16.Spark Streaming源码解读之数据清理机制解析

原创文章，转载请注明:转载自 听风居士博客(http://www.cnblogs.com/zhouyf/)

本期内容：

一、Spark Streaming 数据清理总览

二、Spark Streaming 数据清理过程详解

三、Spark Streaming 数据清理的触发机制

Spark Streaming不像普通Spark 的应用程序，普通Spark程序运行完成后,中间数据会随着SparkContext的关闭而被销毁,而Spark Streaming一直在运行，不断计算，每一秒中在不断运行都会产生大量的中间数据，所以需要对对象及元数据需要定期清理。每个batch duration运行时不断触发job后需要清理rdd和元数据。下面我们就结合源码详细解析一下Spark Streaming程序的数据清理机制。

一、数据清理总览

Spark Streaming 运行过程中,随着时间不断产生Job,当job运行结束后,需要清理相应的数据(RDD,元数据信息,Checkpoint数据),Job由JobGenerator定时产生,数据的清理也是有JobGenerator负责。

JobGenerator负责数据清理控制的代码位于一个消息循环体eventLoop中：

    eventLoop = new EventLoop[JobGeneratorEvent]("JobGenerator") {
      override protected def onReceive(event: JobGeneratorEvent): Unit = processEvent(event)

      override protected def onError(e: Throwable): Unit = {
        jobScheduler.reportError("Error in job generator", e)
      }
    }
    eventLoop.start()

其中的核心逻辑位于processEvent(event)函数中:

  /** 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(time, clearCheckpointDataLater)
      case ClearCheckpointData(time) => clearCheckpointData(time)
    }
  }

可以看到当JobGenerator收到ClearMetadata(time) 和 ClearCheckpointData(time)是会进行相应的数据清理,其中 clearMetadata(time)会清理RDD数据和一些元数据信息, ClearCheckpointData(time)会清理Checkpoint数据。

二、数据清理过程详解

    2.1 ClearMetaData 过程详解

首先看一下clearMetaData函数的处理逻辑：

  /** Clear DStream metadata for the given `time`. */
  private def clearMetadata(time: Time) {
    ssc.graph.clearMetadata(time)

    // If checkpointing is enabled, then checkpoint,
    // else mark batch to be fully processed
    if (shouldCheckpoint) {
      eventLoop.post(DoCheckpoint(time, clearCheckpointDataLater = true))
    } else {
      // If checkpointing is not enabled, then delete metadata information about
      // received blocks (block data not saved in any case). Otherwise, wait for
      // checkpointing of this batch to complete.
      val maxRememberDuration = graph.getMaxInputStreamRememberDuration()
      jobScheduler.receiverTracker.cleanupOldBlocksAndBatches(time - maxRememberDuration)
      jobScheduler.inputInfoTracker.cleanup(time - maxRememberDuration)
      markBatchFullyProcessed(time)
    }
  }

首先调用了DStreamGraph的clearMetadata方法:

  def clearMetadata(time: Time) {
    logDebug("Clearing metadata for time " + time)
    this.synchronized {
      outputStreams.foreach(_.clearMetadata(time))
    }
    logDebug("Cleared old metadata for time " + time)
  }

这里调用了所有OutputDStream (关于DStream 的分类请参考http://blog.csdn.net/zhouzx2010/article/details/51460790)的clearMetadata方法

  private[streaming] def clearMetadata(time: Time) {
    val unpersistData = ssc.conf.getBoolean("spark.streaming.unpersist", true)
     //获取需要清理的RDD
    val oldRDDs = generatedRDDs.filter(_._1 <= (time - rememberDuration))
    logDebug("Clearing references to old RDDs: [" +
      oldRDDs.map(x => s"${x._1} -> ${x._2.id}").mkString(", ") + "]")


    //将要清除的RDD从generatedRDDs 中清除 
    generatedRDDs --= oldRDDs.keys
    if (unpersistData) {
      logDebug(s"Unpersisting old RDDs: ${oldRDDs.values.map(_.id).mkString(", ")}")
      oldRDDs.values.foreach { rdd =>
      //将RDD 从persistence列表中移除
        rdd.unpersist(false)
        // Explicitly remove blocks of BlockRDD
        rdd match {
          case b: BlockRDD[_] =>
            logInfo(s"Removing blocks of RDD $b of time $time")
            //移除RDD的block 数据
            b.removeBlocks()
          case _ =>
        }
      }
    }
    logDebug(s"Cleared ${oldRDDs.size} RDDs that were older than " +
      s"${time - rememberDuration}: ${oldRDDs.keys.mkString(", ")}")
    //清除依赖的DStream
    dependencies.foreach(_.clearMetadata(time))
  }

