Akka源码分析-Remote-收消息

　　上一遍博客中，我们分析了网络链接建立的过程，一旦建立就可以正常的收发消息了。发送消息的细节不再分析，因为对于本地的actor来说这个过程相对简单，它只是创立链接然后给指定的netty网路服务发送消息就好了。接收消息就比较麻烦了，因为这对于actor来说是透明的，netty收到消息后如何把消息分发给指定的actor呢？这个分发的过程值得研究研究。

　　之前分析过，在监听创立的过程中，有一个对象非常关键：TcpServerHandler。它负责链接建立、消息收发等功能。TcpServerHandler继承了ServerHandler

private[netty] abstract class ServerHandler(
  protected final val transport:               NettyTransport,
  private final val associationListenerFuture: Future[AssociationEventListener])
  extends NettyServerHelpers with CommonHandlers

　　ServerHandler继承了NettyServerHelpers

private[netty] trait NettyServerHelpers extends SimpleChannelUpstreamHandler with NettyHelpers {

  final override def messageReceived(ctx: ChannelHandlerContext, e: MessageEvent): Unit = {
    super.messageReceived(ctx, e)
    onMessage(ctx, e)
  }

  final override def exceptionCaught(ctx: ChannelHandlerContext, e: ExceptionEvent): Unit = transformException(ctx, e)

  final override def channelConnected(ctx: ChannelHandlerContext, e: ChannelStateEvent): Unit = {
    super.channelConnected(ctx, e)
    onConnect(ctx, e)
  }

  final override def channelOpen(ctx: ChannelHandlerContext, e: ChannelStateEvent): Unit = {
    super.channelOpen(ctx, e)
    onOpen(ctx, e)
  }

  final override def channelDisconnected(ctx: ChannelHandlerContext, e: ChannelStateEvent): Unit = {
    super.channelDisconnected(ctx, e)
    onDisconnect(ctx, e)
  }
}

　　很明显NettyServerHelpers有一个messageReceived应该就是收到消息时回调的方法，那onMessage在哪里实现呢？TcpServerHandler还继承了TcpHandlers，我们来看看TcpHandlers的onMessage方法。

 override def onMessage(ctx: ChannelHandlerContext, e: MessageEvent): Unit = {
    val bytes: Array[Byte] = e.getMessage.asInstanceOf[ChannelBuffer].array()
    if (bytes.length > 0) notifyListener(e.getChannel, InboundPayload(ByteString(bytes)))
  }

　　它最终用InboundPayload封装了收到的数据，并调用了ChannelLocalActor.notifyListener方法。

private[remote] object ChannelLocalActor extends ChannelLocal[Option[HandleEventListener]] {
  override def initialValue(channel: Channel): Option[HandleEventListener] = None
  def notifyListener(channel: Channel, msg: HandleEvent): Unit = get(channel) foreach { _ notify msg }
}

　　ChannelLocalActor可以先把它理解成一个ThreadLocal对象，其他的技术细节读者可以自行谷歌。notifyListener只调用了get，那具体是在哪里set的呢？通过channel变量get到的Option[HandleEventListener]又是在哪里赋值的呢？

  override def registerListener(
    channel:             Channel,
    listener:            HandleEventListener,
    msg:                 ChannelBuffer,
    remoteSocketAddress: InetSocketAddress): Unit = ChannelLocalActor.set(channel, Some(listener))

