spark join shuffle 数据文件的读取

我们看下在shuffle过程中数据文件的读取过程中调用的类对象

// 下面就是对这个shuffler中的分片数据进行读取并进行相关的aggregate操作了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)

上面类就是负责调用远程和本地的shuffle分片的数据来的
在类ShuffleBlockFetcherIterator.initialize 方法中

private[this] def initialize(): Unit = {// Add a task completion callback (called in both success case and failure case) to cleanup.context.addTaskCompletionListener(_ => cleanup())// Split local and remote blocks.// 拿到要远程拉取的数据信息了val remoteRequests = splitLocalRemoteBlocks()// Add the remote requests into our queue in a random orderfetchRequests ++= Utils.randomize(remoteRequests)// Send out initial requests for blocks, up to our maxBytesInFlight// 从队列里面拿出请求任务进行请求了fetchUpToMaxBytes()val numFetches = remoteRequests.size - fetchRequests.sizelogInfo("Started " + numFetches + " remote fetches in" + Utils.getUsedTimeMs(startTime))// Get Local Blocks// 拉取本地数据了fetchLocalBlocks()logDebug("Got local blocks in " + Utils.getUsedTimeMs(startTime))}

上面可以看到数据的拉取分成远程和本地两种类型的数据。下面先对要拉取的数据进行分类，分成本地和远程的

private[this] def splitLocalRemoteBlocks(): ArrayBuffer[FetchRequest] = {// Make remote requests at most maxBytesInFlight / 5 in length; the reason to keep them// smaller than maxBytesInFlight is to allow multiple, parallel fetches from up to 5// nodes, rather than blocking on reading output from one node.val targetRequestSize = math.max(maxBytesInFlight / 5, 1L)logDebug("maxBytesInFlight: " + maxBytesInFlight + ", targetRequestSize: " + targetRequestSize)// Split local and remote blocks. Remote blocks are further split into FetchRequests of size// at most maxBytesInFlight in order to limit the amount of data in flight.val remoteRequests = new ArrayBuffer[FetchRequest]// Tracks total number of blocks (including zero sized blocks)var totalBlocks = 0for ((address, blockInfos) <- blocksByAddress) {  // 一个一个地址去拉取数据了  totalBlocks += blockInfos.size  if (address.executorId == blockManager.blockManagerId.executorId) {    // Filter out zero-sized blocks    // 如果这个数据是在本地的    localBlocks ++= blockInfos.filter(_._2 != 0).map(_._1)    numBlocksToFetch += localBlocks.size  } else {    // 这里就是一些非本地的数据了    val iterator = blockInfos.iterator    var curRequestSize = 0L    var curBlocks = new ArrayBuffer[(BlockId, Long)]    while (iterator.hasNext) {      val (blockId, size) = iterator.next()      // Skip empty blocks      if (size > 0) {        curBlocks += ((blockId, size))        remoteBlocks += blockId        numBlocksToFetch += 1        curRequestSize += size      } else if (size < 0) {        throw new BlockException(blockId, "Negative block size " + size)      }      if (curRequestSize >= targetRequestSize) {        // Add this FetchRequest        // block 是 ShuffleBlockId        remoteRequests += new FetchRequest(address, curBlocks)        curBlocks = new ArrayBuffer[(BlockId, Long)]        logDebug(s"Creating fetch request of $curRequestSize at $address")        curRequestSize = 0      }    }    // Add in the final request    if (curBlocks.nonEmpty) {      remoteRequests += new FetchRequest(address, curBlocks)    }  }}logInfo(s"Getting $numBlocksToFetch non-empty blocks out of $totalBlocks blocks")// 返回要远程拉取的数据了remoteRequests}

可以看到上面根据当前的 shuffleBlockId 的执行进程executorId是否和当前的一致，如果是一致，说明是本地数据，否则是远程数据。
当分片完后，远程数据就从队列中拿出任务进行执行拉取了

 private def fetchUpToMaxBytes(): Unit = {// Send fetch requests up to maxBytesInFlightwhile (fetchRequests.nonEmpty &&  (bytesInFlight == 0 || bytesInFlight + fetchRequests.front.size <= maxBytesInFlight)) {  // 通过这里进行限流请求了  sendRequest(fetchRequests.dequeue())}}

下面就是具体的远程拉取方法

private[this] def sendRequest(req: FetchRequest) {logDebug("Sending request for %d blocks (%s) from %s".format(  req.blocks.size, Utils.bytesToString(req.size), req.address.hostPort))bytesInFlight += req.size// so we can look up the size of each blockIDval sizeMap = req.blocks.map { case (blockId, size) => (blockId.toString, size) }.toMapval blockIds = req.blocks.map(_._1.toString)// 去这些地址拉取数据了,同时注意block对象是 ShuffleBlockId 里面包含着当前请求的是那个分片数据//  在拉取的时候，还要对block块数据进行分片val address = req.address// NettyBlockTransferService 和 ExternalShuffleClientshuffleClient.fetchBlocks(address.host, address.port, address.executorId, blockIds.toArray,  new BlockFetchingListener {    override def onBlockFetchSuccess(blockId: String, buf: ManagedBuffer): Unit = {      // Only add the buffer to results queue if the iterator is not zombie,      // i.e. cleanup() has not been called yet.      if (!isZombie) {        // Increment the ref count because we need to pass this to a different thread.        // This needs to be released after use.        buf.retain()        results.put(new SuccessFetchResult(BlockId(blockId), address, sizeMap(blockId), buf))        shuffleMetrics.incRemoteBytesRead(buf.size)        shuffleMetrics.incRemoteBlocksFetched(1)      }      logTrace("Got remote block " + blockId + " after " + Utils.getUsedTimeMs(startTime))    }    override def onBlockFetchFailure(blockId: String, e: Throwable): Unit = {      logError(s"Failed to get block(s) from ${req.address.host}:${req.address.port}", e)      results.put(new FailureFetchResult(BlockId(blockId), address, e))    }  })} 

