Spark Shuffle模块——Suffle Read过程分析

在阅读本文之前。请先阅读Spark Sort Based Shuffle内存分析

Spark Shuffle Read调用栈例如以下：
1. org.apache.spark.rdd.ShuffledRDD#compute()
2. org.apache.spark.shuffle.ShuffleManager#getReader()
3. org.apache.spark.shuffle.hash.HashShuffleReader#read()
4. org.apache.spark.storage.ShuffleBlockFetcherIterator#initialize()
5. org.apache.spark.storage.ShuffleBlockFetcherIterator#splitLocalRemoteBlocks()
org.apache.spark.storage.ShuffleBlockFetcherIterator#sendRequest()
org.apache.spark.storage.ShuffleBlockFetcherIterator#fetchLocalBlocks()

以下是fetchLocalBlocks()方法运行时涉及到的类和相应方法：
6. org.apache.spark.storage.BlockManager#getBlockData()
org.apache.spark.shuffle.hash.ShuffleManager#shuffleBlockResolver()
ShuffleManager有两个子类。假设是HashShuffle 则相应的是org.apache.spark.shuffle.hash.HashShuffleManager#shuffleBlockResolver()方法，该方法返回的是org.apache.spark.shuffle.FileShuffleBlockResolver。再调用FileShuffleBlockResolver#getBlockData()方法返回Block数据
；假设是Sort Shuffle，则相应的是
org.apache.spark.shuffle.hash.SortShuffleManager#shuffleBlockResolver()，该方法返回的是org.apache.spark.shuffle.IndexShuffleBlockResolver。然后再调用IndexShuffleBlockResolver#getBlockData()返回Block数据。

以下是org.apache.spark.storage.ShuffleBlockFetcherIterator#sendRequest()方法运行时涉及到的类和相应方法
7.

org.apache.spark.network.shuffle.ShuffleClient#fetchBlocks
org.apache.spark.network.shuffle.ShuffleClient有两个子类，各自是ExternalShuffleClient及BlockTransferService
。其中org.apache.spark.network.shuffle.BlockTransferService又有两个子类，各自是NettyBlockTransferService和NioBlockTransferService。相应两种不同远程获取Block数据方式。Spark 1.5.2中已经将NioBlockTransferService方式设置为deprecated。在兴许版本号中将被移除

以下按上述调用栈对各方法进行说明，这里仅仅讲脉络，细节后面再讨论

ShuffledRDD#compute()代码

Task运行时。调用ShuffledRDD的compute方法，其代码例如以下：

//org.apache.spark.rdd.ShuffledRDD#compute()
override def compute(split: Partition, context: TaskContext): Iterator[(K, C)] = {
    val dep = dependencies.head.asInstanceOf[ShuffleDependency[K, V, C]]
    //通过org.apache.spark.shuffle.ShuffleManager#getReader()方法
    //不管是Sort Shuffle 还是 Hash Shuffle。使用的都是
    //org.apache.spark.shuffle.hash.HashShuffleReader
    SparkEnv.get.shuffleManager.getReader(dep.shuffleHandle, split.index, split.index + 1, context)
      .read()
      .asInstanceOf[Iterator[(K, C)]]
  }

能够看到，其核心逻辑是通过调用ShuffleManager#getReader()方法得到HashShuffleReader对象。然后调用HashShuffleReader#read()方法完毕前一Stage中ShuffleMapTask生成的Shuffle 数据的读取。须要说明的是，不管是Hash Shuffle还是Sort Shuffle。使用的都是HashShuffleReader。

HashShuffleReader#read()

跳到HashShuffleReader#read()方法其中。其源代码例如以下：

/** Read the combined key-values for this reduce task */
  override def read(): Iterator[Product2[K, C]] = {
    //创建ShuffleBlockFetcherIterator对象，在其构造函数中会调用initialize()方法
    //该方法中会运行splitLocalRemoteBlocks()，确定数据的读取策略
    //远程数据调用sendRequest()方法读取
    //本地数据调用fetchLocalBlocks()方法读取
    val blockFetcherItr = new ShuffleBlockFetcherIterator(
      context,
      blockManager.shuffleClient,
      blockManager,
      mapOutputTracker.getMapSizesByExecutorId(handle.shuffleId, startPartition),
      // Note: we use getSizeAsMb when no suffix is provided for backwards compatibility
      SparkEnv.get.conf.getSizeAsMb("spark.reducer.maxSizeInFlight", "48m") * 1024 * 1024)

    // Wrap the streams for compression based on configuration
    val wrappedStreams = blockFetcherItr.map { case (blockId, inputStream) =>
      blockManager.wrapForCompression(blockId, inputStream)
    }

    val ser = Serializer.getSerializer(dep.serializer)
    val serializerInstance = ser.newInstance()

