标签:style color 使用 os strong 文件 数据 io
欢迎转载,转载请注明出处,徽沪一郎。
ShuffleMapTask的计算结果保存在哪,随后Stage中的task又是如何知道从哪里去读取的呢,这个过程一直让我困惑不已。
用比较通俗一点的说法来解释一下Shuffle数据的写入和读取过程
如果能够明白上述的过程,并对应到相应的代码,那就无须看下述的详细解释了。
好了,让我们开始代码跟踪吧。
数据写入动作最原始的触发点是ShuffleMapTask.runTask函数,看一看源码先。
override def runTask(context: TaskContext): MapStatus = {
metrics = Some(context.taskMetrics)
var writer: ShuffleWriter[Any, Any] = null
try {
val manager = SparkEnv.get.shuffleManager
writer = manager.getWriter[Any, Any](dep.shuffleHandle, partitionId, context)
writer.write(rdd.iterator(split, context).asInstanceOf[Iterator[_
if (writer != null) {
writer.stop(success = false)
}
throw e
} finally {
context.executeOnCompleteCallbacks()
}
}
managerGetWriter返回的是HashShuffleWriter,所以调用过程是ShuffleMapTask.runTask->HashShuffleWriter.write->BlockObjectWriter.write. 注意dep.mapSideCombine这一分支判断。ReduceByKey(_ + _)中的(_ + _)在此处被执行一次,另一次执行是在read过程。
override def write(records: Iterator[_ <: Product2[K, V]]): Unit = {
val iter = if (dep.aggregator.isDefined) {
if (dep.mapSideCombine) {
dep.aggregator.get.combineValuesByKey(records, context)
} else {
records
}
} else if (dep.aggregator.isEmpty && dep.mapSideCombine) {
throw new IllegalStateException("Aggregator is empty for map-side combine")
} else {
records
}
for (elem <- iter) {
val bucketId = dep.partitioner.getPartition(elem._1)
shuffle.writers(bucketId).write(elem)
}
HashShuffleWriter.write中主要处理两件事
Partitioner是在什么时候注入的,RDD抽象类中,Partitioner为空?以reduceByKey为例,HashPartitioner会在后面combineByKey的代码创建ShuffledRDD的时候作为ShuffledRDD的构造函数传入。
def reduceByKey(func: (V, V) => V, numPartitions: Int): RDD[(K, V)] = {
reduceByKey(new HashPartitioner(numPartitions), func)
}
Stage在创建的时候通过构造函数入参明确需要从多少Partition读取数据,生成的Partition会有多少。看一看Stage的构造函数,读取的分区数目由RDD.partitions.size决定,输出的partitions由shuffleDep决定。
private[spark] class Stage(
val id: Int,
val rdd: RDD[_],
val numTasks: Int,
val shuffleDep: Option[ShuffleDependency[_, _, _]], // Output shuffle if stage is a map stage
val parents: List[Stage],
val jobId: Int,
val callSite: CallSite)
extends Logging {
val isShuffleMap = shuffleDep.isDefined
val numPartitions = rdd.partitions.size
val outputLocs = Array.fill[List[MapStatus]](numPartitions)(Nil)
var numAvailableOutputs = 0
private var nextAttemptId = 0
回到数据写入的问题上来,结果写入时的一个主要问题就是已经知道shuffle_id, map_id和要写入的elem,如何找到对应的写入文件。每一个临时文件由三元组(shuffle_id,map_id,reduce_id)来决定,当前已经知道了两个,还剩下一下reduce_id待确定。
reduce_id是使用partitioner计算出来的结果,输入的是elem的键值。也就是dep.partitioner.getPartition(elem._1)。 根据计算出来的bucketid找到对应的writer,然后真正写入。
在HashShuffleWriter.write中使用到的shuffle由ShuffleBlockManager的forMapTask函数生成,注意forMapTask中产生writers的代码逻辑。
每个writer分配一下文件, 文件名由三元组(shuffle_id,map_id,reduce_id)组成,如果知道了这个三元组就可以找到对应的文件。
如果consolidation没有打开,那么在一个task中,有多少个输出的partition就会有多少个中间文件。
val writers: Array[BlockObjectWriter] = if (consolidateShuffleFiles) {
fileGroup = getUnusedFileGroup()
Array.tabulate[BlockObjectWriter](numBuckets) { bucketId =>
val blockId = ShuffleBlockId(shuffleId, mapId, bucketId)
blockManager.getDiskWriter(blockId, fileGroup(bucketId), serializer, bufferSize)
}
} else {
Array.tabulate[BlockObjectWriter](numBuckets) { bucketId =>
val blockId = ShuffleBlockId(shuffleId, mapId, bucketId)
val blockFile = blockManager.diskBlockManager.getFile(blockId)
// Because of previous failures, the shuffle file may already exist on this machine.
