Spark Streaming性能优化: 如何在生产环境下应对流数据峰值巨变

1、为什么引入Backpressure

默认情况下，Spark Streaming通过Receiver以生产者生产数据的速率接收数据，计算过程中会出现batch processing time > batch interval的情况，其中batch processing time 为实际计算一个批次花费时间， batch interval为Streaming应用设置的批处理间隔。这意味着Spark Streaming的数据接收速率高于Spark从队列中移除数据的速率，也就是数据处理能力低，在设置间隔内不能完全处理当前接收速率接收的数据。如果这种情况持续过长的时间，会造成数据在内存中堆积，导致Receiver所在Executor内存溢出等问题（如果设置StorageLevel包含disk, 则内存存放不下的数据会溢写至disk, 加大延迟）。Spark 1.5以前版本，用户如果要限制Receiver的数据接收速率，可以通过设置静态配制参数“spark.streaming.receiver.maxRate
”的值来实现，此举虽然可以通过限制接收速率，来适配当前的处理能力，防止内存溢出，但也会引入其它问题。比如：producer数据生产高于maxRate，当前集群处理能力也高于maxRate，这就会造成资源利用率下降等问题。为了更好的协调数据接收速率与资源处理能力，Spark Streaming 从v1.5开始引入反压机制（back-pressure）,通过动态控制数据接收速率来适配集群数据处理能力。
2、Backpressure
Spark Streaming Backpressure: 根据JobScheduler反馈作业的执行信息来动态调整Receiver数据接收率。通过属性“spark.streaming.backpressure.enabled”来控制是否启用backpressure机制，默认值false，即不启用。
2.1 Streaming架构如下图所示（详见Streaming数据接收过程文档和Streaming 源码解析）

2.2 BackPressure执行过程如下图所示:
　　在原架构的基础上加上一个新的组件RateController,这个组件负责监听“OnBatchCompleted”事件，然后从中抽取processingDelay 及schedulingDelay信息. Estimator依据这些信息估算出最大处理速度（rate），最后由基于Receiver的Input Stream将rate通过ReceiverTracker与ReceiverSupervisorImpl转发给BlockGenerator（继承自RateLimiter）.

3、BackPressure 源码解析
3.1 RateController类体系
RatenController 继承自StreamingListener. 用于处理BatchCompleted事件。核心代码为：

**
 * A StreamingListener that receives batch completion     updates, and maintains
 * an estimate of the speed at which this stream should ingest messages,
 * given an estimate computation from a `RateEstimator`
 */
private[streaming] abstract class RateController(val streamUID: Int, rateEstimator: RateEstimator)
extends StreamingListener with Serializable {
  /**
   * Compute the new rate limit and publish it asynchronously.
   */
  private def computeAndPublish(time: Long, elems: Long, workDelay: Long, waitDelay: Long): Unit =
Future[Unit] {
  val newRate = rateEstimator.compute(time, elems, workDelay, waitDelay)
  newRate.foreach { s =>
    rateLimit.set(s.toLong)
    publish(getLatestRate())
  }
}
def getLatestRate(): Long = rateLimit.get()

override def onBatchCompleted(batchCompleted: StreamingListenerBatchCompleted) {
val elements = batchCompleted.batchInfo.streamIdToInputInfo
for {
  processingEnd <- batchCompleted.batchInfo.processingEndTime
  workDelay <- batchCompleted.batchInfo.processingDelay
  waitDelay <- batchCompleted.batchInfo.schedulingDelay
  elems <- elements.get(streamUID).map(_.numRecords)
} computeAndPublish(processingEnd, elems, workDelay, waitDelay)
}
}

3.2 RateController的注册
JobScheduler启动时会抽取在DStreamGraph中注册的所有InputDstream中的rateController，并向ListenerBus注册监听. 此部分代码如下：

def start(): Unit = synchronized {
   if (eventLoop != null) return // scheduler has already been started

   logDebug("Starting JobScheduler")
   eventLoop = new EventLoop[JobSchedulerEvent]("JobScheduler") {
   override protected def onReceive(event: JobSchedulerEvent): Unit = processEvent(event)

