Akka（8）： 分布式运算：Remoting-远程查找式

  Akka是一种消息驱动运算模式，它实现跨JVM程序运算的方式是通过能跨JVM的消息系统来调动分布在不同JVM上ActorSystem中的Actor进行运算，前题是Akka的地址系统可以支持跨JVM定位。Akka的消息系统最高境界可以实现所谓的Actor位置透明化，这样在Akka编程中就无须关注Actor具体在哪个JVM上运行，分布式Actor编程从方式上跟普通Actor编程就不会有什么区别了。Akka的Remoting是一种点对点的跨JVM消息通道，让一个JVM上ActorSystem中的某个Actor可以连接另一个JVM上ActorSystem中的另一个Actor。两个JVM上的ActorSystem之间只需具备TCP网络连接功能就可以实现Akka Remoting了。Akka-Remoting还没有实现完全的位置透明化，因为用户还必须在代码里或者配置文件里指明目标Actor的具体地址。

Akka-Remoting提供了两种Actor之间的沟通方法：

1、远程查找：通过路径Path查找在远程机上已经创建存在的Actor，获取ActorRef后进行沟通

2、远程创建：在远程机上直接创建Actor作为沟通对象

Akka-Remoting的主要应用应该是把一些任务部署到远程机上去运算。发起方（Local JVM）在这里面的主要作用是任务分配，有点像Akka-Router。我们可以用下面的例子来示范：模拟一个计算器，可以进行连续的加减乘除，保留累计结果。我们会把这个计算器部署到远程机上，然后从本机与之沟通分配运算任务及获取运算结果。这个计算器就是个简单的Actor：

import akka.actor._

object Calculator {
  sealed trait MathOps
  case class Num(dnum: Double) extends MathOps
  case class Add(dnum: Double) extends MathOps
  case class Sub(dnum: Double) extends MathOps
  case class Mul(dnum: Double) extends MathOps
  case class Div(dnum: Double) extends MathOps

  sealed trait CalcOps
  case object Clear extends CalcOps
  case object GetResult extends CalcOps
  
}

class Calcultor extends Actor {
  import Calculator._
  var result: Double = 0.0   //internal state

  override def receive: Receive = {
    case Num(d) => result = d
    case Add(d) => result += d
    case Sub(d) => result -= d
    case Mul(d) => result *= d
    case Div(d) => result = result / d

    case Clear => result = 0.0
    case GetResult =>
      sender() ! s"Result of calculation is: $result"
  }


}

就是一个简单的Actor实现，跟Remoting没什么关系。

下面我们会在一个远程机上部署这个Calculator Actor。 先看看这个示范的项目结构：remoteLookup/build.sbt

lazy val commonSettings = seq (
  name := "RemoteLookupDemo",
  version := "1.0",
  scalaVersion := "2.11.8",
  libraryDependencies := Seq(
    "com.typesafe.akka" %% "akka-actor" % "2.5.2",
    "com.typesafe.akka" %% "akka-remote" % "2.5.2"
  )
)

lazy val local = (project in file("."))
     .settings(commonSettings)
     .settings(
       name := "localSystem"
     ).aggregate(messages,remote).dependsOn(messages)

lazy val messages = (project in file("messages"))
    .settings(commonSettings)
    .settings(
      name := "commands"
    )

lazy val remote = (project in file("remote"))
    .settings(commonSettings)
    .settings(
      name := "remoteSystem"
    ).aggregate(messages).dependsOn(messages)

 在这里我们分了三个项目：local是主项目，messages和remote是分项目（subprojects）。messages里只有OpsMessages.scala一个源文件：

package remoteLookup.messages

object Messages {
  sealed trait MathOps
  case class Num(dnum: Double) extends MathOps
  case class Add(dnum: Double) extends MathOps
  case class Sub(dnum: Double) extends MathOps
  case class Mul(dnum: Double) extends MathOps
  case class Div(dnum: Double) extends MathOps

  sealed trait CalcOps
  case object Clear extends CalcOps
  case object GetResult extends CalcOps