　　很显然是在registerListener时set的值，那registerListener在哪里调用呢？如果读过上一篇的文章，一定会知道ServerHandler.initInbound函数，这个函数调用了CommonHandlers.init

final protected def init(channel: Channel, remoteSocketAddress: SocketAddress, remoteAddress: Address, msg: ChannelBuffer)(
    op: (AssociationHandle ⇒ Any)): Unit = {
    import transport._
    NettyTransport.addressFromSocketAddress(channel.getLocalAddress, schemeIdentifier, system.name, Some(settings.Hostname), None) match {
      case Some(localAddress) ⇒
        val handle = createHandle(channel, localAddress, remoteAddress)
        handle.readHandlerPromise.future.foreach {
          listener ⇒
            registerListener(channel, listener, msg, remoteSocketAddress.asInstanceOf[InetSocketAddress])
            channel.setReadable(true)
        }
        op(handle)

      case _ ⇒ NettyTransport.gracefulClose(channel)
    }
  }

　　看到没，上面的函数中调用了registerListener，那listener具体在哪里创建的呢，或者是哪个变量对应的值呢？这就需要研究createHandle对象及其返回值是什么了。经过分析还是找到了TcpHandlers这个trait，里面有createHandle的具体实现。

  override def createHandle(channel: Channel, localAddress: Address, remoteAddress: Address): AssociationHandle =
    new TcpAssociationHandle(localAddress, remoteAddress, transport, channel)

　　TcpAssociationHandle源码如下

private[remote] class TcpAssociationHandle(
  val localAddress:    Address,
  val remoteAddress:   Address,
  val transport:       NettyTransport,
  private val channel: Channel)
  extends AssociationHandle {
  import transport.executionContext

  override val readHandlerPromise: Promise[HandleEventListener] = Promise()

  override def write(payload: ByteString): Boolean =
    if (channel.isWritable && channel.isOpen) {
      channel.write(ChannelBuffers.wrappedBuffer(payload.asByteBuffer))
      true
    } else false

  override def disassociate(): Unit = NettyTransport.gracefulClose(channel)
}

　　由此可见，readHandlerPromise是一个Promise[HandleEventListener]，并没有具体赋值的逻辑，这就要去使用TcpAssociationHandle的相关代码找相关的赋值逻辑了。TcpAssociationHandle在哪里使用呢？还记得handleInboundAssociation建立连接的过程吗？它最终调用了createAndRegisterEndpoint

  private def createAndRegisterEndpoint(handle: AkkaProtocolHandle): Unit = {
    val writing = settings.UsePassiveConnections && !endpoints.hasWritableEndpointFor(handle.remoteAddress)
    eventPublisher.notifyListeners(AssociatedEvent(handle.localAddress, handle.remoteAddress, inbound = true))

    val endpoint = createEndpoint(
      handle.remoteAddress,
      handle.localAddress,
      transportMapping(handle.localAddress),
      settings,
      Some(handle),
      writing)

    if (writing)
      endpoints.registerWritableEndpoint(handle.remoteAddress, Some(handle.handshakeInfo.uid), endpoint)
    else {
      endpoints.registerReadOnlyEndpoint(handle.remoteAddress, endpoint, handle.handshakeInfo.uid)
      if (!endpoints.hasWritableEndpointFor(handle.remoteAddress))
        endpoints.removePolicy(handle.remoteAddress)
    }
  }

　　createAndRegisterEndpoint拿着一个连接实例AkkaProtocolHandle创建了一个endpoint，其中有个很关键的字段writing，它是true还是false呢？UsePassiveConnections默认为true，且经分析!endpoints.hasWritableEndpointFor(handle.remoteAddress)应该也是true，所以writing是true

 # Reuse inbound connections for outbound messages
    use-passive-connections = on

　　ReliableDeliverySupervisor其实是对EndpointWriter的代理。在创建ReliableDeliverySupervisor的过程中AkkaProtocolHandle是作为参数传入的，也就监听到连接消息后创建的handle。而在创建EndpointWriter的过程中，这个handle又是作为第一个参数传入了EndpointWriter。我们来看看EndpointWriter是如何使用这个handle的。

 override def preStart(): Unit = {
    handle match {
      case Some(h) ⇒
        reader = startReadEndpoint(h)
      case None ⇒
        transport.associate(remoteAddress, refuseUid).map(Handle(_)) pipeTo self
    }
  }