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

 // attach rate controllers of input streams to receive batch completion updates
 for {
   inputDStream <- ssc.graph.getInputStreams
   rateController <- inputDStream.rateController
 } ssc.addStreamingListener(rateController)</span>

 listenerBus.start()
 receiverTracker = new ReceiverTracker(ssc)
 inputInfoTracker = new InputInfoTracker(ssc)
 receiverTracker.start()
 jobGenerator.start()
 logInfo("Started JobScheduler")
}

3.3 BackPressure执行过程分析
BackPressure 执行过程分为BatchCompleted事件触发时机和事件处理两个过程
3.3.1 BatchCompleted触发过程
对BatchedCompleted的分析，应该从JobGenerator入手，因为BatchedCompleted是批次处理结束的标志，也就是JobGenerator产生的作业执行完成时触发的，因此进行作业执行分析。
Streaming 应用中JobGenerator每个Batch Interval都会为应用中的每个Output Stream建立一个Job, 该批次中的所有Job组成一个Job Set.使用JobScheduler的submitJobSet进行批量Job提交。此部分代码结构如下所示

 /** 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)

  // Checkpoint all RDDs marked for checkpointing to ensure their lineages are
  // truncated periodically. Otherwise, we may run into stack overflows (SPARK-6847).
  ssc.sparkContext.setLocalProperty(RDD.CHECKPOINT_ALL_MARKED_ANCESTORS, "true")
  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)
}
eventLoop.post(DoCheckpoint(time, clearCheckpointDataLater = false))
}

其中，sumitJobSet会创建固定数量的后台线程（具体由“spark.streaming.concurrentJobs”指定），去处理Job Set中的Job. 具体实现逻辑为：

def submitJobSet(jobSet: JobSet) {
  if (jobSet.jobs.isEmpty) {
    logInfo("No jobs added for time " + jobSet.time)
  } else {
    listenerBus.post(StreamingListenerBatchSubmitted(jobSet.toBatchInfo))
    jobSets.put(jobSet.time, jobSet)
    jobSet.jobs.foreach(job => jobExecutor.execute(new JobHandler(job)))
    logInfo("Added jobs for time " + jobSet.time)
  }
}

其中JobHandler用于执行Job及处理Job执行结果信息。当Job执行完成时会产生JobCompleted事件. JobHandler的具体逻辑如下面代码所示：

当Job执行完成时，向eventLoop发送JobCompleted事件。EventLoop事件处理器接到JobCompleted事件后将调用handleJobCompletion 来处理Job完成事件。handleJobCompletion使用Job执行信息创建StreamingListenerBatchCompleted事件并通过StreamingListenerBus向监听器发送。实现如下：

private def handleJobCompletion(job: Job, completedTime: Long) {
   val jobSet = jobSets.get(job.time)
   jobSet.handleJobCompletion(job)
   job.setEndTime(completedTime)
   listenerBus.post(StreamingListenerOutputOperationCompleted(job.toOutputOperationInfo))
   logInfo("Finished job " + job.id + " from job set of time " + jobSet.time)
   if (jobSet.hasCompleted) {
     jobSets.remove(jobSet.time)
     jobGenerator.onBatchCompletion(jobSet.time)
     logInfo("Total delay: %.3f s for time %s (execution: %.3f s)".format(
     jobSet.totalDelay / 1000.0, jobSet.time.toString,
     jobSet.processingDelay / 1000.0
   ))
 listenerBus.post(StreamingListenerBatchCompleted(jobSet.toBatchInfo))
 }
 job.result match {
   case Failure(e) =>
       reportError("Error running job " + job, e)
   case _ =>
 }
}

3.3.2、BatchCompleted事件处理过程
StreamingListenerBus将事件转交给具体的StreamingListener，因此BatchCompleted将交由RateController进行处理。RateController接到BatchCompleted事件后将调用onBatchCompleted对事件进行处理。

override def onBatchCompleted(batchCompleted: StreamingListenerBatchCompleted) {
  val elements = batchCompleted.batchInfo.streamIdToInputInfo

  for {
    processingEnd <- batchCompleted.batchInfo.processingEndTime
    workDelay <- batchCompleted.batchInfo.processingDelay
    waitDelay <- batchCompleted.batchInfo.schedulingDelay
    elems <- elements.get(streamUID).map(_.numRecords)
  } computeAndPublish(processingEnd, elems, workDelay, waitDelay)
}

onBatchCompleted会从完成的任务中抽取任务的执行延迟和调度延迟，然后用这两个参数用RateEstimator（目前存在唯一实现PIDRateEstimator，proportional-integral-derivative (PID) controller， PID控制器）估算出新的rate并发布。代码如下：