关键的清理逻辑在代码中做了详细注释,首先清理DStream对应的RDD的元数据信息,然后清理RDD的数据,最后对DStream所依赖的DStream进行清理。

回到JobGenerator的clearMetadata函数：

    if (shouldCheckpoint) {
      eventLoop.post(DoCheckpoint(time, clearCheckpointDataLater = true))
    } else {
      // If checkpointing is not enabled, then delete metadata information about
      // received blocks (block data not saved in any case). Otherwise, wait for
      // checkpointing of this batch to complete.
      val maxRememberDuration = graph.getMaxInputStreamRememberDuration()
      jobScheduler.receiverTracker.cleanupOldBlocksAndBatches(time - maxRememberDuration)
      jobScheduler.inputInfoTracker.cleanup(time - maxRememberDuration)
      markBatchFullyProcessed(time)
    }

调用了ReceiverTracker的 cleanupOldBlocksAndBatches方法，最后调用了clearupOldBatches方法：

  def cleanupOldBatches(cleanupThreshTime: Time, waitForCompletion: Boolean): Unit = synchronized {
    require(cleanupThreshTime.milliseconds < clock.getTimeMillis())
    val timesToCleanup = timeToAllocatedBlocks.keys.filter { _ < cleanupThreshTime }.toSeq
    logInfo(s"Deleting batches: ${timesToCleanup.mkString(" ")}")
    if (writeToLog(BatchCleanupEvent(timesToCleanup))) {
      //将要删除的Batch数据清除
      timeToAllocatedBlocks --= timesToCleanup
      //清理WAL日志
      writeAheadLogOption.foreach(_.clean(cleanupThreshTime.milliseconds, waitForCompletion))
    } else {
      logWarning("Failed to acknowledge batch clean up in the Write Ahead Log.")
    }
  }

可以看到ReceiverTracker的clearupOldBatches方法清理了Receiver数据,也就是Batch数据和WAL日志数据。

最后对InputInfoTracker信息进行清理：

  def cleanup(batchThreshTime: Time): Unit = synchronized {
    val timesToCleanup = batchTimeToInputInfos.keys.filter(_ < batchThreshTime)
    logInfo(s"remove old batch metadata: ${timesToCleanup.mkString(" ")}")
    batchTimeToInputInfos --= timesToCleanup
  }

这简单的清除了batchTimeToInputInfos 的输入信息。

2.2 ClearCheckPoint 过程详解

看一下clearCheckpointData的处理逻辑:

  /** Clear DStream checkpoint data for the given `time`. */
  private def clearCheckpointData(time: Time) {
    ssc.graph.clearCheckpointData(time)

    // All the checkpoint information about which batches have been processed, etc have
    // been saved to checkpoints, so its safe to delete block metadata and data WAL files
    val maxRememberDuration = graph.getMaxInputStreamRememberDuration()
    jobScheduler.receiverTracker.cleanupOldBlocksAndBatches(time - maxRememberDuration)
    jobScheduler.inputInfoTracker.cleanup(time - maxRememberDuration)
    markBatchFullyProcessed(time)
  }

后面的ReceiverTraker和InputInforTracker的清理逻辑和ClearMetaData的相同,这分析DStreamGraph的clearCheckpointData方法:

  def clearCheckpointData(time: Time) {
    logInfo("Clearing checkpoint data for time " + time)
    this.synchronized {
      outputStreams.foreach(_.clearCheckpointData(time))
    }
    logInfo("Cleared checkpoint data for time " + time)
  }

同样的调用了DStreamGraph中所有OutputDStream的clearCheckPiontData 方法:

  private[streaming] def clearCheckpointData(time: Time) {
    logDebug("Clearing checkpoint data")
    checkpointData.cleanup(time)
    dependencies.foreach(_.clearCheckpointData(time))
    logDebug("Cleared checkpoint data")
  }

这里的核心逻辑在checkpointData.cleanup(time)方法,这里的CheckpointData 是 DStreamCheckpointData对象, DStreamCheckpointData的clearup方法如下:

def cleanup(time: Time) {
    // 获取需要清理的Checkpoint 文件 时间
    timeToOldestCheckpointFileTime.remove(time) match {
      case Some(lastCheckpointFileTime) =>
        //获取需要删除的文件
        val filesToDelete = timeToCheckpointFile.filter(_._1 < lastCheckpointFileTime)
        logDebug("Files to delete:\n" + filesToDelete.mkString(","))
        filesToDelete.foreach {
          case (time, file) =>
            try {
              val path = new Path(file)
              if (fileSystem == null) {
                fileSystem = path.getFileSystem(dstream.ssc.sparkContext.hadoopConfiguration)
              }
          //  删除文件   
              fileSystem.delete(path, true)
              timeToCheckpointFile -= time
              logInfo("Deleted checkpoint file '" + file + "' for time " + time)
            } catch {
              case e: Exception =>
                logWarning("Error deleting old checkpoint file '" + file + "' for time " + time, e)
                fileSystem = null
            }
        }
      case None =>
        logDebug("Nothing to delete")
    }
  }

