Spark源码分析之五:Task调度(一)

        在前四篇博文中,我们分析了Job提交运行总流程的第一阶段Stage划分与提交,它又被细化为三个分阶段:

        1、Job的调度模型与运行反馈;

        2、Stage划分;

        3、Stage提交:对应TaskSet的生成。

        Stage划分与提交阶段主要是由DAGScheduler完成的,而DAGScheduler负责Job的逻辑调度,主要职责也即DAG图的分解,按照RDD间是否为shuffle dependency,将整个Job划分为一个个stage,并将每个stage转化为tasks的集合--TaskSet。

        接下来我们要讲的第二阶段Task调度与执行,则是Spark中Job的物理调度,它实际上分为两个主要阶段:

        1、Task调度;

        2、Task运行。

        下面,我们分析下Task的调度。我们知道,在第一阶段的末尾,stage被提交后,每个stage被转化为一组task的集合--TaskSet,而紧接着,则调用taskScheduler.submitTasks()提交这些tasks,而TaskScheduler的主要职责,则是负责Job物理调度阶段--Task调度。TaskScheduler为scala中的一个trait,你可以简单的把它理解为Java中的接口,目前它仅仅有一个实现类TaskSchedulerImpl。

        TaskScheduler负责低层次任务的调度,每个TaskScheduler为一个特定的SparkContext调度tasks。这些调度器获取到由DAGScheduler为每个stage提交至他们的一组Tasks,并负责将这些tasks发送到集群,以执行它们,在它们失败时重试,并减轻掉队情况(类似MapReduce的推测执行原理吧,在这里留个疑问)。这些调度器返回一些事件events给DAGScheduler。其源码如下:

/**
 * Low-level task scheduler interface, currently implemented exclusively by
 * [[org.apache.spark.scheduler.TaskSchedulerImpl]].
 * This interface allows plugging in different task schedulers. Each TaskScheduler schedules tasks
 * for a single SparkContext. These schedulers get sets of tasks submitted to them from the
 * DAGScheduler for each stage, and are responsible for sending the tasks to the cluster, running
 * them, retrying if there are failures, and mitigating stragglers. They return events to the
 * DAGScheduler.
 * 
 */
private[spark] trait TaskScheduler {

  private val appId = "spark-application-" + System.currentTimeMillis

  def rootPool: Pool

  def schedulingMode: SchedulingMode

  def start(): Unit

  // Invoked after system has successfully initialized (typically in spark context).
  // Yarn uses this to bootstrap allocation of resources based on preferred locations,
  // wait for slave registrations, etc.
  def postStartHook() { }

  // Disconnect from the cluster.
  def stop(): Unit

  // Submit a sequence of tasks to run.
  def submitTasks(taskSet: TaskSet): Unit

  // Cancel a stage.
  def cancelTasks(stageId: Int, interruptThread: Boolean)

  // Set the DAG scheduler for upcalls. This is guaranteed to be set before submitTasks is called.
  def setDAGScheduler(dagScheduler: DAGScheduler): Unit

  // Get the default level of parallelism to use in the cluster, as a hint for sizing jobs.
  def defaultParallelism(): Int

  /**
   * Update metrics for in-progress tasks and let the master know that the BlockManager is still
   * alive. Return true if the driver knows about the given block manager. Otherwise, return false,
   * indicating that the block manager should re-register.
   */
  def executorHeartbeatReceived(execId: String, taskMetrics: Array[(Long, TaskMetrics)],
    blockManagerId: BlockManagerId): Boolean

  /**
   * Get an application ID associated with the job.
   *
   * @return An application ID
   */
  def applicationId(): String = appId

  /**
   * Process a lost executor
   */
  def executorLost(executorId: String, reason: ExecutorLossReason): Unit

  /**
   * Get an application's attempt ID associated with the job.
   *
   * @return An application's Attempt ID
   */
  def applicationAttemptId(): Option[String]

