第6课：Spark Streaming源码解读之Job动态生成和深度思考

Posted 2020-07-07 snail_gesture

一：Spark Streaming Job生成深度思考
1. 做大数据例如Hadoop,Spark等，如果不是流处理的话，一般会有定时任务。例如10分钟触发一次，1个小时触发一次，这就是做流处理的感觉，一切不是流处理，或者与流处理无关的数据都将是没有价值的数据，以前做批处理的时候其实也是隐形的在做流处理。
2. JobGenerator构造的时候有一个核心的参数是jobScheduler, jobScheduler是整个作业的生成和提交给集群的核心，JobGenerator会基于DStream生成Job。这里面的Job就相当于Java中线程要处理的Runnable里面的业务逻辑封装。Spark的Job就是运行的一个作业。
3. Spark Streaming除了基于定时操作以外参数Job，还可以通过各种聚合操作，或者基于状态的操作。
4. 每5秒钟JobGenerator都会产生Job，此时的Job是逻辑级别的，也就是说有这个Job，并且说这个Job具体该怎么去做，此时并没有执行。具体执行的话是交给底层的RDD的action去触发，此时的action也是逻辑级别的。底层物理级别的，Spark Streaming他是基于DStream构建的依赖关系导致的Job是逻辑级别的，底层是基于RDD的逻辑级别的。

val ssc = new StreamingContext(conf, Seconds(5))

 5. Spark Streaming的触发器是以时间为单位的，storm是以事件为触发器，也就是基于一个又一个record. Spark Streaming基于时间，这个时间是Batch Duractions

从逻辑级别翻译成物理级别，最后一个操作肯定是RDD的action，但是并不想一翻译立马就触发job。这个时候怎么办？
6. action触发作业，这个时候作为Runnable接口封装，他会定义一个方法，这个方法里面是基于DStream的依赖关系生成的RDD。翻译的时候是将DStream的依赖关系翻译成RDD的依赖关系，由于DStream的依赖关系最后一个是action级别的，翻译成RDD的时候，RDD的最后一个操作也应该是action级别的，如果翻译的时候直接执行的话，就直接生成了Job，就没有所谓的队列，所以会将翻译的事件放到一个函数中或者一个方法中，因此，如果这个函数没有指定的action触发作业是执行不了的。
7. Spark Streaming根据时间不断的去管理我们的生成的作业，所以这个时候我们每个作业又有action级别的操作，这个action操作是对DStream进行逻辑级别的操作，他生成每个Job放到队列的时候，他一定会被翻译为RDD的操作，那基于RDD操作的最后一个一定是action级别的，如果翻译的话直接就是触发action的话整个Spark Streaming的Job就不受管理了。因此我们既要保证他的翻译，又要保证对他的管理，把DStream之间的依赖关系转变为RDD之间的依赖关系，最后一个DStream使得action的操作，翻译成一个RDD之间的action操作，整个翻译后的内容他是一块内容，他这一块内容是放在一个函数体中的，这个函数体，他会函数的定义，这个函数由于他只是定义还没有执行，所以他里面的RDD的action不会执行，不会触发Job,当我们的JobScheduler要调度Job的时候，转过来在线程池中拿出一条线程执行刚才的封装的方法。

二：Spark Streaming Job生成源码解析
Spark 作业动态生成三大核心：
JobGenerator: 负责Job生成。
JobSheduler： 负责Job调度。
ReceiverTracker: 获取元数据。
1. JobScheduler的start方法被调用的时候，会启动JobGenerator的start方法。

/** Start generation of jobs */
def start(): Unit = synchronized {
//eventLoop是消息循环体，因为不断的生成Job
  if (eventLoop != null) return // generator has already been started

  // Call checkpointWriter here to initialize it before eventLoop uses it to avoid a deadlock.
  // See SPARK-10125
  checkpointWriter
//匿名内部类
  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)
    }
  }
//调用start方法。
  eventLoop.start()

  if (ssc.isCheckpointPresent) {
    restart()
  } else {
    startFirstTime()
  }
}

