Task的运行过程分析

时间:2024-01-04 14:52:44

Task的运行过程分析

Task的运行通过Worker启动时生成的Executor实例进行,

caseRegisteredExecutor(sparkProperties)=>

logInfo("Successfullyregistered with driver")

//Make this host instead of hostPort ?

executor=
newExecutor(executorId, Utils.parseHostPort(hostPort)._1,sparkProperties)

通过executor实例的launchTask启动task的运行操作。

deflaunchTask(context: ExecutorBackend,
taskId: Long, serializedTask:ByteBuffer) {

valtr
= newTaskRunner(context, taskId, serializedTask)

runningTasks.put(taskId,tr)

threadPool.execute(tr)

}

生成TaskRunner线程,把task与当前的Wroker的启动的executorBackend传入,

onyarn模式为CoarseGrainedExecutorBackend.

通过threadPool线程池运行生成TaskRunner线程。

TaskRunner.run函数:

用于运行task任务的线程

overridedefrun():
Unit = SparkHadoopUtil.get.runAsUser(sparkUser){ () =>

valstartTime
=System.currentTimeMillis()

SparkEvn后面在进行分析。

SparkEnv.set(env)

Thread.currentThread.setContextClassLoader(replClassLoader)

valser
=SparkEnv.get.closureSerializer.newInstance()

logInfo("Runningtask ID " + taskId)

通过execBackend更新此task的状态。设置task的状态为RUNNING.向master发送StatusUpdate事件。

execBackend.statusUpdate(taskId,TaskState.RUNNING,EMPTY_BYTE_BUFFER)

varattemptedTask:Option[Task[Any]]
= None

vartaskStart:Long
=

defgcTime =ManagementFactory.getGarbageCollectorMXBeans.map(_.getCollectionTime).sum

valstartGCTime=
gcTime

try{

SparkEnv.set(env)

Accumulators.clear()

解析出task的资源信息。包含要运行的jar,其他资源,task定义信息

val(taskFiles,taskJars,taskBytes)=
Task.deserializeWithDependencies(serializedTask)

更新资源信息,并把task运行的jar更新到当前Thread的ClassLoader中。

updateDependencies(taskFiles,taskJars)

通过SparkEnv中配置的Serialize实现对task定义进行反serialize,得到Task实例。

Task的详细实现为ShuffleMapTask或者ResultTask

task=
ser.deserialize[Task[Any]](taskBytes,Thread.currentThread.getContextClassLoader)

假设killed的值为true,不运行当前task任务,进入catch处理。

//If this task has been killed before we
deserializedit, let's quit now. Otherwise,

//continue executing the task.

if(killed)
{

//Throw an exception rather than returning, because returning within atry{} block

//causes a NonLocalReturnControl exception to be thrown. TheNonLocalReturnControl

//exception will be caught by the catch block, leading to an incorrectExceptionFailure

//for the task.

throwTaskKilledException

}

attemptedTask= Some(task)

logDebug("Task" + taskId +"'sepoch
is " + task.epoch)

env.mapOutputTracker.updateEpoch(task.epoch)

生成TaskContext实例,通过Task.runTask运行task的任务,等待task运行完毕。

//Run the actual task and measure its runtime.

taskStart= System.currentTimeMillis()

valvalue
=task.run(taskId.toInt)

valtaskFinish=
System.currentTimeMillis()

此时task运行结束,检查假设task是被killed的结果,进入catch处理。

//If the task has been killed, let's fail it.

if(task.killed){

throwTaskKilledException

}

对task运行的返回结果进行serialize操作。

valresultSer
=SparkEnv.get.serializer.newInstance()

valbeforeSerialization=
System.currentTimeMillis()

valvalueBytes=
resultSer.serialize(value)

valafterSerialization=
System.currentTimeMillis()

发送监控指标

for(m
<-task.metrics){

m.hostname=
Utils.localHostName()

m.executorDeserializeTime=
(taskStart-
startTime).toInt

m.executorRunTime=
(taskFinish-
taskStart).toInt

m.jvmGCTime=
gcTime - startGCTime

m.resultSerializationTime=
(afterSerialization-
beforeSerialization).toInt

}

valaccumUpdates=
Accumulators.values

把Task的返回结果生成DirectTaskResult实例。并对其进行serialize操作。

valdirectResult=
newDirectTaskResult(valueBytes,accumUpdates,task.metrics.getOrElse(null))

valserializedDirectResult=
ser.serialize(directResult)

logInfo("Serializedsize of result for " + taskId +
"is " +serializedDirectResult.limit)

