Apache Kafka源码分析 – Broker Server

1. Kafka.scala

在Kafka的main入口中startup KafkaServerStartable, 而KafkaServerStartable这是对KafkaServer的封装

val kafkaServerStartble = new KafkaServerStartable(serverConfig)
kafkaServerStartble.startup

package kafka.server
class KafkaServerStartable(val serverConfig: KafkaConfig) extends Logging {
  private var server : KafkaServer = null

  private def init() {
    server = new KafkaServer(serverConfig)
  }

  def startup() {
    try {
      server.startup()
    }
    catch {...}
  }
}

2. KafkaServer

KafkaServer代表一个kafka broker, 这是kafka的核心.
只需要看看里面startup了哪些modules, 就知道broker做了哪些工作, 后面一个个具体分析吧

package kafka.server
/**
 * Represents the lifecycle of a single Kafka broker. Handles all functionality required
 * to start up and shutdown a single Kafka node.
 */
class KafkaServer(val config: KafkaConfig, time: Time = SystemTime) extends Logging {
  var socketServer: SocketServer = null
  var requestHandlerPool: KafkaRequestHandlerPool = null
  var logManager: LogManager = null
  var kafkaHealthcheck: KafkaHealthcheck = null
  var topicConfigManager: TopicConfigManager = null
  var replicaManager: ReplicaManager = null
  var apis: KafkaApis = null
  var kafkaController: KafkaController = null
  val kafkaScheduler = new KafkaScheduler(config.backgroundThreads)
  var zkClient: ZkClient = null

  /**
   * Start up API for bringing up a single instance of the Kafka server.
   * Instantiates the LogManager, the SocketServer and the request handlers - KafkaRequestHandlers
   */
  def startup() {
    /* start scheduler */
    kafkaScheduler.startup()

    /* setup zookeeper */
    zkClient = initZk()

    /* start log manager */
    logManager = createLogManager(zkClient)
    logManager.startup()

    socketServer = new SocketServer(config.brokerId,
                                    config.hostName,
                                    config.port,
                                    config.numNetworkThreads,
                                    config.queuedMaxRequests,
                                    config.socketSendBufferBytes,
                                    config.socketReceiveBufferBytes,
                                    config.socketRequestMaxBytes)
    socketServer.startup()

    replicaManager = new ReplicaManager(config, time, zkClient, kafkaScheduler, logManager, isShuttingDown)
    kafkaController = new KafkaController(config, zkClient)

    /* start processing requests */
    apis = new KafkaApis(socketServer.requestChannel, replicaManager, zkClient, config.brokerId, config, kafkaController)
    requestHandlerPool = new KafkaRequestHandlerPool(config.brokerId, socketServer.requestChannel, apis, config.numIoThreads)

    replicaManager.startup()

    kafkaController.startup()

    topicConfigManager = new TopicConfigManager(zkClient, logManager)
    topicConfigManager.startup()

    /* tell everyone we are alive */
    kafkaHealthcheck = new KafkaHealthcheck(config.brokerId, config.advertisedHostName, config.advertisedPort, config.zkSessionTimeoutMs, zkClient)
    kafkaHealthcheck.startup()
  }

2.1 KafkaScheduler

KafkaSchduler用于在后台执行一些任务，用ScheduledThreadPoolExecutor实现

package kafka.utils

/**
 * A scheduler based on java.util.concurrent.ScheduledThreadPoolExecutor
 *
 * It has a pool of kafka-scheduler- threads that do the actual work.
 *
 * @param threads The number of threads in the thread pool
 * @param threadNamePrefix The name to use for scheduler threads. This prefix will have a number appended to it.
 * @param daemon If true the scheduler threads will be "daemon" threads and will not block jvm shutdown.
 */
@threadsafe
class KafkaScheduler(val threads: Int,
                     val threadNamePrefix: String = "kafka-scheduler-",
                     daemon: Boolean = true) extends Scheduler with Logging {
  @volatile private var executor: ScheduledThreadPoolExecutor = null
  override def startup() {
    this synchronized {
      executor = new ScheduledThreadPoolExecutor(threads) //创建ScheduledThreadPoolExecutor
      executor.setContinueExistingPeriodicTasksAfterShutdownPolicy(false)
      executor.setExecuteExistingDelayedTasksAfterShutdownPolicy(false)
      executor.setThreadFactory(new ThreadFactory() {
                                  def newThread(runnable: Runnable): Thread =
                                    Utils.newThread(threadNamePrefix + schedulerThreadId.getAndIncrement(), runnable, daemon)
                                })
    }
  }

