Spark分析之DAGScheduler

val sc = new SparkContext("local[2]", "WordCount", System.getenv("SPARK_HOME"), Seq(System.getenv("SPARK_TEST_JAR")))
val textFile = sc.textFile("xxx")
val result = textFile.flatMap(line => line.split("\t")).map(word => (word, 1)).reduceByKey(_ + _)
result.collect

def receive = {
    case JobSubmitted(jobId, rdd, func, partitions, allowLocal...) =>
      dagScheduler.handleJobSubmitted(jobId, rdd, func, partitions, allowLocal...)
}


//完成job到stage的转换，生成finalStage并提交
private[scheduler] def handleJobSubmitted(jobId: Int,
      finalRDD: RDD[_],
      func: (TaskContext, Iterator[_]) => _,
      partitions: Array[Int],
      allowLocal: Boolean...){
     //注意：该RDD是final RDD，而不是一系列的RDD，用finalRDD来创建finalStage
     //newStage操作对应会生成新的result stage或者shuffle stage：内部有一个isShuffleMap变量来标识该stage是shuffle or result
     var finalStage: Stage = newStage(rdd, partitions.size, None, jobId, Some(callSite))
 
    //使用finalStage来构建job
    val job = new ActiveJob(jobId, finalStage, func, partitions, callSite, listener, properties)

    //对于简单的job，没有依赖关系并且只有一个partition，该类job会使用local thread处理而并非提交到TaskScheduler上处理
    if (allowLocal && finalStage.parents.size == 0 && partitions.length == 1) {
        runLocally(job)
    } else {
        submitStage(finalStage)
    }
}

private def newStage(
      rdd: RDD[_],  numTasks: Int,     //task个数就是partitions个数
      shuffleDep: Option[ShuffleDependency[_,_]],
      jobId: Int, callSite: Option[String] = None) : Stage = {
    val id = nextStageId.getAndIncrement() 
    val stage = new Stage(id, rdd, numTasks, shuffleDep, getParentStages(rdd, jobId), jobId, callSite)
   ......
}


private def getParentStages(rdd: RDD[_], jobId: Int): List[Stage] = {
    val parents = new HashSet[Stage]
    val visited = new HashSet[RDD[_]]
    def visit(r: RDD[_]) {
      if (!visited(r)) {
        visited += r
        for (dep <- r.dependencies) {
          dep match {
            case shufDep: ShuffleDependency[_,_] =>
              parents += getShuffleMapStage(shufDep, jobId)
            case _ =>
              visit(dep.rdd)
          }
        }
      }
    }
    visit(rdd)
    parents.toList
}

private def getShuffleMapStage(shuffleDep: ShuffleDependency[_,_], jobId: Int): Stage = {
    shuffleToMapStage.get(shuffleDep.shuffleId) match {
      case Some(stage) => stage
      case None =>
        val stage =
          newOrUsedStage(shuffleDep.rdd, shuffleDep.rdd.partitions.size, shuffleDep, jobId)
        shuffleToMapStage(shuffleDep.shuffleId) = stage
        stage
    }

private def submitStage(stage: Stage) { 
 if (!waiting(stage) && !running(stage) && !failed(stage)) {
  val missing = getMissingParentStages(stage).sortBy(_.id)  //根据final stage发现是否有parent stage
  if (missing == Nil) { // 如果计算中发现当前的stage没有任何依赖或者所有的依赖都已经准备完毕，则提交task
   submitMissingTasks(stage, jobId.get)
   running += stage //设置当前的stage为running，因为当前的stage没有未处理完的依赖的stage
  } else { //如果有parent stage,需要先submit parent, 因为stage之间需要顺序执行
   for (parent <- missing) {
    submitStage(parent)
   }
   waiting += stage   //当前stage放入到waiting列表中，表示该stage需要等待parent先执行完成
  }
 }
}

//根据final stage的parents找出所有的parent stage
private def getMissingParentStages(stage: Stage): List[Stage] = { 
 ......
 dep match {
  //如果是ShuffleDependency，则新建一个shuffle map stage，且该stage是可用的话则加入missing中
  case shufDep: ShuffleDependency[_,_] =>  //ShuffleDependecy
   val mapStage = getShuffleMapStage(shufDep, stage.jobId)
   if (!mapStage.isAvailable) {
    missing += mapStage
   }
  case narrowDep: NarrowDependency[_] =>  //NarrowDependecy
   visit(narrowDep.rdd)
 }
}

private def submitMissingTasks(stage: Stage, jobId: Int) {
 //首先根据stage所依赖的RDD的partition的分布，会产生出与partition数量相等的task
 var tasks = ArrayBuffer[Task[_]]()

 //对于finalStage或是mapStage会产生不同的task。
 //检查该stage时是否ShuffleMap，如果是则生成ShuffleMapTask
 if (stage.isShuffleMap) { //mapStage：表示还有其他stage依赖此stage
  for (p <- 0 until stage.numPartitions if stage.outputLocs(p) == Nil) {
   //task根据partition的locality进行分布
   val locs = getPreferredLocs(stage.rdd, p)
   tasks += new ShuffleMapTask(stage.id, stage.rdd, stage.shuffleDep.get, p, locs)
  }
 } else { //finalStage：该类型stage直接输出结果生成ResultTask
  val job = resultStageToJob(stage)
  for (id <- 0 until job.numPartitions if !job.finished(id)) {
   val partition = job.partitions(id)
   val locs = getPreferredLocs(stage.rdd, partition)
   //由于是ResultTask，因此需要传入定义的func，也就是如果处理结果返回
   tasks += new ResultTask(stage.id, stage.rdd, job.func, partition, locs, id)
  }
 }
 //向TaskSchuduler提交任务，以stage为单位，一个stage对应一个TaskSet
 taskSched.submitTasks(new TaskSet(tasks.toArray, stage.id, stage.newAttemptId(), stage.jobId, properties))
}