Spark作业执行流程源码解析

本文梳理一下Spark作业执行的流程。

Spark作业和任务调度系统是其核心,通过内部RDD的依赖DAG,使得模块之间的调用和处理变得游刃有余。

相关概念

Job(作业):通过行动操作生成的一个或多个调度阶段

Stage:根据依赖关系划分的多个任务集,称为调度阶段,也叫做TaskSet(任务集)。划分Stage是由DAGScheduler进行的,任务阶段分为Shuffle Map Stage和Result Stage。

Task:是Spark执行计算的最小单位,会被分发到Executor中执行。

DAGScheduler:是面向调度阶段的任务调度器,接收Spark应用提交的作业,根据依赖关系划分stage,并提交给TaskScheduler。

TaskScheduler:是面向任务的 调度器,接收DAGScheduler划分好的stage,发送给Worker节点的Executor运行任务。

关于RDD相关知识、行动操作、宽窄依赖请参考Spark RDD基本概念、宽窄依赖、转换行为操作

概述

Spark作业主要是根据我们编写的业务处理代码,生成一系列相互依赖的调度阶段,之后将调度阶段中的任务提交Executor的执行的过程。
Spark作业执行流程

上图是spark作业运行流程图。主要分为四块:

  • 构建DAG

    行动操作触发提交作业,提交之后根据依赖关系构造DAG。

  • 划分调度阶段、提交调度阶段

    DAGScheduler中根据宽依赖划分调度阶段(stage)。每个stage包含多个task,组成taskset提交给TaskScheduler执行

  • 通过集群管理器启动任务

    TaskScheduler收到DAGScheduler提交的任务集,以任务的形式一个个分发到Executor中进行执行。

  • Executor端执行任务,完成后存储报告结果

    Executor接到任务后,扔到线程池中执行任务。任务完成后,报告结果给Driver。

源码解析

从以下的代码展开叙述:

def main(args: Array[String]): Unit = {
    val sc = new SparkContext("local", "word-count", new SparkConf())
    val words = Seq("hello spark", "hello scala", "hello java")
    val rdd = sc.makeRDD(words)
    rdd
    .flatMap(_.split(" "))
    .map((_, 1))
    .reduceByKey(_ + _)
    .sortByKey()
    .foreach(println(_))
}

这是一个简单的WordCount案例。首先根据序列生成RDD,再经过一系列算子调用计算word的个数,之后再进行排序,输出结果。

作业提交

上面的代码中,flatMap、map、reduceByKey、sortByKey都是转化算子,不会触发计算;foreach是行动算子,会提交作业,触发计算。

看看foreach的内部的实现:

def foreach(f: T => Unit): Unit = withScope {
    val cleanF = sc.clean(f)
    // 将当前rdd引用和我们编写的函数传给sc.runJob
    sc.runJob(this, (iter: Iterator[T]) => iter.foreach(cleanF))
}
// 以下runJob函数都是SparkContext内部的重载函数
def runJob[T, U: ClassTag](rdd: RDD[T], func: Iterator[T] => U): Array[U] = {
    // 添加分区信息
    runJob(rdd, func, 0 until rdd.partitions.length)
}
def runJob[T, U: ClassTag](
    rdd: RDD[T],
    func: Iterator[T] => U,
    partitions: Seq[Int]): Array[U] = {
    val cleanedFunc = clean(func)
    runJob(rdd, (ctx: TaskContext, it: Iterator[T]) => cleanedFunc(it), partitions)
}
def runJob[T, U: ClassTag](
    rdd: RDD[T],
    func: (TaskContext, Iterator[T]) => U,
    partitions: Seq[Int]): Array[U] = {
    // 创建一个数组来保存结果
    val results = new Array[U](partitions.size)
    runJob[T, U](rdd, func, partitions, (index, res) => results(index) = res)
    results
}
// 多次调用runJob,之后将调用DAGScheduler的runJob提交作业
def runJob[T, U: ClassTag](
    rdd: RDD[T],
    func: (TaskContext, Iterator[T]) => U,
    partitions: Seq[Int],
    // 任务成功后的处理函数
    resultHandler: (Int, U) => Unit): Unit = {
    if (stopped.get()) {
        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)
    }
    // 调用DAGScheduler.runJob提交作业
    dagScheduler.runJob(rdd, cleanedFunc, partitions, callSite, resultHandler, localProperties.get)
    progressBar.foreach(_.finishAll())
    rdd.doCheckpoint()
}

foreach内部调用了SparkContext.runJob()提交作业,SparkContext内部反复调用了几次重载的runJob方法。

runJob最终的参数中有当前rdd的引用处理逻辑函数分区数等,之后调用DagScheduler.runJob()提交作业。

现在再来到DagScheduler.runJob(),看看内部调用:

def runJob[T, U](
    rdd: RDD[T],
    func: (TaskContext, Iterator[T]) => U,
    partitions: Seq[Int],
    callSite: CallSite,
    resultHandler: (Int, U) => Unit,
    properties: Properties): Unit = {
    val start = System.nanoTime
    // 提交作业
    // waiter是等待DAGScheduler作业完成的对象。
    // 任务完成后,它将结果传递给给定的处理函数
    val waiter = submitJob(rdd, func, partitions, callSite, resultHandler, properties)
    ThreadUtils.awaitReady(waiter.completionFuture, Duration.Inf)
    waiter.completionFuture.value.get match {
        case scala.util.Success(_) =>
        case scala.util.Failure(exception) =>
        val callerStackTrace = Thread.currentThread().getStackTrace.tail
        exception.setStackTrace(exception.getStackTrace ++ callerStackTrace)
        throw exception
    }
}
// 提交job,划分调度阶段
def submitJob[T, U](
    rdd: RDD[T],
    func: (TaskContext, Iterator[T]) => U,
    partitions: Seq[Int],
    callSite: CallSite,
    resultHandler: (Int, U) => Unit,
    properties: Properties): JobWaiter[U] = {
    // 检查以确保我们没有在不存在的分区上启动任务。
    val maxPartitions = rdd.partitions.length
    partitions.find(p => p >= maxPartitions || p < 0).foreach { p =>
        throw new IllegalArgumentException(
            "Attempting to access a non-existent partition: " + p + ". " +
            "Total number of partitions: " + maxPartitions)
    }
	// 为当前job获取id
    val jobId = nextJobId.getAndIncrement()
    // 如果分区为0,返回一个空job
    if (partitions.size == 0) {
        return new JobWaiter[U](this, jobId, 0, resultHandler)
    }
    assert(partitions.size > 0)
    val func2 = func.asInstanceOf[(TaskContext, Iterator[_]) => _]
    // 封装waiter,用于在执行结束时,回调处理结果
    val waiter = new JobWaiter(this, jobId, partitions.size, resultHandler)
    // eventProcessLoop是用于提交/接收DAG调度事件的事件环
    // 提交作业,告知DAGScheduler开始划分调度阶段。
    eventProcessLoop.post(JobSubmitted(
        jobId, rdd, func2, partitions.toArray, callSite, waiter,
        SerializationUtils.clone(properties)))
    waiter
}

内部调用了submitJob(),发送提交作业的消息到DAGScheduler的eventProcessLoop事件环中。

划分&提交调度阶段

eventProcessLoop是用于接收调度事件的调度环,对应的类是DAGSchedulerEventProcessLoop。

内部通过模式匹配接收消息,作出相应处理。接收到提交作业的消息后,调用dagScheduler.handleJobSubmitted()开始划分调度阶段、提交调度阶段。

private def doOnReceive(event: DAGSchedulerEvent): Unit = event match {
    // 匹配提交作业的消息
    case JobSubmitted(jobId, rdd, func, partitions, callSite, listener, properties) =>
    dagScheduler.handleJobSubmitted(jobId, rdd, func, partitions, callSite, listener, properties)
}

看看dagScheduler.handleJobSubmitted()内部:

private[scheduler] def handleJobSubmitted(jobId: Int,
                                          finalRDD: RDD[_],
                                          func: (TaskContext, Iterator[_]) => _,
                                          partitions: Array[Int],
                                          callSite: CallSite,
                                          listener: JobListener,
                                          properties: Properties) {
    var finalStage: ResultStage = null
    try {
        // 根据依赖关系创建ResultStage
        finalStage = createResultStage(finalRDD, func, partitions, jobId, callSite)
    } catch {
        ...
    }
    // 提交作业,清除内部数据
    barrierJobIdToNumTasksCheckFailures.remove(jobId)
	// 通过jobId, finalStage创建job
    val job = new ActiveJob(jobId, finalStage, callSite, listener, properties)
    val jobSubmissionTime = clock.getTimeMillis()
    // 将job存入jobId映射到job的map中
    jobIdToActiveJob(jobId) = job
    activeJobs += job
    finalStage.setActiveJob(job)
    val stageIds = jobIdToStageIds(jobId).toArray
    val stageInfos = stageIds.flatMap(id => stageIdToStage.get(id).map(_.latestInfo))
    listenerBus.post(
        SparkListenerJobStart(job.jobId, jobSubmissionTime, stageInfos, properties))
    // 提交调度阶段
    submitStage(finalStage)
}

