Spark Shuffle（二）Executor、Driver之间Shuffle结果消息传递、追踪(转载)

1. 前言

在博客里介绍了 ShuffleWrite关于shuffleMapTask如何运行，输出Shuffle结果到Shuffle_shuffleId_mapId_0.data数据文件中，每个executor需要向Driver汇报当前节点的Shuffle结果状态，Driver保存结果信息进行下个Task的调度。

2. StatusUpdate消息

当Executor运行完Task的时候需要向Driver汇报StatusUpdate的消息

  override def statusUpdate(taskId: Long, state: TaskState, data: ByteBuffer) {
    val msg = StatusUpdate(executorId, taskId, state, data)
    driver match {
      case Some(driverRef) => driverRef.send(msg)
      case None => logWarning(s"Drop $msg because has not yet connected to driver")
    }
  }

整个结构体中包含了

• ExecutorId: Executor自己的ID
• TaskId: task分配的ID
• State: Task的运行状态

  LAUNCHING, RUNNING, FINISHED, FAILED, KILLED, LOST

• Data: 保存序列化的Result

2.1 Executor端发送

在Task运行后的结果，Executor会将结果首先序列化成ByteBuffer封装成DirectTaskResult，再次序列化DirectTaskResult成ByteBuffer，很显然序列化的结果的大小会决定不同的传递策略。在这里会有两个筏值来控制

• 最大的返回结果大小，如果超过设定的最大返回结果时，返回的结果内容会被丢弃，只是返回序列化的InDirectTaskResult，里面包含着BlockID和序列化后的结果大小

spark.driver.maxResultSize

• 最大的直接返回结果大小：如果返回的结果大于最大的直接返回结果大小，小于最大的返回结果大小，采用了保存的折中的策略，将序列化DirectTaskResult保存到BlockManager中，关于BlockManager可以参考前面写的BlockManager系列，返回InDirectTaskResult，里面包含着BlockID和序列化的结果大小

spark.task.maxDirectResultSize

• 直接返回：如果返回的结果小于等于最大的直接返回结果大小，将直接将序列化的DirectTaskResult返回给Driver端

val serializedResult: ByteBuffer = {
          if (maxResultSize > 0 && resultSize > maxResultSize) {
            logWarning(s"Finished $taskName (TID $taskId). Result is larger than maxResultSize " +
              s"(${Utils.bytesToString(resultSize)} > ${Utils.bytesToString(maxResultSize)}), " +
              s"dropping it.")
            ser.serialize(new IndirectTaskResult[Any](TaskResultBlockId(taskId), resultSize))
          } else if (resultSize > maxDirectResultSize) {
            val blockId = TaskResultBlockId(taskId)
            env.blockManager.putBytes(
              blockId,
              new ChunkedByteBuffer(serializedDirectResult.duplicate()),
              StorageLevel.MEMORY_AND_DISK_SER)
            logInfo(
              s"Finished $taskName (TID $taskId). $resultSize bytes result sent via BlockManager)")
            ser.serialize(new IndirectTaskResult[Any](blockId, resultSize))
          } else {
            logInfo(s"Finished $taskName (TID $taskId). $resultSize bytes result sent to driver")
            serializedDirectResult
          }
        }

2.2 Driver端接收

Driver端处理StatusUpdate的消息的代码如下：

case StatusUpdate(executorId, taskId, state, data) =>
        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 =>
              // Ignoring the update since we don't know about the executor.
              logWarning(s"Ignored task status update ($taskId state $state) " +
                s"from unknown executor with ID $executorId")
          }
        }

scheduler实例是TaskSchedulerImpl.scala

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)
              }
            }

statusUpdate函数调用了enqueueSuccessfulTask方法

def enqueueSuccessfulTask(
      taskSetManager: TaskSetManager,
      tid: Long,
      serializedData: ByteBuffer): Unit = {
    getTaskResultExecutor.execute(new Runnable {
      override def run(): Unit = Utils.logUncaughtExceptions {
        try {
          val (result, size) = serializer.get().deserialize[TaskResult[_]](serializedData) match {
            case directResult: DirectTaskResult[_] =>
              if (!taskSetManager.canFetchMoreResults(serializedData.limit())) {
                return
              }
              // deserialize "value" without holding any lock so that it won't block other threads.
              // We should call it here, so that when it's called again in
              // "TaskSetManager.handleSuccessfulTask", it does not need to deserialize the value.
              directResult.value(taskResultSerializer.get())
              (directResult, serializedData.limit())
            case IndirectTaskResult(blockId, size) =>
              if (!taskSetManager.canFetchMoreResults(size)) {
                // dropped by executor if size is larger than maxResultSize
                sparkEnv.blockManager.master.removeBlock(blockId)
                return
              }
              logDebug("Fetching indirect task result for TID %s".format(tid))
              scheduler.handleTaskGettingResult(taskSetManager, tid)
              val serializedTaskResult = sparkEnv.blockManager.getRemoteBytes(blockId)
              if (!serializedTaskResult.isDefined) {
                /* We won't be able to get the task result if the machine that ran the task failed
                 * between when the task ended and when we tried to fetch the result, or if the
                 * block manager had to flush the result. */
                scheduler.handleFailedTask(
                  taskSetManager, tid, TaskState.FINISHED, TaskResultLost)
                return
              }
              val deserializedResult = serializer.get().deserialize[DirectTaskResult[_]](
                serializedTaskResult.get.toByteBuffer)
              // force deserialization of referenced value
              deserializedResult.value(taskResultSerializer.get())
              sparkEnv.blockManager.master.removeBlock(blockId)
              (deserializedResult, size)
          }

