spark源码分析之任务调度篇

DAG的生成

概述

DAG(Directed Acyclic Graph)叫做有向无环图，原始的RDD通过一系列的转换就就形成了DAG，根据RDD之间的依赖关系的不同将DAG划分成不同的Stage，对于窄依赖，partition的转换处理在Stage中完成计算。对于宽依赖，由于有Shuffle的存在，只能在parent RDD处理完成后，才能开始接下来的计算，因此宽依赖是划分Stage的依据。

窄依赖 指的是每一个父RDD的Partition最多被子RDD的一个Partition使用
宽依赖 指的是多个子RDD的Partition会依赖同一个父RDD的Partition

DAGScheduler调度队列

当我们看完Executor的创建与启动流程后，我们继续在SparkContext的构造方法中继续查看

class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationClient {
  。。。。。。
    
 private[spark] def createSparkEnv(
      conf: SparkConf,
      isLocal: Boolean,
      listenerBus: LiveListenerBus): SparkEnv = {
    //通过SparkEnv来创建createDriverEnv
    SparkEnv.createDriverEnv(conf, isLocal, listenerBus)
  }
  //在这里调用了createSparkEnv，返回一个SparkEnv对象，这个对象里面有很多重要属性，最重要的ActorSystem
  private[spark] val env = createSparkEnv(conf, isLocal, listenerBus)
  SparkEnv.set(env)


  //创建taskScheduler
  // Create and start the scheduler
  private[spark] var (schedulerBackend, taskScheduler) =
    SparkContext.createTaskScheduler(this, master)

  //创建DAGScheduler
  dagScheduler = new DAGScheduler(this)

  //启动TaksScheduler
  taskScheduler.start()
    。。。。。
}

在构造方法中还创建了一个DAGScheduler对象，这个类的任务就是用来划分Stage任务的，构造方法中初始化了 private[scheduler] val eventProcessLoop = new DAGSchedulerEventProcessLoop(this)
DAGSchedulerEventProcessLoop是一个事件总线对象，用来负责任务的分发，在构造方法eventProcessLoop.start()被调用，该方法是父类EventLoop的start

  def start(): Unit = {
    if (stopped.get) {
      throw new IllegalStateException(name + " has already been stopped")
    }
    // Call onStart before starting the event thread to make sure it happens before onReceive
    onStart()
    eventThread.start()
  }

调用了eventThread的start方法，开启了一个线程

  private val eventThread = new Thread(name) {
    setDaemon(true)

    override def run(): Unit = {
      try {
        while (!stopped.get) {
          val event = eventQueue.take()
          try {
            onReceive(event)
          } catch {
            case NonFatal(e) => {
              try {
                onError(e)
              } catch {
                case NonFatal(e) => logError("Unexpected error in " + name, e)
              }
            }
          }
        }
      } catch {
        case ie: InterruptedException => // exit even if eventQueue is not empty
        case NonFatal(e) => logError("Unexpected error in " + name, e)
      }
    }
  }

run方法中不断的从LinkedBlockingDeque阻塞队列中取消息，然后调用onReceive(event)方法，该方法是由子类DAGSchedulerEventProcessLoop实现的

  override def onReceive(event: DAGSchedulerEvent): Unit = event match {
    case JobSubmitted(jobId, rdd, func, partitions, allowLocal, callSite, listener, properties) =>
      //调用dagScheduler来出来提交任务
      dagScheduler.handleJobSubmitted(jobId, rdd, func, partitions, allowLocal, callSite,
        listener, properties)

    case StageCancelled(stageId) =>
      dagScheduler.handleStageCancellation(stageId)

    case JobCancelled(jobId) =>
      dagScheduler.handleJobCancellation(jobId)

    case JobGroupCancelled(groupId) =>
      dagScheduler.handleJobGroupCancelled(groupId)

    case AllJobsCancelled =>
      dagScheduler.doCancelAllJobs()

    case ExecutorAdded(execId, host) =>
      dagScheduler.handleExecutorAdded(execId, host)

    case ExecutorLost(execId) =>
      dagScheduler.handleExecutorLost(execId, fetchFailed = false)

    case BeginEvent(task, taskInfo) =>
      dagScheduler.handleBeginEvent(task, taskInfo)

    case GettingResultEvent(taskInfo) =>
      dagScheduler.handleGetTaskResult(taskInfo)

    case completion @ CompletionEvent(task, reason, _, _, taskInfo, taskMetrics) =>
      dagScheduler.handleTaskCompletion(completion)

    case TaskSetFailed(taskSet, reason) =>
      dagScheduler.handleTaskSetFailed(taskSet, reason)

    case ResubmitFailedStages =>
      dagScheduler.resubmitFailedStages()
  }

onReceive中会匹配到传入的任务类型，执行相应的逻辑。到此DAGScheduler的调度队列会一直挂起，不断轮询队列中的任务。

DAG提交Task任务流程

当RDD经过一系列的转换Transformation方法后，最终要执行Action动作方法，这里比如WordCount程序中最后调用collect()方法时会将数据提交到Master上运行，任务真正的被执行，这里的方法执行过程如下