handleJobSubmitted主要分为两块,一块是根据依赖生成ResultStage,一块是提交ResultStage

生成ResultStage

先看一下生成ResultStage,也就是createResultStage方法。

private def createResultStage(
    rdd: RDD[_],
    func: (TaskContext, Iterator[_]) => _,
    partitions: Array[Int],
    jobId: Int,
    callSite: CallSite): ResultStage = {
    checkBarrierStageWithDynamicAllocation(rdd)
    checkBarrierStageWithNumSlots(rdd)
    checkBarrierStageWithRDDChainPattern(rdd, partitions.toSet.size)
    // 先获取当前rdd的父调度阶段
    val parents = getOrCreateParentStages(rdd, jobId)
    val id = nextStageId.getAndIncrement()
    val stage = new ResultStage(id, rdd, func, partitions, parents, jobId, callSite)
    stageIdToStage(id) = stage
    updateJobIdStageIdMaps(jobId, stage)
    stage
}

会首先获取当前RDD的父阶段,获取后根据父阶段,创建ResultStage

这里注意一下,这里的rdd是ShuffledRDD的引用。因为我们foreach触发计算的时候,将调用rdd的引用传了进来,也就是sortByKey生成的ShuffledRDD的引用。

接着看getOrCreateParentStages()是怎么获取当前RDD的父阶段的:

private def getOrCreateParentStages(rdd: RDD[_], firstJobId: Int): List[Stage] = {
    // 获取宽依赖,之后根据获取的宽依赖,创建对应的ShuffleMapStage
    getShuffleDependencies(rdd).map { shuffleDep =>
        getOrCreateShuffleMapStage(shuffleDep, firstJobId)
    }.toList
}
// 获取当前RDD的宽依赖
// 返回作为给定RDD的直接父级的shuffle依赖项
// 此函数将不会返回更远的祖先。例如,如果C对B具有宽依赖性,而B对A具有宽依赖性
// A <-- B <-- C
// 用rdd C调用此函数只会返回B <-C依赖项。
private[scheduler] def getShuffleDependencies(
    rdd: RDD[_]): HashSet[ShuffleDependency[_, _, _]] = {
    val parents = new HashSet[ShuffleDependency[_, _, _]]
    val visited = new HashSet[RDD[_]]
    val waitingForVisit = new ArrayStack[RDD[_]]
    waitingForVisit.push(rdd)
    while (waitingForVisit.nonEmpty) {
        val toVisit = waitingForVisit.pop()
        if (!visited(toVisit)) {
            visited += toVisit
            toVisit.dependencies.foreach {
                case shuffleDep: ShuffleDependency[_, _, _] =>
                parents += shuffleDep
                case dependency =>
                waitingForVisit.push(dependency.rdd)
            }
        }
    }
    parents
}
// 如果shuffle map stage已在shuffleIdToMapStage中存在,则获取
// 不存在的话,将创建shuffle map stage 
private def getOrCreateShuffleMapStage(
    shuffleDep: ShuffleDependency[_, _, _],
    firstJobId: Int): ShuffleMapStage = {
    shuffleIdToMapStage.get(shuffleDep.shuffleId) match {
        case Some(stage) =>
        	stage
        case None =>
            // 查找尚未在shuffleToMapStage中注册的祖先shuffle依赖项,
            // 并为它创建shuffle map stage
            getMissingAncestorShuffleDependencies(shuffleDep.rdd).foreach { dep =>
                if (!shuffleIdToMapStage.contains(dep.shuffleId)) {
                    createShuffleMapStage(dep, firstJobId)
                }
            }
            // 为当前shuffle依赖创建shuffle map stage
            createShuffleMapStage(shuffleDep, firstJobId)
    }
}

getOrCreateParentStages中先调用getShuffleDependencies(),获取当前RDD的宽依赖;获取后,调用getOrCreateShuffleMapStage()为宽依赖创建stage(如果stage已存在就直接获取)。

先说一下getShuffleDependencies方法,如代码注释所说:返回作为给定RDD的直接父级的shuffle依赖项,不会返回整个DAG上所有的宽依赖。另外说一下,getShuffleDependencies这种写法感觉极度舒适,之后还有一个方法也是这么写。

我们传入的RDD是sortByKey生成的ShuffleRDD实例,调用getShuffleDependencies就会返回ShuffleDependency


再说一下getOrCreateShuffleMapStage方法,它为返回的ShuffleDependency创建shuffle map stage。

它内部会在shuffleIdToMapStage中找当前ShuffleDependency是否存在stage,如果存在则返回,不存在则创建。

在创建之前,首先会调用getMissingAncestorShuffleDependencies()获取当前依赖的所有祖先宽依赖,并判断他们是否存在对应的调度阶段,如果不存在则调用createShuffleMapStage()创建。确保所有祖先宽依赖都存在对应的调度阶段后,调用createShuffleMapStage()为当前ShuffleDependency创建stage。