　　在preStart时，handle应该是有值的，如果有值，就调用了startReadEndpoint(h)方法。

private def startReadEndpoint(handle: AkkaProtocolHandle): Some[ActorRef] = {
    val newReader =
      context.watch(context.actorOf(
        RARP(context.system).configureDispatcher(EndpointReader.props(localAddress, remoteAddress, transport, settings, codec,
          msgDispatch, inbound, handle.handshakeInfo.uid, reliableDeliverySupervisor, receiveBuffers)).withDeploy(Deploy.local),
        "endpointReader-" + AddressUrlEncoder(remoteAddress) + "-" + readerId.next()))
    handle.readHandlerPromise.success(ActorHandleEventListener(newReader))
    Some(newReader)
  }

　　startReadEndpoint做了什么呢？它又创建了一个Actor：EndpointReader！！！好多中间的actor创建。创建之后，调用了handle.readHandlerPromise.success(ActorHandleEventListener(newReader))给handle.readHandlerPromise。还记得ActorHandleEventListener吗，它就是把收到的消息转发了对应的actor，此处就是newReader。

　　EndpointReader如何处理InboundPayload消息呢？首先解码收到的消息，然后给创建它的reliableDelivery发送ack消息。

  override def decodeMessage(
    raw:          ByteString,
    provider:     RemoteActorRefProvider,
    localAddress: Address): (Option[Ack], Option[Message]) = {
    val ackAndEnvelope = AckAndEnvelopeContainer.parseFrom(raw.toArray)

    val ackOption = if (ackAndEnvelope.hasAck) {
      import scala.collection.JavaConverters._
      Some(Ack(SeqNo(ackAndEnvelope.getAck.getCumulativeAck), ackAndEnvelope.getAck.getNacksList.asScala.map(SeqNo(_)).toSet))
    } else None

    val messageOption = if (ackAndEnvelope.hasEnvelope) {
      val msgPdu = ackAndEnvelope.getEnvelope
      Some(Message(
        recipient = provider.resolveActorRefWithLocalAddress(msgPdu.getRecipient.getPath, localAddress),
        recipientAddress = AddressFromURIString(msgPdu.getRecipient.getPath),
        serializedMessage = msgPdu.getMessage,
        senderOption =
          if (msgPdu.hasSender) OptionVal(provider.resolveActorRefWithLocalAddress(msgPdu.getSender.getPath, localAddress))
          else OptionVal.None,
        seqOption =
          if (msgPdu.hasSeq) Some(SeqNo(msgPdu.getSeq)) else None))
    } else None

    (ackOption, messageOption)
  }

　　上面是decodeMessage的源码，消息最终被decode成了Message对象。

  final case class Message(
    recipient:         InternalActorRef,
    recipientAddress:  Address,
    serializedMessage: SerializedMessage,
    senderOption:      OptionVal[ActorRef],
    seqOption:         Option[SeqNo]) extends HasSequenceNumber {

    def reliableDeliveryEnabled = seqOption.isDefined

    override def seq: SeqNo = seqOption.get
  }

　　默认情况下reliableDeliveryEnabled是false的，因为发送出去的msgPdu是没有getSeq的，因为默认的tcp是保证消息发送的。所以EndpointReader收到消息后调用了msgDispatch.dispatch把消息分发出去了。根据上下文msgDispatch是在EndpointWriter创建的，代码如下。

val msgDispatch = new DefaultMessageDispatcher(extendedSystem, provider, markLog)

　　DefaultMessageDispatcher.dispatch不再具体分析，它就是把消息tell给了Message.recipient，而recipient是一个InternalActorRef，对的，你没有看错，这就是一个InternalActorRef，是不是很神奇，payload解码之后直接就有目标actor的InternalActorRef了？？！！那我们就得好好看看是如何对payload进行解码的了。

