Spark Streaming源码解读之Driver容错安全性

本篇博文的目标如下：
1. ReceiverBlockTracker容错安全性
2. DStream和JobGenerator容错安全性

文章的组织思路如下：
考虑Driver容错安全性，我们要思考什么？
再详细分析ReceiverBlockTracker，DStream和JobGenerator容错安全性

一：容错安全性
1. ReceivedBlockTracker负责管理Spark Streaming运行程序的元数据。数据层面
2. DStream和JobGenerator是作业调度的核心层面，也就是具体调度到什么程度了，从运行的考虑的。DStream是逻辑层面。
3. 作业生存层面，JobGenerator是Job调度层面，具体调度到什么程度了。从运行的角度的。
谈Driver容错你要考虑Driver中有那些需要维持状态的运行。
1. ReceivedBlockTracker跟踪了数据，因此需要容错。通过WAL方式容错。
2. DStreamGraph表达了依赖关系，恢复状态的时候需要根据DStream恢复计算逻辑级别的依赖关系。通过checkpoint方式容错。
3. JobGenerator表面你是怎么基于ReceiverBlockTracker中的数据，以及DStream构成的依赖关系不断的产生Job的过程。你消费了那些数据，进行到什么程度了。

总结如下：

ReceivedBlockTracker：
1. ReceivedBlockTracker会管理Spark Streaming运行过程中所有的数据。并且把数据分配给需要的batches，所有的动作都会被WAL写入到Log中，Driver失败的话，就可以根据历史恢复tracker状态，在ReceivedBlockTracker创建的时候，使用checkpoint保存历史目录。

/**
 * This class manages the execution of the receivers of ReceiverInputDStreams. Instance of
 * this class must be created after all input streams have been added and StreamingContext.start()
 * has been called because it needs the final set of input streams at the time of instantiation.
 *
 * @param skipReceiverLaunch Do not launch the receiver. This is useful for testing.
 */
private[streaming]
class ReceiverTracker(ssc: StreamingContext, skipReceiverLaunch: Boolean = false) extends Logging {

下面就从Receiver收到数据之后，怎么处理的开始。
2. ReceiverBlockTracker.addBlock源码如下：
Receiver接收到数据，把元数据信息汇报上来，然后通过ReceiverSupervisorImpl就将数据汇报上来，就直接通过WAL进行容错.
当Receiver的管理者，ReceiverSupervisorImpl把元数据信息汇报给Driver的时候，正在处理是交给ReceiverBlockTracker. ReceiverBlockTracker将数据写进WAL文件中，然后才会写进内存中，被当前的Spark Streaming程序的调度器使用的，也就是JobGenerator使用的。JobGenerator不可能直接使用WAL。WAL的数据在磁盘中，这里JobGenerator使用的内存中缓存的数据结构

/** Add received block. This event will get written to the write ahead log (if enabled). */
def addBlock(receivedBlockInfo: ReceivedBlockInfo): Boolean = {
try {
// writeToLog
val writeResult = writeToLog(BlockAdditionEvent(receivedBlockInfo))
if (writeResult) {
      synchronized {
//数据汇报上来的时候只有成功写进WAL的时候，才会把ReceivedBlockInfo元数据信息放进Queue
        getReceivedBlockQueue(receivedBlockInfo.streamId) += receivedBlockInfo
      }
      logDebug(s"Stream ${receivedBlockInfo.streamId} received " +
        s"block ${receivedBlockInfo.blockStoreResult.blockId}")
    } else {
      logDebug(s"Failed to acknowledge stream ${receivedBlockInfo.streamId} receiving " +
        s"block ${receivedBlockInfo.blockStoreResult.blockId} in the Write Ahead Log.")
    }
    writeResult
  } catch {
case NonFatal(e) =>
      logError(s"Error adding block $receivedBlockInfo", e)
false
  }
}

此时的数据结构就是streamIdToUnallocatedBlockQueues，Driver端接收到的数据保存在streamIdToUnallocatedBlockQueues中。

private val streamIdToUnallocatedBlockQueues = new mutable.HashMap[Int, ReceivedBlockQueue]

