Kafka网络模型和通信流程剖析

1.概述

最近有同学在学习Kafka的网络通信这块内容时遇到一些疑问，关于网络模型和通信流程的相关内容，这里笔者将通过这篇博客为大家来剖析一下这部分内容。

2.内容

Kafka系统作为一个Message Queue，涉及到的网络通信主要包含以下两个方面：

• Pull：Consumer从消息队列中拉取消息数据；
• Push：Producer往消息队列中推送消息数据。

要实现高性能的网络通信，可以使用更加底层的TCP协议或者UDP协议来实现。Kafka在Producer、Broker、Consumer之间设计了一套基于TCP层的通信协议，这套协议完全是为了Kafka系统自身需求而定制实现的。

提示：
这里需要注意的是，由于UDP协议是一种不可靠的传输协议，所以Kafka系统采用TCP协议作为服务间的通信协议。

2.1 基本数据类型

通信协议中的基本数据类型分为以下几种：

• 定长数据类型：例如，int8、int16、int32和、int64，对应到Java语言中，分别是byte、short、int和long
• 可变数据类型：例如，Java语言中Map、List等
• 数组：例如，Java语言中的int[]、String[]等

2.2 通信模型

Kafka系统采用的是Reactor多线程模型，即通过一个Acceptor线程处理所有的新连接，通过多个Processor线程对请求进行处理（比如解析协议、封装请求、、转发等）。

提示：
Reactor是一种事件模型，可以将请求提交到一个或者多个服务程序中进行处理。
当收到Client的请求后，Server处理程序使用多路分发策略，由一个非阻塞的线程来接收所有的请求，然后将这些请求转发到对应的工作线程中进行处理。

之后，在Kafka的版本迭代中，新增了一个Handler模块，它通过指定的线程数对请求进行处理。Handler和Processor之间通过一个Block Queue进行连接。如下图所示：

这里 Acceptor是一个继承于AbstractServerThread的线程类，Acceptor的主要目的是监听并且接收Client的请求，同时，建立数据传输通道（SocketChannel），然后通过轮询的方式交给一个Processor处理。其核心代码在Acceptor的run方法中，代码如下：

def run() {
    serverChannel.register(nioSelector, SelectionKey.OP_ACCEPT)
    startupComplete()
    try {
      var currentProcessor = 0
      while (isRunning) {
        try {
          val ready = nioSelector.select(500)
          if (ready > 0) {
            val keys = nioSelector.selectedKeys()
            val iter = keys.iterator()
            while (iter.hasNext && isRunning) {
              try {
                val key = iter.next
                iter.remove()
                if (key.isAcceptable)
                  accept(key, processors(currentProcessor))
                else
                  throw new IllegalStateException("Unrecognized key state for acceptor thread.")

                // round robin to the next processor thread
                currentProcessor = (currentProcessor + 1) % processors.length
              } catch {
                case e: Throwable => error("Error while accepting connection", e)
              }
            }
          }
        }
        catch {
          // We catch all the throwables to prevent the acceptor thread from exiting on exceptions due
          // to a select operation on a specific channel or a bad request. We don't want
          // the broker to stop responding to requests from other clients in these scenarios.
          case e: ControlThrowable => throw e
          case e: Throwable => error("Error occurred", e)
        }
      }
    } finally {
      debug("Closing server socket and selector.")
      swallowError(serverChannel.close())
      swallowError(nioSelector.close())
      shutdownComplete()
    }
  }