检查taskresult的大小是否超过了akka的发送消息大小,

假设是通过BlockManager来管理结果,设置RDD的存储级别为MEMORY与DISK,否则表示没有达到actor消息大小,

直接使用TaskResult,此部分信息主要是须要通过状态更新向Scheduler向送StatusUpdate事件调用。

valserializedResult=
{

if(serializedDirectResult.limit>=
akkaFrameSize-
) {

logInfo("Storingresult for " + taskId +
"in local BlockManager")

valblockId
=TaskResultBlockId(taskId)

env.blockManager.putBytes(

blockId,serializedDirectResult,StorageLevel.MEMORY_AND_DISK_SER)

ser.serialize(newIndirectTaskResult[Any](blockId))

}
else{

logInfo("Sendingresult for " + taskId +
"directly to driver")

serializedDirectResult

}

}

通过execBackend更新此task的状态。设置task的状态为FINISHED.向master发送StatusUpdate事件。

execBackend.statusUpdate(taskId,TaskState.FINISHED,serializedResult)

logInfo("Finishedtask ID " + taskId)

}
catch{

出现异常,发送FAILED事件。

caseffe:FetchFailedException
=> {

valreason
=ffe.toTaskEndReason

execBackend.statusUpdate(taskId,TaskState.FAILED,ser.serialize(reason))

}

caseTaskKilledException => {

logInfo("Executorkilled task " + taskId)

execBackend.statusUpdate(taskId,TaskState.KILLED,ser.serialize(TaskKilled))

}

caset:Throwable
=> {

valserviceTime=
(System.currentTimeMillis() - taskStart).toInt

valmetrics
=attemptedTask.flatMap(t=> t.metrics)

for(m
<-metrics) {

m.executorRunTime=
serviceTime

m.jvmGCTime=
gcTime - startGCTime

}

valreason
=ExceptionFailure(t.getClass.getName,t.toString,t.getStackTrace,metrics)

execBackend.statusUpdate(taskId,TaskState.FAILED,ser.serialize(reason))

//TODO: Should we exit the whole executor here? On the one hand, thefailed task may

//have left some weird state around depending on when the exception wasthrown, but on

//the other hand, maybe we could detect that when future tasks fail andexit then.

logError("Exceptionin task ID " + taskId,
t)

//System.exit(1)

}

}
finally{

从shuffleMemoryMap中移出此线程相应的shuffle的内存空间

//TODO: Unregister shuffle memory only for ResultTask

valshuffleMemoryMap=
env.shuffleMemoryMap

shuffleMemoryMap.synchronized{

shuffleMemoryMap.remove(Thread.currentThread().getId)

}

从runningTasks中移出此task

runningTasks.remove(taskId)

}

}

}

Task运行过程的状态更新

ExecutorBackend.statusUpdate

onyarn模式实现类CoarseGrainedExecutorBackend,通过master的actor发送StatusUpdate事件。

overridedefstatusUpdate(taskId:
Long, state: TaskState, data: ByteBuffer) {

driver! StatusUpdate(executorId, taskId, state,
data)

}

master中的ExecutorBackend处理状态更新操作:

实现类:CoarseGrainedSchedulerBackend.DriverActor

caseStatusUpdate(executorId,taskId,state,data)
=>

通过TaskSchedulerImpl的statusUpdate处理状态更新。

scheduler.statusUpdate(taskId,state,data.value)

假设Task状态为完毕状态,完毕状态包括(FINISHED,FAILED,KILLED,LOST)

if(TaskState.isFinished(state)){

if(executorActor.contains(executorId)){

每个task占用一个cpucore,此时task完毕,把可用的core值加一

freeCores(executorId) +=

在此executor上接着运行其于的task任务,此部分可參见scheduler调度过程分析中的部分说明。

makeOffers(executorId)

}
else{

//Ignoring the update since we don't know about the executor.

valmsg
="Ignored task status update (%dstate %s) from unknown executor %s with ID %s"

logWarning(msg.format(taskId,state,sender,
executorId))

}

}

TaskSchedulerImpl.statusUpdate函数处理流程

defstatusUpdate(tid: Long, state: TaskState,
serializedData: ByteBuffer){

varfailedExecutor:Option[String]
= None

synchronized {

try{

假设Task的状态传入为Task的运行丢失,同一时候task在executor列表中存在

if(state == TaskState.LOST&&
taskIdToExecutorId.contains(tid)){

得到此task运行的worker所属的executorID,

//We lost this entire executor, so remember that it's gone

valexecId
=taskIdToExecutorId(tid)

假设此executor是active的Executor,运行scheduler的executorLost操作。

包括TaskSetManager,会运行TaskSetManager.executorLost操作.