          // Set the task result size in the accumulator updates received from the executors.
          // We need to do this here on the driver because if we did this on the executors then
          // we would have to serialize the result again after updating the size.
          result.accumUpdates = result.accumUpdates.map { a =>
            if (a.name == Some(InternalAccumulator.RESULT_SIZE)) {
              val acc = a.asInstanceOf[LongAccumulator]
              assert(acc.sum == 0L, "task result size should not have been set on the executors")
              acc.setValue(size.toLong)
              acc
            } else {
              a
            }
          }

          scheduler.handleSuccessfulTask(taskSetManager, tid, result)
        } catch {
          case cnf: ClassNotFoundException =>
            val loader = Thread.currentThread.getContextClassLoader
            taskSetManager.abort("ClassNotFound with classloader: " + loader)
          // Matching NonFatal so we don't catch the ControlThrowable from the "return" above.
          case NonFatal(ex) =>
            logError("Exception while getting task result", ex)
            taskSetManager.abort("Exception while getting task result: %s".format(ex))
        }
      }
    })
  }

在函数中，反序列化的过程是通过线程池里的线程来运行的，Netty的接收数据线程是不能被堵塞（同时还接受着别的消息），反序列化是耗时的任务，不能在Netty的消息处理线程中运行。

2.2.1 DirectTaskResult处理过程

• 直接反序列化成DirectTaskResult，反序列化后进行了整体返回内容的大小的判断，在前面的2.1中介绍参数：spark.driver.maxResultSize，这个参数是Driver端的参数控制的，在Spark中会启动多个Task，参数的控制是一个整体的控制所有的Tasks的返回结果的数量大小，当然单个task使用该筏值的控制也是没有问题，因为只要有一个任务返回的结果超过maxResultSize，整体返回的数据也会超过maxResultSize。
• 对DirectTaskResult里的result进行了反序列化。

2.2.2 InDirectTaskResult处理过程

• 通过size判断大小是否超过spark.driver.maxResultSize筏值控制
• 通过BlockManager来获取BlockID的内容反序列化成DirectTaskResult
• 对DirectTaskResult里的result进行了反序列化

最后调用handleSuccessfulTask方法

sched.dagScheduler.taskEnded(tasks(index), Success, result.value(), result.accumUpdates, info)

回到了Dag的调度，向eventProcessLoop的队列里提交了CompletionEvent的事件

def taskEnded(
      task: Task[_],
      reason: TaskEndReason,
      result: Any,
      accumUpdates: Seq[AccumulatorV2[_, _]],
      taskInfo: TaskInfo): Unit = {
    eventProcessLoop.post(
      CompletionEvent(task, reason, result, accumUpdates, taskInfo))
  }

处理eventProcessLoop队列的event是在DAG的线程处理的，在这里我们不讨论DAG的任务调度。

 

2.3 MapOutputTracker

MapOutputTracker是当运行完ShuffleMapTask的时候，ShuffleWrite会生成Shuffle_shuffleId_mapId_0.data、index文件，Executor需要将具体的信息返回给Driver，当Driver进行下一步的Task运算的时候，Executor也需要获取具体Shuffle数据文件的信息进行下一步的action算子的运算，结构的保存、管理就是通过MapOutputTracker跟踪器进行追踪的。

 

2.3.1 RegisterMapOutput

Execute端

在ShuffleMapTask中运行后会生成一个MapStatus，也就是上图的Map0结构，ComressedMapStatus、HighlyCompressedMapStatus这里的两个区别主要是增对Partition1...的size long的压缩，但这里的压缩算法并不准确比，如CompressedMapStatus的算法：

  def compressSize(size: Long): Byte = {
    if (size == 0) {
      0
    } else if (size <= 1L) {
      1
    } else {
      math.min(255, math.ceil(math.log(size) / math.log(LOG_BASE)).toInt).toByte
    }
  }

