Spark Streaming源码分析 – DStream

A Discretized Stream (DStream), the basic abstraction in Spark Streaming, is a continuous sequence of RDDs (of the same type) representing a continuous stream of data.
Dstream本质就是离散化的stream，将stream离散化成一组RDD的list，所以基本的操作仍然是以RDD为基础
下面看到DStream的基本定义，对于普通的RDD而言，时间对于DStream是更为重要的因素
将stream切分成RDD的interval时间，stream开始的时间，DStream需要保留的RDD的时间，每个RDD所对于的时间key……

DStream抽象定义

/**
 * A Discretized Stream (DStream), the basic abstraction in Spark Streaming, is a continuous
 * sequence of RDDs (of the same type) representing a continuous stream of data (see
 * org.apache.spark.rdd.RDD in the Spark core documentation for more details on RDDs).
 * DStreams can either be created from live data (such as, data from TCP sockets, Kafka, Flume,
 * etc.) using a [[org.apache.spark.streaming.StreamingContext]] or it can be generated by
 * transforming existing DStreams using operations such as `map`,
 * `window` and `reduceByKeyAndWindow`. While a Spark Streaming program is running, each DStream
 * periodically generates a RDD, either from live data or by transforming the RDD generated by a
 * parent DStream.
 *
 * This class contains the basic operations available on all DStreams, such as `map`, `filter` and
 * `window`. In addition, [[org.apache.spark.streaming.dstream.PairDStreamFunctions]] contains
 * operations available only on DStreams of key-value pairs, such as `groupByKeyAndWindow` and
 * `join`. These operations are automatically available on any DStream of pairs
 * (e.g., DStream[(Int, Int)] through implicit conversions when
 * `org.apache.spark.streaming.StreamingContext._` is imported.
 *
 * DStreams internally is characterized by a few basic properties:
 *  - A list of other DStreams that the DStream depends on
 *  - A time interval at which the DStream generates an RDD
 *  - A function that is used to generate an RDD after each time interval
 */

abstract class DStream[T: ClassTag] (
    @transient private[streaming] var ssc: StreamingContext
  ) extends Serializable with Logging {

  // =======================================================================
  // Methods that should be implemented by subclasses of DStream
  // =======================================================================

  /** Time interval after which the DStream generates a RDD */
  def slideDuration: Duration   // 将stream切分成RDD的interval

  /** List of parent DStreams on which this DStream depends on */
  def dependencies: List[DStream[_]] // 和RDD一样，DStream之间也存在dependency关系

  /** Method that generates a RDD for the given time */
  def compute (validTime: Time): Option[RDD[T]] // RDD的生成逻辑

  // =======================================================================
  // Methods and fields available on all DStreams
  // =======================================================================

  // RDDs generated, marked as private[streaming] so that testsuites can access it
  @transient
  private[streaming] var generatedRDDs = new HashMap[Time, RDD[T]] () // 最为核心的结构，可以看到DStream就是以time为key的RDD的hashmap

  // Time zero for the DStream
  private[streaming] var zeroTime: Time = null // Stream开始的时间

  // Duration for which the DStream will remember each RDD created
  private[streaming] var rememberDuration: Duration = null // Stream是无限的，而在DStream不可能保留所有的RDD，所以设置DStream需要remember的duration

  // Storage level of the RDDs in the stream
  private[streaming] var storageLevel: StorageLevel = StorageLevel.NONE

  // Checkpoint details
  private[streaming] val mustCheckpoint = false
  private[streaming] var checkpointDuration: Duration = null
  private[streaming] val checkpointData = new DStreamCheckpointData(this)

  // Reference to whole DStream graph
  private[streaming] var graph: DStreamGraph = null // DStreamGraph

  // Duration for which the DStream requires its parent DStream to remember each RDD created
  private[streaming] def parentRememberDuration = rememberDuration

  /** Return the StreamingContext associated with this DStream */
  def context = ssc 

  /** Persist the RDDs of this DStream with the given storage level */
  def persist(level: StorageLevel): DStream[T] = {
    this.storageLevel = level
    this
  }

  /** Persist RDDs of this DStream with the default storage level (MEMORY_ONLY_SER) */
  def persist(): DStream[T] = persist(StorageLevel.MEMORY_ONLY_SER)

  /** Persist RDDs of this DStream with the default storage level (MEMORY_ONLY_SER) */
  def cache(): DStream[T] = persist()