}

我们看到：这个文件是把上面的Calculator支持的消息拆了出来。这是因为Calculator Actor会在另一个JVM remote上部署，而我们会从local JVM里向Calculator发送操作消息，所以Messages必须是local和remote共享的。这个要求我们通过dependOn(messages)实现了。现在Calculator是在remote项目里定义的：remote/Calculator.scala

package remoteLookup.remote

import akka.actor._
import remoteLookup.messages.Messages._

object CalcProps {
  def props = Props(new Calcultor)
}

class Calcultor extends Actor with ActorLogging {

  var result: Double = 0.0   //internal state

  override def receive: Receive = {
    case Num(d) => result = d
    case Add(d) => result += d
    case Sub(d) => result -= d
    case Mul(d) => result *= d
    case Div(d) =>
      val _ = result.toInt / d.toInt   //yield ArithmeticException
      result /= d
    case Clear => result = 0.0
    case GetResult =>
      sender() ! s"Result of calculation is: $result"
  }

  override def preRestart(reason: Throwable, message: Option[Any]): Unit = {
    log.info(s"Restarting calculator: ${reason.getMessage}")
    super.preRestart(reason, message)
  }
}

由于ArithmeticException默认的处理策略SupervisorStrategy是Restart，一旦输入Div(0.0)时会重启将result清零。我们可以在remote上加一个Supervisor来把异常处理策略改为Resume。

下面我们先在remote项目本地对Calculator的功能进行测试：remote/CalculatorRunner.scala

package remoteLookup.remote
import akka.actor._
import akka.pattern._
import remoteLookup.messages.Messages._

import scala.concurrent.duration._

class SupervisorActor extends Actor {
  def decider: PartialFunction[Throwable,SupervisorStrategy.Directive] = {
    case _: ArithmeticException => SupervisorStrategy.Resume
  }

  override def supervisorStrategy: SupervisorStrategy =
    OneForOneStrategy(maxNrOfRetries = 5, withinTimeRange = 5 seconds){
      decider.orElse(SupervisorStrategy.defaultDecider)
    }

  val calcActor = context.actorOf(CalcProps.props,"calculator")

  override def receive: Receive = {
    case msg@ _ => calcActor.forward(msg)
  }

}

object CalculatorRunner extends App {

  val remoteSystem = ActorSystem("remoteSystem")
  val calcActor = remoteSystem.actorOf(Props[SupervisorActor],"supervisorActor")

  import remoteSystem.dispatcher

  calcActor ! Clear
  calcActor ! Num(13.0)
  calcActor ! Mul(1.5)

  implicit val timeout = akka.util.Timeout(1 second)

  ((calcActor ? GetResult).mapTo[String]) foreach println
  scala.io.StdIn.readLine()

  calcActor ! Div(0.0)
  calcActor ! Div(1.5)
  calcActor ! Add(100.0)
  ((calcActor ? GetResult).mapTo[String]) foreach println
  scala.io.StdIn.readLine()

  remoteSystem.terminate()

}

测试运算得出以下结果：

Result of calculation is: 19.5

Result of calculation is: 113.0
[WARN] [06/20/2017 19:28:10.720] [remoteSystem-akka.actor.default-dispatcher-4] [akka://remoteSystem/user/parentActor/calculator] / by zero

supervisorActor实现了它应有的功能。

下面进行远程查找示范：首先，remote需要把Calculator向外发布。这可以通过配置文件设置实现：remote/src/main/resources/application.conf

akka {
  actor {
    provider = remote 
  }
  remote {
    enabled-transports = ["akka.remote.netty.tcp"]
    netty.tcp {
      hostname = "127.0.0.1"
      port = 2552
    }
    log-sent-messages = on
    log-received-messages = on
  }
}

上面这段的意思是：所有向外公开Actor的地址前缀为：akka.tcp://[email protected]:2552/user/???