}
        通过源码我们可以知道,TaskScheduler提供了实例化与销毁时必要的start()和stop()方法,并提供了提交Tasks与取消Tasks的submitTasks()和cancelTasks()方法,并且通过executorHeartbeatReceived()周期性的接收executor的心跳,更新运行中tasks的元信息,并让master知晓BlockManager仍然存活。

        好了,结合源码,我们一步步来看吧。

        首先,在DAGScheduler的submitMissingTasks()方法的最后,每个stage生成一组tasks后,即调用用TaskScheduler的submitTasks()方法提交task,代码如下:

// 利用taskScheduler.submitTasks()提交task
      taskScheduler.submitTasks(new TaskSet(
        tasks.toArray, stage.id, stage.latestInfo.attemptId, jobId, properties))
      // 记录提交时间
      stage.latestInfo.submissionTime = Some(clock.getTimeMillis())
        那么我们先来看下TaskScheduler的submitTasks()方法,在其实现类TaskSchedulerImpl中,代码如下:

override def submitTasks(taskSet: TaskSet) {
    
    // 获取TaskSet中的tasks
    val tasks = taskSet.tasks
    logInfo("Adding task set " + taskSet.id + " with " + tasks.length + " tasks")
    
    // 使用synchronized进行同步
    this.synchronized {
      
      // 创建TaskSetManager
      val manager = createTaskSetManager(taskSet, maxTaskFailures)
      
      // 获取taskSet对应的stageId
      val stage = taskSet.stageId
      
      // taskSetsByStageIdAndAttempt存储的是stageId->[taskSet.stageAttemptId->TaskSetManager]
      // 更新taskSetsByStageIdAndAttempt,将上述对应关系存入
      val stageTaskSets =
        taskSetsByStageIdAndAttempt.getOrElseUpdate(stage, new HashMap[Int, TaskSetManager])
      stageTaskSets(taskSet.stageAttemptId) = manager
      
      // 查看是否存在冲突的taskSet,如果存在,抛出IllegalStateException异常
      val conflictingTaskSet = stageTaskSets.exists { case (_, ts) =>
        ts.taskSet != taskSet && !ts.isZombie
      }
      if (conflictingTaskSet) {
        throw new IllegalStateException(s"more than one active taskSet for stage $stage:" +
          s" ${stageTaskSets.toSeq.map{_._2.taskSet.id}.mkString(",")}")
      }
      
      // 将TaskSetManager添加到schedulableBuilder中
      schedulableBuilder.addTaskSetManager(manager, manager.taskSet.properties)

      // 如果不是本地任务,且不再接受任务
      if (!isLocal && !hasReceivedTask) {
        starvationTimer.scheduleAtFixedRate(new TimerTask() {
          override def run() {
            if (!hasLaunchedTask) {
              logWarning("Initial job has not accepted any resources; " +
                "check your cluster UI to ensure that workers are registered " +
                "and have sufficient resources")
            } else {
              this.cancel()
            }
          }
        }, STARVATION_TIMEOUT_MS, STARVATION_TIMEOUT_MS)
      }
      
      // 设置标志位hasReceivedTask为true
      hasReceivedTask = true
    }
    
    // 最后调用SchedulerBackend的reviveOffers()
    backend.reviveOffers()
  }
        该方法首先从入参TaskSet中获取tasks;

        接下来,在synchronized同步代码块内,主要完成以下几件事:

        1、创建TaskSetManager,TaskSetManager主要用来干什么呢,后面我们会分析;

        2、通过taskSet获取stageId;

        3、更新数据结构taskSetsByStageIdAndAttempt,将映射关系stageId->[taskSet.stageAttemptId->TaskSetManager]存入,这里的TaskSetManager就是上面创建的TaskSetManager,taskSet.stageAttemptId是怎么赋值的呢?为了保证叙述的完整性,还是先留个小小的疑问吧;

        4、查看是否存在冲突的taskSet,如果存在,抛出IllegalStateException异常;

        5、将TaskSetManager添加到schedulableBuilder中;