  /**
   * Return an array that contains all of the elements in this RDD.
   */
  def collect(): Array[T] = {
    val results = sc.runJob(this, (iter: Iterator[T]) => iter.toArray)
    Array.concat(results: _*)
  }

sc 是SparkContext对象，这里调用 一个runJob该方法调用多次重载的方法后,该方法最终会调用 dagScheduler.runJob

  def runJob[T, U: ClassTag](
      rdd: RDD[T],
      func: (TaskContext, Iterator[T]) => U,
      partitions: Seq[Int],
      allowLocal: Boolean,
      resultHandler: (Int, U) => Unit) {
    if (stopped) {
      throw new IllegalStateException("SparkContext has been shutdown")
    }
    val callSite = getCallSite
    val cleanedFunc = clean(func)
    logInfo("Starting job: " + callSite.shortForm)
    if (conf.getBoolean("spark.logLineage", false)) {
      logInfo("RDD's recursive dependencies:\n" + rdd.toDebugString)
    }
    //dagScheduler出现了，可以切分stage
    dagScheduler.runJob(rdd, cleanedFunc, partitions, callSite, allowLocal,
      resultHandler, localProperties.get)
    progressBar.foreach(_.finishAll())
    rdd.doCheckpoint()
  }

dagScheduler的runJob是我们比较关心的

 def runJob[T, U: ClassTag](
 
    。。。。。

    val waiter = submitJob(rdd, func, partitions, callSite, allowLocal, resultHandler, properties)
    waiter.awaitResult() match {
      case JobSucceeded => {
        logInfo("Job %d finished: %s, took %f s".format
          (waiter.jobId, callSite.shortForm, (System.nanoTime - start) / 1e9))
      }
      case JobFailed(exception: Exception) =>
        logInfo("Job %d failed: %s, took %f s".format
          (waiter.jobId, callSite.shortForm, (System.nanoTime - start) / 1e9))
        throw exception
    }
  }

这里面的我们主要看的是submitJob(rdd, func, partitions, callSite, allowLocal, resultHandler, properties)提交任务

def submitJob[T, U](
      rdd: RDD[T],
      func: (TaskContext, Iterator[T]) => U,
      partitions: Seq[Int],
      callSite: CallSite,
      allowLocal: Boolean,
      resultHandler: (Int, U) => Unit,
      properties: Properties): JobWaiter[U] = {

     。。。。。。

    //把job加入到任务队列里面
    eventProcessLoop.post(JobSubmitted(
      jobId, rdd, func2, partitions.toArray, allowLocal, callSite, waiter, properties))
    waiter
  }

这里比较关键的地方是 eventProcessLoop.post往任务队列中加入一个JobSubmitted类型的任务，eventProcessLoop是在构造方法中就初始化好的事件总线对象，内部有一个线程不断的轮询队列里的任务

轮询到任务后调用onReceive方法匹配任务类型，在这里我们提交的任务是JobSubmitted类型

    case JobSubmitted(jobId, rdd, func, partitions, allowLocal, callSite, listener, properties) =>
      //调用dagScheduler来出来提交任务
      dagScheduler.handleJobSubmitted(jobId, rdd, func, partitions, allowLocal, callSite,
        listener, properties)

调用了handleJobSubmitted方法，接下来查看该方法

private[scheduler] def handleJobSubmitted(jobId: Int,
      finalRDD: RDD[_],
      func: (TaskContext, Iterator[_]) => _,
      partitions: Array[Int],
      allowLocal: Boolean,
      callSite: CallSite,
      listener: JobListener,
      properties: Properties) {
    var finalStage: Stage = null
    try {
      // New stage creation may throw an exception if, for example, jobs are run on a
      // HadoopRDD whose underlying HDFS files have been deleted.
      //最终的stage
      finalStage = newStage(finalRDD, partitions.size, None, jobId, callSite)
    } catch {
      case e: Exception =>
        logWarning("Creating new stage failed due to exception - job: " + jobId, e)
        listener.jobFailed(e)
        return
    }
        。。。。
        submitStage(finalStage)
   }

上面的代码中，调用了newStage进行任务的划分，该方法是划分任务的核心方法，划分任务的根据最后一个依赖关系作为开始，通过递归，将每个宽依赖做为切分Stage的依据，切分Stage的过程是流程中的一环，但在这里不详细阐述，当任务切分完毕后，代码继续执行来到submitStage(finalStage)这里开始进行任务提交
这里以递归的方式进行任务的提交

//递归的方式提交stage
  private def submitStage(stage: Stage) {
    val jobId = activeJobForStage(stage)
    if (jobId.isDefined) {
      logDebug("submitStage(" + stage + ")")
      if (!waitingStages(stage) && !runningStages(stage) && !failedStages(stage)) {
        val missing = getMissingParentStages(stage).sortBy(_.id)
        logDebug("missing: " + missing)
        if (missing == Nil) {
          logInfo("Submitting " + stage + " (" + stage.rdd + "), which has no missing parents")
            //提交任务
          submitMissingTasks(stage, jobId.get)
        } else {
          for (parent <- missing) {
            submitStage(parent)
          }
          waitingStages += stage
        }
      }
    } else {
      abortStage(stage, "No active job for stage " + stage.id)
    }
  }