那么Calculator的完整地址path应该就是：akka.tcp://[email protected]:2552/user/supervisorActor/calculator

Akka-Remoting提供了两种远程查找方式：actorSelection.resolveOne方法和Identify消息确认。无论如何，local都需要进行Remoting配置: local/src/main/resources/application.conf

akka {
  actor {
    provider = remote
  }
  remote {
    enabled-transports = ["akka.remote.netty.tcp"]
    netty.tcp {
      hostname = "127.0.0.1"
      port = 0
    }
  }
}

port=0的意思是由系统自动选择任何可用的端口。现在我们完成了Remoting设置，也得到了在远程机上Calculator的具体地址，应该足够进行远程Actor沟通了。我们先用actorSelection.resolveOne示范。resolveOne源代码如下：

/**
   * Resolve the [[ActorRef]] matching this selection.
   * The result is returned as a Future that is completed with the [[ActorRef]]
   * if such an actor exists. It is completed with failure [[ActorNotFound]] if
   * no such actor exists or the identification didn't complete within the
   * supplied `timeout`.
   *
   * Under the hood it talks to the actor to verify its existence and acquire its
   * [[ActorRef]].
   */
  def resolveOne()(implicit timeout: Timeout): Future[ActorRef] = {
    implicit val ec = ExecutionContexts.sameThreadExecutionContext
    val p = Promise[ActorRef]()
    this.ask(Identify(None)) onComplete {
      case Success(ActorIdentity(_, Some(ref))) ⇒ p.success(ref)
      case _                                    ⇒ p.failure(ActorNotFound(this))
    }
    p.future
  }

resolveOne返回Future[ActorRef]，我们可以用Future的函数组件（combinator）来操作：localAccessDemo.scala

import akka.actor._
import akka.util.Timeout
import scala.concurrent.duration._
import akka.pattern._
import remoteLookup.messages.Messages._

object LocalSelectionDemo extends App {


  val localSystem = ActorSystem("localSystem")
  import localSystem.dispatcher

  val path = "akka.tcp://[email protected]:2552/user/supervisorActor/calculator"


      implicit val timeout = Timeout(5 seconds)
      for (calcActor : ActorRef <- localSystem.actorSelection(path).resolveOne()) {

        calcActor ! Clear
        calcActor ! Num(13.0)
        calcActor ! Mul(1.5)
        ((calcActor ? GetResult).mapTo[String]) foreach println

        calcActor ! Div(0.0)
        calcActor ! Div(1.5)
        calcActor ! Add(100.0)
        ((calcActor ? GetResult).mapTo[String]) foreach println

      }


  scala.io.StdIn.readLine()
  localSystem.terminate()

}

因为resolveOne返回的是个Future[x]，我们可以用for来对嵌在Future内的x进行操作。现在remoteSystem只需要构建Calculator待用就行了：remote/CalculatorRunner.scala

package remoteLookup.remote
import akka.actor._
import akka.pattern._
import remoteLookup.messages.Messages._

import scala.concurrent.duration._

class SupervisorActor extends Actor {
  def decider: PartialFunction[Throwable,SupervisorStrategy.Directive] = {
    case _: ArithmeticException => SupervisorStrategy.Resume
  }

  override def supervisorStrategy: SupervisorStrategy =
    OneForOneStrategy(maxNrOfRetries = 5, withinTimeRange = 5 seconds){
      decider.orElse(SupervisorStrategy.defaultDecider)
    }

  val calcActor = context.actorOf(CalcProps.props,"calculator")

  override def receive: Receive = {
    case msg@ _ => calcActor.forward(msg)
  }

}

object CalculatorRunner extends App {

  val remoteSystem = ActorSystem("remoteSystem")
  val calcActor = remoteSystem.actorOf(Props[SupervisorActor],"supervisorActor")
/*
  import remoteSystem.dispatcher

  calcActor ! Clear
  calcActor ! Num(13.0)
  calcActor ! Mul(1.5)

  implicit val timeout = akka.util.Timeout(1 second)