看看getMissingAncestorShuffleDependenciescreateShuffleMapStage的实现:

// 查找所有尚未在shuffleToMapStage中注册的祖先shuffle依赖项
private def getMissingAncestorShuffleDependencies(
    rdd: RDD[_]): ArrayStack[ShuffleDependency[_, _, _]] = {
    val ancestors = new ArrayStack[ShuffleDependency[_, _, _]]
    val visited = new HashSet[RDD[_]]
    val waitingForVisit = new ArrayStack[RDD[_]]
    waitingForVisit.push(rdd)
    while (waitingForVisit.nonEmpty) {
        val toVisit = waitingForVisit.pop()
        if (!visited(toVisit)) {
            visited += toVisit
            // 获取宽依赖
            getShuffleDependencies(toVisit).foreach { shuffleDep =>
                if (!shuffleIdToMapStage.contains(shuffleDep.shuffleId)) {
                    ancestors.push(shuffleDep)
                    waitingForVisit.push(shuffleDep.rdd)
                } 
            }
        }
    }
    ancestors
}
// 为shuffle依赖创建shuffle map stage 
def createShuffleMapStage(shuffleDep: ShuffleDependency[_, _, _], jobId: Int): ShuffleMapStage = {
    val rdd = shuffleDep.rdd
    checkBarrierStageWithDynamicAllocation(rdd)
    checkBarrierStageWithNumSlots(rdd)
    checkBarrierStageWithRDDChainPattern(rdd, rdd.getNumPartitions)
    val numTasks = rdd.partitions.length
    val parents = getOrCreateParentStages(rdd, jobId)
    val id = nextStageId.getAndIncrement()
    val stage = new ShuffleMapStage(
        id, rdd, numTasks, parents, jobId, rdd.creationSite, shuffleDep, mapOutputTracker)

    stageIdToStage(id) = stage
    // 创建stage时会将stage放入shuffleId映射到stage的Map中
    shuffleIdToMapStage(shuffleDep.shuffleId) = stage
    updateJobIdStageIdMaps(jobId, stage)
    if (!mapOutputTracker.containsShuffle(shuffleDep.shuffleId)) {
        mapOutputTracker.registerShuffle(shuffleDep.shuffleId, rdd.partitions.length)
    }
    stage
}

getMissingAncestorShuffleDependencies和getShuffleDependencies的实现方法类似,返回所有尚未在shuffleToMapStage中注册的祖先shuffle依赖项。createShuffleMapStage为shuffle dependency创建shuffle map stage。


到此,getOrCreateParentStages的步骤就走完了,也就获取到了当前rdd的父阶段。

视线回到createResultStage方法中来:

val stage = new ResultStage(id, rdd, func, partitions, parents, jobId, callSite)

将stageId、rdd、处理逻辑方法、分区、父调度阶段等作为参数构造ResultStage。ResultStage就生成成功了。

提交ResultStage

在handleJobSubmitted方法中,调用submitStage()将生成的ResultStage提交。

看看submitStage内部:

// 提交阶段,但首先递归提交所有丢失的父阶段
private def submitStage(stage: Stage) {
    val jobId = activeJobForStage(stage)
    if (jobId.isDefined) {
        logDebug("submitStage(" + stage + ")")
        // 如果当前阶段不是在等待&不是在运行&没有结束,开始运行
        if (!waitingStages(stage) && !runningStages(stage) && !failedStages(stage)) {
            val missing = getMissingParentStages(stage).sortBy(_.id)
            logDebug("missing: " + missing)
            if (missing.isEmpty) {
                logInfo("Submitting " + stage + " (" + stage.rdd + "), which has no missing parents")
                submitMissingTasks(stage, jobId.get)
            } else {
                for (parent <- missing) {
                    submitStage(parent)
                }
                waitingStages += stage
            }
        }
    } else {
        abortStage(stage, "No active job for stage " + stage.id, None)
    }
}

submitStage先调用getMissingParentStages获取所有丢失的父阶段

如果没有丢失的父阶段,才会调用submitMissingTasks()提交当前阶段的任务集;如果存在丢失的父阶段,则递归调用submitStage先提交父阶段

getMissingParentStages的实现方式和getShuffleDependencies也类似,这里就不看了,它的作用就是获取所有丢失的父阶段。

再大致说一下submitMissingTasks()是怎么提交任务的:

val tasks: Seq[Task[_]] = try {
    stage match {
        case stage: ShuffleMapStage =>
            stage.pendingPartitions.clear()
            partitionsToCompute.map { id =>
                val locs = taskIdToLocations(id)
                val part = partitions(id)
                stage.pendingPartitions += id
                // 创建shuffleMapTask
                new ShuffleMapTask(stage.id, stage.latestInfo.attemptNumber,
                                   taskBinary, part, locs, properties, serializedTaskMetrics, Option(jobId),
                                   Option(sc.applicationId), sc.applicationAttemptId, stage.rdd.isBarrier())
            }