        6、最后调用SchedulerBackend的reviveOffers()。

        下面慢慢分析上述流程,首先这个TaskSetManager是干什么呢?通过名字可以简单的推论出,它是TaskSet的管理者,主要在TaskSchedulerImpl中调度同一个TaskSet中的tasks。该类追踪每个task,当它们失败时重试(直到限制的最大次数),并通过延迟调度处理位置感知调度。该类最主要的接口就是resourceOffer()方法,该方法会询问TaskSet,它是否想要在一个节点上运行一个task,并在TaskSet中的task状态变更时通知它(比如完成等)。

        现在再来看下taskSet的stageAttemptId,在DAGScheduler的submitMissingTasks()方法中调用TaskScheduler的submitTasks()方法提交task,构造TaskSet对象时,赋值给TaskSet的stageAttemptId字段的是stage.latestInfo.attemptId。而Stage的latestInfo是这样定义的:

/** Returns the StageInfo for the most recent attempt for this stage. */
  // 返回该stage的最新尝试attempt的StageInfo
  def latestInfo: StageInfo = _latestInfo
        即它是由_latestInfo来赋值的,那么_latestInfo呢?代码如下:

/**
   * Pointer to the [StageInfo] object for the most recent attempt. This needs to be initialized
   * here, before any attempts have actually been created, because the DAGScheduler uses this
   * StageInfo to tell SparkListeners when a job starts (which happens before any stage attempts
   * have been created).
   * 指向stage最新一次尝试的StageInfo对象。
   * 在任何尝试实际发生之前,都需要在这里被初始化,因为当一个Job启动时(任何stage尝试发生时)DAGScheduler使用
   * 这个StageInfo告诉SparkListeners。
   */
  private var _latestInfo: StageInfo = StageInfo.fromStage(this, nextAttemptId)
        具体初始化过程我们不做过多讨论,我们只要知道,StageInfo中存在一个成员变量attemptId即可,而这个成员变量就是上面我们所说的taskSet的stageAttemptId。而StageInfo中attemptId的值,则是由Stage中nextAttemptId的值确定的,其定义如下:

/** The ID to use for the next new attempt for this stage. */
  // 该stage下一次新尝试的id
  private var nextAttemptId: Int = 0
         而它值的变化是怎么样的呢?答案就在Stage的makeNewStageAttempt()方法中,代码如下:

/** Creates a new attempt for this stage by creating a new StageInfo with a new attempt ID. */
  // 通过用一个最新的nextAttemptId创建的StageInfo对象来创建该stage的最新的一次尝试
  def makeNewStageAttempt(
      numPartitionsToCompute: Int,
      taskLocalityPreferences: Seq[Seq[TaskLocation]] = Seq.empty): Unit = {
    
    // 构造_latestInfo
    _latestInfo = StageInfo.fromStage(
      this, nextAttemptId, Some(numPartitionsToCompute), taskLocalityPreferences)
    
    // nextAttemptId自增
    nextAttemptId += 1
  }
        什么时候调用makeNewStageAttempt()方法呢?还记得 《Spark源码分析之Stage提交》一文的最后,真正提交stage的方法submitMissingTasks()中第6步,标记新的stage attempt,并发送一个SparkListenerStageSubmitted事件吗,代码如下:

// 标记新的stage attempt
    stage.makeNewStageAttempt(partitionsToCompute.size, taskIdToLocations.values.toSeq)
    // 发送一个SparkListenerStageSubmitted事件
    listenerBus.post(SparkListenerStageSubmitted(stage.latestInfo, properties))
        也就是说,在每次提交stage时,即会调用该方法,创建一个新的_latestInfo对象,并对nextAttemptId进行自增。

        好了,言归正传,继续往下看。第5步便是将TaskSetManager添加到schedulableBuilder中,那么这里就有两个问题:

        1、schedulableBuilder是什么?

        2、为什么要将TaskSetManager添加到schedulableBuilder中呢?