设置当前的executor为failedExecutor,共函数最后使用。

if(activeExecutorIds.contains(execId)){

removeExecutor(execId)

failedExecutor= Some(execId)

}

}

taskIdToTaskSetId.get(tid)match{

caseSome(taskSetId)=>

假设task状态是完毕状态,非RUNNING状态。移出相应的容器中的值

if(TaskState.isFinished(state)) {

taskIdToTaskSetId.remove(tid)

if(taskSetTaskIds.contains(taskSetId)){

taskSetTaskIds(taskSetId)-=
tid

}

taskIdToExecutorId.remove(tid)

}

activeTaskSets.get(taskSetId).foreach{
taskSet =>

假设task是成功完毕,从TaskSet中移出此task,同一时候通过TaskResultGetter获取数据。

if(state == TaskState.FINISHED){

taskSet.removeRunningTask(tid)

taskResultGetter.enqueueSuccessfulTask(taskSet,tid,
serializedData)

}
elseif(Set(TaskState.FAILED,TaskState.KILLED,TaskState.LOST).contains(state)){

task任务运行失败的处理部分:

taskSet.removeRunningTask(tid)

taskResultGetter.enqueueFailedTask(taskSet,tid, state,
serializedData)

}

}

caseNone =>

logInfo("Ignoringupdate with state %s from TID %s because its task set is gone"

.format(state, tid))

}

}
catch{

casee:Exception
=> logError("Exceptionin statusUpdate",
e)

}

}

假设有failed的worker,通过dagScheduler处理此executor.

//Update the DAGScheduler without holding a lock on this, since thatcan deadlock

if(failedExecutor!=
None) {

dagScheduler.executorLost(failedExecutor.get)

发起task运行的分配与任务运行操作。

backend.reviveOffers()

}

}

TaskStatus.LOST状态,同一时候executor在activeExecutorIds中

TaskStatus的状态为LOST时,同一时候executor是活动的executor(也就是有过运行task的情况)

privatedefremoveExecutor(executorId:
String) {

从activeExecutorIds中移出此executor

activeExecutorIds-= executorId

得到此executor相应的worker的host

valhost
=executorIdToHost(executorId)

取出host相应的全部executor,并移出当前的executor

valexecs
=executorsByHost.getOrElse(host,newHashSet)

execs-=
executorId

if(execs.isEmpty){

executorsByHost-=
host

}

从executor相应的host容器中移出此executor

executorIdToHost-= executorId

此处主要是去运行TaskSetManager.executorLost函数。

rootPool.executorLost(executorId,host)

}

TaskSetManager.executorLost函数:

此函数主要处理executor导致task丢失的情况,把executor上的task又一次加入到pending的tasks列表中

overridedefexecutorLost(execId:
String, host: String) {

logInfo("Re-queueingtasks for " + execId +
"from TaskSet " +
taskSet.id)

//Re-enqueue pending tasks for this host based on the status of thecluster -- for example,
a

//task that used to have locations on only this host might now go tothe no-prefslist.
Note

//that it's okay if we add a task to the same queue twice (if it hadmultiple preferred

//locations), because findTaskFromList will skip already-running tasks.

又一次生成此TaskSet中的pending队列,由于当前executor的实例被移出,须要又一次生成。

for(index
<-getPendingTasksForExecutor(execId)) {

addPendingTask(index,readding=true)

}

for(index
<-getPendingTasksForHost(host)) {

addPendingTask(index,readding=true)

}

//Re-enqueue any tasks that ran on the failed executor if this is ashuffle map stage

假设当前的RDD是shuffle的rdd,

).isInstanceOf[ShuffleMapTask]){

for((tid,info)
<-taskInfosifinfo.executorId==
execId) {

valindex
=taskInfos(tid).index

if(successful(index)){

successful(index)=
false

copiesRunning(index) -=

tasksSuccessful-=

addPendingTask(index)

//Tell the DAGScheduler that this task was resubmitted so that itdoesn't think our

//stage finishes when a total of tasks.size tasks finish.