    // Create a key/value iterator for each stream
    val recordIter = wrappedStreams.flatMap { wrappedStream =>
      // Note: the asKeyValueIterator below wraps a key/value iterator inside of a
      // NextIterator. The NextIterator makes sure that close() is called on the
      // underlying InputStream when all records have been read.
      serializerInstance.deserializeStream(wrappedStream).asKeyValueIterator
    }

    // Update the context task metrics for each record read.
    val readMetrics = context.taskMetrics.createShuffleReadMetricsForDependency()
    val metricIter = CompletionIterator[(Any, Any), Iterator[(Any, Any)]](
      recordIter.map(record => {
        readMetrics.incRecordsRead(1)
        record
      }),
      context.taskMetrics().updateShuffleReadMetrics())

    // An interruptible iterator must be used here in order to support task cancellation
    val interruptibleIter = new InterruptibleIterator[(Any, Any)](context, metricIter)

    val aggregatedIter: Iterator[Product2[K, C]] = if (dep.aggregator.isDefined) {
      if (dep.mapSideCombine) { 
        // 读取Map端已经聚合的数据
        val combinedKeyValuesIterator = interruptibleIter.asInstanceOf[Iterator[(K, C)]]
        dep.aggregator.get.combineCombinersByKey(combinedKeyValuesIterator, context)
      } else {
        //读取Reducer端聚合的数据
        val keyValuesIterator = interruptibleIter.asInstanceOf[Iterator[(K, Nothing)]]
        dep.aggregator.get.combineValuesByKey(keyValuesIterator, context)
      }
    } else {
      require(!dep.mapSideCombine, "Map-side combine without Aggregator specified!")
      interruptibleIter.asInstanceOf[Iterator[Product2[K, C]]]
    }

    // 对输出结果进行排序
    dep.keyOrdering match {
      case Some(keyOrd: Ordering[K]) =>
        // Create an ExternalSorter to sort the data. Note that if spark.shuffle.spill is disabled,
        // the ExternalSorter won't spill to disk.
        val sorter = new ExternalSorter[K, C, C](ordering = Some(keyOrd), serializer = Some(ser))
        sorter.insertAll(aggregatedIter)
        context.taskMetrics().incMemoryBytesSpilled(sorter.memoryBytesSpilled)
        context.taskMetrics().incDiskBytesSpilled(sorter.diskBytesSpilled)
        context.internalMetricsToAccumulators(
          InternalAccumulator.PEAK_EXECUTION_MEMORY).add(sorter.peakMemoryUsedBytes)
        sorter.iterator
      case None =>
        aggregatedIter
    }
  }

ShuffleBlockFetcherIterator#splitLocalRemoteBlocks()

splitLocalRemoteBlocks()方法确定数据的读取策略，localBlocks变量记录在本地机器的BlockID，remoteBlocks变量则用于记录全部在远程机器上的BlockID。

远程数据块被切割成最大为maxSizeInFlight大小的FetchRequests

val remoteRequests = new ArrayBuffer[FetchRequest]

splitLocalRemoteBlocks()方法具有源代码例如以下：

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.
    //maxBytesInFlight为每次请求的最大数据量，默认值为48M
    //通过SparkEnv.get.conf.getSizeAsMb("spark.reducer.maxSizeInFlight", "48m") * 1024 * 1024)进行设置
    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 = 0
    for ((address, blockInfos) <- blocksByAddress) {
      totalBlocks += blockInfos.size
      //要获取的数据在本地
      if (address.executorId == blockManager.blockManagerId.executorId) {
        // Filter out zero-sized blocks
        //记录数据在本地的BlockID
        localBlocks ++= blockInfos.filter(_._2 != 0).map(_._1)
        numBlocksToFetch += localBlocks.size
      } else {
       //数据不在本地时
        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))
            //记录数据在远程机器上的BlockID
            remoteBlocks += blockId
            numBlocksToFetch += 1
            curRequestSize += size
          } else if (size < 0) {
            throw new BlockException(blockId, "Negative block size " + size)
          }
          if (curRequestSize >= targetRequestSize) {
            // Add this FetchRequest
            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")
    remoteRequests
  }

ShuffleBlockFetcherIterator#fetchLocalBlocks()

fetchLocalBlocks()方法进行本地Block的读取。调用的是BlockManager的getBlockData方法。其源代码例如以下：

private[this] def fetchLocalBlocks() {
    val iter = localBlocks.iterator
    while (iter.hasNext) {
      val blockId = iter.next()
      try {
        //调用BlockManager的getBlockData方法
        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
      }
    }
  }