        我们首先看下schedulableBuilder的定义及初始化。其定义代码如下:

var schedulableBuilder: SchedulableBuilder = null
        而它的初始化则是在TaskSchedulerImpl的initialize()方法中。如下:

// 初始化
  def initialize(backend: SchedulerBackend) {
    // 赋值SchedulerBackend
    this.backend = backend
    
    // temporarily set rootPool name to empty
    // 临时将rootPool的名字设置为空
    rootPool = new Pool("", schedulingMode, 0, 0)
    
    // 调度构造器,分两种,FIFO和FAIR
    schedulableBuilder = {
      schedulingMode match {
        case SchedulingMode.FIFO =>
          new FIFOSchedulableBuilder(rootPool)
        case SchedulingMode.FAIR =>
          new FairSchedulableBuilder(rootPool, conf)
      }
    }
    schedulableBuilder.buildPools()
  }
        这个方法同时也初始化了TaskSchedulerImpl中SchedulerBackend类型的backend对象,这个对象在最后一步会用到,我们稍后再说。

        继续看schedulableBuilder,通过代码我们就能知道,这个schedulableBuilder是调度构造器,分FIFO和FAIR两种。至于这两种构造器的含义和区别,我们以后再分析。下面看下SchedulableBuilder的源码:

/**
 * An interface to build Schedulable tree
 * buildPools: build the tree nodes(pools)
 * addTaskSetManager: build the leaf nodes(TaskSetManagers)
 */
private[spark] trait SchedulableBuilder {
  def rootPool: Pool

  def buildPools()

  def addTaskSetManager(manager: Schedulable, properties: Properties)
}
        从上面的英文注释我们就能知道,SchedulableBuilder是一个构造调度树的接口,它提供了一个成员变量Pool类型的rootPool和两个主要方法:
        1、buildPools()方法:构造调度树节点(调度池);

        2、addTaskSetManager()方法:构造叶子节点(TaskSetManagers)。

        下面,我们以FIFOSchedulableBuilder为例,简单说下。FIFOSchedulableBuilder中buildPools()是个空方法,没什么可说的,我们主要分析下它的buildPools()方法,代码如下:

override def addTaskSetManager(manager: Schedulable, properties: Properties) {
    rootPool.addSchedulable(manager)
  }
        可以看到,它实际上是调用的Pool的addSchedulable()方法。继续追踪:

override def addSchedulable(schedulable: Schedulable) {
    require(schedulable != null)
    
    // 将schedulable加入到schedulableQueue队列,队列为ConcurrentLinkedQueue类型
    schedulableQueue.add(schedulable)
    
    // 将schedulable的name与schedulable的对应关系添加到schedulableNameToSchedulable集合,集合为ConcurrentHashMap类型
    schedulableNameToSchedulable.put(schedulable.name, schedulable)
    
    // 将this赋值给schedulable的parent,即形成schedulable为this子节点(即截至目前时点的叶子节点)的树形结构
    schedulable.parent = this
  }
        而翻看TaskSetManager的源码可以知道,TaskSetManager就实现了Schedulable这个trait(特质,类似java的接口),也就意味着TaskSetManager是可以被调度的,这也就回答了上面的问题2。

        好了,我们继续看最后一步,调用SchedulerBackend的reviveOffers()。问题又来了,问题不断啊。

        1、SchedulerBackend是什么?

        2、SchedulerBackend如何被初始化?

        3、SchedulerBackend的reviveOffers()到底做了什么?