def schedule(name: String, fun: ()=>Unit, delay: Long, period: Long, unit: TimeUnit) = {
  val runnable = new Runnable { //将fun封装成Runnable
    def run() = {
      try {
        fun()
      } catch {...}
      finally {...}
    }
  }
  if(period >= 0) //在pool中进行delay schedule
    executor.scheduleAtFixedRate(runnable, delay, period, unit)
  else
    executor.schedule(runnable, delay, unit)
}

2.2 Zookeeper Client

由于Kafka是基于zookeeper进行配置管理的， 所以需要创建zkclient和zookeeper集群通信

2.3 logManager

The entry point to the kafka log management subsystem. The log manager is responsible for log creation, retrieval, and cleaning.
Apache Kafka源码分析 – Log Management

2.4 ReplicaManager

在0.8中新加入的replica相关模块

Apache Kafka Replication Design – High level
kafka Detailed Replication Design V3
Apache Kafka源码分析 – ReplicaManager

2.5 Kafka Socket Server

首先broker server是socket server，所有和broker的交互都是通过往socket端口发送request来实现的

socketServer = new SocketServer(config.brokerId...)

KafkaApis
该类封装了所有request的处理逻辑

/**
 * Logic to handle the various Kafka requests
 */
class KafkaApis(val requestChannel: RequestChannel,
                val replicaManager: ReplicaManager,
                val zkClient: ZkClient,
                val brokerId: Int,
                val config: KafkaConfig,
                val controller: KafkaController) extends Logging {
  /**
   * Top-level method that handles all requests and multiplexes to the right api
   */
  def handle(request: RequestChannel.Request) {
    try{
      request.requestId match {
        case RequestKeys.ProduceKey => handleProducerRequest(request)
        case RequestKeys.FetchKey => handleFetchRequest(request)
        case RequestKeys.OffsetsKey => handleOffsetRequest(request)
        case RequestKeys.MetadataKey => handleTopicMetadataRequest(request)
        case RequestKeys.LeaderAndIsrKey => handleLeaderAndIsrRequest(request)
        case RequestKeys.StopReplicaKey => handleStopReplicaRequest(request)
        case RequestKeys.UpdateMetadataKey => handleUpdateMetadataRequest(request)
        case RequestKeys.ControlledShutdownKey => handleControlledShutdownRequest(request)
        case RequestKeys.OffsetCommitKey => handleOffsetCommitRequest(request)
        case RequestKeys.OffsetFetchKey => handleOffsetFetchRequest(request)
        case requestId => throw new KafkaException("Unknown api code " + requestId)
      }
    } catch {
    } finally
      request.apiLocalCompleteTimeMs = SystemTime.milliseconds
  }
}

KafkaRequestHandler
基于线程池的KafkaRequestHandler

/**
 * A thread that answers kafka requests.
 */
class KafkaRequestHandler(id: Int, brokerId: Int, val requestChannel: RequestChannel, apis: KafkaApis) extends Runnable with Logging {
  def run() { 
    while(true) {
      try {
        val req = requestChannel.receiveRequest() //从socketChannel接受request
        if(req eq RequestChannel.AllDone) {
          debug("Kafka request handler %d on broker %d received shut down command".format(
            id, brokerId))
          return
        }
        req.requestDequeueTimeMs = SystemTime.milliseconds
        apis.handle(req) //使用kafkaApis来处理request
      } catch {
        case e: Throwable => error("Exception when handling request", e)
      }
    }
  }

  def shutdown(): Unit = requestChannel.sendRequest(RequestChannel.AllDone)
}

class KafkaRequestHandlerPool(val brokerId: Int,
                              val requestChannel: RequestChannel,
                              val apis: KafkaApis,
                              numThreads: Int) extends Logging {
  val threads = new Array[Thread](numThreads) //线程池
  val runnables = new Array[KafkaRequestHandler](numThreads)
  for(i <- 0 until numThreads) {
    runnables(i) = new KafkaRequestHandler(i, brokerId, requestChannel, apis)
    threads(i) = Utils.daemonThread("kafka-request-handler-" + i, runnables(i))
    threads(i).start()
  }

  def shutdown() {
    info("shutting down")
    for(handler <- runnables)
      handler.shutdown
    for(thread <- threads)
      thread.join
    info("shut down completely")
  }
}