// If so, remove it.
if (blockFile.exists) {
if (blockFile.delete()) {
logInfo(s"Removed existing shuffle file $blockFile")
} else {
logWarning(s"Failed to remove existing shuffle file $blockFile")
}
}
blockManager.getDiskWriter(blockId, blockFile, serializer, bufferSize)
}
}
getFile负责将三元组(shuffle_id,map_id,reduce_id)映射到文件名
def getFile(filename: String): File = {
// Figure out which local directory it hashes to, and which subdirectory in that
val hash = Utils.nonNegativeHash(filename)
val dirId = hash % localDirs.length
val subDirId = (hash / localDirs.length) % subDirsPerLocalDir
// Create the subdirectory if it doesn‘t already exist
var subDir = subDirs(dirId)(subDirId)
if (subDir == null) {
subDir = subDirs(dirId).synchronized {
val old = subDirs(dirId)(subDirId)
if (old != null) {
old
} else {
val newDir = new File(localDirs(dirId), "%02x".format(subDirId))
newDir.mkdir()
subDirs(dirId)(subDirId) = newDir
newDir
}
}
}
new File(subDir, filename)
}
def getFile(blockId: BlockId): File = getFile(blockId.name)
产生的文件在哪呢,如果没有更改默认的配置,生成的目录结构类似于下
/tmp/spark-local-20140723092540-7f24
/tmp/spark-local-20140723092540-7f24/0d
/tmp/spark-local-20140723092540-7f24/0d/shuffle_0_0_1
/tmp/spark-local-20140723092540-7f24/0d/shuffle_0_1_0
/tmp/spark-local-20140723092540-7f24/0c
/tmp/spark-local-20140723092540-7f24/0c/shuffle_0_0_0
/tmp/spark-local-20140723092540-7f24/0e
/tmp/spark-local-20140723092540-7f24/0e/shuffle_0_1_1
当所有的数据写入文件并提交以后,还需要生成MapStatus汇报给driver application. MapStatus在哪生成的呢?commitWritesAndBuildStatus就干这活。
调用关系HashShuffleWriter.stop->commitWritesAndBuildStatus
private def commitWritesAndBuildStatus(): MapStatus = {
// Commit the writes. Get the size of each bucket block (total block size).
var totalBytes = 0L
var totalTime = 0L
val compressedSizes = shuffle.writers.map { writer: BlockObjectWriter =>
writer.commit()
writer.close()
val size = writer.fileSegment().length
totalBytes += size
totalTime += writer.timeWriting()
MapOutputTracker.compressSize(size)
}
// Update shuffle metrics.
val shuffleMetrics = new ShuffleWriteMetrics
shuffleMetrics.shuffleBytesWritten = totalBytes
shuffleMetrics.shuffleWriteTime = totalTime
metrics.shuffleWriteMetrics = Some(shuffleMetrics)
new MapStatus(blockManager.blockManagerId, compressedSizes)
}
compressedSize是一个非常让人疑惑的地方,原因慢慢道来,先看一下MapStatus的构造函数
class MapStatus(var location: BlockManagerId, var compressedSizes: Array[Byte])
compressedSize是一个byte数组,每一个byte反应了该partiton中的数据大小。如Array(0)=128就表示在data partition 0中有128byte数据。
问题的问题是一个byte只能表示255,如果超过255怎么办呢?
当当当,数学闪亮登场了,注意到compressSize没,通过转换将2^8变换为1.1^256。一下子由255byte延伸到近35G.