跳转到BlockManager的getBlockData方法。能够看到其源代码例如以下：

override def getBlockData(blockId: BlockId): ManagedBuffer = {
          if (blockId.isShuffle) {   
//先调用的是ShuffleManager的shuffleBlockResolver方法。得到ShuffleBlockResolver
//然后再调用其getBlockData方法   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)
      }
    }
  }

org.apache.spark.shuffle.hash.ShuffleManager#shuffleBlockResolver()方法获取相应的ShuffleBlockResolver，假设是Hash Shuffle，则
是org.apache.spark.shuffle.FileShuffleBlockResolver，假设是Sort Shuffle则org.apache.spark.shuffle.IndexShuffleBlockResolver。

然后调用相应ShuffleBlockResolver的getBlockData方法，返回相应的FileSegment。

FileShuffleBlockResolver#getBlockData方法源代码例如以下：

override def getBlockData(blockId: ShuffleBlockId): ManagedBuffer = {
    //相应Hash Shuffle中的Shuffle Consolidate Files机制生成的文件
    if (consolidateShuffleFiles) { 
      // Search all file groups associated with this shuffle.
      val shuffleState = shuffleStates(blockId.shuffleId)
      val iter = shuffleState.allFileGroups.iterator
      while (iter.hasNext) {
        val segmentOpt = iter.next.getFileSegmentFor(blockId.mapId, blockId.reduceId)
        if (segmentOpt.isDefined) {
          val segment = segmentOpt.get
          return new FileSegmentManagedBuffer(
            transportConf, segment.file, segment.offset, segment.length)
        }
      }
      throw new IllegalStateException("Failed to find shuffle block: " + blockId)
    } else {
      //普通的Hash Shuffle机制生成的文件
      val file = blockManager.diskBlockManager.getFile(blockId)
      new FileSegmentManagedBuffer(transportConf, file, 0, file.length)
    }
  }

IndexShuffleBlockResolver#getBlockData方法源代码例如以下：

override def getBlockData(blockId: ShuffleBlockId): ManagedBuffer = {
    // The block is actually going to be a range of a single map output file for this map, so
    // find out the consolidated file, then the offset within that from our index
    //使用shuffleId和mapId，获取相应索引文件
    val indexFile = getIndexFile(blockId.shuffleId, blockId.mapId)

    val in = new DataInputStream(new FileInputStream(indexFile))
    try {
      //定位到本次Block相应的数据位置
      ByteStreams.skipFully(in, blockId.reduceId * 8)
      //数据起始位置
      val offset = in.readLong()
      //数据结束位置
      val nextOffset = in.readLong()
      //返回FileSegment
      new FileSegmentManagedBuffer(
        transportConf,
        getDataFile(blockId.shuffleId, blockId.mapId),
        offset,
        nextOffset - offset)
    } finally {
      in.close()
    }
  }

ShuffleBlockFetcherIterator#sendRequest()

sendRequest()方法用于从远程机器上获取数据

 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 blockID
    val sizeMap = req.blocks.map { case (blockId, size) => (blockId.toString, size) }.toMap
    val blockIds = req.blocks.map(_._1.toString)

    val address = req.address
    //使用ShuffleClient的fetchBlocks方法获取数据
    //有两种ShuffleClient。各自是ExternalShuffleClient和BlockTransferService
    //默觉得BlockTransferService
    shuffleClient.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))
        }
      }
    )
  }

通过上面的代码能够看到，代码使用的是shuffleClient.fetchBlocks进行远程Block数据的获取。org.apache.spark.network.shuffle.ShuffleClient有两个子类，各自是ExternalShuffleClient和BlockTransferService，而org.apache.spark.network.shuffle.BlockTransferService又有两个子类。各自是NettyBlockTransferService和NioBlockTransferService，shuffleClient 对象在 org.apache.spark.storage.BlockManager定义，其源代码例如以下：

// org.apache.spark.storage.BlockManager中定义的shuffleClient 
 private[spark] val shuffleClient = if (externalShuffleServiceEnabled) {
    //使用ExternalShuffleClient获取远程Block数据
    val transConf = SparkTransportConf.fromSparkConf(conf, numUsableCores)
    new ExternalShuffleClient(transConf, securityManager, securityManager.isAuthenticationEnabled(),
      securityManager.isSaslEncryptionEnabled())
  } else {
    //使用NettyBlockTransferService或NioBlockTransferService获取远程Block数据
    blockTransferService
  }

代码中的blockTransferService在SparkEnv中被初始化，详细例如以下：

 //org.apache.spark.SparkEnv中初始化blockTransferService 
 val blockTransferService =
      conf.get("spark.shuffle.blockTransferService", "netty").toLowerCase match {
        case "netty" =>
          new NettyBlockTransferService(conf, securityManager, numUsableCores)
        case "nio" =>
          logWarning("NIO-based block transfer service is deprecated, " +
            "and will be removed in Spark 1.6.0.")
          new NioBlockTransferService(conf, securityManager)
      }