求Log1.1(size)的整数转为byte，也就是支持最大1.1^255=35G左右

为何不需要计算精准的尺寸？

还记得前面博客里提到的Shuffle_shuffleId_mapId_reduceId.index文件么，这里才是精准的位置，当读取本地文件的时候，并不使用MapStatus里的Size

Size有何用？

有存在别的Execute获取别的Execute的Shuffle结果文件，此时的size是获取文件的大概位置。

 

MapStatus是ShuffleMapTask运行的结果，被序列化成DirectTaskResult中的value，通过StatusUpdate消息传递

 

Driver端

DAG线程调度处理CompletionEvent的事件

private[scheduler] def handleTaskCompletion(event: CompletionEvent) {
............
case smt: ShuffleMapTask =>
            val shuffleStage = stage.asInstanceOf[ShuffleMapStage]
            updateAccumulators(event)
            val status = event.result.asInstanceOf[MapStatus]
            val execId = status.location.executorId
            logDebug("ShuffleMapTask finished on " + execId)
            if (failedEpoch.contains(execId) && smt.epoch <= failedEpoch(execId)) {
              logInfo(s"Ignoring possibly bogus $smt completion from executor $execId")
            } else {
              shuffleStage.addOutputLoc(smt.partitionId, status)
            }

            if (runningStages.contains(shuffleStage) && shuffleStage.pendingPartitions.isEmpty) {
              markStageAsFinished(shuffleStage)
              logInfo("looking for newly runnable stages")
              logInfo("running: " + runningStages)
              logInfo("waiting: " + waitingStages)
              logInfo("failed: " + failedStages)

              // We supply true to increment the epoch number here in case this is a
              // recomputation of the map outputs. In that case, some nodes may have cached
              // locations with holes (from when we detected the error) and will need the
              // epoch incremented to refetch them.
              // TODO: Only increment the epoch number if this is not the first time
              //       we registered these map outputs.
              mapOutputTracker.registerMapOutputs(
                shuffleStage.shuffleDep.shuffleId,
                shuffleStage.outputLocInMapOutputTrackerFormat(),
                changeEpoch = true)

              clearCacheLocs()

              if (!shuffleStage.isAvailable) {
                // Some tasks had failed; let's resubmit this shuffleStage
                // TODO: Lower-level scheduler should also deal with this
                logInfo("Resubmitting " + shuffleStage + " (" + shuffleStage.name +
                  ") because some of its tasks had failed: " +
                  shuffleStage.findMissingPartitions().mkString(", "))
                submitStage(shuffleStage)
              } else {
                // Mark any map-stage jobs waiting on this stage as finished
                if (shuffleStage.mapStageJobs.nonEmpty) {
                  val stats = mapOutputTracker.getStatistics(shuffleStage.shuffleDep)
                  for (job <- shuffleStage.mapStageJobs) {
                    markMapStageJobAsFinished(job, stats)
                  }
                }
                submitWaitingChildStages(shuffleStage)
              }
            }

当处理shuffleMapTask的结果的时候，mapOutputTracker.registerMapOutputs进行了MapOutputs的注册

  protected val mapStatuses = new ConcurrentHashMap[Int, Array[MapStatus]]().asScala
def registerMapOutputs(shuffleId: Int, statuses: Array[MapStatus], changeEpoch: Boolean = false) {
    mapStatuses.put(shuffleId, statuses.clone())
    if (changeEpoch) {
      incrementEpoch()
    }
  }

在Driver端保存了一个Map是以ShuffldId为Key的MapStatus的数组

 

2.3.2 获取MapStatus

在ResultTask中，通过获取反序列化的ShuffledRDD，在Fetch Shuffle数据文件的时候

val blockFetcherItr = new ShuffleBlockFetcherIterator(
      context,
      blockManager.shuffleClient,
      blockManager,
      mapOutputTracker.getMapSizesByExecutorId(handle.shuffleId, startPartition, endPartition),
      // Note: we use getSizeAsMb when no suffix is provided for backwards compatibility
      SparkEnv.get.conf.getSizeAsMb("spark.reducer.maxSizeInFlight", "48m") * 1024 * 1024,
      SparkEnv.get.conf.getInt("spark.reducer.maxReqsInFlight", Int.MaxValue))

通过getMapSizesByExecutorId获取ShuffledId所对应的MapStatus

def getMapSizesByExecutorId(shuffleId: Int, startPartition: Int, endPartition: Int)
      : Seq[(BlockManagerId, Seq[(BlockId, Long)])] = {
    logDebug(s"Fetching outputs for shuffle $shuffleId, partitions $startPartition-$endPartition")
    val statuses = getStatuses(shuffleId)
    // Synchronize on the returned array because, on the driver, it gets mutated in place
    statuses.synchronized {
      return MapOutputTracker.convertMapStatuses(shuffleId, startPartition, endPartition, statuses)
    }
  }