        case stage: ResultStage =>
            partitionsToCompute.map { id =>
                val p: Int = stage.partitions(id)
                val part = partitions(p)
                val locs = taskIdToLocations(id)
                // 创建ResultTask
                new ResultTask(stage.id, stage.latestInfo.attemptNumber,
                               taskBinary, part, locs, id, properties, serializedTaskMetrics,
                               Option(jobId), Option(sc.applicationId), sc.applicationAttemptId,
                               stage.rdd.isBarrier())
            }
    }
}


if (tasks.size > 0) {
    // 调用taskScheduler.submitTasks()提交task
    taskScheduler.submitTasks(new TaskSet(
        tasks.toArray, stage.id, stage.latestInfo.attemptNumber, jobId, properties))
}

submitMissingTasks内部根据ShuffleMapStage和ResultStage分别生成ShuffleMapTask和ResultTask。

之后将task封装为TaskSet,调用TaskScheduler.submitTasks()提交任务。

到这里,划分和提交调度阶段已经走完了。接下来开始看提交任务的源码。

提交任务

上面调用了TaskScheduler.submitTasks()提交任务,TaskScheduler是特质,真正方法实现在类TaskSchedulerImpl中,我们看看内部实现:

override def submitTasks(taskSet: TaskSet) {
    val tasks = taskSet.tasks
    this.synchronized {
        // 为该TaskSet创建TaskSetManager,管理这个任务集的生命周期
        val manager = createTaskSetManager(taskSet, maxTaskFailures)
        val stage = taskSet.stageId
        val stageTaskSets =
        taskSetsByStageIdAndAttempt.getOrElseUpdate(stage, new HashMap[Int, TaskSetManager])
        stageTaskSets.foreach { case (_, ts) =>
            ts.isZombie = true
        }
        stageTaskSets(taskSet.stageAttemptId) = manager
        // 将该任务集的管理器加入到系统调度池中去,由系统统一调度
        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_MS, STARVATION_TIMEOUT_MS)
        }
        hasReceivedTask = true
    }
    backend.reviveOffers()
}

会首先为每个TaskSet创建TaskSetManager用于管理整个TaskSet的生命周期,并调用schedulableBuilder.addTaskSetManager将任务集管理器添加到系统调度池中去,之后调用SchedulerBackend.reviveOffers()分配资源并运行

看一下SchedulerBackend的其中一个子类CoarseGrainedSchedulerBackend的实现:

override def reviveOffers() {
	// 向Driver发送ReviveOffsers的消息
    driverEndpoint.send(ReviveOffers)
}

内部会向Driver终端点发送ReviveOffers的消息,分配资源并运行。

CoarseGrainedSchedulerBackend的实例就是代表Driver端的守护进程,其实也相当于自己发给自己。

接收到ReviveOffers的消息后,会调用makeOffers()

看看makeOffers()实现:

private def makeOffers() {
    val taskDescs = withLock {
        // 获取集群中可用的Executor列表
        val activeExecutors = executorDataMap.filterKeys(executorIsAlive)
        val workOffers = activeExecutors.map {
            case (id, executorData) =>
            new WorkerOffer(id, executorData.executorHost, executorData.freeCores,
                            Some(executorData.executorAddress.hostPort))
        }.toIndexedSeq
        // 分配运行资源
        scheduler.resourceOffers(workOffers)
    }
    if (!taskDescs.isEmpty) {
        // 提交任务
        launchTasks(taskDescs)
    }
}

makeOffers()内部会先获取所有可用的Executor列表,然后调用TaskSchedulerImpl.resourceOffers()分配资源,分配资源完成后,调用launchTask()提交任务

看看TaskSchedulerImpl.resourceOffers()的实现:

// 由集群管理器调用以在slave上提供资源。
def resourceOffers(offers: IndexedSeq[WorkerOffer]): Seq[Seq[TaskDescription]] = synchronized {
    //将每个slave标记为活动并记住其主机名, 还跟踪是否添加了新的Executor
    var newExecAvail = false
    for (o <- offers) {
        if (!hostToExecutors.contains(o.host)) {
            hostToExecutors(o.host) = new HashSet[String]()
        }
        if (!executorIdToRunningTaskIds.contains(o.executorId)) {
            hostToExecutors(o.host) += o.executorId
            executorAdded(o.executorId, o.host)
            executorIdToHost(o.executorId) = o.host
            executorIdToRunningTaskIds(o.executorId) = HashSet[Long]()
            newExecAvail = true
        }
        for (rack <- getRackForHost(o.host)) {
            hostsByRack.getOrElseUpdate(rack, new HashSet[String]()) += o.host
        }
    }
    // 移除黑名单中的节点
    blacklistTrackerOpt.foreach(_.applyBlacklistTimeout())
    val filteredOffers = blacklistTrackerOpt.map { blacklistTracker =>
        offers.filter { offer =>
            !blacklistTracker.isNodeBlacklisted(offer.host) &&
            !blacklistTracker.isExecutorBlacklisted(offer.executorId)
        }
    }.getOrElse(offers)
    // 为任务随机分配Executor,避免任务集中分配到Worker上
    val shuffledOffers = shuffleOffers(filteredOffers)
    // 存储已分配好的任务
    val tasks = shuffledOffers.map(o => new ArrayBuffer[TaskDescription](o.cores / CPUS_PER_TASK))
    val availableCpus = shuffledOffers.map(o => o.cores).toArray
    val availableSlots = shuffledOffers.map(o => o.cores / CPUS_PER_TASK).sum
    // 获取按照调度策略排序好的TaskSetManager
    val sortedTaskSets = rootPool.getSortedTaskSetQueue
    for (taskSet <- sortedTaskSets) {
        logDebug("parentName: %s, name: %s, runningTasks: %s".format(
            taskSet.parent.name, taskSet.name, taskSet.runningTasks))
        if (newExecAvail) {
            taskSet.executorAdded()
        }
    }

    // 为排好序的TaskSetManager列表进行分配资源。分配的原则是就近原则,按照顺序为PROCESS_LOCAL、NODE_LOCAL、NO_PREF、RACK_LOCAL、ANY
    for (taskSet <- sortedTaskSets) {
        if (taskSet.isBarrier && availableSlots < taskSet.numTasks) {
            ...
        } else {
            var launchedAnyTask = false
            val addressesWithDescs = ArrayBuffer[(String, TaskDescription)]()
            for (currentMaxLocality <- taskSet.myLocalityLevels) {
                var launchedTaskAtCurrentMaxLocality = false
                do {
                    launchedTaskAtCurrentMaxLocality = resourceOfferSingleTaskSet(taskSet,
                                                                                  currentMaxLocality, shuffledOffers, availableCpus, tasks, addressesWithDescs)
                    launchedAnyTask |= launchedTaskAtCurrentMaxLocality
                } while (launchedTaskAtCurrentMaxLocality)
            }
            ... 
        }
    }

    if (tasks.size > 0) {
        hasLaunchedTask = true
    }
    return tasks
}

resourceOffers中按照调度策略、就近原则为Task分配资源,返回分配好资源的Task。

分配好资源后,调用launchTasks()提交任务

private def launchTasks(tasks: Seq[Seq[TaskDescription]]) {
    for (task <- tasks.flatten) {
        // 序列化任务
        val serializedTask = ser.serialize(task)
        if (serializedTask.limit >= maxRpcMessageSize) {
           ...
        }
        else {
            val executorData = executorDataMap(task.executorId)
            executorData.freeCores -= scheduler.CPUS_PER_TASK
			// 向Executor所在节点终端发送LaunchTask的消息
            executorData.executorEndpoint.send(LaunchTask(new SerializableBuffer(serializedTask)))
        }
    }
}

launchTasks内部先将任务序列化,之后把任务一个个的发送到对应的CoarseGrainedExecutorBackend进行执行。

至此任务就提交完成了,接下来看Executor是如何执行任务的。

执行任务

CoarseGrainedExecutorBackend接收到LaunchTask消息后,会调用Executor.launchTask()执行任务

override def receive: PartialFunction[Any, Unit] = {
    case LaunchTask(data) =>
    if (executor == null) {
        exitExecutor(1, "Received LaunchTask command but executor was null")
    } else {
        val taskDesc = TaskDescription.decode(data.value)
        logInfo("Got assigned task " + taskDesc.taskId)
        // 调用Executor.launchTask执行任务
        executor.launchTask(this, taskDesc)
    }
}