3.  allocateBlocksToBatch把接收到的数据但是没有分配，分配给batch，根据streamId取出Block，由此就知道Spark Streaming处理数据的时候可以有不同的数据来源，例如Kafka,Socket。 

到底什么是batchTime?
batchTime是上一个Job分配完数据之后，开始再接收到的数据的时间。

/**
 * Allocate all unallocated blocks to the given batch.
 * This event will get written to the write ahead log (if enabled).
 */
def allocateBlocksToBatch(batchTime: Time): Unit = synchronized {
if (lastAllocatedBatchTime == null || batchTime > lastAllocatedBatchTime) {
// streamIdToBlocks获得了所有分配的数据 
val streamIdToBlocks = streamIds.map { streamId =>
// getReceivedBlockQueue就把streamId获得的数据存储了。如果要分配给batch，//让数据出队列就OK了。
        (streamId, getReceivedBlockQueue(streamId).dequeueAll(x => true))
    }.toMap
val allocatedBlocks = AllocatedBlocks(streamIdToBlocks)
//获得元数据信息之后并没有立即分配给作业，还是进行WAL
//所以如果Driver出错之后，再恢复就可以将作业的正常的分配那些Block状态
//这里指的是针对于Batch Time分配那些Block状态都可以恢复回来。
if (writeToLog(BatchAllocationEvent(batchTime, allocatedBlocks))) {
//JobGenerator就是从timeToAllocatedBlocks中获取数据。
// 这个时间段batchTime就知道了要处理那些数据allocatedBlocks
      timeToAllocatedBlocks.put(batchTime, allocatedBlocks)
//
      lastAllocatedBatchTime = batchTime
    } else {
      logInfo(s"Possibly processed batch $batchTime need to be processed again in WAL recovery")
    }
  } else {
// This situation occurs when:
// 1. WAL is ended with BatchAllocationEvent, but without BatchCleanupEvent,
// possibly processed batch job or half-processed batch job need to be processed again,
// so the batchTime will be equal to lastAllocatedBatchTime.
// 2. Slow checkpointing makes recovered batch time older than WAL recovered
// lastAllocatedBatchTime.
// This situation will only occurs in recovery time.
    logInfo(s"Possibly processed batch $batchTime need to be processed again in WAL recovery")
  }
}

4.  timeToAllocatedBlocks可以有很多的时间窗口的Blocks，也就是Batch Duractions的Blocks。这里面就维护了很多Batch Duractions分配的数据，假设10秒是一个Batch Duractions也就是10s产生一个Job的话，如果此时想算过去的数据，只需要根据时间进行聚合操作即可。

private val timeToAllocatedBlocks = new mutable.HashMap[Time, AllocatedBlocks]

5.  根据streamId获取Block信息

/** Class representing the blocks of all the streams allocated to a batch */
private[streaming]
case class AllocatedBlocks(streamIdToAllocatedBlocks: Map[Int, Seq[ReceivedBlockInfo]]) {
def getBlocksOfStream(streamId: Int): Seq[ReceivedBlockInfo] = {
    streamIdToAllocatedBlocks.getOrElse(streamId, Seq.empty)
  }
}

6.  cleanupOldBatches：因为时间的推移会不断的生成RDD，RDD会不断的处理数据，
因此不可能一直保存历史数据。

/**
 * Clean up block information of old batches. If waitForCompletion is true, this method
 * returns only after the files are cleaned up.
 */
def cleanupOldBatches(cleanupThreshTime: Time, waitForCompletion: Boolean): Unit = synchronized {
  require(cleanupThreshTime.milliseconds < clock.getTimeMillis())
  val timesToCleanup = timeToAllocatedBlocks.keys.filter { _ < cleanupThreshTime }.toSeq
  logInfo("Deleting batches " + timesToCleanup)
//WAL
if (writeToLog(BatchCleanupEvent(timesToCleanup))) {
    timeToAllocatedBlocks --= timesToCleanup
    writeAheadLogOption.foreach(_.clean(cleanupThreshTime.milliseconds, waitForCompletion))
  } else {
    logWarning("Failed to acknowledge batch clean up in the Write Ahead Log.")
  }
}