  ((calcActor ? GetResult).mapTo[String]) foreach println
  scala.io.StdIn.readLine()

  calcActor ! Div(0.0)
  calcActor ! Div(1.5)
  calcActor ! Add(100.0)
  ((calcActor ? GetResult).mapTo[String]) foreach println
  */
  scala.io.StdIn.readLine()
  remoteSystem.terminate()

}

注意：注销的操作转移到了localSelectionDemo里。

先运行remote项目：

INFO] [06/20/2017 21:24:37.955] [main] [akka.remote.Remoting] Starting remoting
[INFO] [06/20/2017 21:24:38.091] [main] [akka.remote.Remoting] Remoting started; listening on addresses :[akka.tcp://[email protected]:2552]
[INFO] [06/20/2017 21:24:38.092] [main] [akka.remote.Remoting] Remoting now listens on addresses: [akka.tcp://[email protected]:2552]

remoteSystem开始监视配置的公开地址。

用sbt run 运行local：

Result of calculation is: 19.5
Result of calculation is: 113.0

结果正确。supervisorActor的SupervisorStrategy起到了应有的作用。

remote项目输出显示也能证明：

[INFO] [06/20/2017 21:24:37.955] [main] [akka.remote.Remoting] Starting remoting
[INFO] [06/20/2017 21:24:38.091] [main] [akka.remote.Remoting] Remoting started; listening on addresses :[akka.tcp://[email protected]:2552]
[INFO] [06/20/2017 21:24:38.092] [main] [akka.remote.Remoting] Remoting now listens on addresses: [akka.tcp://[email protected]:2552]
[WARN] [06/20/2017 21:27:06.330] [remoteSystem-akka.actor.default-dispatcher-4] [akka://remoteSystem/user/supervisorActor/calculator] / by zero
[ERROR] [06/20/2017 21:27:34.176] [remoteSystem-akka.remote.default-remote-dispatcher-5] [akka.tcp://[email protected]:2552/system/endpointManager/reliableEndpointWriter-akka.tcp%3A%2F%2FlocalSystem%40127.0.0.1%3A60601-0/endpointWriter] AssociationError [akka.tcp://[email protected]:2552] <- [akka.tcp://[email protected]:60601]: Error [Shut down address: akka.tcp://[email protected]:60601] [
akka.remote.ShutDownAssociation: Shut down address: akka.tcp://[email protected]:60601
Caused by: akka.remote.transport.Transport$InvalidAssociationException: The remote system terminated the association because it is shutting down.
]&nbsp;

 下面我们试着用Identify消息确认方式来复演上述例子。Akka是如下这样定义有关Identify消息确认的：

/**
 * A message all Actors will understand, that when processed will reply with
 * [[akka.actor.ActorIdentity]] containing the `ActorRef`. The `messageId`
 * is returned in the `ActorIdentity` message as `correlationId`.
 */
@SerialVersionUID(1L)
final case class Identify(messageId: Any) extends AutoReceivedMessage with NotInfluenceReceiveTimeout

/**
 * Reply to [[akka.actor.Identify]]. Contains
 * `Some(ref)` with the `ActorRef` of the actor replying to the request or
 * `None` if no actor matched the request.
 * The `correlationId` is taken from the `messageId` in
 * the `Identify` message.
 */
@SerialVersionUID(1L)
final case class ActorIdentity(correlationId: Any, ref: Option[ActorRef]) {
  if (ref.isDefined && ref.get == null) {
    throw new IllegalArgumentException("ActorIdentity created with ref = Some(null) is not allowed, " +
      "this could happen when serializing with Scala 2.12 and deserializing with Scala 2.11 which is not supported.")
  }

  /**
   * Java API: `ActorRef` of the actor replying to the request or
   * null if no actor matched the request.
   */
  @deprecated("Use getActorRef instead", "2.5.0")
  def getRef: ActorRef = ref.orNull

  /**
   * Java API: `ActorRef` of the actor replying to the request or
   * not defined if no actor matched the request.
   */
  def getActorRef: Optional[ActorRef] = {
    import scala.compat.java8.OptionConverters._
    ref.asJava
  }
}