EvenLoop: 的start方法被调用，首先会调用onstart方法。然后就启动线程。

/**
 * An event loop to receive events from the caller and process all events in the event thread. It
 * will start an exclusive event thread to process all events.
 *
 * Note: The event queue will grow indefinitely. So subclasses should make sure `onReceive` can
 * handle events in time to avoid the potential OOM.
 */
private[spark] abstract class EventLoop[E](name: String) extends Logging {

  private val eventQueue: BlockingQueue[E] = new LinkedBlockingDeque[E]()

  private val stopped = new AtomicBoolean(false)
//开启后台线程。   
  private val eventThread = new Thread(name) {
    setDaemon(true)

    override def run(): Unit = {
      try {
//不断的从BlockQueue中拿消息。
        while (!stopped.get) {
//线程的start方法调用就会不断的循环队列，而我们将消息放到eventQueue中。
          val event = eventQueue.take()
          try {
//
            onReceive(event)
          } catch {
            case NonFatal(e) => {
              try {
                onError(e)
              } catch {
                case NonFatal(e) => logError("Unexpected error in " + name, e)
              }
            }
          }
        }
      } catch {
        case ie: InterruptedException => // exit even if eventQueue is not empty
        case NonFatal(e) => logError("Unexpected error in " + name, e)
      }
    }

  }

  def start(): Unit = {
    if (stopped.get) {
      throw new IllegalStateException(name + " has already been stopped")
    }
    // Call onStart before starting the event thread to make sure it happens before onReceive

    onStart()
    eventThread.start()
  }

onReceive:不断的从消息队列中获得消息，一旦获得消息就会处理。
不要在onReceive中添加阻塞的消息，如果这样的话会不断的阻塞消息。
消息循环器一般都不会处理具体的业务逻辑，一般消息循环器发现消息以后都会将消息路由给其他的线程去处理。

/**
 * Invoked in the event thread when polling events from the event queue.
 *
 * Note: Should avoid calling blocking actions in `onReceive`, or the event thread will be blocked
 * and cannot process events in time. If you want to call some blocking actions, run them in
 * another thread.
 */
protected def onReceive(event: E): Unit

消息队列接收到事件后具体处理如下：

/** 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)
  }
}

基于Batch Duractions生成Job，并完成checkpoint.
Job生成的5个步骤。

/** 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 {
//第一步：获取当前时间段里面的数据。根据分配的时间来分配具体要处理的数据。
    jobScheduler.receiverTracker.allocateBlocksToBatch(time) // allocate received blocks to batch
//第二步：生成Job，获取RDD的DAG依赖关系。在此基于DStream生成了RDD实例。
    graph.generateJobs(time) // generate jobs using allocated block
  } match {
    case Success(jobs) =>
//第三步：获取streamIdToInputInfos的信息。BacthDuractions要处理的数据，以及我们要处理的业务逻辑。
      val streamIdToInputInfos = jobScheduler.inputInfoTracker.getInfo(time)
//第四步：将生成的Job交给jobScheduler
      jobScheduler.submitJobSet(JobSet(time, jobs, streamIdToInputInfos))
    case Failure(e) =>
      jobScheduler.reportError("Error generating jobs for time " + time, e)
  }
//第五步：进行checkpoint
  eventLoop.post(DoCheckpoint(time, clearCheckpointDataLater = false))
}

此时的outputStream是整个DStream中的最后一个DStream，也就是foreachDStream.

def generateJobs(time: Time): Seq[Job] = {
  logDebug("Generating jobs for time " + time)
  val jobs = this.synchronized {
    outputStreams.flatMap { outputStream =>
//根据最后一个DStream，然后根据时间生成Job.
      val jobOption = outputStream.generateJob(time)
      jobOption.foreach(_.setCallSite(outputStream.creationSite))
      jobOption
    }
  }
  logDebug("Generated " + jobs.length + " jobs for time " + time)
  jobs
}

此时的JobFunc就是我们前面提到的用函数封装了Job。
generateJob基于给定的时间生成Spark Streaming 的Job，这个方法会基于我们的DStream的操作物化成了RDD，由此可以看出，DStream是逻辑级别的，RDD是物理级别的。

/**
 * Generate a SparkStreaming job for the given time. This is an internal method that
 * should not be called directly. This default implementation creates a job
 * that materializes the corresponding RDD. Subclasses of DStream may override this
 * to generate their own jobs.
 */
private[streaming] def generateJob(time: Time): Option[Job] = {
  getOrCompute(time) match {
    case Some(rdd) => {
      val jobFunc = () => {
        val emptyFunc = { (iterator: Iterator[T]) => {} }
//rdd => 就是RDD的依赖关系
        context.sparkContext.runJob(rdd, emptyFunc)
      }
//此时的
      Some(new Job(time, jobFunc))
    }
    case None => None
  }
}

Job这个类就代表了Spark业务逻辑，可能包含很多Spark Jobs.