可以看到checkpoint的清理,就是删除了指定时间以前的checkpoint文件。

三、数据清理的触发

    3.1 ClearMetaData 过程的触发

JobGenerator 生成job后,交给JobHandler执行, JobHandler的run方法中,会在job执行完后给JobScheduler 发送JobCompleted消息:

          _eventLoop = eventLoop
          if (_eventLoop != null) {
            _eventLoop.post(JobCompleted(job, clock.getTimeMillis()))

          }

JobScheduler 收到JobCompleted 消息调用 handleJobCompletion 方法,源码如下:

  private def processEvent(event: JobSchedulerEvent) {
    try {
      event match {
        case JobStarted(job, startTime) => handleJobStart(job, startTime)
        case JobCompleted(job, completedTime) => handleJobCompletion(job, completedTime)
        case ErrorReported(m, e) => handleError(m, e)
      }
    } catch {
      case e: Throwable =>
        reportError("Error in job scheduler", e)
    }
  }

在 JobScheduler 的handleJobCompletion方法中会调用JobGenerator的onBatchCompletion方法,我们看一下JobGenerator的 onBatchCompletion 方法的源码:

  def onBatchCompletion(time: Time) {
    eventLoop.post(ClearMetadata(time))
  }

可以看到JobGenerator的onBatchCompletion方法给自己发送了ClearMetadata消息从而触发了ClearMetaData操作。

3.2 ClearCheckPoint 过程的触发

    清理CheckPoint数据发生在CheckPoint完成之后,我们先看一下CheckPointHandler的run方法:

          // All done, print success
          val finishTime = System.currentTimeMillis()
          logInfo("Checkpoint for time " + checkpointTime + " saved to file '" + checkpointFile +
            "', took " + bytes.length + " bytes and " + (finishTime - startTime) + " ms")
          //调用JobGenerator的方法进行checkpoint数据清理
          jobGenerator.onCheckpointCompletion(checkpointTime, clearCheckpointDataLater)
          return

可以看到在checkpoint完成后,会调用JobGenerator的onCheckpointCompletion方法进行checkpoint数据清理,我查看JobGenerator的onCheckpointCompletion方法源码:

  def onCheckpointCompletion(time: Time, clearCheckpointDataLater: Boolean) {
    if (clearCheckpointDataLater) {
      eventLoop.post(ClearCheckpointData(time))
    }
  }

可以看到JobGenerator的onCheckpointCompletion方法中首先对传进来的 clearCheckpointDataLater 参数进行判断,如果该参数为true,就会给JobGenerator的eventLoop循环体发送ClearCheckpointData消息,从而触发clearCheckpointData 方法的调用,进行Checkpoint数据的清理。

什么时候该参数会true呢？

我们回到JobGenerator的 ClearMetadata 方法：

  private def clearMetadata(time: Time) {
    ssc.graph.clearMetadata(time)

    if (shouldCheckpoint) {
      //发送DoCheckpoint消息,并进行相应的Checkpoint数据清理
      eventLoop.post(DoCheckpoint(time, clearCheckpointDataLater = true))
    } else {
      val maxRememberDuration = graph.getMaxInputStreamRememberDuration()
      jobScheduler.receiverTracker.cleanupOldBlocksAndBatches(time - maxRememberDuration)
      jobScheduler.inputInfoTracker.cleanup(time - maxRememberDuration)
      markBatchFullyProcessed(time)
    }
  }

可以看到在clearMetadata方法中,发送了DoCheckpoint消息,其中参数 clearCheckpointDataLater 为ture。Generator的eventLoop收到该消息后调用 doCheckpoint 方法：

  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)
    }
  }

这里关键一步：调用了CheckpointWriter的write方法,注意此时参数 clearCheckpointDataLater 为true。我们进入该方法：

  def write(checkpoint: Checkpoint, clearCheckpointDataLater: Boolean) {
    try {
      val bytes = Checkpoint.serialize(checkpoint, conf)
//将参数clearCheckpointDataLater传入CheckpoitWriteHandler
      executor.execute(new CheckpointWriteHandler(
        checkpoint.checkpointTime, bytes, clearCheckpointDataLater))
      logInfo("Submitted checkpoint of time " + checkpoint.checkpointTime + " writer queue")
    } catch {
      case rej: RejectedExecutionException =>
        logError("Could not submit checkpoint task to the thread pool executor", rej)
    }
  }

可以看到此时参数 clearCheckpointDataLater 传入CheckpointWriteHandler 。这样Checkpoint完成之后就会发送ClearCheckpointData消息给JobGenerator进行Checkpoint数据的清理。

原创文章，转载请注明:转载自 听风居士博客(http://www.cnblogs.com/zhouyf/)

来自为知笔记(Wiz)