　　在decodeMessage函数中，有两处代码非常关键：“recipient = provider.resolveActorRefWithLocalAddress(msgPdu.getRecipient.getPath, localAddress)”、“if (msgPdu.hasSender) OptionVal(provider.resolveActorRefWithLocalAddress(msgPdu.getSender.getPath, localAddress))”。都是调用provider.resolveActorRefWithLocalAddress函数通过actor的path转化成了对应actor的ActorRef，很显然provider就是RemoteActorRefProvider。

/**
   * INTERNAL API
   * Called in deserialization of incoming remote messages where the correct local address is known.
   */
  private[akka] def resolveActorRefWithLocalAddress(path: String, localAddress: Address): InternalActorRef = {
    path match {
      case ActorPathExtractor(address, elems) ⇒
        if (hasAddress(address))
          local.resolveActorRef(rootGuardian, elems)
        else try {
          new RemoteActorRef(transport, localAddress, RootActorPath(address) / elems, Nobody, props = None, deploy = None)
        } catch {
          case NonFatal(e) ⇒
            log.warning("Error while resolving ActorRef [{}] due to [{}]", path, e.getMessage)
            new EmptyLocalActorRef(this, RootActorPath(address) / elems, eventStream)
        }
      case _ ⇒
        log.debug("Resolve (deserialization) of unknown (invalid) path [{}], using deadLetters.", path)
        deadLetters
    }
  }

　　 resolveActorRefWithLocalAddress也很简单，如果目标address包含在本机范围，就调用local.resolveActorRef，否则就创建RemoteActorRef，关于RemoteActorRef的作用这里不再讲解。

  /**
   * INTERNAL API
   */
  private[akka] def resolveActorRef(ref: InternalActorRef, pathElements: Iterable[String]): InternalActorRef =
    if (pathElements.isEmpty) {
      log.debug("Resolve (deserialization) of empty path doesn't match an active actor, using deadLetters.")
      deadLetters
    } else ref.getChild(pathElements.iterator) match {
      case Nobody ⇒
        if (log.isDebugEnabled)
          log.debug(
            "Resolve (deserialization) of path [{}] doesn't match an active actor. " +
              "It has probably been stopped, using deadLetters.",
            pathElements.mkString("/"))
        new EmptyLocalActorRef(system.provider, ref.path / pathElements, eventStream)
      case x ⇒ x
    }

　　LocalActorRefProvider.resolveActorRef也比较简单，就是调用ref.getChild，而ref是LocalActorRefProvider.rootGuardian，其实就是在本地范围内从root向下查找对应的ActorRef。

　　既然在收到消息时，是通过ActorPath找到对应的ActorRef的，那么发送消息的时候一定有把ActorRef转化成ActorPath的地方，关于这点我也带领大家验证一下。在之前的文章，我们分析过，发送消息是通过EndpointWriter.writeSend发送的，那就再来回顾一下这个函数。

def writeSend(s: Send): Boolean = try {
    handle match {
      case Some(h) ⇒
        if (provider.remoteSettings.LogSend && log.isDebugEnabled) {
          def msgLog = s"RemoteMessage: [${s.message}] to [${s.recipient}]<+[${s.recipient.path}] from [${s.senderOption.getOrElse(extendedSystem.deadLetters)}]"
          log.debug("sending message {}", msgLog)
        }

        val pdu = codec.constructMessage(
          s.recipient.localAddressToUse,
          s.recipient,
          serializeMessage(s.message),
          s.senderOption,
          seqOption = s.seqOpt,
          ackOption = lastAck)

        val pduSize = pdu.size
        remoteMetrics.logPayloadBytes(s.message, pduSize)

        if (pduSize > transport.maximumPayloadBytes) {
          val reason = new OversizedPayloadException(s"Discarding oversized payload sent to ${s.recipient}: max allowed size ${transport.maximumPayloadBytes} bytes, actual size of encoded ${s.message.getClass} was ${pdu.size} bytes.")
          log.error(reason, "Transient association error (association remains live)")
          true
        } else {
          val ok = h.write(pdu)
          if (ok) {
            ackDeadline = newAckDeadline
            lastAck = None
          }
          ok
        }