如果拿上面的例子，我们就会向远程机上的Calculator地址发送Identify(path)，而Calculator返回ActorIdentity消息，参数包括correlationId = path, ref = Calculator的ActorRef。 下面是使用示范代码：

object LocalIdentifyDemo extends App {

  class RemoteCalc extends Actor with ActorLogging {

    val path = "akka.tcp://[email protected]:2552/user/supervisorActor/calculator"

    context.actorSelection(path) ! Identify(path)  //send req for ActorRef

    import context.dispatcher
    implicit val timeout = Timeout(5 seconds)

    override def receive: Receive = {
      case ActorIdentity(p,someRef) if p.equals(path) => 
        someRef foreach { calcActor =>

          calcActor ! Clear
          calcActor ! Num(13.0)
          calcActor ! Mul(1.5)
          ((calcActor ? GetResult).mapTo[String]) foreach println

          calcActor ! Div(0.0)
          calcActor ! Div(1.5)
          calcActor ! Add(100.0)
          ((calcActor ? GetResult).mapTo[String]) foreach println

        }
    }

  }

  val localSystem = ActorSystem("localSystem")
  val localActor = localSystem.actorOf(Props[RemoteCalc],"localActor")

  scala.io.StdIn.readLine()
  localSystem.terminate()

}

Identify消息确认机制是一种Actor沟通模式，所以我们需要构建一个RemoteCalc Actor，把程序包嵌在这个Actor里面。当receive收到确认消息ActorIdentity后获取ActorRef运算程序。

查看运算结果，正确。

下面是这次示范的完整源代码：

build.sbt

lazy val commonSettings = seq (
  name := "RemoteLookupDemo",
  version := "1.0",
  scalaVersion := "2.11.8",
  libraryDependencies := Seq(
    "com.typesafe.akka" %% "akka-actor" % "2.5.2",
    "com.typesafe.akka" %% "akka-remote" % "2.5.2"
  )
)


lazy val local = (project in file("."))
     .settings(commonSettings)
     .settings(
       name := "remoteLookupDemo"
     ).aggregate(messages,remote).dependsOn(messages)

lazy val messages = (project in file("messages"))
    .settings(commonSettings)
    .settings(
      name := "commands"
    )

lazy val remote = (project in file("remote"))
    .settings(commonSettings)
    .settings(
      name := "remoteSystem"
    ).aggregate(messages).dependsOn(messages)

messages/OpsMessages.scala

package remoteLookup.messages

object Messages {
  sealed trait MathOps
  case class Num(dnum: Double) extends MathOps
  case class Add(dnum: Double) extends MathOps
  case class Sub(dnum: Double) extends MathOps
  case class Mul(dnum: Double) extends MathOps
  case class Div(dnum: Double) extends MathOps

  sealed trait CalcOps
  case object Clear extends CalcOps
  case object GetResult extends CalcOps

}

remote/src/main/resources/application.conf

akka {
  actor {
    provider = remote
  }
  remote {
    enabled-transports = ["akka.remote.netty.tcp"]
    netty.tcp {
      hostname = "127.0.0.1"
      port = 2552
    }
    log-sent-messages = on
    log-received-messages = on
  }
}

remote/Calculator.scala

package remoteLookup.remote

import akka.actor._
import remoteLookup.messages.Messages._

object CalcProps {
  def props = Props(new Calcultor)
}

class Calcultor extends Actor with ActorLogging {

  var result: Double = 0.0   //internal state

  override def receive: Receive = {
    case Num(d) => result = d
    case Add(d) => result += d
    case Sub(d) => result -= d
    case Mul(d) => result *= d
    case Div(d) =>
      val _ = result.toInt / d.toInt   //yield ArithmeticException
      result /= d
    case Clear => result = 0.0
    case GetResult =>
      sender() ! s"Result of calculation is: $result"
  }