  /**
   * Enable periodic checkpointing of RDDs of this DStream
   * @param interval Time interval after which generated RDD will be checkpointed
   */
  def checkpoint(interval: Duration): DStream[T] = {
    persist()
    checkpointDuration = interval
    this
  }
}

getOrCompute
注意的是，这里是产生RDD对象，而不是真正的进行计算，只有在runjob时才会做真正的计算
Spark RDD本身是不包含具体数据的，只是定义了workflow（依赖关系），处理逻辑

  /**
   * Retrieve a precomputed RDD of this DStream, or computes the RDD. This is an internal
   * method that should not be called directly.
   */
  private[streaming] def getOrCompute(time: Time): Option[RDD[T]] = {
    // If this DStream was not initialized (i.e., zeroTime not set), then do it
    // If RDD was already generated, then retrieve it from HashMap
    generatedRDDs.get(time) match {

      // If an RDD was already generated and is being reused, then
      // probably all RDDs in this DStream will be reused and hence should be cached
      case Some(oldRDD) => Some(oldRDD)

      // if RDD was not generated, and if the time is valid
      // (based on sliding time of this DStream), then generate the RDD
      case None => { // 需要compute
        if (isTimeValid(time)) { // invalid的定义，(time <= zeroTime || ! (time - zeroTime).isMultipleOf(slideDuration)
          compute(time) match { // 使用compute生成RDD对象
            case Some(newRDD) =>
              if (storageLevel != StorageLevel.NONE) {
                newRDD.persist(storageLevel) // 设置persist level
              }
              if (checkpointDuration != null &&
                (time - zeroTime).isMultipleOf(checkpointDuration)) {
                newRDD.checkpoint() // 设置checkpoint
              }
              generatedRDDs.put(time, newRDD) // 将产生的RDD对象放入generatedRDDs
              Some(newRDD)
            case None =>
              None
          }
        } else {
          None
        }
      }
    }
  }

generateJob
对于用getOrCompute产生的RDD对象，需要封装成job
而Job的关键，jobFunc，其实就是想Spark集群提交一个job
这里只是使用了emptyFunc，具体的output逻辑是需要被具体的outputDStream改写的

  /**
   * 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]) => {} }
          context.sparkContext.runJob(rdd, emptyFunc)
        }
        Some(new Job(time, jobFunc))
      }
      case None => None
    }
  }

clearMetadata
清除过时的RDD对象，其中还会做unpersist，以及调用dependencies的clearMetadata

  /**
   * Clear metadata that are older than `rememberDuration` of this DStream.
   * This is an internal method that should not be called directly. This default
   * implementation clears the old generated RDDs. Subclasses of DStream may override
   * this to clear their own metadata along with the generated RDDs.
   */
  private[streaming] def clearMetadata(time: Time) {
    val oldRDDs = generatedRDDs.filter(_._1 <= (time - rememberDuration))
    generatedRDDs --= oldRDDs.keys
    if (ssc.conf.getBoolean("spark.streaming.unpersist", false)) {
      oldRDDs.values.foreach(_.unpersist(false))
    }
    dependencies.foreach(_.clearMetadata(time))
  }

具体DStream的定义

FilteredDStream

package org.apache.spark.streaming.dstream

private[streaming]
class FilteredDStream[T: ClassTag](
    parent: DStream[T],
    filterFunc: T => Boolean
  ) extends DStream[T](parent.ssc) {

  override def dependencies = List(parent)

  override def slideDuration: Duration = parent.slideDuration

  override def compute(validTime: Time): Option[RDD[T]] = {
    parent.getOrCompute(validTime).map(_.filter(filterFunc))
  }
}

WindowedDStream

private[streaming]
class WindowedDStream[T: ClassTag](
    parent: DStream[T],
    _windowDuration: Duration,
    _slideDuration: Duration)
  extends DStream[T](parent.ssc) {