这里还有一个块通道（BlockingChannel），用于连接Processor和Handler，其代码如下所示：

class BlockingChannel( val host: String, 
                       val port: Int, 
                       val readBufferSize: Int, 
                       val writeBufferSize: Int, 
                       val readTimeoutMs: Int ) extends Logging {
  private var connected = false
  private var channel: SocketChannel = null
  private var readChannel: ReadableByteChannel = null
  private var writeChannel: GatheringByteChannel = null
  private val lock = new Object()
  private val connectTimeoutMs = readTimeoutMs
  private var connectionId: String = ""

  def connect() = lock synchronized  {
    if(!connected) {
      try {
        channel = SocketChannel.open()
        if(readBufferSize > 0)
          channel.socket.setReceiveBufferSize(readBufferSize)
        if(writeBufferSize > 0)
          channel.socket.setSendBufferSize(writeBufferSize)
        channel.configureBlocking(true)
        channel.socket.setSoTimeout(readTimeoutMs)
        channel.socket.setKeepAlive(true)
        channel.socket.setTcpNoDelay(true)
        channel.socket.connect(new InetSocketAddress(host, port), connectTimeoutMs)

        writeChannel = channel
        // Need to create a new ReadableByteChannel from input stream because SocketChannel doesn't implement read with timeout
        // See: http://stackoverflow.com/questions/2866557/timeout-for-socketchannel-doesnt-work
        readChannel = Channels.newChannel(channel.socket().getInputStream)
        connected = true
        val localHost = channel.socket.getLocalAddress.getHostAddress
        val localPort = channel.socket.getLocalPort
        val remoteHost = channel.socket.getInetAddress.getHostAddress
        val remotePort = channel.socket.getPort
        connectionId = localHost + ":" + localPort + "-" + remoteHost + ":" + remotePort
        // settings may not match what we requested above
        val msg = "Created socket with SO_TIMEOUT = %d (requested %d), SO_RCVBUF = %d (requested %d), SO_SNDBUF = %d (requested %d), connectTimeoutMs = %d."
        debug(msg.format(channel.socket.getSoTimeout,
                         readTimeoutMs,
                         channel.socket.getReceiveBufferSize, 
                         readBufferSize,
                         channel.socket.getSendBufferSize,
                         writeBufferSize,
                         connectTimeoutMs))

      } catch {
        case _: Throwable => disconnect()
      }
    }
  }
  
  def disconnect() = lock synchronized {
    if(channel != null) {
      swallow(channel.close())
      swallow(channel.socket.close())
      channel = null
      writeChannel = null
    }
    // closing the main socket channel *should* close the read channel
    // but let's do it to be sure.
    if(readChannel != null) {
      swallow(readChannel.close())
      readChannel = null
    }
    connected = false
  }

  def isConnected = connected

  def send(request: RequestOrResponse): Long = {
    if(!connected)
      throw new ClosedChannelException()

    val send = new RequestOrResponseSend(connectionId, request)
    send.writeCompletely(writeChannel)
  }
  
  def receive(): NetworkReceive = {
    if(!connected)
      throw new ClosedChannelException()

    val response = readCompletely(readChannel)
    response.payload().rewind()

    response
  }

  private def readCompletely(channel: ReadableByteChannel): NetworkReceive = {
    val response = new NetworkReceive
    while (!response.complete())
      response.readFromReadableChannel(channel)
    response
  }

}

3.通信过程

Kafka系统的通信框架也是经过了不同的版本迭代的。例如，在Kafka老的版本中，以NIO作为网络通信的基础，通过将多个Socket连接注册到一个Selector上进行监听，只用一个线程就能管理多个连接，这极大的节省了多线程的资源开销。

在Kafka之后的新版本中，依然以NIO作为网络通信的基础，也使用了Reactor多线程模型，不同的是，新版本将具体的业务处理模块（Handler模块）独立出去了，并用单独的线程池进行控制。如下图所示：

 通过上图，我们可以总结一下Kafka的通信流程：

• Client向Server发送请求时，Acceptor负责接收TCP请求，连接成功后传递给Processor线程；
• Processor线程接收到新的连接后，将其注册到自身的Selector中，并监听READ事件
• 当Client在当前连接对象上写入数据时，会触发READ事件，根据TCP协议调用Handler进行处理
• Handler处理完成后，可能会有返回值给Client，并将Handler返回的结果绑定Response端进行发送