  override def preRestart(reason: Throwable, message: Option[Any]): Unit = {
    log.info(s"Restarting calculator: ${reason.getMessage}")
    super.preRestart(reason, message)
  }
}

remote/CalculatorRunner.scala

package remoteLookup.remote
import akka.actor._
import akka.pattern._
import remoteLookup.messages.Messages._

import scala.concurrent.duration._

class SupervisorActor extends Actor {
  def decider: PartialFunction[Throwable,SupervisorStrategy.Directive] = {
    case _: ArithmeticException => SupervisorStrategy.Resume
  }

  override def supervisorStrategy: SupervisorStrategy =
    OneForOneStrategy(maxNrOfRetries = 5, withinTimeRange = 5 seconds){
      decider.orElse(SupervisorStrategy.defaultDecider)
    }

  val calcActor = context.actorOf(CalcProps.props,"calculator")

  override def receive: Receive = {
    case msg@ _ => calcActor.forward(msg)
  }

}

object CalculatorRunner extends App {

  val remoteSystem = ActorSystem("remoteSystem")
  val calcActor = remoteSystem.actorOf(Props[SupervisorActor],"supervisorActor")
/*
  import remoteSystem.dispatcher

  calcActor ! Clear
  calcActor ! Num(13.0)
  calcActor ! Mul(1.5)

  implicit val timeout = akka.util.Timeout(1 second)

  ((calcActor ? GetResult).mapTo[String]) foreach println
  scala.io.StdIn.readLine()

  calcActor ! Div(0.0)
  calcActor ! Div(1.5)
  calcActor ! Add(100.0)
  ((calcActor ? GetResult).mapTo[String]) foreach println
  */
  scala.io.StdIn.readLine()
  remoteSystem.terminate()

}

local/src/main/resources/application.conf

akka {
  actor {
    provider = remote
  }
  remote {
    enabled-transports = ["akka.remote.netty.tcp"]
    netty.tcp {
      hostname = "127.0.0.1"
      port = 0
    }
  }
}

local/localAccessDemo.scala

import akka.actor._
import akka.util.Timeout
import scala.concurrent.duration._
import akka.pattern._
import remoteLookup.messages.Messages._

object LocalSelectionDemo extends App {


  val localSystem = ActorSystem("localSystem")
  import localSystem.dispatcher

  val path = "akka.tcp://[email protected]:2552/user/supervisorActor/calculator"


      implicit val timeout = Timeout(5 seconds)
      for (calcActor : ActorRef <- localSystem.actorSelection(path).resolveOne()) {

        calcActor ! Clear
        calcActor ! Num(13.0)
        calcActor ! Mul(1.5)
        ((calcActor ? GetResult).mapTo[String]) foreach println

        calcActor ! Div(0.0)
        calcActor ! Div(1.5)
        calcActor ! Add(100.0)
        ((calcActor ? GetResult).mapTo[String]) foreach println

      }

  scala.io.StdIn.readLine()
  localSystem.terminate()

}

object LocalIdentifyDemo extends App {

  class RemoteCalc extends Actor with ActorLogging {

    val path = "akka.tcp://[email protected]:2552/user/supervisorActor/calculator"

    context.actorSelection(path) ! Identify(path)  //semd req for ActorRef

    import context.dispatcher
    implicit val timeout = Timeout(5 seconds)

    override def receive: Receive = {
      case ActorIdentity(p,someRef) if p.equals(path) =>
        someRef foreach { calcActor =>

          calcActor ! Clear
          calcActor ! Num(13.0)
          calcActor ! Mul(1.5)
          ((calcActor ? GetResult).mapTo[String]) foreach println

          calcActor ! Div(0.0)
          calcActor ! Div(1.5)
          calcActor ! Add(100.0)
          ((calcActor ? GetResult).mapTo[String]) foreach println

        }
    }

  }

  val localSystem = ActorSystem("localSystem")
  val localActor = localSystem.actorOf(Props[RemoteCalc],"localActor")

  scala.io.StdIn.readLine()
  localSystem.terminate()

}