/**
 * Class representing a Spark computation. It may contain multiple Spark jobs.
 */
private[streaming]
class Job(val time: Time, func: () => _) {
  private var _id: String = _
  private var _outputOpId: Int = _
  private var isSet = false
  private var _result: Try[_] = null
  private var _callSite: CallSite = null
  private var _startTime: Option[Long] = None
  private var _endTime: Option[Long] = None

  def run() {
//调用func函数，此时这个func就是我们前面generateJob中的func
    _result = Try(func())
  }

此时put函数中的RDD是最后一个RDD，虽然触发Job是基于时间，但是也是基于DStream的action的。

/**
 * Get the RDD corresponding to the given time; either retrieve it from cache
 * or compute-and-cache it.
 */
private[streaming] final def getOrCompute(time: Time): Option[RDD[T]] = {
  // If RDD was already generated, then retrieve it from HashMap,
  // or else compute the RDD
//基于时间生成RDD
  generatedRDDs.get(time).orElse {
    // Compute the RDD if time is valid (e.g. correct time in a sliding window)
    // of RDD generation, else generate nothing.
    if (isTimeValid(time)) {

      val rddOption = createRDDWithLocalProperties(time, displayInnerRDDOps = false) {
        // Disable checks for existing output directories in jobs launched by the streaming
        // scheduler, since we may need to write output to an existing directory during checkpoint
        // recovery; see SPARK-4835 for more details. We need to have this call here because
        // compute() might cause Spark jobs to be launched.
        PairRDDFunctions.disableOutputSpecValidation.withValue(true) {
//
          compute(time)
        }
      }
//然后对generated RDD进行checkpoint
      rddOption.foreach { case newRDD =>
        // Register the generated RDD for caching and checkpointing
        if (storageLevel != StorageLevel.NONE) {
          newRDD.persist(storageLevel)
          logDebug(s"Persisting RDD ${newRDD.id} for time $time to $storageLevel")
        }
        if (checkpointDuration != null && (time - zeroTime).isMultipleOf(checkpointDuration)) {
          newRDD.checkpoint()
          logInfo(s"Marking RDD ${newRDD.id} for time $time for checkpointing")
        }
//以时间为Key,RDD为Value,此时的RDD为最后一个RDD
        generatedRDDs.put(time, newRDD)
      }
      rddOption
    } else {
      None
    }
  }
}

回到JobGenerator中的start方法。

  if (ssc.isCheckpointPresent) {
//如果不是第一次启动的话，就需要从checkpoint中恢复。
    restart()
  } else {
//否则的话，就是第一次启动。
    startFirstTime()
  }
}

StartFirstTime的源码如下：

/** Starts the generator for the first time */
private def startFirstTime() {
  val startTime = new Time(timer.getStartTime())
//告诉DStreamGraph第一个Batch启动时间。
  graph.start(startTime - graph.batchDuration)
//timer启动，整个job不断生成就开始了。
  timer.start(startTime.milliseconds)
  logInfo("Started JobGenerator at " + startTime)
}

这里的timer是RecurringTimer。RecurringTimer的start方法会启动内置线程thread.

private val timer = new RecurringTimer(clock, ssc.graph.batchDuration.milliseconds,
  longTime => eventLoop.post(GenerateJobs(new Time(longTime))), "JobGenerator")

Timer.start源码如下：

/**
 * Start at the given start time.
 */
def start(startTime: Long): Long = synchronized {
  nextTime = startTime //每次调用的
  thread.start()
  logInfo("Started timer for " + name + " at time " + nextTime)
  nextTime
}

调用thread启动后台进程。

private val thread = new Thread("RecurringTimer - " + name) {
  setDaemon(true)
  override def run() { loop }
}

loop源码如下：

  /**
   * Repeatedly call the callback every interval.
   */
  private def loop() {
    try {
      while (!stopped) {
        triggerActionForNextInterval()
      }
      triggerActionForNextInterval()
    } catch {
      case e: InterruptedException =>
    }
  }
}

tiggerActionForNextInterval源码如下：

private def triggerActionForNextInterval(): Unit = {
  clock.waitTillTime(nextTime)
  callback(nextTime)
  prevTime = nextTime
  += period
  logDebug("Callback for " + name + " called at time " + prevTime)
}

此时的callBack是RecurringTimer传入的。下面就去找callBack是谁传入的，这个时候就应该找RecurringTimer什么时候实例化的。

private[streaming]
class RecurringTimer(clock: Clock, period: Long, callback: (Long) => Unit, name: String)
  extends Logging {

  private val thread = new Thread("RecurringTimer - " + name) {
    setDaemon(true)
    override def run() { loop }
  }