        带着问题去学习终究是好的,它让我们有了暂时的目标。下面,我们一步步来分析。

        SchedulerBackend是Spark中一个可插拔组件,可插拔意味着它可以有多种实现方式,后续我们会概略讲讲。按照字面意思,它就是调度器的一个后台服务或者实现,其主要作用就是在物理机器或者说worker就绪后,能够提供其上的资源并将tasks加载到那些机器或者worker上。

        上文中我们已经预先提到过,在TaskSchedulerImpl的initialize()方法初始化schedulableBuilder时,同时也初始化了SchedulerBackend,即:

// 赋值SchedulerBackend
    this.backend = backend
        这个SchedulerBackend是被传递进来的,那么这时我们就要追溯到TaskSchedulerImpl实例化的时候了。在Spark应用环境的初始化时,其上下文信息SparkContext中存在以下代码:

// Create and start the scheduler
    val (sched, ts) = SparkContext.createTaskScheduler(this, master)
    _schedulerBackend = sched
    _taskScheduler = ts
        createTaskScheduler()方法主要就是根据给定的Master URL创建一个TaskScheduler。大致代码如下:

/**
   * Create a task scheduler based on a given master URL.
   * Return a 2-tuple of the scheduler backend and the task scheduler.
   * 根据给定的Master URL创建一个TaskScheduler。
   */
  private def createTaskScheduler(
      sc: SparkContext,
      master: String): (SchedulerBackend, TaskScheduler) = {
    import SparkMasterRegex._

    // When running locally, don't try to re-execute tasks on failure.
    val MAX_LOCAL_TASK_FAILURES = 1

    master match {
      case "local" =>
        val scheduler = new TaskSchedulerImpl(sc, MAX_LOCAL_TASK_FAILURES, isLocal = true)
        val backend = new LocalBackend(sc.getConf, scheduler, 1)
        scheduler.initialize(backend)
        (backend, scheduler)

      case LOCAL_N_REGEX(threads) =>
        def localCpuCount: Int = Runtime.getRuntime.availableProcessors()
        // local[*] estimates the number of cores on the machine; local[N] uses exactly N threads.
        val threadCount = if (threads == "*") localCpuCount else threads.toInt
        if (threadCount <= 0) {
          throw new SparkException(s"Asked to run locally with $threadCount threads")
        }
        val scheduler = new TaskSchedulerImpl(sc, MAX_LOCAL_TASK_FAILURES, isLocal = true)
        val backend = new LocalBackend(sc.getConf, scheduler, threadCount)
        scheduler.initialize(backend)
        (backend, scheduler)

      case LOCAL_N_FAILURES_REGEX(threads, maxFailures) =>
        def localCpuCount: Int = Runtime.getRuntime.availableProcessors()
        // local[*, M] means the number of cores on the computer with M failures
        // local[N, M] means exactly N threads with M failures
        val threadCount = if (threads == "*") localCpuCount else threads.toInt
        val scheduler = new TaskSchedulerImpl(sc, maxFailures.toInt, isLocal = true)
        val backend = new LocalBackend(sc.getConf, scheduler, threadCount)
        scheduler.initialize(backend)
        (backend, scheduler)

      // Standalone模式
      case SPARK_REGEX(sparkUrl) =>
      
        // 初始化TaskSchedulerImpl实例scheduler
        val scheduler = new TaskSchedulerImpl(sc)
        val masterUrls = sparkUrl.split(",").map("spark://" + _)
        
        // 初始化一个SparkDeploySchedulerBackend实例backend
        val backend = new SparkDeploySchedulerBackend(scheduler, sc, masterUrls)
        
        // 调用TaskSchedulerImpl的initialize()方法,
        // 为其成员变量SchedulerBackend赋值SparkDeploySchedulerBackend
        scheduler.initialize(backend)
        // 返回backend和scheduler
        (backend, scheduler)

      case LOCAL_CLUSTER_REGEX(numSlaves, coresPerSlave, memoryPerSlave) =>
        // Check to make sure memory requested <= memoryPerSlave. Otherwise Spark will just hang.
        val memoryPerSlaveInt = memoryPerSlave.toInt
        if (sc.executorMemory > memoryPerSlaveInt) {
          throw new SparkException(
            "Asked to launch cluster with %d MB RAM / worker but requested %d MB/worker".format(
              memoryPerSlaveInt, sc.executorMemory))
        }

