为了账号安全,请及时绑定邮箱和手机立即绑定

[spark streaming] ReceiverTracker 数据产生与存储

标签:
Spark

前言

在Spark Streaming里,总体负责任务的动态调度是JobScheduler,而JobScheduler有两个很重要的成员:JobGeneratorReceiverTrackerJobGenerator 负责将每个 batch 生成具体的 RDD DAG ,而ReceiverTracker负责数据的来源。

需要在executor上运行的receiver接收数据的InputDStream都需要继承ReceiverInputDStream,ReceiverInputDStream有一个def getReceiver(): Receiver[T]方法,子类都需要实现这个方法。如KafkaInputDStream对应KafkaReceiverFlumeInputDStream对应FlumeReceiverTwitterInputDStream对应TwitterReceiver等。

流程概述:

  • ReceiverTracker 启动,获取所有InputDStreams对应的receivers

  • 根据调度策略确定每个Receiver的优先位置(能在哪些executor上执行)

  • 将Receiver包装成RDD并通过sc提交一个job,执行函数为创建supervisor实例,调用start()方法,也即调用了Receiver的onStart()方法

  • Receiver的onStart不断接收数据,通过store方法最终调用supervisor来存储块

  • 存储后通知ReceiverTracker此Block的信息

  • ReceiverTracker将Block消息交给ReceivedBlockTracker管理

启动 Receiver

先看看receiverTracker的启动过程:

ssc.start()
    scheduler.start()
        receiverTracker.start()
        jobGenerator.start()
---- def start(): Unit = synchronized {    if (isTrackerStarted) {      throw new SparkException("ReceiverTracker already started")
    }    if (!receiverInputStreams.isEmpty) {
      endpoint = ssc.env.rpcEnv.setupEndpoint(        "ReceiverTracker", new ReceiverTrackerEndpoint(ssc.env.rpcEnv))      if (!skipReceiverLaunch) launchReceivers()
      logInfo("ReceiverTracker started")
      trackerState = Started
    }
  }

在start方法中先创建了ReceiverTracker的Endpoint,接着调用launchReceivers()方法来启动Recivers:

 private def launchReceivers(): Unit = {
    val receivers = receiverInputStreams.map { nis =>
      val rcvr = nis.getReceiver()
      rcvr.setReceiverId(nis.id)
      rcvr
    }

    runDummySparkJob()

    logInfo("Starting " + receivers.length + " receivers")
    endpoint.send(StartAllReceivers(receivers))
  }

遍历所有的InputStream,并得到所对应的Receiver集合receivers。并向ReceiverTrackerEndpoint发送了StartAllReceivers消息,看看接收到该消息后是如何处理的:

 case StartAllReceivers(receivers) =>
        val scheduledLocations = schedulingPolicy.scheduleReceivers(receivers, getExecutors)        for (receiver <- receivers) {
          val executors = scheduledLocations(receiver.streamId)
          updateReceiverScheduledExecutors(receiver.streamId, executors)
          receiverPreferredLocations(receiver.streamId) = receiver.preferredLocation
          startReceiver(receiver, executors)
        }

通过调度策略来计算决定每个receiver的一组优先位置,即一个Receiver改在哪个executor节点上启动,调度的主要原则是:

  • 满足Receiver的preferredLocation。

  • 其次保证将Receiver分布的尽量均匀。

接着遍历所有receivers调用了startReceiver(receiver, executors)方法来启动receiver:

 private def startReceiver(
        receiver: Receiver[_],
        scheduledLocations: Seq[TaskLocation]): Unit = {
      def shouldStartReceiver: Boolean = {        // It's okay to start when trackerState is Initialized or Started
        !(isTrackerStopping || isTrackerStopped)
      }

      val receiverId = receiver.streamId      if (!shouldStartReceiver) {
        onReceiverJobFinish(receiverId)        return
      }

      val checkpointDirOption = Option(ssc.checkpointDir)
      val serializableHadoopConf =        new SerializableConfiguration(ssc.sparkContext.hadoopConfiguration)      // Function to start the receiver on the worker node
      val startReceiverFunc: Iterator[Receiver[_]] => Unit =
        (iterator: Iterator[Receiver[_]]) => {          if (!iterator.hasNext) {            throw new SparkException(              "Could not start receiver as object not found.")
          }          if (TaskContext.get().attemptNumber() == 0) {
            val receiver = iterator.next()
            assert(iterator.hasNext == false)
            val supervisor = new ReceiverSupervisorImpl(
              receiver, SparkEnv.get, serializableHadoopConf.value, checkpointDirOption)
            supervisor.start()
            supervisor.awaitTermination()
          } else {            // It's restarted by TaskScheduler, but we want to reschedule it again. So exit it.
          }
        }      // Create the RDD using the scheduledLocations to run the receiver in a Spark job
      val receiverRDD: RDD[Receiver[_]] =        if (scheduledLocations.isEmpty) {
          ssc.sc.makeRDD(Seq(receiver), 1)
        } else {
          val preferredLocations = scheduledLocations.map(_.toString).distinct
          ssc.sc.makeRDD(Seq(receiver -> preferredLocations))
        }
      receiverRDD.setName(s"Receiver $receiverId")
      ssc.sparkContext.setJobDescription(s"Streaming job running receiver $receiverId")
      ssc.sparkContext.setCallSite(Option(ssc.getStartSite()).getOrElse(Utils.getCallSite()))

