implicits.global
import scala.concurrent.blocking
import scala.concurrent.duration._
def f(item: Int): Future[Unit] = Future{
print("Waiting " + item + " seconds ...")
Console.flush
blocking{Thread.sleep((item seconds).toMillis)}
println("Done")
}
val itemIterable: Iterable[Int] = List[Int](4, 16, 2, 8)
val empty = Future[Unit]{()}
def serialize(f1: Future[Unit], f2: Future[Unit]) = f1 flatMap(res1 => f2)
//val fSerialized = itemIterable.iterator.foldLeft(empty){(fAccum, item) => serialize(fAccum, f(item))}
val fSerialized = itemIterable.iterator.foldRight(empty){(item, fAccum) => serialize(fAccum, f(item))}
fSerialized.onComplete{case resTry => println("!!!! That's a wrap !!!! Success=" + resTry.isSuccess)}
但这有效(涉及var):
import scala.concurrent.Future
import scala.concurrent.ExecutionContext.Implicits.global
import scala.concurrent.blocking
import scala.concurrent.duration._
def f(item: Int): Future[Unit] = Future{
print("Waiting " + item + " seconds ...")
Console.flush
blocking{Thread.sleep((item seconds).toMillis)}
println("Done")
}
val itemIterable: Iterable[Int] = List[Int](4, 16, 2, 8)
var fSerial = Future{()}
for {nextItem <- itemIterable} fSerial = fSerial.flatMap(accumRes => f(nextItem))
答案 1 :(得分:34)
def seqFutures[T, U](items: TraversableOnce[T])(yourfunction: T => Future[U]): Future[List[U]] = {
items.foldLeft(Future.successful[List[U]](Nil)) {
(f, item) => f.flatMap {
x => yourfunction(item).map(_ :: x)
}
} map (_.reverse)
}
如果您按顺序运行,因为资源限制阻止一次运行多个Future
,则创建和使用仅包含单个线程的自定义ExecutionContext
可能更容易。
答案 2 :(得分:5)
另一个选择是使用Akka Streams:
val doneFuture = Source
.fromIterator(() => it)
.mapAsync(parallelism = 1)(f)
.runForeach{identity}
答案 3 :(得分:0)
此代码向您展示如何使用简单的承诺按顺序运行期货来完成它。
代码包含两个顺序器,一个逐个执行工作,另一个允许您指定同时运行的数量。
例外情况无法保持简单。
import scala.concurrent.{Await, Future, Promise}
import scala.util.Try
import scala.concurrent.ExecutionContext.Implicits.global
import scala.concurrent.duration.Duration
/**
* Simple class to encapsulate work, the important element here is the future
* you can ignore the rest
*/
case class Work(id:String, workTime:Long = 100) {
def doWork(): Future[String] = Future {
println(s"Starting $id")
Thread.sleep(workTime)
println(s"End $id")
s"$id ready"
}
}
/**
* SimpleSequencer is the one by one execution, the promise is the element
* who allow to the sequencer to work, pay attention to it.
*
* Exceptions are ignore, this is not production code
*/
object SimpleSequencer {
private def sequence(works:Seq[Work], results:Seq[String], p:Promise[Seq[String]]) : Unit = {
works match {
case Nil => p.tryComplete(Try(results))
case work::tail => work.doWork() map {
result => sequence(tail, results :+ result, p)
}
}
}
def sequence(works:Seq[Work]) : Future[Seq[String]] = {
val p = Promise[Seq[String]]()
sequence(works, Seq.empty, p)
p.future
}
}
/**
* MultiSequencer fire N works at the same time
*/
object MultiSequencer {
private def sequence(parallel:Int, works:Seq[Work], results:Seq[String], p:Promise[Seq[String]]) : Unit = {
works match {
case Nil => p.tryComplete(Try(results))
case work =>
val parallelWorks: Seq[Future[String]] = works.take(parallel).map(_.doWork())
Future.sequence(parallelWorks) map {
result => sequence(parallel, works.drop(parallel), results ++ result, p)
}
}
}
def sequence(parallel:Int, works:Seq[Work]) : Future[Seq[String]] = {
val p = Promise[Seq[String]]()
sequence(parallel, works, Seq.empty, p)
p.future
}
}
object Sequencer {
def main(args: Array[String]): Unit = {
val works = Seq.range(1, 10).map(id => Work(s"w$id"))
val p = Promise[Unit]()
val f = MultiSequencer.sequence(4, works) map {
resultFromMulti =>
println(s"MultiSequencer Results: $resultFromMulti")
SimpleSequencer.sequence(works) map {
resultsFromSimple =>
println(s"MultiSequencer Results: $resultsFromSimple")
p.complete(Try[Unit]())
}
}
Await.ready(p.future, Duration.Inf)
}
}
答案 4 :(得分:0)
也许更优雅的解决方案是使用递归,如下面详述。
这可以作为返回Future的长操作的示例:
def longOperation(strToReturn: String): Future[String] = Future {
Thread.sleep(5000)
strToReturn
}
以下是遍历要处理的项目的递归函数,并按顺序处理它们:
def processItems(strToReturn: Seq[String]): Unit = strToReturn match {
case x :: xs => longOperation(x).onComplete {
case Success(str) =>
println("Got: " + str)
processItems(xs)
case Failure(_) =>
println("Something went wrong")
processItems(xs)
}
case Nil => println("Done")
}
这是通过让函数递归调用自身,并在Future完成或失败后处理其余项目来完成的。
要开始此活动,请调用'processItems'函数,其中包含一些要处理的项目,如下所示:
processItems(Seq("item1", "item2", "item3"))
答案 5 :(得分:0)
仅仅扩展@ wingedsubmariner的答案,因为最后.reverse
让我烦恼(并为完整性添加了导入语句)
import scala.collection.mutable
import scala.concurrent.{ExecutionContext, Future}
def seqFutures[T, U](xs: TraversableOnce[T])(f: T => Future[U])
(implicit ec: ExecutionContext): Future[List[U]] = {
val resBase = Future.successful(mutable.ListBuffer.empty[U])
xs
.foldLeft(resBase) { (futureRes, x) =>
futureRes.flatMap {
res => f(x).map(res += _)
}
}
.map(_.toList)
}
注意: ListBuffer
有固定时间+=
和.toList
操作
答案 6 :(得分:-6)
您可以使用Await.result :(未经测试的代码)
“等待:用于阻止未来的单例对象(将其结果传输到当前线程)。”
val result = item map {it => Await.result(f(it), Duration.Inf) }