调用submitMissingTasks(stage, jobId.get)提交任务，将每一个Stage和jobId传入

  private def submitMissingTasks(stage: Stage, jobId: Int) {
   。。。。。

    if (tasks.size > 0) {
      logInfo("Submitting " + tasks.size + " missing tasks from " + stage + " (" + stage.rdd + ")")
      stage.pendingTasks ++= tasks
      logDebug("New pending tasks: " + stage.pendingTasks)
      //taskScheduler提交task
      taskScheduler.submitTasks(
        new TaskSet(tasks.toArray, stage.id, stage.newAttemptId(), stage.jobId, properties))
      stage.latestInfo.submissionTime = Some(clock.getTimeMillis())
    } else {
      // Because we posted SparkListenerStageSubmitted earlier, we should mark
      // the stage as completed here in case there are no tasks to run
      markStageAsFinished(stage, None)
      logDebug("Stage " + stage + " is actually done; %b %d %d".format(
        stage.isAvailable, stage.numAvailableOutputs, stage.numPartitions))
    }
  }

这里的代码我们需要关注的是`taskScheduler.submitTasks(

    new TaskSet(tasks.toArray, stage.id, stage.newAttemptId(), stage.jobId, properties))`

创建了一个TaskSet对象，将所有任务的信息封装，包括task任务列表，stageId,任务id,分区数参数等

Task任务调度

override def submitTasks(taskSet: TaskSet) {
    val tasks = taskSet.tasks
    logInfo("Adding task set " + taskSet.id + " with " + tasks.length + " tasks")
    this.synchronized {
        //创建TaskSetManager保存了taskSet任务列表
      val manager = createTaskSetManager(taskSet, maxTaskFailures)
      activeTaskSets(taskSet.id) = manager
     //将任务加入调度池
      schedulableBuilder.addTaskSetManager(manager, manager.taskSet.properties)

      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, STARVATION_TIMEOUT)
      }
      hasReceivedTask = true
    }
    //接受任务
    backend.reviveOffers()
  }

该方法比较重要，主要将任务加入调度池，最后调用了backend.reviveOffers()这里的backend是CoarseGrainedSchedulerBackend一个Executor任务调度对象

  override def reviveOffers() {
    //自己给自己发消息
    driverActor ! ReviveOffers
  }

这里用了内部的DriverActor对象发送了一个内部消息给自己，接下来查看receiver方法接受的消息

      case ReviveOffers =>
        makeOffers()

收到消息后调用了 makeOffers()方法

    def makeOffers() {
      launchTasks(scheduler.resourceOffers(executorDataMap.map { case (id, executorData) =>
        new WorkerOffer(id, executorData.executorHost, executorData.freeCores)
      }.toSeq))
    }

makeOffers方法中，将Executor的信息集合与调度池中的Tasks封装成WokerOffers列表传给了
launchTasks

    def launchTasks(tasks: Seq[Seq[TaskDescription]]) {
      for (task <- tasks.flatten) {
           。。。。。。
        //把task序列化
        val serializedTask = ser.serialize(task)

            。。。。。
          val executorData = executorDataMap(task.executorId)
          executorData.freeCores -= scheduler.CPUS_PER_TASK
          //把序列化好的task发送给Executor
          executorData.executorActor ! LaunchTask(new SerializableBuffer(serializedTask))
        }
      }
    }

launchTasks方法将遍历Tasks集合,每个Task任务序列化，发送启动Task执行消息的给Executor
Executor的onReceive方法

  //DriverActor发送给Executor的启动Task的消息
    case LaunchTask(data) =>
      if (executor == null) {
        logError("Received LaunchTask command but executor was null")
        System.exit(1)
      } else {
        val ser = env.closureSerializer.newInstance()
        //把Task反序列化
        val taskDesc = ser.deserialize[TaskDescription](data.value)
        logInfo("Got assigned task " + taskDesc.taskId)
        //启动task
        executor.launchTask(this, taskId = taskDesc.taskId, attemptNumber = taskDesc.attemptNumber,
          taskDesc.name, taskDesc.serializedTask)
      }

Executor收到DriverActor发送的启动Task的消息，这里才开始真正执行任务了，将收到的Task序列化信息反序列化,调用Executor的launchTask方法执行任务

  def launchTask(
      context: ExecutorBackend,
      taskId: Long,
      attemptNumber: Int,
      taskName: String,
      serializedTask: ByteBuffer) {
    //把task的描述信息放到了一份TaskRunner
    val tr = new TaskRunner(context, taskId = taskId, attemptNumber = attemptNumber, taskName,
      serializedTask)
    runningTasks.put(taskId, tr)
    //然后把TaskRunner丢到线程池里面
    threadPool.execute(tr)
  }

launchTask内将Task提交到线程池去运行，TaskRunner是Runnable对象，里面的run方法执行了我们app生成的每一个RDD的链上的逻辑。