/**
   * Compute the new rate limit and publish it asynchronously.
   */
  private def computeAndPublish(time: Long, elems: Long, workDelay: Long, waitDelay: Long): Unit =
Future[Unit] {
  val newRate = rateEstimator.compute(time, elems, workDelay, waitDelay)
  newRate.foreach { s =>
    rateLimit.set(s.toLong)
    publish(getLatestRate())
  }
}

其中publish()由RateController的子类ReceiverRateController来定义。具体逻辑如下（ReceiverInputDStream中定义）：

/**
   * A RateController that sends the new rate to receivers, via the receiver tracker.
   */
 private[streaming] class ReceiverRateController(id: Int, estimator: RateEstimator)
  extends RateController(id, estimator) {
  override def publish(rate: Long): Unit =
    ssc.scheduler.receiverTracker.sendRateUpdate(id, rate)
}

publish的功能为新生成的rate 借助ReceiverTracker进行转发。ReceiverTracker将rate包装成UpdateReceiverRateLimit事交ReceiverTrackerEndpoint

/** Update a receiver's maximum ingestion rate */
def sendRateUpdate(streamUID: Int, newRate: Long):   Unit = synchronized {
  if (isTrackerStarted) {
    endpoint.send(UpdateReceiverRateLimit(streamUID, newRate))
  }
}

ReceiverTrackerEndpoint接到消息后，其将会从receiverTrackingInfos列表中获取Receiver注册时使用的endpoint(实为ReceiverSupervisorImpl)，再将rate包装成UpdateLimit发送至endpoint.其接到信息后，使用updateRate更新BlockGenerators(RateLimiter子类),来计算出一个固定的令牌间隔。

其中RateLimiter的updateRate实现如下：

/**
  * Set the rate limit to `newRate`. The new rate will not exceed the maximum rate configured by
  * {{{spark.streaming.receiver.maxRate}}}, even if `newRate` is higher than that.
  *
  * @param newRate A new rate in events per second. It has no effect if it's 0 or negative.
  */
 private[receiver] def updateRate(newRate: Long): Unit =
   if (newRate > 0) {
   if (maxRateLimit > 0) {
     rateLimiter.setRate(newRate.min(maxRateLimit))
   } else {
     rateLimiter.setRate(newRate)
   }
 }

setRate的实现如下：

public final void setRate(double permitsPerSecond) {
  Preconditions.checkArgument(permitsPerSecond > 0.0
    && !Double.isNaN(permitsPerSecond), "rate must be positive");
  synchronized (mutex) {
    resync(readSafeMicros());
    double stableIntervalMicros = TimeUnit.SECONDS.toMicros(1L) / permitsPerSecond;  //固定间隔
    this.stableIntervalMicros = stableIntervalMicros;
    doSetRate(permitsPerSecond, stableIntervalMicros);
  }
}

到此，backpressure反压机制调整rate结束。

4．流量控制点
　　当Receiver开始接收数据时，会通过supervisor.pushSingle()方法将接收的数据存入currentBuffer等待BlockGenerator定时将数据取走，包装成block. 在将数据存放入currentBuffer之时，要获取许可（令牌）。如果获取到许可就可以将数据存入buffer, 否则将被阻塞，进而阻塞Receiver从数据源拉取数据。

  /**
   * Push a single data item into the buffer.
   */
  def addData(data: Any): Unit = {
      if (state == Active) {
         waitToPush()  //获取令牌
        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")
}

其令牌投放采用令牌桶机制进行，原理如下图所示:

令牌桶机制：大小固定的令牌桶可自行以恒定的速率源源不断地产生令牌。如果令牌不被消耗，或者被消耗的速度小于产生的速度，令牌就会不断地增多，直到把桶填满。后面再产生的令牌就会从桶中溢出。最后桶中可以保存的最大令牌数永远不会超过桶的大小。当进行某操作时需要令牌时会从令牌桶中取出相应的令牌数，如果获取到则继续操作，否则阻塞。用完之后不用放回。
　　Streaming 数据流被Receiver接收后，按行解析后存入iterator中。然后逐个存入Buffer，在存入buffer时会先获取token，如果没有token存在，则阻塞；如果获取到则将数据存入buffer. 然后等价后续生成block操作。