看一看这神奇的compressSize函数吧,只是聊聊几行代码而已。
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
}
}
ShuffleMapTask运行结束时,会将MapStatus结果封装在StatusUpdate消息中汇报给SchedulerBackend, 由DAGScheduler在handleTaskCompletion函数中将MapStatus加入到相应的Stage。这一过程略过,不再详述。
MapOutputTrackerMaster会保存所有最新的MapStatus.
只画张图来表示存储之后的示意。
ShuffledRDD.compute函数是读取过程的触发点。
override def compute(split: Partition, context: TaskContext): Iterator[P] = {
val dep = dependencies.head.asInstanceOf[ShuffleDependency[K, V, C]]
SparkEnv.get.shuffleManager.getReader(dep.shuffleHandle, split.index, split.index + 1, context)
.read()
.asInstanceOf[Iterator[P]]
}
shuffleManager.getReader返回的是HashShuffleReader,所以看一看HashShuffleReader中的read函数的具体实现。
read函数处理逻辑中需要注意到一点即combine过程有可能会被再次执行。注意dep.aggregator.isDefined这一分支判断。ReduceByKey(_ + _)中的(_ + _)在此处被执行。
override def read(): Iterator[Product2[K, C]] = {
val iter = BlockStoreShuffleFetcher.fetch(handle.shuffleId, startPartition, context,
Serializer.getSerializer(dep.serializer))
if (dep.aggregator.isDefined) {
if (dep.mapSideCombine) {
new InterruptibleIterator(context, dep.aggregator.get.combineCombinersByKey(iter, context))
} else {
new InterruptibleIterator(context, dep.aggregator.get.combineValuesByKey(iter, context))
}
} else if (dep.aggregator.isEmpty && dep.mapSideCombine) {
throw new IllegalStateException("Aggregator is empty for map-side combine")
} else {
iter
}
}
一路辗转,终于来到了读取过程中非常关键的所在BlockStoreShuffleFetcher。
BlockStoreShuffleFetcher需要回答如下问题
val blockManager = SparkEnv.get.blockManager
val startTime = System.currentTimeMillis
val statuses = SparkEnv.get.mapOutputTracker.getServerStatuses(shuffleId, reduceId)
logDebug("Fetching map output location for shuffle %d, reduce %d took %d ms".format(
shuffleId, reduceId, System.currentTimeMillis - startTime))
val splitsByAddress = new HashMap[BlockManagerId, ArrayBuffer[(Int, Long)]]
for (((address, size), index)
(address, splits.map(s => (ShuffleBlockId(shuffleId, s._1, reduceId), s._2)))
}
val blockFetcherItr = blockManager.getMultiple(blocksByAddress, serializer)
val itr = blockFetcherItr.flatMap(unpackBlock)
一个ShuffleMapTask会生成一个MapStatus,MapStatus中含有当前ShuffleMapTask产生的数据落到各个Partition中的大小。如果大小为0,则表示该分区没有数据产生。MapStatus中另一个重要的成员变量就是BlockManagerId,该变量表示目标数据在哪个BlockManager当中。
MapoutputTrackerMaster拥有最新的MapStatus信息,为了执行效率,MapoutputTrackerWorker会定期更新数据到本地,所以MapoutputTracker先从本地查找,如果找不到再从MapoutputTrackerMaster上同步最新数据。
索引即是reduceId,如果array(0) == 0,就表示上一个ShuffleMapTask中生成的数据中没有任意的内容可以作为reduceId为0的ResultTask的输入。如果不能理解,返回仔细看一下MapStatus的结构图。
BlockManager.getMultiple用于读取BlockManager中的数据,根据配置确定生成tNettyBlockFetcherIterator还是BasicBlockFetcherIterator。
如果所要获取的文件落在本地,则调用getLocal读取,否则发送请求到远端blockmanager。看一下BlockFetcherIterator的initialize函数
override def initialize() {
// Split local and remote blocks.