看看Executor.launchTask的实现:

def launchTask(context: ExecutorBackend, taskDescription: TaskDescription): Unit = {
    // 将Task封装到TaskRunner中
    val tr = new TaskRunner(context, taskDescription)
    runningTasks.put(taskDescription.taskId, tr)
    // 将TaskRunner扔到线程池中进行执行
    threadPool.execute(tr)
}

launchTask中会将Task封装到TaskRunner中,然后把TaskRunner扔到线程池中进行执行。

TaskRunner是一个线程类,看一下它run方法的操作:

override def run(): Unit = {
    threadId = Thread.currentThread.getId
    Thread.currentThread.setName(threadName)
    val threadMXBean = ManagementFactory.getThreadMXBean
    val taskMemoryManager = new TaskMemoryManager(env.memoryManager, taskId)
    val deserializeStartTime = System.currentTimeMillis()
    val deserializeStartCpuTime = if (threadMXBean.isCurrentThreadCpuTimeSupported) {
        threadMXBean.getCurrentThreadCpuTime
    } else 0L
    Thread.currentThread.setContextClassLoader(replClassLoader)
    val ser = env.closureSerializer.newInstance()
    // 开始运行
    execBackend.statusUpdate(taskId, TaskState.RUNNING, EMPTY_BYTE_BUFFER)
    var taskStartTime: Long = 0
    var taskStartCpu: Long = 0
    startGCTime = computeTotalGcTime()

    try {
        // 反序列化任务
        task = ser.deserialize[Task[Any]](
            taskDescription.serializedTask, Thread.currentThread.getContextClassLoader)
        // value是返回结果
        val value = Utils.tryWithSafeFinally {
            // 调用Task.run运行Task,并获取返回结果
            val res = task.run(
                taskAttemptId = taskId,
                attemptNumber = taskDescription.attemptNumber,
                metricsSystem = env.metricsSystem)
            threwException = false
            res
        } {
            val releasedLocks = env.blockManager.releaseAllLocksForTask(taskId)
            val freedMemory = taskMemoryManager.cleanUpAllAllocatedMemory()     
        }

        val resultSer = env.serializer.newInstance()
        val beforeSerialization = System.currentTimeMillis()
        val valueBytes = resultSer.serialize(value)
        val afterSerialization = System.currentTimeMillis()
        val directResult = new DirectTaskResult(valueBytes, accumUpdates)
        val serializedDirectResult = ser.serialize(directResult)
        val resultSize = serializedDirectResult.limit()

        // 执行结果的处理
        val serializedResult: ByteBuffer = {
            // 结果大于maxResultSize,直接丢弃;这个值通过spark.driver.maxResultSize进行设置
            if (maxResultSize > 0 && resultSize > maxResultSize) {
            	ser.serialize(new IndirectTaskResult[Any](TaskResultBlockId(taskId), resultSize))
            
            }
            // 结果大于maxDirectResultSize,存放到BlockManager中,然后将BlockId发送到Driver
            else if (resultSize > maxDirectResultSize) {
                val blockId = TaskResultBlockId(taskId)
                env.blockManager.putBytes(
                    blockId,
                    new ChunkedByteBuffer(serializedDirectResult.duplicate()),
                    StorageLevel.MEMORY_AND_DISK_SER)
                ser.serialize(new IndirectTaskResult[Any](blockId, resultSize))
            }
            // 直接将结果发到Driver
            else {
                serializedDirectResult
            }
        }
		// 任务执行完成,调用CoarseGrainedExecutorBackend.statusUpdate
        execBackend.statusUpdate(taskId, TaskState.FINISHED, serializedResult)
    } catch {
        ...
    } finally {
        runningTasks.remove(taskId)
    }
}

run方法中,会将任务反序列化,然后调用Task.run()执行Task;执行完成后获取执行结果,根据结果的大小分情况处理,之后调用CoarseGrainedExecutorBackend.statusUpdate()向Driver汇报执行结果


Task的run方法中,会调用runTask()执行任务

Task是抽象类,没有对runTask()进行实现。具体的实现是由ShuffleMapTask和ResultTask进行的。

先看看ShuffleMapTask的runTask的实现:

override def runTask(context: TaskContext): MapStatus = {
    val threadMXBean = ManagementFactory.getThreadMXBean
    val deserializeStartTime = System.currentTimeMillis()
    val deserializeStartCpuTime = if (threadMXBean.isCurrentThreadCpuTimeSupported) {
        threadMXBean.getCurrentThreadCpuTime
    } else 0L
    // 反序列化
    val ser = SparkEnv.get.closureSerializer.newInstance()
    val (rdd, dep) = ser.deserialize[(RDD[_], ShuffleDependency[_, _, _])](
        ByteBuffer.wrap(taskBinary.value), Thread.currentThread.getContextClassLoader)
    _executorDeserializeTime = System.currentTimeMillis() - deserializeStartTime
    _executorDeserializeCpuTime = if (threadMXBean.isCurrentThreadCpuTimeSupported) {
        threadMXBean.getCurrentThreadCpuTime - deserializeStartCpuTime
    } else 0L
	
    var writer: ShuffleWriter[Any, Any] = null
    try {
        val manager = SparkEnv.get.shuffleManager
        writer = manager.getWriter[Any, Any](dep.shuffleHandle, partitionId, context)
       // 执行计算,并将结果写入本地系统的BlockManager中
        writer.write(rdd.iterator(partition, context).asInstanceOf[Iterator[_ <: Product2[Any, Any]]])
        // 关闭writer,返回计算结果
        // 返回包含了数据的location和size元数据信息的MapStatus信息
        writer.stop(success = true).get
    } catch {
    }
}