      val future = ssc.sparkContext.submitJob[Receiver[_], Unit, Unit](
        receiverRDD, startReceiverFunc, Seq(0), (_, _) => Unit, ())      // We will keep restarting the receiver job until ReceiverTracker is stopped
      future.onComplete {        case Success(_) =>          if (!shouldStartReceiver) {
            onReceiverJobFinish(receiverId)
          } else {
            logInfo(s"Restarting Receiver $receiverId")            self.send(RestartReceiver(receiver))
          }        case Failure(e) =>          if (!shouldStartReceiver) {
            onReceiverJobFinish(receiverId)
          } else {
            logError("Receiver has been stopped. Try to restart it.", e)
            logInfo(s"Restarting Receiver $receiverId")            self.send(RestartReceiver(receiver))
          }
      }(ThreadUtils.sameThread)
      logInfo(s"Receiver ${receiver.streamId} started")
    }

注意这里巧妙的将receiver包装成了RDD,并把scheduledLocations作为RDD的优先位置locationPrefs。

然后通过sc提交了一个Spark Core Job,执行函数是startReceiverFunc(也就是要在executor上执行的),在该方法中创建一个ReceiverSupervisorImpl对象,并调用了start()方法,在该方法中会调用 receiver的onStart 后立即返回。

receiver的onStart 方法一般会新建线程或线程池来接收数据,比如在 KafkaReceiver 中,就新建了线程池,在线程池中接收 topics 的数据。

supervisor.start() 返回后,由 supervisor.awaitTermination() 阻塞住线程,以让这个 task 一直不退出,从而可以源源不断接收数据。

Receiver 数据处理

前面提到receiver的onStart()方法会新建线程或线程池来接收数据,那接收的数据怎么处理的呢?都会调用receiver的store(),而store方法又调用了supervisor的方法。对应的store方法有多种形式:

  • pushSingle: 对应单条小数据,需要通过BlockGenerator聚集多条数据后再成块的存储

  • pushArrayBuffer: 对应数组形式的数据

  • pushIterator: 对应 iterator 形式数据

  • pushBytes: 对应 ByteBuffer 形式的块数据

除了pushSingle需要通过BlockGenerator将数据聚集成一个块的时候再存储,其他方法都是直接成块存储。

看看pushSingle是怎么通过聚集的方式存储块的:

def pushSingle(data: Any) {
    defaultBlockGenerator.addData(data)
  }
------def addData(data: Any): Unit = {    if (state == Active) {
      waitToPush()      synchronized {        if (state == Active) {
          currentBuffer += data
        } else {          throw new SparkException(            "Cannot add data as BlockGenerator has not been started or has been stopped")
        }
      }
    } else {      throw new SparkException(        "Cannot add data as BlockGenerator has not been started or has been stopped")
    }
  }

这里的先调用waitToPush(),会有rateLimiter检查速率,防止加入过快,如果过快会block住等到下一秒再添加,一秒能添加的条数受spark.streaming.receiver.maxRate控制,即一个Receiver每秒能添加的条数。
检查完后会将数据添加到一个变长数组currentBuffer中。

另外,BlockGenerator被初始化的时候就创建了一个定时器:

private val blockIntervalMs = conf.getTimeAsMs("spark.streaming.blockInterval", "200ms")  require(blockIntervalMs > 0, s"'spark.streaming.blockInterval' should be a positive value")  private val blockIntervalTimer =    new RecurringTimer(clock, blockIntervalMs, updateCurrentBuffer, "BlockGenerator")

定时间隔默认200ms,可通过spark.streaming.blockInterval配置,每次定时执行的是updateCurrentBuffer方法:

private def updateCurrentBuffer(time: Long): Unit = {    try {      var newBlock: Block = null
      synchronized {        if (currentBuffer.nonEmpty) {
          val newBlockBuffer = currentBuffer
          currentBuffer = new ArrayBuffer[Any]
          val blockId = StreamBlockId(receiverId, time - blockIntervalMs)
          listener.onGenerateBlock(blockId)
          newBlock = new Block(blockId, newBlockBuffer)
        }
      }      if (newBlock != null) {
        blocksForPushing.put(newBlock)  // put is blocking when queue is full
      }
    } catch {      case ie: InterruptedException =>
        logInfo("Block updating timer thread was interrupted")      case e: Exception =>
        reportError("Error in block updating thread", e)
    }
  }
  • 将 currentBuffer 赋值给 newBlockBuffer

  • 重新为currentBuffer分配一个新对象,以供存储新的数据

  • 将currentBuffer封装为Block后添加到blocksForPushing中,blocksForPushing是一个默认长度为10的Queue,可通过spark.streaming.blockQueueSize配置

BlockGenerator初始化的时候还启动了一个线程来从blocksForPushing队列中取出Block通过supervisor来存储块的:

private val blockPushingThread = new Thread() { override def run() { keepPushingBlocks() } }

supervisor 存储数据块

先存储再向上报告:

#pushAndReportBlockval blockStoreResult = receivedBlockHandler.storeBlock(blockId, receivedBlock)
logDebug(s"Pushed block $blockId in ${(System.currentTimeMillis - time)} ms")
val numRecords = blockStoreResult.numRecords
val blockInfo = ReceivedBlockInfo(streamId, numRecords, metadataOption, blockStoreResult) trackerEndpoint.askWithRetry[Boolean](AddBlock(blockInfo))

存储数据块有对应的receivedBlockHandler,在启用了WAL(spark.streaming.receiver.writeAheadLog.enable为true)的情况下对应的是WriteAheadLogBasedBlockHandler(启用了WAL的情况下在应用程序挂掉后可以从WAL恢复数据),未启用的情况下对应的是BlockManagerBasedBlockHandler。

private val receivedBlockHandler: ReceivedBlockHandler = {    if (WriteAheadLogUtils.enableReceiverLog(env.conf)) {      if (checkpointDirOption.isEmpty) {        throw new SparkException(          "Cannot enable receiver write-ahead log without checkpoint directory set. " +            "Please use streamingContext.checkpoint() to set the checkpoint directory. " +            "See documentation for more details.")
      }      new WriteAheadLogBasedBlockHandler(env.blockManager, env.serializerManager, receiver.streamId,
        receiver.storageLevel, env.conf, hadoopConf, checkpointDirOption.get)
    } else {      new BlockManagerBasedBlockHandler(env.blockManager, receiver.storageLevel)
    }

storeBlock方法部分代码:

case ArrayBufferBlock(arrayBuffer) =>
    numRecords = Some(arrayBuffer.size.toLong)
    blockManager.putIterator(blockId, arrayBuffer.iterator, storageLevel,tellMaster = true)case IteratorBlock(iterator) =>
    val countIterator = new CountingIterator(iterator)
    val putResult = blockManager.putIterator(blockId, countIterator, storageLevel,tellMaster = true)
    numRecords = countIterator.count
    putResultcase ByteBufferBlock(byteBuffer) =>
    blockManager.putBytes(blockId, new ChunkedByteBuffer(byteBuffer.duplicate()), storageLevel, tellMaster = true)

两种handler都是通过blockManager来存储block到内存或者磁盘,存储的细节可见BlockManager 解析

通知 ReceiverTracker

存储了block后,接着创建了ReceivedBlockInfo实例,对应该block的一些信息,包括streamId(一个InputDStream对应一个Receiver,一个Receiver对应一个streamId)、block中数据的条数、storeResult等信息。

接着将receivedBlockInfo作为参数和ReceiverTracker通信发送AddBlock消息,ReceiverTracker收到消息后的处理如下:

 case AddBlock(receivedBlockInfo) =>        if (WriteAheadLogUtils.isBatchingEnabled(ssc.conf, isDriver = true)) {
          walBatchingThreadPool.execute(new Runnable {            override def run(): Unit = Utils.tryLogNonFatalError {              if (active) {
                context.reply(addBlock(receivedBlockInfo))
              } else {                throw new IllegalStateException("ReceiverTracker RpcEndpoint shut down.")
              }
            }
          })
        } else {
          context.reply(addBlock(receivedBlockInfo))
        }

都会调用addBlock(receivedBlockInfo)方法:

private def addBlock(receivedBlockInfo: ReceivedBlockInfo): Boolean = {
    receivedBlockTracker.addBlock(receivedBlockInfo)
  }

ReceiverTracker有个专门管理block的成员receivedBlockTracker,通过addBlock(receivedBlockInfo)来添加block信息:

def addBlock(receivedBlockInfo: ReceivedBlockInfo): Boolean = {
    try {
      val writeResult = writeToLog(BlockAdditionEvent(receivedBlockInfo))      if (writeResult) {
        synchronized {
          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
    }
  }

若启用WAL则会先将block信息以WAL保存,之后都会将block信息保存到streamIdToUnallocatedBlockQueuesmutable.HashMap[Int, ReceivedBlockQueue]中,其中key为InputDStream唯一id,value为已存储但未分配的block信息。之后为 batch 分配blocks,会访问该结构来获取每个 InputDStream 对应的未消费的 blocks。



作者:BIGUFO
链接:https://www.jianshu.com/p/a8c05c4c8130

点击查看更多内容
TA 点赞

若觉得本文不错,就分享一下吧!

评论

作者其他优质文章

正在加载中
JAVA开发工程师
手记
粉丝
39
获赞与收藏
123

关注作者,订阅最新文章

阅读免费教程

  • 推荐
  • 评论
  • 收藏
  • 共同学习,写下你的评论
感谢您的支持,我会继续努力的~
扫码打赏,你说多少就多少
赞赏金额会直接到老师账户
支付方式
打开微信扫一扫,即可进行扫码打赏哦
今天注册有机会得

100积分直接送

付费专栏免费学

大额优惠券免费领

立即参与 放弃机会
意见反馈 帮助中心 APP下载
官方微信

举报

0/150
提交
取消