通过DAGScheduler发送CompletionEvent处理事件,事件类型为Resubmitted,

sched.dagScheduler.taskEnded(tasks(index),Resubmitted,
null,null,info,
null)

}

}

}

假设task还处于running状态,同一时候此task在lost的executor上执行,

//Also re-enqueue any tasks that were running on the node

for((tid,info)
<-taskInfosifinfo.running&&
info.executorId== execId) {

设置task的failed值为true,移出此task的running列表中的值,又一次加入task到pendingtasks队列中。

handleFailedTask(tid,TaskState.FAILED,None)

}

}

DAGScheduler处理CompletionEvent事件。

...........................

casecompletion@
CompletionEvent(task,reason,_,
_, taskInfo,taskMetrics)=>

listenerBus.post(SparkListenerTaskEnd(task,reason,taskInfo,taskMetrics))

handleTaskCompletion(completion)

.........................

caseResubmitted =>

logInfo("Resubmitted" +
task+
", so marking it as stillrunning")

pendingTasks(stage)+=
task

(TaskState.FAILED,TaskState.KILLED,TaskState.LOST)状态

.........................

}
elseif(Set(TaskState.FAILED,TaskState.KILLED,TaskState.LOST).contains(state)){

把task从running容器中移出

taskSet.removeRunningTask(tid)

此函数主要是解析出出错的信息。并通过TaskSchedulerImpl.handleFailedTask处理exception

taskResultGetter.enqueueFailedTask(taskSet,tid, state,
serializedData)

}

TaskSchedulerImpl.handleFailedTask函数:

defhandleFailedTask(

taskSetManager: TaskSetManager,

tid: Long,

taskState: TaskState,

reason: Option[TaskEndReason]) =synchronized {

taskSetManager.handleFailedTask(tid,taskState, reason)

假设task不是被KILLED掉的task,又一次发起task的分配与运行操作。

if(taskState != TaskState.KILLED){

//Need to revive offers again now that the task set manager state hasbeen updated to

//reflect failed tasks that need to be re-run.

backend.reviveOffers()

}

}

TaskSetManager.handleFailedTask函数流程

TaskSetManager.handleFailedTask,函数,处理task运行的exception信息。

defhandleFailedTask(tid: Long, state:
TaskState, reason:Option[TaskEndReason]) {

valinfo
=taskInfos(tid)

if(info.failed){

return

}

removeRunningTask(tid)

valindex
=info.index

info.markFailed()

varfailureReason=
"unknown"

if(!successful(index)){

logWarning("LostTID %s (task %s:%d)".format(tid,taskSet.id,index))

copiesRunning(index) -=

假设是通过TaskSetManager.executorLost函数发起的此函数调用(Task.LOST),以下的case部分不会运行,

否则是task的运行exception情况,也就是状态更新中非Task.LOST状态时。

//Check if the problem is a map output fetch failure. In that case,this

//task will never succeed on any node, so tell the scheduler about it.

reason.foreach {

casefetchFailed:FetchFailed
=>

读取失败,移出全部此taskset的task运行。并从scheduler中移出此taskset的调度,不再运行以下流程

logWarning("Losswas due to fetch failure from " +fetchFailed.bmAddress)

sched.dagScheduler.taskEnded(tasks(index),fetchFailed,null,null,info,
null)

successful(index)=
true

tasksSuccessful+=

sched.taskSetFinished(this)

removeAllRunningTasks()

return

caseTaskKilled =>

task被kill掉,移出此task,同一时候不再运行以下流程

logWarning("Task%d was killed.".format(tid))

sched.dagScheduler.taskEnded(tasks(index),reason.get,
null,null,info,
null)

return

caseef:ExceptionFailure
=>

sched.dagScheduler.taskEnded(

tasks(index),ef,
null,null,info,ef.metrics.getOrElse(null))

if(ef.className==
classOf[NotSerializableException].getName()) {

//If the task result wasn't
rerializable,there's no point in trying to re-execute it.

logError("Task%s:%s had a not serializable result: %s; not retrying".format(

taskSet.id,index,ef.description))

abort("Task%s:%s had a not serializable result: %s".format(

taskSet.id,index,ef.description))

return

}

valkey
=ef.description

failureReason=
"Exception failure:%s".format(ef.description)

valnow
=clock.getTime()

val(printFull,dupCount)
={

if(recentExceptions.contains(key)){

val(dupCount,printTime)=
recentExceptions(key)

if(now
-printTime >EXCEPTION_PRINT_INTERVAL){

recentExceptions(key)=
(,
now)