      case None ⇒
        throw new EndpointException("Internal error: Endpoint is in state Writing, but no association handle is present.")
    }
  } catch {
    case e: NotSerializableException ⇒
      log.error(e, "Serializer not defined for message type [{}]. Transient association error (association remains live)", s.message.getClass)
      true
    case e: IllegalArgumentException ⇒
      log.error(e, "Serializer not defined for message type [{}]. Transient association error (association remains live)", s.message.getClass)
      true
    case e: MessageSerializer.SerializationException ⇒
      log.error(e, "{} Transient association error (association remains live)", e.getMessage)
      true
    case e: EndpointException ⇒
      publishAndThrow(e, Logging.ErrorLevel)
    case NonFatal(e) ⇒
      publishAndThrow(new EndpointException("Failed to write message to the transport", e), Logging.ErrorLevel)
  }

　　在发送之前，调用了codec.constructMessage把消息相关的数据都编码进了pdu，具体如何进行编码的呢？

override def constructMessage(
    localAddress:      Address,
    recipient:         ActorRef,
    serializedMessage: SerializedMessage,
    senderOption:      OptionVal[ActorRef],
    seqOption:         Option[SeqNo]       = None,
    ackOption:         Option[Ack]         = None): ByteString = {

    val ackAndEnvelopeBuilder = AckAndEnvelopeContainer.newBuilder

    val envelopeBuilder = RemoteEnvelope.newBuilder

    envelopeBuilder.setRecipient(serializeActorRef(recipient.path.address, recipient))
    senderOption match {
      case OptionVal.Some(sender) ⇒ envelopeBuilder.setSender(serializeActorRef(localAddress, sender))
      case OptionVal.None         ⇒
    }

    seqOption foreach { seq ⇒ envelopeBuilder.setSeq(seq.rawValue) }
    ackOption foreach { ack ⇒ ackAndEnvelopeBuilder.setAck(ackBuilder(ack)) }
    envelopeBuilder.setMessage(serializedMessage)
    ackAndEnvelopeBuilder.setEnvelope(envelopeBuilder)

    ByteString.ByteString1C(ackAndEnvelopeBuilder.build.toByteArray) //Reuse Byte Array (naughty!)
  }

　　看到serializeActorRef了吗，它把ActorRef（这里分别是recipient和sender）进行了序列化。

private def serializeActorRef(defaultAddress: Address, ref: ActorRef): ActorRefData = {
    ActorRefData.newBuilder.setPath(
      if (ref.path.address.host.isDefined) ref.path.toSerializationFormat else ref.path.toSerializationFormatWithAddress(defaultAddress)).build()
  }

　　其实serializeActorRef也比较简单，如果当前ActorRef是本地（有host字段）则直接调用path.toSerializationFormat，否则调用toSerializationFormatWithAddress(defaultAddress)

  /**
   * Generate full String representation including the
   * uid for the actor cell instance as URI fragment.
   * This representation should be used as serialized
   * representation instead of `toString`.
   */
  def toSerializationFormat: String

  /**
   * Generate full String representation including the uid for the actor cell
   * instance as URI fragment, replacing the Address in the RootActor Path
   * with the given one unless this path’s address includes host and port
   * information. This representation should be used as serialized
   * representation instead of `toStringWithAddress`.
   */
  def toSerializationFormatWithAddress(address: Address): String

　　toSerializationFormat和toSerializationFormatWithAddress的功能官网注释已经解释的很清楚，我就不啰嗦了，不过这直接验证了在发送消息时把ActorRef序列化成对应ActorPath的String的猜测。那么在收到消息时就可以通过ActorPath找到具体的ActorRef了。

　　至此remote模式下收发消息的过程我们就分析清楚了，如果还有不清楚的小伙伴就再把之前的文章复习一下，当然还可以在下面留言讨论。