在jobGenerator中，匿名函数会随着时间不断的推移反复被调用。

private val timer = new RecurringTimer(clock, ssc.graph.batchDuration.milliseconds,
//匿名函数，复制给callback。
  longTime => eventLoop.post(GenerateJobs(new Time(longTime))), "JobGenerator")

而此时的eventLoop就是JobGenerator的start方法中eventLoop.eventLoop是一个消息循环体当收到generateJobs，就会将消息放到线程池中去执行。
至此，就知道了基于时间怎么生成作业的流程就贯通了。

Jobs: 此时的jobs就是jobs的业务逻辑，就类似于RDD之间的依赖关系，保存最后一个job，然后根据依赖关系进行回溯。
streamIdToInputInfos：基于Batch Duractions以及要处理的业务逻辑，然后就生成了JobSet.

jobScheduler.submitJobSet(JobSet(time, jobs, streamIdToInputInfos))

此时的JobSet就包含了数据以及对数据处理的业务逻辑。

/** Class representing a set of Jobs
  * belong to the same batch.
  */
private[streaming]
case class JobSet(
    time: Time,
    jobs: Seq[Job],
    streamIdToInputInfo: Map[Int, StreamInputInfo] = Map.empty) {

  private val incompleteJobs = new HashSet[Job]()
  private val submissionTime = System.currentTimeMillis() // when this jobset was submitted
  private var processingStartTime = -1L // when the first job of this jobset started processing
  private var processingEndTime = -1L // when the last job of this jobset finished processing

  jobs.zipWithIndex.foreach { case (job, i) => job.setOutputOpId(i) }
  incompleteJobs ++= jobs

  def handleJobStart(job: Job) {
    if (processingStartTime < 0) processingStartTime = System.currentTimeMillis()
  }

submitJobSet:

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)
//jobHandler
    jobSet.jobs.foreach(job => jobExecutor.execute(new JobHandler(job)))
    logInfo("Added jobs for time " + jobSet.time)
  }
}

JobHandle是一个Runnable接口，Job就是我们业务逻辑，代表的就是一系列RDD的依赖关系,job.run方法就导致了func函数的调用。

  private class JobHandler(job: Job) extends Runnable with Logging {
    import JobScheduler._

    def run() {
      try {
        val formattedTime = UIUtils.formatBatchTime(
          job.time.milliseconds, ssc.graph.batchDuration.milliseconds, showYYYYMMSS = false)
        val batchUrl = s"/streaming/batch/?id=${job.time.milliseconds}"
        val batchLinkText = s"[output operation ${job.outputOpId}, batch time ${formattedTime}]"

        ssc.sc.setJobDescription(
          s"""Streaming job from <a href="$batchUrl">$batchLinkText</a>""")
        ssc.sc.setLocalProperty(BATCH_TIME_PROPERTY_KEY, job.time.milliseconds.toString)
        ssc.sc.setLocalProperty(OUTPUT_OP_ID_PROPERTY_KEY, job.outputOpId.toString)

        // We need to assign `eventLoop` to a temp variable. Otherwise, because
        // `JobScheduler.stop(false)` may set `eventLoop` to null when this method is running, then
        // it's possible that when `post` is called, `eventLoop` happens to null.
        var _eventLoop = eventLoop
        if (_eventLoop != null) {
          _eventLoop.post(JobStarted(job, clock.getTimeMillis()))
          // Disable checks for existing output directories in jobs launched by the streaming
          // scheduler, since we may need to write output to an existing directory during checkpoint
          // recovery; see SPARK-4835 for more details.
          PairRDDFunctions.disableOutputSpecValidation.withValue(true) {
//
            job.run()
          }
          _eventLoop = eventLoop
          if (_eventLoop != null) {
            _eventLoop.post(JobCompleted(job, clock.getTimeMillis()))
          }
        } else {
          // JobScheduler has been stopped.
        }
      } finally {
        ssc.sc.setLocalProperty(JobScheduler.BATCH_TIME_PROPERTY_KEY, null)
        ssc.sc.setLocalProperty(JobScheduler.OUTPUT_OP_ID_PROPERTY_KEY, null)
      }
    }
  }
}

此时的func就是基于DStream的业务逻辑。也就是RDD之间依赖的业务逻辑。

def run() {
  _result = Try(func())
}

总体架构如下：