可以看到主要使用NettyBlockTransferService 和 ExternalShuffleClient 两种客户端进行数据的拉取。
本地数据则如下

private[this] def fetchLocalBlocks() {val iter = localBlocks.iteratorwhile (iter.hasNext) {  val blockId = iter.next()  try {    // 拉取数据了    val buf = blockManager.getBlockData(blockId)    shuffleMetrics.incLocalBlocksFetched(1)    shuffleMetrics.incLocalBytesRead(buf.size)    buf.retain()    results.put(new SuccessFetchResult(blockId, blockManager.blockManagerId, 0, buf))  } catch {    case e: Exception =>      // If we see an exception, stop immediately.      logError(s"Error occurred while fetching local blocks", e)      results.put(new FailureFetchResult(blockId, blockManager.blockManagerId, e))      return  }}}

使用blockManager对象进行拉取数据

override def getBlockData(blockId: BlockId): ManagedBuffer = {if (blockId.isShuffle) {  // 当是shuffle数据时  shuffleManager.shuffleBlockResolver.getBlockData(blockId.asInstanceOf[ShuffleBlockId])} else {  val blockBytesOpt = doGetLocal(blockId, asBlockResult = false)    .asInstanceOf[Option[ByteBuffer]]  if (blockBytesOpt.isDefined) {    val buffer = blockBytesOpt.get    new NioManagedBuffer(buffer)  } else {    throw new BlockNotFoundException(blockId.toString)  }}}

然后当前数据是 shuffle数据，所以调用了shuffleBlockResolver对象，然后实现类为FileShuffleBlockResolver和IndexShuffleBlockResolver 两个，现在查看FileShuffleBlockResolver 对象。

 override def getBlockData(blockId: ShuffleBlockId): ManagedBuffer = {// 拿到数据文件回来了val file = blockManager.diskBlockManager.getFile(blockId)// 返回这个文件读取对象了  new FileSegmentManagedBuffer(transportConf, file, 0, file.length)}

可以看到，该方法先去拿到该分片的文件，然后创建FileSegmentManagedBuffer 对象。
具体的 DiskBlockManager实现搜索文件为

def getFile(filename: String): File = {// filename 包含着 shuffleid_index_partvar attempts = 0val numLocalDirs = localDirs.lengthval maxAttempts = numLocalDirsval subDirsMap = new mutable.HashMap[Int, Array[File]]()subDirs.zipWithIndex.foreach(s => subDirsMap.put(s._2, s._1))// Figure out which local directory it hashes to, and which subdirectory in that// 文件名称的hashval hash = Utils.nonNegativeHash(filename)// hash 到这些目录下面var dirId = hash % subDirsMap.size// 子目录下面var subDirId = (hash / subDirsMap.size) % subDirsPerLocalDirvar dir: File = nullwhile (dir == null) {  attempts += 1  if (attempts > maxAttempts) {    /*    throw new IOException("Failed to create local dir after " +      maxAttempts + " attempts!")    */    logError("Failed to create local dir in root " + dirId + " after " +      maxAttempts + " attempts!")    subDirsMap.remove(dirId)    if (subDirsMap.size == 0) {      throw new IOException("Failed to create local dir after try in all root dir")    }    // 有多个盘下面可以存储数据，所以这里这样操作了    dirId = hash % subDirsMap.size    subDirId = (hash / subDirsMap.size) % subDirsPerLocalDir  }  try {    // Create the subdirectory if it doesn't already exist    dir = subDirs(dirId).synchronized {      val old = subDirs(dirId)(subDirId)      if (old != null) {        new File(old, filename)      } else {        val newDir = new File(localDirs(dirId), "%02x".format(subDirId))        if (newDir.exists() || newDir.mkdir()) {          subDirs(dirId)(subDirId) = newDir          new File(newDir, filename)        }        else {          logWarning(s"Failed to create local dir in $newDir.")          null        }      }    }  } catch { case e: SecurityException => dir = null; }}dir}

从这里可以看到，当是spark的数据文件存放在多硬盘时，是通过hash到多个目录下面进行数据文件的存放的。
然后FileSegmentManagedBuffer 文件就就是纯粹的nio文件的读取

 @Overridepublic ByteBuffer nioByteBuffer() throws IOException {FileChannel channel = null;try {  channel = new RandomAccessFile(file, "r").getChannel();  // Just copy the buffer if it's sufficiently small, as memory mapping has a high overhead.  if (length < conf.memoryMapBytes()) {    ByteBuffer buf = ByteBuffer.allocate((int) length);    channel.position(offset);    while (buf.remaining() != 0) {      // 不断读取到直接内存中去      if (channel.read(buf) == -1) {        throw new IOException(String.format("Reached EOF before filling buffer\n" +          "offset=%s\nfile=%s\nbuf.remaining=%s",          offset, file.getAbsoluteFile(), buf.remaining()));      }    }    buf.flip();    return buf;  } else {    // 直接文件的读取了    return channel.map(FileChannel.MapMode.READ_ONLY, offset, length);  }

从上面可以看到，spark的数据文件存放在多个硬盘的原理。思想是通用的，存放其它的临时文件也一样。

总结

1. 读取指定shuffleid的part分片的数据
2. 通过拉取该分片数据的mapstatus信息，存放在那些节点当中
3. 然后对这数据进行分类是远程还是本地数据
4. 如果是远程数据就通用netty进行拉取，本地数据就读取数据文件
5. 当是本地数据时，就通过 shuffleid + mapid + part 定位到具体存放的数据文件然后通过nio方式读取