)

}
else{

recentExceptions(key)=
(dupCount+
,printTime)

(false,dupCount
+)

}

}
else{

recentExceptions(key)=
(,
now)

)

}

}

if(printFull){

vallocs
=ef.stackTrace.map(loc=>
"\tat%s".format(loc.toString))

logWarning("Losswas due to %s\n%s\n%s".format(

ef.className,ef.description,locs.mkString("\n")))

}
else{

logInfo("Losswas due to %s [duplicate %d]".format(ef.description,dupCount))

}

caseTaskResultLost =>

failureReason=
"Lost result for TID %s onhost %s".format(tid,
info.host)

logWarning(failureReason)

sched.dagScheduler.taskEnded(tasks(index),TaskResultLost,
null,null,info,
null)

case_ => {}

}

又一次把task加入到pending的运行队列中,同一时候假设状态非KILLED的状态,设置并检查是否达到重试的最大次数

//On non-fetch failures, re-enqueue the task as pending for a maxnumber of retries

addPendingTask(index)

if(state != TaskState.KILLED){

numFailures(index) +=

if(numFailures(index)>=
maxTaskFailures){

logError("Task%s:%d failed %d times; aborting job".format(

taskSet.id,index,maxTaskFailures))

abort("Task%s:%d failed %d times (most recent failure: %s)".format(

taskSet.id,index,maxTaskFailures,failureReason))

}

}

}else{

logInfo("Ignoringtask-lost event for TID " + tid +

"because task " +
index+
" is already finished")

}

}

DAGScheduler处理taskEnded流程:

deftaskEnded(

task: Task[_],

reason: TaskEndReason,

result: Any,

accumUpdates: Map[Long, Any],

taskInfo: TaskInfo,

taskMetrics: TaskMetrics) {

eventProcessActor! CompletionEvent(task, reason,
result, accumUpdates, taskInfo,taskMetrics)

}

处理CompletionEvent事件:

casecompletion@
CompletionEvent(task,reason, _,_,
taskInfo,taskMetrics)=>

listenerBus.post(SparkListenerTaskEnd(task,reason,taskInfo,taskMetrics))

handleTaskCompletion(completion)

DAGScheduler.handleTaskCompletion

读取失败的case,

caseFetchFailed(bmAddress,shuffleId,mapId,reduceId)=>

//Mark the stage that the reducer was in as
unrunnable

valfailedStage=
stageIdToStage(task.stageId)

running-=
failedStage

failed+=
failedStage

..............................

//Mark the map whose fetch failed as broken in the map stage

valmapStage
=shuffleToMapStage(shuffleId)

if(mapId
!=-) {

mapStage.removeOutputLoc(mapId,bmAddress)

mapOutputTracker.unregisterMapOutput(shuffleId,mapId,bmAddress)

}

...........................

failed+=
mapStage

//Remember that a fetch failed now; this is used to resubmit the broken

//stages later, after a small wait (to give other tasks the chance tofail)

lastFetchFailureTime= System.currentTimeMillis()
// TODO:Use pluggableclock

//TODO: mark the executor as failed only if there were lots of fetchfailures on it

if(bmAddress!=
null){

把stage中可运行的partition中相应的executorid的location所有移出。

handleExecutorLost(bmAddress.executorId,Some(task.epoch))

}

caseExceptionFailure(className,description,stackTrace,metrics)
=>

//Do nothing here, left up to the TaskScheduler to decide how to handleuser failures

caseTaskResultLost =>

//Do nothing here; the TaskScheduler handles these failures andresubmits the task.

TaskStatus.FINISHED状态

此状态表示task正常完毕,

if(state == TaskState.FINISHED){

移出taskSet中的running队列中移出此task

taskSet.removeRunningTask(tid)

获取task的响应数据。

taskResultGetter.enqueueSuccessfulTask(taskSet,tid,
serializedData)

TaskResultGetter.enqueueSuccessfulTask函数:

defenqueueSuccessfulTask(

taskSetManager: TaskSetManager,tid: Long, serializedData: ByteBuffer) {

getTaskResultExecutor.execute(newRunnable
{

overridedefrun()
{

try{

从响应的结果中得到数据,须要先运行deserialize操作。

valresult
=serializer.get().deserialize[TaskResult[_]](serializedData)match{

假设result的结果小于akka的actor传输的大小,直接返回task的运行结果

casedirectResult:DirectTaskResult[_]
=> directResult

caseIndirectTaskResult(blockId)=>

否则,result结果太大,通过BlockManager管理,通过blockManager拿到result的数据

logDebug("Fetchingindirect task result for TID %s".format(tid))