        val scheduler = new TaskSchedulerImpl(sc)
        val localCluster = new LocalSparkCluster(
          numSlaves.toInt, coresPerSlave.toInt, memoryPerSlaveInt, sc.conf)
        val masterUrls = localCluster.start()
        val backend = new SparkDeploySchedulerBackend(scheduler, sc, masterUrls)
        scheduler.initialize(backend)
        backend.shutdownCallback = (backend: SparkDeploySchedulerBackend) => {
          localCluster.stop()
        }
        (backend, scheduler)

      case "yarn-standalone" | "yarn-cluster" =>
        if (master == "yarn-standalone") {
          logWarning(
            "\"yarn-standalone\" is deprecated as of Spark 1.0. Use \"yarn-cluster\" instead.")
        }
        val scheduler = try {
          val clazz = Utils.classForName("org.apache.spark.scheduler.cluster.YarnClusterScheduler")
          val cons = clazz.getConstructor(classOf[SparkContext])
          cons.newInstance(sc).asInstanceOf[TaskSchedulerImpl]
        } catch {
          // TODO: Enumerate the exact reasons why it can fail
          // But irrespective of it, it means we cannot proceed !
          case e: Exception => {
            throw new SparkException("YARN mode not available ?", e)
          }
        }
        val backend = try {
          val clazz =
            Utils.classForName("org.apache.spark.scheduler.cluster.YarnClusterSchedulerBackend")
          val cons = clazz.getConstructor(classOf[TaskSchedulerImpl], classOf[SparkContext])
          cons.newInstance(scheduler, sc).asInstanceOf[CoarseGrainedSchedulerBackend]
        } catch {
          case e: Exception => {
            throw new SparkException("YARN mode not available ?", e)
          }
        }
        scheduler.initialize(backend)
        (backend, scheduler)

      case "yarn-client" =>
        val scheduler = try {
          val clazz = Utils.classForName("org.apache.spark.scheduler.cluster.YarnScheduler")
          val cons = clazz.getConstructor(classOf[SparkContext])
          cons.newInstance(sc).asInstanceOf[TaskSchedulerImpl]

        } catch {
          case e: Exception => {
            throw new SparkException("YARN mode not available ?", e)
          }
        }

        val backend = try {
          val clazz =
            Utils.classForName("org.apache.spark.scheduler.cluster.YarnClientSchedulerBackend")
          val cons = clazz.getConstructor(classOf[TaskSchedulerImpl], classOf[SparkContext])
          cons.newInstance(scheduler, sc).asInstanceOf[CoarseGrainedSchedulerBackend]
        } catch {
          case e: Exception => {
            throw new SparkException("YARN mode not available ?", e)
          }
        }

        scheduler.initialize(backend)
        (backend, scheduler)

      case MESOS_REGEX(mesosUrl) =>
        MesosNativeLibrary.load()
        val scheduler = new TaskSchedulerImpl(sc)
        val coarseGrained = sc.conf.getBoolean("spark.mesos.coarse", defaultValue = true)
        val backend = if (coarseGrained) {
          new CoarseMesosSchedulerBackend(scheduler, sc, mesosUrl, sc.env.securityManager)
        } else {
          new MesosSchedulerBackend(scheduler, sc, mesosUrl)
        }
        scheduler.initialize(backend)
        (backend, scheduler)

      case SIMR_REGEX(simrUrl) =>
        val scheduler = new TaskSchedulerImpl(sc)
        val backend = new SimrSchedulerBackend(scheduler, sc, simrUrl)
        scheduler.initialize(backend)
        (backend, scheduler)

      case zkUrl if zkUrl.startsWith("zk://") =>
        logWarning("Master URL for a multi-master Mesos cluster managed by ZooKeeper should be " +
          "in the form mesos://zk://host:port. Current Master URL will stop working in Spark 2.0.")
        createTaskScheduler(sc, "mesos://" + zkUrl)

      case _ =>
        throw new SparkException("Could not parse Master URL: '" + master + "'")
    }
  }
        可以看出,它是根据Spark的部署模式来确定创建何种TaskScheduler及SchedulerBackend的。我们就以常见的Standalone模式来说明,关键代码如下:

// Standalone模式
      case SPARK_REGEX(sparkUrl) =>
      
        // 初始化TaskSchedulerImpl实例scheduler
        val scheduler = new TaskSchedulerImpl(sc)
        val masterUrls = sparkUrl.split(",").map("spark://" + _)
        
        // 初始化一个SparkDeploySchedulerBackend实例backend
        val backend = new SparkDeploySchedulerBackend(scheduler, sc, masterUrls)
        
        // 调用TaskSchedulerImpl的initialize()方法,
        // 为其成员变量SchedulerBackend赋值SparkDeploySchedulerBackend
        scheduler.initialize(backend)
        // 返回backend和scheduler
        (backend, scheduler)
        Standalone模式模式中,TaskScheduler的实现为TaskSchedulerImpl,而SchedulerBackend的实现为SparkDeploySchedulerBackend,并且在TaskScheduler生成后,随即调用其initialize()方法完成了初始化,也就确定了SchedulableBuilder和SchedulerBackend。

        至此,前两个是什么以及如何初始化的问题我们都已得到答案,下面再看最后一个关于做什么的问题:SchedulerBackend的reviveOffers()到底做了什么?还是以Standalone模式来说明。SparkDeploySchedulerBackend中没有提供此方法,我们只能寄希望于其父类CoarseGrainedSchedulerBackend,果不其然,在CoarseGrainedSchedulerBackend中我们找到了reviveOffers()方法。但是,代码很简单:

override def reviveOffers() {
    // 调用driverEndpoint的send()方法,发送一个ReviveOffers消息
    driverEndpoint.send(ReviveOffers)
  }
        我们继续看driverEndpoint是什么鬼。driverEndpoint是RPC中driver端Endpoint的引用,其类型为RpcEndpointRef。在CoarseGrainedSchedulerBackend启动时的start()方法中,对driverEndpoint进行了赋值:

// TODO (prashant) send conf instead of properties
    driverEndpoint = rpcEnv.setupEndpoint(ENDPOINT_NAME, createDriverEndpoint(properties))
        这个RpcEnv只是一个抽象类,它有两种实现,一个是基于AKKa的AkkaRpcEnv,另外一个则是基于Netty的NettyRpcEnv,默认的实现是Netty。通过下述RpcEnv的代码即可看出:

private def getRpcEnvFactory(conf: SparkConf): RpcEnvFactory = {
    
    // 两种实现方式:
    // akka:org.apache.spark.rpc.akka.AkkaRpcEnvFactory
    // netty:org.apache.spark.rpc.netty.NettyRpcEnvFactory
    val rpcEnvNames = Map(
      "akka" -> "org.apache.spark.rpc.akka.AkkaRpcEnvFactory",
      "netty" -> "org.apache.spark.rpc.netty.NettyRpcEnvFactory")
    
    // 通过参数spark.rpc配置,默认为netty
    val rpcEnvName = conf.get("spark.rpc", "netty")
    val rpcEnvFactoryClassName = rpcEnvNames.getOrElse(rpcEnvName.toLowerCase, rpcEnvName)
    Utils.classForName(rpcEnvFactoryClassName).newInstance().asInstanceOf[RpcEnvFactory]
  }

        下面,我们就看下Netty的概要实现,在NettyRpcEnv的setupEndpoint()方法中:

override def setupEndpoint(name: String, endpoint: RpcEndpoint): RpcEndpointRef = {
    