7.  writeToLog源码如下：

/** Write an update to the tracker to the write ahead log */
private def writeToLog(record: ReceivedBlockTrackerLogEvent): Boolean = {
if (isWriteAheadLogEnabled) {
    logTrace(s"Writing record: $record")
try {
      writeAheadLogOption.get.write(ByteBuffer.wrap(Utils.serialize(record)),
        clock.getTimeMillis())
true
    } catch {
case NonFatal(e) =>
        logWarning(s"Exception thrown while writing record: $record to the WriteAheadLog.", e)
false
    }
  } else {
true
  }
}

总结：
WAL对数据的管理包括数据的生成，数据的销毁和消费。上述在操作之后都要先写入到WAL的文件中.

JobGenerator:
Checkpoint会有时间间隔Batch Duractions，Batch执行前和执行后都会进行checkpoint。
doCheckpoint被调用的前后流程：

1. generateJobs：

/** Generate jobs and perform checkpoint for the given `time`. */
private def generateJobs(time: Time) {
  // Set the SparkEnv in this thread, so that job generation code can access the environment
 // Example: BlockRDDs are created in this thread, and it needs to access BlockManager
 // Update: This is probably redundant after threadlocal stuff in SparkEnv has been removed.
 SparkEnv.set(ssc.env)
 Try {
 jobScheduler.receiverTracker.allocateBlocksToBatch(time) // allocate received blocks to batch
 graph.generateJobs(time) // generate jobs using allocated block
 } match {
case Success(jobs) =>
 val streamIdToInputInfos = jobScheduler.inputInfoTracker.getInfo(time)
 jobScheduler.submitJobSet(JobSet(time, jobs, streamIdToInputInfos))
case Failure(e) =>
 jobScheduler.reportError("Error generating jobs for time " + time, e)
 }
//上面自学就那个完之后就需要进行checkpoint
 eventLoop.post(DoCheckpoint(time, clearCheckpointDataLater = false))
}

2.  processEvent接收到消息

/** Processes all events */
private def processEvent(event: JobGeneratorEvent) {
  logDebug("Got event " + event)
  event match {
case GenerateJobs(time) => generateJobs(time)
case ClearMetadata(time) => clearMetadata(time)
case DoCheckpoint(time, clearCheckpointDataLater) =>
// doCheckpoint被调用
      doCheckpoint(time, clearCheckpointDataLater)
case ClearCheckpointData(time) => clearCheckpointData(time)
  }
}

3.  把当前的状态进行Checkpoint.

/** Perform checkpoint for the give `time`. */
private def doCheckpoint(time: Time, clearCheckpointDataLater: Boolean) {
if (shouldCheckpoint && (time - graph.zeroTime).isMultipleOf(ssc.checkpointDuration)) {
    logInfo("Checkpointing graph for time " + time)
    ssc.graph.updateCheckpointData(time)
    checkpointWriter.write(new Checkpoint(ssc, time), clearCheckpointDataLater)
  }
}

4.  DStream中的updateCheckpointData源码如下：最终导致RDD的Checkpoint

/**
 * Refresh the list of checkpointed RDDs that will be saved along with checkpoint of
 * this stream. This is an internal method that should not be called directly. This is
 * a default implementation that saves only the file names of the checkpointed RDDs to
 * checkpointData. Subclasses of DStream (especially those of InputDStream) may override
 * this method to save custom checkpoint data.
 */
private[streaming] def updateCheckpointData(currentTime: Time) {
  logDebug("Updating checkpoint data for time " + currentTime)
  checkpointData.update(currentTime)
  dependencies.foreach(_.updateCheckpointData(currentTime))
  logDebug("Updated checkpoint data for time " + currentTime + ": " + checkpointData)
}

5.  shouldCheckpoint是状态变量。

// This is marked lazy so that this is initialized after checkpoint duration has been set
// in the context and the generator has been started.
private lazy val shouldCheckpoint = ssc.checkpointDuration != null && ssc.checkpointDir != null

JobGenerator容错安全性如下图：