通过总结和分析，我们可以知道Kafka新版中独立Handler模块，用这样以下几点优势：

• 能够单独指定Handler的线程数，便于调优和管理
• 防止一个过大的请求阻塞一个Processor线程
• Request、Handler、Response之间都是通过队列来进行连接的，这样它们彼此之间不存在耦合现象，对提升Kafka系统的性能很有帮助

这里需要注意的是，在Kafka的网络通信中，RequestChannel为Processor线程与Handler线程之间数据交换提供了一个缓冲区，是通信中Request和Response缓存的地方。因此，其作用就是在通信中起到了一个数据缓冲队列的作用。Processor线程将读取到的请求添加至RequestChannel的全局队列（requestQueue）中，Handler线程从请求队列中获取并处理，处理完成后将Response添加至RequestChannel的响应队列（responseQueues）中，通过responseListeners唤醒对应的Processor线程，最后Processor线程从响应队列中取出后发送到Client。实现代码如下：

class RequestChannel(val numProcessors: Int, val queueSize: Int) extends KafkaMetricsGroup {
  private var responseListeners: List[(Int) => Unit] = Nil
  private val requestQueue = new ArrayBlockingQueue[RequestChannel.Request](queueSize)
  private val responseQueues = new Array[BlockingQueue[RequestChannel.Response]](numProcessors)
  for(i <- 0 until numProcessors)
    responseQueues(i) = new LinkedBlockingQueue[RequestChannel.Response]()

  newGauge(
    "RequestQueueSize",
    new Gauge[Int] {
      def value = requestQueue.size
    }
  )

  newGauge("ResponseQueueSize", new Gauge[Int]{
    def value = responseQueues.foldLeft(0) {(total, q) => total + q.size()}
  })

  for (i <- 0 until numProcessors) {
    newGauge("ResponseQueueSize",
      new Gauge[Int] {
        def value = responseQueues(i).size()
      },
      Map("processor" -> i.toString)
    )
  }

  /** Send a request to be handled, potentially blocking until there is room in the queue for the request */
  def sendRequest(request: RequestChannel.Request) {
    requestQueue.put(request)
  }

  /** Send a response back to the socket server to be sent over the network */
  def sendResponse(response: RequestChannel.Response) {
    responseQueues(response.processor).put(response)
    for(onResponse <- responseListeners)
      onResponse(response.processor)
  }

  /** No operation to take for the request, need to read more over the network */
  def noOperation(processor: Int, request: RequestChannel.Request) {
    responseQueues(processor).put(RequestChannel.Response(processor, request, null, RequestChannel.NoOpAction))
    for(onResponse <- responseListeners)
      onResponse(processor)
  }

  /** Close the connection for the request */
  def closeConnection(processor: Int, request: RequestChannel.Request) {
    responseQueues(processor).put(RequestChannel.Response(processor, request, null, RequestChannel.CloseConnectionAction))
    for(onResponse <- responseListeners)
      onResponse(processor)
  }

  /** Get the next request or block until specified time has elapsed */
  def receiveRequest(timeout: Long): RequestChannel.Request =
    requestQueue.poll(timeout, TimeUnit.MILLISECONDS)

  /** Get the next request or block until there is one */
  def receiveRequest(): RequestChannel.Request =
    requestQueue.take()

  /** Get a response for the given processor if there is one */
  def receiveResponse(processor: Int): RequestChannel.Response = {
    val response = responseQueues(processor).poll()
    if (response != null)
      response.request.responseDequeueTimeMs = Time.SYSTEM.milliseconds
    response
  }

  def addResponseListener(onResponse: Int => Unit) {
    responseListeners ::= onResponse
  }

  def shutdown() {
    requestQueue.clear()
  }
}

4.总结

通过认真阅读和分析Kafka的网络通信层代码，可以收获不少关于NIO的网络通信知识。通过对Kafka的源代码进行阅读和学习，这对大规模Kafka集群性能的调优和问题定位排查是很有帮助的。

5.结束语

这篇博客就和大家分享到这里，如果大家在研究学习的过程当中有什么问题，可以加群进行讨论或发送邮件给我，我会尽我所能为您解答，与君共勉！