    // 调用Dispatcher的registerRpcEndpoint()方法完成注册
    dispatcher.registerRpcEndpoint(name, endpoint)
  }
        它是通过dispatcher来完成endpoint注册的,name为“CoarseGrainedScheduler”,RpcEndpoint为CoarseGrainedSchedulerBackend中通过createDriverEndpoint()方法创建的DriverEndpoint对象。代码如下:

protected def createDriverEndpoint(properties: Seq[(String, String)]): DriverEndpoint = {
    new DriverEndpoint(rpcEnv, properties)
  }
        那么这个DriverEndpoint是什么类呢?我们发现它继承自ThreadSafeRpcEndpoint,继而继承RpcEndpoint这个类。这里,我们只要知道这个RpcEndpoint是进程间消息传递调用的一个端点,定义了消息触发的函数。当一个消息到来时,方法调用顺序为  onStart, receive, onStop。它的生命周期为constructor -> onStart -> receive* -> onStop。

        为什么要用RpcEndpoint呢?很简单,Task的调度与执行是在一个分布式集群上进行的,自然需要进程间的通讯。

        继续分析,那么上面提到的driverEndpoint是如何赋值的呢?我们继续看Dispatcher的registerRpcEndpoint()方法,因为最终是由它向上返回RpcEndpointRef来完成driverEndpoint的赋值的。代码如下:

// 注册RpcEndpoint
  // name为“Master”,endpoint为Master对象
  def registerRpcEndpoint(name: String, endpoint: RpcEndpoint): NettyRpcEndpointRef = {
    
    // 创建RpcEndpointAddress
    val addr = RpcEndpointAddress(nettyEnv.address, name)
    
    // 创建NettyRpcEndpointRef
    val endpointRef = new NettyRpcEndpointRef(nettyEnv.conf, addr, nettyEnv)
    
    // 同步代码块
    synchronized {
      if (stopped) {
        throw new IllegalStateException("RpcEnv has been stopped")
      }
      
      // ConcurrentHashMap的putIfAbsent()方法确保不会重复创建EndpointData
      if (endpoints.putIfAbsent(name, new EndpointData(name, endpoint, endpointRef)) != null) {
        throw new IllegalArgumentException(s"There is already an RpcEndpoint called $name")
      }
      
      
      val data = endpoints.get(name)
      endpointRefs.put(data.endpoint, data.ref)
      receivers.offer(data)  // for the OnStart message
    }
    endpointRef
  }
        返回的RpcEndpointRef为NettyRpcEndpointRef类型,而RpcEndpointRef则是一个远程RpcEndpoint的引用,通过它可以给远程RpcEndpoint发送消息,可以是同步可以是异步,它映射一个地址。这么看来,我们在远端(ps:另外的机器或者进程)注册了一个RpcEndpoint,即DriverEndpoint,而在本地端(当前机器或者进程)则持有一个RpcEndpoint的引用,即NettyRpcEndpointRef,可以由它来往远端发送消息,那么发送的是什么消息呢?我们现在返回CoarseGrainedSchedulerBackend中的reviveOffers()方法,发现发送的是ReviveOffers消息。这里只是发送,具体处理还要看远端的RpcEndpoint,即DriverEndpoint。通过上面我们可以知道,RpcEndpoint的服务流程为onStart()-->receive()--> onStop(),每当消息来临时,DriverEndpoint都会调用receive()方法来处理。关键代码如下:

// 如果是ReviveOffers事件,则调用makeOffers()方法
      case ReviveOffers =>
        makeOffers()
        继续追踪其makeOffers()方法,代码如下:

// Make fake resource offers on all executors
    // 在所有的executors上提供假的资源(抽象的资源,也就是资源的对象信息,我是这么理解的)
    private def makeOffers() {
      // Filter out executors under killing
      // 过滤掉under killing的executors
      val activeExecutors = executorDataMap.filterKeys(executorIsAlive)
      
      // 获取workOffers,即资源
      val workOffers = activeExecutors.map { case (id, executorData) =>
        // 创建WorkerOffer对象
        new WorkerOffer(id, executorData.executorHost, executorData.freeCores)
      }.toSeq
      
      // 调用scheduler的resourceOffers()方法,分配资源
      // 调用launchTasks()方法,启动tasks
      launchTasks(scheduler.resourceOffers(workOffers))
    }
        好了,留个尾巴,明天再继续分析吧~





        
        





         











评论 5
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值