2.6 offsetManager

offsetManager = createOffsetManager()
定期清除过期的offset数据，即compact操作，

scheduler.schedule(name = "offsets-cache-compactor",
                     fun = compact,
                     period = config.offsetsRetentionCheckIntervalMs,
                     unit = TimeUnit.MILLISECONDS)

以及consumer相关的一些offset操作，不细究了，因为我们不用highlevel consumer

2.7 KafkaController

kafkaController = new KafkaController(config, zkClient, brokerState)

Apache Kafka源码分析 – Controller

0.8后，为了处理replica，会用一个broker作为master，即controller，用于协调replica的一致性

2.8 TopicConfigManager

topicConfigManager = new TopicConfigManager(zkClient, logManager)

TopicConfigManager用于处理topic config的change，kafka除了全局的配置，还有一种叫Topic-level configuration

> bin/kafka-topics.sh --zookeeper localhost:2181 --alter --topic my-topic 
    --config max.message.bytes=128000

比如你可以这样设置，那么这些topic config如何生效的？

topic-level config默认是被存储在，

/brokers/topics/<topic_name>/config
但是topic很多的情况下，为了避免创建太多的watcher，

所以单独创建一个目录

/brokers/config_changes

来触发配置的变化
所以上面的命令除了，把配置写入topic/config，还有增加一个通知，告诉watcher哪个topic的config发生了变化

/brokers/config_changes/config_change_13321

并且这个通知有个suffix，用于区别是否已处理过

/**
   * Process the given list of config changes
   */
  private def processConfigChanges(notifications: Seq[String]) {
    if (notifications.size > 0) {
      info("Processing config change notification(s)...")
      val now = time.milliseconds
      val logs = logManager.logsByTopicPartition.toBuffer
      val logsByTopic = logs.groupBy(_._1.topic).mapValues(_.map(_._2))
      for (notification <- notifications) {
        val changeId = changeNumber(notification)
        if (changeId > lastExecutedChange) {  //未处理过
          val changeZnode = ZkUtils.TopicConfigChangesPath + "/" + notification
          val (jsonOpt, stat) = ZkUtils.readDataMaybeNull(zkClient, changeZnode)
          if(jsonOpt.isDefined) {
            val json = jsonOpt.get
            val topic = json.substring(1, json.length - 1) // hacky way to dequote，从通知中获取topic name
            if (logsByTopic.contains(topic)) {
              /* combine the default properties with the overrides in zk to create the new LogConfig */
              val props = new Properties(logManager.defaultConfig.toProps)
              props.putAll(AdminUtils.fetchTopicConfig(zkClient, topic))
              val logConfig = LogConfig.fromProps(props)
              for (log <- logsByTopic(topic))
                log.config = logConfig    //真正的更新log配置
              info("Processed topic config change %d for topic %s, setting new config to %s.".format(changeId, topic, props))
              purgeObsoleteNotifications(now, notifications) //删除过期的notification，10分钟
            }
          }
          lastExecutedChange = changeId
        }
      }
    }
  }

这个failover也没问题，反正配置设置多次也是无害的，每次启动都会把所有没过期的notification处理一遍

并且broker重启后是会从zk中， loading完整的配置的，所以也ok的，这个主要用于实时更新topic的配置

2.8 KafkaHealthcheck

kafkaHealthcheck = new KafkaHealthcheck(config.brokerId, config.advertisedHostName, config.advertisedPort, config.zkSessionTimeoutMs, zkClient)

这个很简单，就像注释的，告诉所有人我还活着。。。

实现就是在，

 /brokers/[0...N] --> advertisedHost:advertisedPort

register一个ephemeral znode，当SessionExpired时，再去register，典型zk应用
所以只需要watch这个路径就是知道broker是否还活着

2.9 ContolledShutdown

对于0.8之前，broker的startup和shutdown都很简单，把上面这些组件初始化，或stop就可以了

但是0.8后，增加replica，所以broker不能自己直接shutdown，需要先通知controller，controller做完处理后，比如partition leader的迁移，或replica offline，然后才能shutdown

private def controlledShutdown()

挺长的，逻辑就是找到controller，发送ControlledShutdownRequest，然后等待返回，如果失败，就是unclean shutdown