akka-stream与actor系统集成

2017-09-24  本文已影响0人  hangscer

一共有四个api:

Source.actorRef

  val stringSourceinFuture=Source.actorRef[String](100,OverflowStrategy.fail) // 缓存最大为100,超出的话,将以失败告终
  val hahaStrSource=stringSourceinFuture.filter(str=>str.startsWith("haha")) //source数据流中把不是以"haha"开头的字符串过滤掉
  val actor=hahaStrSource.to(Sink.foreach(println)).run()
  actor!"asdsadasd"
  actor!"hahaasd"
  actor!Success("ok")// 数据流成功完成并关闭

"how to create a Source that can receive elements later via a method call?"在akka-http中经常遇见Source[T,N]的地方就是对文件上传和下载的功能的编码(文件IO)中,完成file=>Source[ByteString,_]的转化,或者Source(List(1,2,3,4,5))这种hello-world级别的玩具代码中,这些代码中在定义Source时,已经确定流中数据是什么了。那么如何先定义流,而后给流传递数据呢?答案就是Source.actorRef。说明:Source.actorRef没有背压策略(背压简单说就是生产者的生成速率大于消费者处理速率,导致数据积压)。

Sink.actorRef

class MyActor extends Actor{
  override def receive: Receive = {
    case "FIN"=>
      println("完成了哇!!!")
      context.stop(self)
    case str:String =>
      println("msgStr:"+str)
  }
}
......
  val actor=system.actorOf(Props[MyActor],"myActor")
  val sendToActor=Sink.actorRef(actor,onCompleteMessage = "FIN")
  val hahaStringSource=Source.actorRef[String](100,OverflowStrategy.dropHead).filter(str=>str.startsWith("haha"))
  val actorReceive=hahaStringSource.to(sendToActor).run()
  actorReceive!"hahasdsadsa1"
  actorReceive!"hahasdsadsa2"
  actorReceive!"hahasdsadsa3"
  actorReceive!"hahasdsadsa4"
  actorReceive!Success("ok")
//output
msgStr:hahasdsadsa1
msgStr:hahasdsadsa2
msgStr:hahasdsadsa3
msgStr:hahasdsadsa4
完成了哇!!!

Sink作为数据流终端消费节点,常见用法比如Sink.foreach[T](t:T=>Unit)Sink.fold[U,T](z:U)((u:U,t:T)=>U)等等。Sink.actorRef用于指定某个actorRef实例,把本该数据流终端处理的数据全部发送给这个actorRef实例去处理。解释上述程序,Sink,actorRef需要说明哪一个actorRef来接收消息,并且在数据流上游完成时,这个actorRef会接收到什么样的消息作为完成的信号。我们可以看到onCompleteMessage这条消息并没有受到str=>str.startsWith("haha")这过滤条件的作用(同样的,Sink.actorRef没有处理背压功能,数据挤压过多只能按某些策略舍弃,或者直接失败)。

背压处理

以上Source.actorRefSink.actorRef均不支持背压策略。我们可以借助Source.actorPublisher或者Sink.actorPublisher在数据流的上游或者下游处理背压问题,但是需要去继承ActorPublisher[T]ActorSubscriber实现了处理逻辑。

Source.actorPublisher

在数据流上游处自己手动实现背压处理逻辑:

case object JobAccepted
case object JobDenied
case class Job(msg:String)
...
class MyPublisherActor extends ActorPublisher[Job]{
  import akka.stream.actor.ActorPublisherMessage._
  val MAXSize=10
  var buf=Vector.empty[Job]
  override def receive: Receive = {
    case job:Job if buf.size==MAXSize =>
      sender()!JobDenied //超出缓存 拒绝处理
    case job:Job =>
      sender()!JobAccepted //确认处理该任务
      buf.isEmpty&&totalDemand>0 match {
        case true =>
          onNext(job)
        case false=>
          buf:+=job //先向缓存中存放job
          deliverBuf() //当下游存在需求时,再去从缓存中消费job
      }
    case req@Request(n)=>
      deliverBuf()
    case Cancel=>
      context.stop(self)
  }

  def deliverBuf():Unit= totalDemand>0 match {
    case true =>
      totalDemand<=Int.MaxValue match {
        case true =>
          val (use,keep)=buf.splitAt(totalDemand.toInt) //相当于(buf.take(n),buf.drop(n))
          buf=keep
          use.foreach(onNext(_)) //把buf一份两半,前一半发送给下游节点消费,后一半保留
        case false=>
          buf.take(Int.MaxValue).foreach(onNext(_))
          buf=buf.drop(Int.MaxValue)
          deliverBuf() //递归
      }
    case false=>
  }
}
...
val jobSource=Source.actorPublisher[Job](Props[MyPublisherActor])
val jobSourceActor=jobSource.via(Flow[Job].map(job=>Job(job.msg*2))).to(Sink.foreach(println)).run()
jobSourceActor!Job("ha")
jobSourceActor!Job("he")

actorPublisher的函数签名def actorPublisher[T](props: Props): Source[T, ActorRef]。上述代码中totalDemand是由下游消费节点确定。onNext(e)方法在ActorPublisher中定义,作用是将数据传输给下游节点。当然还有onComplete()onError(ex)函数,也是用于通知下游节点作出相应处理。

Sink.actorSubscriber

case class Reply(id:Int)
...
class Worker extends Actor{
  override def receive: Receive = {
    case (id:Int,job:Job)=>
      println("finish job:"+job)
      sender()!Reply(id)
  }
}
...
class CenterSubscriber extends ActorSubscriber{
  val router={ //路由组
    val routees=Vector.fill(3){ActorRefRoutee(context.actorOf(Props[Worker]))}
    Router(RoundRobinRoutingLogic(),routees)
  }
  var buf=Map.empty[Int,Job]
  override def requestStrategy: RequestStrategy = WatermarkRequestStrategy.apply(100)
  import akka.stream.actor.ActorSubscriberMessage._
  override def receive: Receive = {
    case OnNext(job:Job)=>
      val temp=(Random).nextInt(10000)->job
      buf+=temp //记录并下发任务
      router.route(temp,self)
    case OnError(ex)=>
      println("上游发生错误了::"+ex.getMessage)
    case OnComplete=>
      println("该数据流完成使命..")
    case Reply(id)=>
      buf-=id//当处理完成时,删去记录
  }
}
...
val actor=Source.actorPublisher[Job](Props[MyPublisherActor]).to(Sink.actorSubscriber[Job](Props[CenterSubscriber])).run()
actor!Job("job1")
actor!Job("job2")
actor!Job("job3")

ActorSubscriber可以接收如下几种消息类型:OnNext上游来的新消息、OnComplete上游已经结束数据流、OnError上游发生错误以及其他普通类型的消息。继承ActorSubscriber的子类都需要覆写requestStrategy以此来提供请求策略去控制数据流的背压(围绕requestDemand展开,何时向上游请求数据,一次请求多少数据等等问题)。

上一篇下一篇

猜你喜欢

热点阅读