给DAGScheduler发送GettingResultEvent事件处理,

见以下TaskSchedulerImpl.handleTaskGettingResult函数

scheduler.handleTaskGettingResult(taskSetManager,tid)

得到task的运行结果

valserializedTaskResult=
sparkEnv.blockManager.getRemoteBytes(blockId)

task运行完毕,并拿结果失败,见上面的错误处理中的TaskResultLost部分。

if(!serializedTaskResult.isDefined){

/*We won't be able to get the task result if the machine that ran thetask failed

* between when the taskended and when we tried to fetch the result, or if the

* block manager had toflush the result. */

scheduler.handleFailedTask(

taskSetManager, tid,TaskState.FINISHED,Some(TaskResultLost))

return

}

对task的运行结果进行deserialized操作。

valdeserializedResult=
serializer.get().deserialize[DirectTaskResult[_]](

serializedTaskResult.get)

拿到运行结果,移出相应的blockid

sparkEnv.blockManager.master.removeBlock(blockId)

deserializedResult

}

result.metrics.resultSize=
serializedData.limit()

见以下的TaskSchedulerImpl.handleSuccessfulTask处理函数。

scheduler.handleSuccessfulTask(taskSetManager,tid,
result)

}
catch{

casecnf:ClassNotFoundException
=>

valloader
=Thread.currentThread.getContextClassLoader

taskSetManager.abort("ClassNotFoundwith classloader: "
+ loader)

caseex:Throwable
=>

taskSetManager.abort("Exceptionwhile deserializing and fetching task: %s".format(ex))

}

}

})

}

TaskSchedulerImpl.handleTaskGettingResult函数:

defhandleTaskGettingResult(taskSetManager:
TaskSetManager, tid: Long) {

taskSetManager.handleTaskGettingResult(tid)

}

taskSetManager中

defhandleTaskGettingResult(tid: Long)
= {

valinfo
=taskInfos(tid)

info.markGettingResult()

sched.dagScheduler.taskGettingResult(tasks(info.index),info)

}

通过DAGScheduler发起GettingResultEvent事件。

deftaskGettingResult(task: Task[_],
taskInfo: TaskInfo) {

eventProcessActor! GettingResultEvent(task, taskInfo)

}

对GettingResultEvent事件的处理:事实上就是打个酱油,无实际处理操作。

caseGettingResultEvent(task,taskInfo)=>

listenerBus.post(SparkListenerTaskGettingResult(task,taskInfo))

TaskSchedulerImpl.handleSuccessfulTask处理函数:

defhandleSuccessfulTask(

taskSetManager: TaskSetManager,

tid: Long,

taskResult: DirectTaskResult[_])= synchronized {

taskSetManager.handleSuccessfulTask(tid,taskResult)

}

TastSetManager中

defhandleSuccessfulTask(tid: Long,
result: DirectTaskResult[_]) = {

valinfo
=taskInfos(tid)

valindex
=info.index

info.markSuccessful()

从running队列中移出此task

removeRunningTask(tid)

if(!successful(index)){

logInfo("FinishedTID %s in %d ms on %s (progress: %d/%d)".format(

tid,
info.duration,info.host,tasksSuccessful,numTasks))

向dagscheduler发送success消息,

sched.dagScheduler.taskEnded(

tasks(index),Success,
result.value,result.accumUpdates,info,result.metrics)

设置成功完毕的task个数加一,同一时候在successful容器中设置task相应的执行状态为true,表示成功。

//Mark successful and stop if all the tasks have succeeded.

tasksSuccessful+=

successful(index)=
true

假设完毕的task个数,达到task的总个数,完毕此taskset,也就相当于完毕了一个rdd

if(tasksSuccessful==
numTasks){

sched.taskSetFinished(this)

}

}else{

logInfo("Ignorningtask-finished event for TID " + tid+
" because task "+

index+
" has already completedsuccessfully")

}

}

DAGScheduler处理CompletionEvent的Success,,,,

caseSuccess
=>

logInfo("Completed" +
task)

if(event.accumUpdates!=
null){

Accumulators.add(event.accumUpdates)//
TODO: do this only if task wasn'tresubmitted