在getStatuses方法中

 private def getStatuses(shuffleId: Int): Array[MapStatus] = {
    val statuses = mapStatuses.get(shuffleId).orNull
    if (statuses == null) {
      logInfo("Don't have map outputs for shuffle " + shuffleId + ", fetching them")
      val startTime = System.currentTimeMillis
      var fetchedStatuses: Array[MapStatus] = null
      fetching.synchronized {
        // Someone else is fetching it; wait for them to be done
        while (fetching.contains(shuffleId)) {
          try {
            fetching.wait()
          } catch {
            case e: InterruptedException =>
          }
        }

        // Either while we waited the fetch happened successfully, or
        // someone fetched it in between the get and the fetching.synchronized.
        fetchedStatuses = mapStatuses.get(shuffleId).orNull
        if (fetchedStatuses == null) {
          // We have to do the fetch, get others to wait for us.
          fetching += shuffleId
        }
      }

      if (fetchedStatuses == null) {
        // We won the race to fetch the statuses; do so
        logInfo("Doing the fetch; tracker endpoint = " + trackerEndpoint)
        // This try-finally prevents hangs due to timeouts:
        try {
          val fetchedBytes = askTracker[Array[Byte]](GetMapOutputStatuses(shuffleId))
          fetchedStatuses = MapOutputTracker.deserializeMapStatuses(fetchedBytes)
          logInfo("Got the output locations")
          mapStatuses.put(shuffleId, fetchedStatuses)
        } finally {
          fetching.synchronized {
            fetching -= shuffleId
            fetching.notifyAll()
          }
        }
      }
      logDebug(s"Fetching map output statuses for shuffle $shuffleId took " +
        s"${System.currentTimeMillis - startTime} ms")

      if (fetchedStatuses != null) {
        return fetchedStatuses
      } else {
        logError("Missing all output locations for shuffle " + shuffleId)
        throw new MetadataFetchFailedException(
          shuffleId, -1, "Missing all output locations for shuffle " + shuffleId)
      }
    } else {
      return statuses
    }
  }

• 封装了一层缓存mapStatus，对同一个Executor来说，里面的线程都是运行同一个Driver的提交的任务，对相同的shuffeID，MapStatus是一样的
• 对同一个Executor、ShuffeID来说，通过Driver获取信息只需要一次，Driver里保存的Shuffle的结果是单点的，对同一个Executor来说获取同一个ShuffleID只需要请求一次，在Traker里面保存了一个队列fetching，里面保存的ShuffeID代表的是有线程正在从Driver端获取ShuffleID的MapStatus，如果发现有值，当前线程会等待，直到其他的线程获取ShuffleID状态并保存到缓存结束，当前线程直接从缓存中获取当前状态
• Executor 向Driver发送GetMapOutputStatuses(shuffleId)消息
• Driver收到GetMapOutputStatuses消息后保存到消息队列mapOutputRequests，Map-Output-Dispatcher-x多线程处理消息队列，返回序列化的MapStatus
• Executor反序列化成MapStatus

2.2.3 以BlockManagerId为key的Shuffle的序列

在前面的博客里提到过 Driver分配Task的数量的策略是依赖于Partition，在单个任务ShuffledMapTask对Data进行分片也是依赖于Partition

前面一个的Partition 是MapId，后面一个Partition 指的是ReduceId

在ResultTask里所取的Shuffle数据文件中的Partition是ReduceId，而不是MapId

 

 

 

也就是每个ResultTask会去获取所有不同的MapId中相同的PartitionID部分Shuffle文件，而不是继续按前面的Map进行分配，那意味着ResultTask将会去获取所有Shuffle文件

Shuffle_shuffleId_mapId_0.data中的Partition那部分进行Action操作，这样可以适当避免在ShuffledMapTask中分配的数据不均衡，导致单个Shuffle_shuffleId_mapId_0.data文件数据过大的问题。

具体的代码实现如下：

private def convertMapStatuses(
      shuffleId: Int,
      startPartition: Int,
      endPartition: Int,
      statuses: Array[MapStatus]): Seq[(BlockManagerId, Seq[(BlockId, Long)])] = {
    assert (statuses != null)
    val splitsByAddress = new HashMap[BlockManagerId, ArrayBuffer[(BlockId, Long)]]
    for ((status, mapId) <- statuses.zipWithIndex) {
      if (status == null) {
        val errorMessage = s"Missing an output location for shuffle $shuffleId"
        logError(errorMessage)
        throw new MetadataFetchFailedException(shuffleId, startPartition, errorMessage)
      } else {
        for (part <- startPartition until endPartition) {
          splitsByAddress.getOrElseUpdate(status.location, ArrayBuffer()) +=
            ((ShuffleBlockId(shuffleId, mapId, part), status.getSizeForBlock(part)))
        }
      }
    }

    splitsByAddress.toSeq
  }