val remoteRequests = splitLocalRemoteBlocks()
// Add the remote requests into our queue in a random order
fetchRequests ++= Utils.randomize(remoteRequests)
// Send out initial requests for blocks, up to our maxBytesInFlight
while (!fetchRequests.isEmpty &&
(bytesInFlight == 0 || bytesInFlight + fetchRequests.front.size <= maxBytesInFlight)) {
sendRequest(fetchRequests.dequeue())
}
val numFetches = remoteRequests.size - fetchRequests.size
logInfo("Started " + numFetches + " remote fetches in" + Utils.getUsedTimeMs(startTime))
// Get Local Blocks
startTime = System.currentTimeMillis
getLocalBlocks()
logDebug("Got local blocks in " + Utils.getUsedTimeMs(startTime) + " ms")
}
至此,数据读取的正常流程讲述完毕。
如果数据读取中碰到异常怎么办?比如,
如果无法获取目标数据,就会上报FetchFailedException.
def unpackBlock(blockPair: (BlockId, Option[Iterator[Any]])) : Iterator[T] = {
val blockId = blockPair._1
val blockOption = blockPair._2
blockOption match {
case Some(block) => {
block.asInstanceOf[Iterator[T]]
}
case None => {
blockId match {
case ShuffleBlockId(shufId, mapId, _) =>
val address = statuses(mapId.toInt)._1
throw new FetchFailedException(address, shufId.toInt, mapId.toInt, reduceId)
case _ =>
throw new SparkException(
"Failed to get block " + blockId + ", which is not a shuffle block")
}
}
}
}
FetchFailedExecption会被包装在StatutsUpdate上报给SchedulerBackend,然后一路处理下去,最终将丢失目标数据的归属Task重新提交。比如当前是(stage_1, task_0),需要读取(stage_2, task_1)产生的目标数据,但是对应的目标数据丢失,这个时候就需要将(stage_2, task_1)重新提交运行。
注意DAGScheduler中的FetchFailed处理分支,一路跟踪下去就会看到任务被重新提交了
case FetchFailed(bmAddress, shuffleId, mapId, reduceId) =>
// Mark the stage that the reducer was in as unrunnable
val failedStage = stageIdToStage(task.stageId)
runningStages -= failedStage
// TODO: Cancel running tasks in the stage
logInfo("Marking " + failedStage + " (" + failedStage.name +
") for resubmision due to a fetch failure")
// Mark the map whose fetch failed as broken in the map stage
val mapStage = shuffleToMapStage(shuffleId)
if (mapId != -1) {
mapStage.removeOutputLoc(mapId, bmAddress)
mapOutputTracker.unregisterMapOutput(shuffleId, mapId, bmAddress)
}
logInfo("The failed fetch was from " + mapStage + " (" + mapStage.name +
"); marking it for resubmission")
if (failedStages.isEmpty && eventProcessActor != null) {
// Don‘t schedule an event to resubmit failed stages if failed isn‘t empty, because
// in that case the event will already have been scheduled. eventProcessActor may be
// null during unit tests.
import env.actorSystem.dispatcher
env.actorSystem.scheduler.scheduleOnce(
RESUBMIT_TIMEOUT, eventProcessActor, ResubmitFailedStages)
}
failedStages += failedStage
failedStages += mapStage
// TODO: mark the executor as failed only if there were lots of fetch failures on it
if (bmAddress != null) {
handleExecutorLost(bmAddress.executorId, Some(task.epoch))
}
生成的中间数据是在什么时候被清除的呢?
当Driver Application退出的时候,该Application生成的临时文件将会被一一清除,注意是application结束生命,不是job。一个application可以包含一至多个job。
以local-cluster方式运行spark-shell,观察/tmp/spark-local*目录下的文件变化,具体指令如下
MASTER=local-cluster[2,2,512] bin/spark-shell
#进入spark-shell之后,输入
sc.textFile("README.md").flatMap(_.split(" ")).map(w=>(w,1)).reduceByKey(_ + _)
Shuffle数据的写入和读取是Spark Core这一部分最为复杂的内容,彻底了解该部分内容才能深刻意识到Spark实现的精髓所在。
Apache Spark源码走读之20 -- ShuffleMapTask计算结果的保存与读取,布布扣,bubuko.com
Apache Spark源码走读之20 -- ShuffleMapTask计算结果的保存与读取
标签:style color 使用 os strong 文件 数据 io
原文地址:http://www.cnblogs.com/hseagle/p/3863966.html