}

把等待运行队列中移出此task

pendingTasks(stage)-=
task

stageToInfos(stage).taskInfos+=
event.taskInfo-> event.taskMetrics

依据task的运行类型,处理两个类型的Task

taskmatch{

假设task是ResultTask,表示不须要shuffle操作

casert:ResultTask[_,
_] =>

resultStageToJob.get(stage)match{

caseSome(job)=>

假设此运行的stage的ActiveJob中相应此task的partition存储的finished标志为false,

if(!job.finished(rt.outputId)){

设置task的完毕标志为true

job.finished(rt.outputId)=
true

把job中完毕的task个数加一,同一时候检查是否全部的task都完毕,假设全部task都完毕,

从相关的容器中移出此job与相应的stage.

job.numFinished+=

//If the whole job has finished, remove it

if(job.numFinished==
job.numPartitions){

idToActiveJob-=
stage.jobId

activeJobs-=
job

resultStageToJob-=
stage

markStageAsFinished(stage)

jobIdToStageIdsRemove(job.jobId)

listenerBus.post(SparkListenerJobEnd(job,JobSucceeded))

}

调用ActiveJob内的JobWaiter.taskSucceeded函数,更新此task为完毕,同一时候把result传入进行输出处理。

job.listener.taskSucceeded(rt.outputId,event.result)

}

caseNone =>

logInfo("Ignoringresult from " +
rt+
" because its job has finished")

}

针对shuffle的task的运行完毕,处理流程:

casesmt:ShuffleMapTask
=>

valstatus
=event.result.asInstanceOf[MapStatus]

valexecId
=status.location.executorId

logDebug("ShuffleMapTaskfinished on " +
execId)

if(failedEpoch.contains(execId)&&
smt.epoch<=
failedEpoch(execId)){

logInfo("Ignoringpossibly bogus ShuffleMapTask completion from "+
execId)

}
else{

把shuffle的result(MapStatus)写入到stage的outputLoc中。每加入一个会把numAvailableOutputs的值加一,

当numAvailableOutputs的值==numPartitions的值时,表示shuffle的map运行完毕。

stage.addOutputLoc(smt.partitionId,status)

}

假设此stage还处在running状态,同一时候pendingTasks中全部的task已经处理完毕

if(running.contains(stage)&&
pendingTasks(stage).isEmpty){

更新stage的状态

markStageAsFinished(stage)

.......................................

此处表示shuffle的stage处理完毕,把shuffleid与stage的outputLocs注冊到mapOutputTracker中。

把每个shuffletaks运行的executor与host等信息,每个task运行完毕的大小。注冊到mapoutput中。

每个task的shuffle的writer都会有shuffleid的信息,注冊成功后,

下一个stage会依据mapoutputtracker中此shuffleid的信息读取数据。

mapOutputTracker.registerMapOutputs(

stage.shuffleDep.get.shuffleId,

stage.outputLocs.map(list=>
if(list.isEmpty)
nullelselist.head).toArray,

changeEpoch =
true)

}

clearCacheLocs()

stage中每个partition的outputLoc默认值为Nil,假设发现有partition的值为Nil,表示有task处理失败,

又一次提交此stage.此时会把没有成功的task又一次运行。

if(stage.outputLocs.exists(_==
Nil)) {

.........................................

submitStage(stage)

}
else{

valnewlyRunnable=
newArrayBuffer[Stage]

for(stage
<-waiting) {

logInfo("Missingparents for " +
stage+
": "+ getMissingParentStages(stage))

}

此处检查以下未运行的全部的stage,假设stage(RDD)的上级shuffle依赖完毕,

或者后面全部的stage不再有shuffle的stage的全部stage,拿到这些个stage.

for(stage
<-waiting
ifgetMissingParentStages(stage) == Nil) {

newlyRunnable+=
stage

}

运行此stage后面的全部可运行的stage,把waiting中移出要运行的stage,

waiting--=
newlyRunnable

在running队列中加入要运行的新的stage.

running++=
newlyRunnable

for{

stage<-
newlyRunnable.sortBy(_.id)

jobId<- activeJobForStage(stage)

} {

提交下一个stage的task分配与运行。

logInfo("Submitting" +
stage+
" ("+ stage.rdd+
"), which is now runnable")

submitMissingTasks(stage,jobId)