ShuffleMapTask会将计算结果写入到BlockManager中,最终会返回包含相关元数据信息的MapStatus。MapStatus将成为下一阶段获取输入数据时的依据。

再看看ResultTask的runTask的实现:

override def runTask(context: TaskContext): U = {
    // Deserialize the RDD and the func using the broadcast variables.
    val threadMXBean = ManagementFactory.getThreadMXBean
    val deserializeStartTime = System.currentTimeMillis()
    val deserializeStartCpuTime = if (threadMXBean.isCurrentThreadCpuTimeSupported) {
        threadMXBean.getCurrentThreadCpuTime
    } else 0L
    val ser = SparkEnv.get.closureSerializer.newInstance()
    // 反序列化
    val (rdd, func) = ser.deserialize[(RDD[T], (TaskContext, Iterator[T]) => U)](
        ByteBuffer.wrap(taskBinary.value), Thread.currentThread.getContextClassLoader)
    _executorDeserializeTime = System.currentTimeMillis() - deserializeStartTime
    _executorDeserializeCpuTime = if (threadMXBean.isCurrentThreadCpuTimeSupported) {
        threadMXBean.getCurrentThreadCpuTime - deserializeStartCpuTime
    } else 0L
	
    // 执行func进行计算
    func(context, rdd.iterator(partition, context))
}

ResultTask会直接执行封装进来的func函数,返回计算结果。


执行完成后,调用CoarseGrainedExecutorBackend.statusUpdate()。statusUpdate方法中向Driver终端点发送StatusUpdate的消息汇报任务执行结果。

结果处理

Driver接到StatusUpdate消息后,调用TaskSchedulerImpl.statusUpdate()进行处理

override def receive: PartialFunction[Any, Unit] = {
    case StatusUpdate(executorId, taskId, state, data) =>
    	// 调用statusUpdate处理
        scheduler.statusUpdate(taskId, state, data.value)
        if (TaskState.isFinished(state)) {
            executorDataMap.get(executorId) match {
                case Some(executorInfo) =>
                    executorInfo.freeCores += scheduler.CPUS_PER_TASK
                    makeOffers(executorId)
                case None =>
  					》。 
            }
        }
}

看看statusUpdate方法:

def statusUpdate(tid: Long, state: TaskState, serializedData: ByteBuffer) {
    var failedExecutor: Option[String] = None
    var reason: Option[ExecutorLossReason] = None
    synchronized {
        try {
            taskIdToTaskSetManager.get(tid) match {
                case Some(taskSet) =>
                // 如果FINISHED,调用taskResultGetter.enqueueSuccessfulTask()
                if (TaskState.isFinished(state)) {
                    cleanupTaskState(tid)
                    taskSet.removeRunningTask(tid)
                    if (state == TaskState.FINISHED) {
                        taskResultGetter.enqueueSuccessfulTask(taskSet, tid, serializedData)
                    } else if (Set(TaskState.FAILED, TaskState.KILLED, TaskState.LOST).contains(state)) {
                        taskResultGetter.enqueueFailedTask(taskSet, tid, state, serializedData)
                    }
                }
                case None =>
                	....
            }
        } catch {
        }
    }
}

statusUpdate内部会根据任务的状态不同做不同处理,这里只说一下任务是FINISHED的情况。

如果状态是TaskState.FINISHED,调用TaskResultGetter的enqueueSuccessfulTask方法处理

enqueueSuccessfulTask内部根据结果类型进行处理。如果是IndirectTaskResult,通过blockid从远程获取结果;如果DirectTaskResult,那么无需远程获取。


如果任务是ShuffleMapTask,需要将结果告知下游调度阶段,以便作为后续调度阶段的输入。

这个是在DAGScheduler的handleTaskCompletion中实现的,将MapStatus注册到MapOutputTrackerMaster中,从而完成ShuffleMapTask的处理

如果任务是ResultTask,如果完成,直接标记作业已经完成。


至此整个流程就走了一遍了。

在任务资源分配和结果处理说的有点不清晰,但对于了解整个任务执行流程没有很大影响。

end.

以上是结合看书以及看源码写的流程,如有偏差,欢迎交流指正。


Reference

《图解Spark核心技术与案例实战》



个人公众号:码农峰,定时推送行业资讯,持续发布原创技术文章,欢迎大家关注。

posted @ 2020-02-23 13:47  upupfeng  阅读(1610)  评论(0编辑  收藏  举报