}

}

}

}

JobWaiter.taskSucceeded函数,

task完毕后的处理函数。

overridedef taskSucceeded(index:
Int,result: Any): Unit = synchronized {

if(_jobFinished){

thrownewUnsupportedOperationException("taskSucceeded()called
on a finished JobWaiter")

}

通过resultHandler函数把结果进行处理。此函数是生成JobWaiter时传入

resultHandler(index,result.asInstanceOf[T])

把完毕的task值加一

finishedTasks+=

if(finishedTasks==
totalTasks) {

假设完毕的task个数等于全部的task的个数时,设置job的完毕状态为true,并设置状态为JobSucceeded

假设设置为true,表示job运行完毕,前面的等待运行完毕结束等待。

_jobFinished=
true

jobResult= JobSucceeded

this.notifyAll()

}

}

Task.runTask函数实现

Task的实现分为两类,

须要进行shuffle操作的ShuffleMapTask,

不须要进行shuffle操作的ResultTask.

ResulitTask.runTask

overridedef runTask(context:TaskContext):
U = {

metrics= Some(context.taskMetrics)

try{

此处通过生成task实例时也就是DAGScheduler的runJob时传入的function进行处理

比方在PairRDDFunction.saveAsHadoopDataset中定义的writeToFile函数

rdd.iterator中会依据不现的RDD的实现,运行其compute函数,

而compute函数详细运行通过业务代码中定义的如map函数传入的定义的function进行运行,

func(context,rdd.iterator(split,context))

}finally{

context.executeOnCompleteCallbacks()

}

}

ShuffleMapTask.runTask

overridedef runTask(context:TaskContext):
MapStatus = {

valnumOutputSplits=
dep.partitioner.numPartitions

metrics= Some(context.taskMetrics)

valblockManager=
SparkEnv.get.blockManager

valshuffleBlockManager=
blockManager.shuffleBlockManager

varshuffle:ShuffleWriterGroup
= null

varsuccess
=false

try{

通过shuffleId拿到一个shuffle的写入实例

//Obtain all the block writers for shuffle blocks.

valser
=SparkEnv.get.serializerManager.get(dep.serializerClass,SparkEnv.get.conf)

shuffle=
shuffleBlockManager.forMapTask(dep.shuffleId,partitionId,numOutputSplits,ser)

运行rdd.iterator操作,生成Pair,也就是Product2,依据key又一次shuffle到不同的文件里。

当全部的shuffle的task完毕后,会把此stage注冊到mapOutputTracker中,

等待下一个stage从中读取数据并运行其他操作,每个shuffle的task完毕后会生成一个MapStatus实例,

此实例主要包括有shuffle运行的executor与host等信息,每个task运行完毕的大小。

详细的shuffle数据读取可參见后面的shufle分析.

//Write the map output to its associated buckets.

for(elem
<-rdd.iterator(split,context)) {

valpair
=elem.asInstanceOf[Product2[Any,Any]]

valbucketId
=dep.partitioner.getPartition(pair._1)

shuffle.writers(bucketId).write(pair)

}

//Commit the writes. Get the size of each bucket block (total blocksize).

vartotalBytes=
0L

vartotalTime
=0L

valcompressedSizes:Array[Byte]
= shuffle.writers.map{ writer: BlockObjectWriter =>

writer.commit()

writer.close()

valsize
=writer.fileSegment().length

totalBytes+=
size

totalTime+= writer.timeWriting()

MapOutputTracker.compressSize(size)

}

//Update shuffle metrics.

valshuffleMetrics=
newShuffleWriteMetrics

shuffleMetrics.shuffleBytesWritten=
totalBytes

shuffleMetrics.shuffleWriteTime=
totalTime

metrics.get.shuffleWriteMetrics=
Some(shuffleMetrics)

success=
true

newMapStatus(blockManager.blockManagerId,compressedSizes)

}catch{
casee:Exception =>

//If there is an exception from running the task, revert the partialwrites

//and throw the exception upstream to Spark.

if(shuffle
!=null&&
shuffle.writers!=
null){

for(writer
<-shuffle.writers){

writer.revertPartialWrites()

writer.close()

}

}

throwe

}finally{

//Release the writers back to the shuffle block manager.

if(shuffle
!=null&&
shuffle.writers!=
null){

shuffle.releaseWriters(success)

}

//Execute the
callbackson task completion.

context.executeOnCompleteCallbacks()

}

}