  // Persist parent level by default, as those RDDs are going to be obviously reused.
  parent.persist(StorageLevel.MEMORY_ONLY_SER) //默认将parentRDD设置persist,因为parent RDD会在window slide中被反复读到

  def windowDuration: Duration =  _windowDuration // Windows大小

  override def dependencies = List(parent)

  override def slideDuration: Duration = _slideDuration // Windows滑动

  override def parentRememberDuration: Duration = rememberDuration + windowDuration // 保证RememberDuratioin一定大于windowDuration

  override def persist(level: StorageLevel): DStream[T] = {
    // Do not let this windowed DStream be persisted as windowed (union-ed) RDDs share underlying
    // RDDs and persisting the windowed RDDs would store numerous copies of the underlying data.
    // Instead control the persistence of the parent DStream.
    // 不要直接persist windowed RDDS，而是去persist parent RDD，原因是各个windows RDDs之间有大量的重复数据，直接persist浪费空间
    parent.persist(level)
    this
  }

  override def compute(validTime: Time): Option[RDD[T]] = {
    val currentWindow = new Interval(validTime - windowDuration + parent.slideDuration, validTime) //计算窗口inteval
    val rddsInWindow = parent.slice(currentWindow)
    val windowRDD = if (rddsInWindow.flatMap(_.partitioner).distinct.length == 1) {
      new PartitionerAwareUnionRDD(ssc.sc, rddsInWindow)
    } else {
      new UnionRDD(ssc.sc,rddsInWindow) //本质就是把parent DStream窗口内的RDD做union
    }
    Some(windowRDD)
  }
}

ShuffledDStream

private[streaming]
class ShuffledDStream[K: ClassTag, V: ClassTag, C: ClassTag](
    parent: DStream[(K,V)],
    createCombiner: V => C,
    mergeValue: (C, V) => C,
    mergeCombiner: (C, C) => C,
    partitioner: Partitioner,
    mapSideCombine: Boolean = true
  ) extends DStream[(K,C)] (parent.ssc) {

  override def dependencies = List(parent)

  override def slideDuration: Duration = parent.slideDuration

  override def compute(validTime: Time): Option[RDD[(K,C)]] = {
    parent.getOrCompute(validTime) match {
      case Some(rdd) => Some(rdd.combineByKey[C](
          createCombiner, mergeValue, mergeCombiner, partitioner, mapSideCombine))
      case None => None
    }
  }
}

PairDStreamFunctions
以groupByKey为例，和普通Spark里面没啥区别，依赖是基于combineByKey实现
比较有特点是提供groupByKeyAndWindow，其实就是先使用WindowedDStream将windows中的RDD union，然后再使用combineByKey

/**
 * Extra functions available on DStream of (key, value) pairs through an implicit conversion.
 * Import `org.apache.spark.streaming.StreamingContext._` at the top of your program to use
 * these functions.
 */
class PairDStreamFunctions[K: ClassTag, V: ClassTag](self: DStream[(K,V)])
  extends Serializable {

  private[streaming] def ssc = self.ssc

  private[streaming] def defaultPartitioner(numPartitions: Int = self.ssc.sc.defaultParallelism)
  = {new HashPartitioner(numPartitions)}

  /**
   * Return a new DStream by applying `groupByKey` on each RDD. The supplied
   * [[org.apache.spark.Partitioner]] is used to control the partitioning of each RDD.
   */
  def groupByKey(partitioner: Partitioner): DStream[(K, Seq[V])] = {
    val createCombiner = (v: V) => ArrayBuffer[V](v)
    val mergeValue = (c: ArrayBuffer[V], v: V) => (c += v)
    val mergeCombiner = (c1: ArrayBuffer[V], c2: ArrayBuffer[V]) => (c1 ++ c2)
    combineByKey(createCombiner, mergeValue, mergeCombiner, partitioner)
      .asInstanceOf[DStream[(K, Seq[V])]]
  }

  /**
   * Combine elements of each key in DStream's RDDs using custom functions. This is similar to the
   * combineByKey for RDDs. Please refer to combineByKey in
   * org.apache.spark.rdd.PairRDDFunctions in the Spark core documentation for more information.
   */
  def combineByKey[C: ClassTag](
    createCombiner: V => C,
    mergeValue: (C, V) => C,
    mergeCombiner: (C, C) => C,
    partitioner: Partitioner,
    mapSideCombine: Boolean = true): DStream[(K, C)] = {
    new ShuffledDStream[K, V, C](self, createCombiner, mergeValue, mergeCombiner, partitioner,
      mapSideCombine)
  }
}

groupByKeyAndWindow

  /**
   * Create a new DStream by applying `groupByKey` over a sliding window on `this` DStream.
   * Similar to `DStream.groupByKey()`, but applies it over a sliding window.
   * @param windowDuration width of the window; must be a multiple of this DStream's
   *                       batching interval
   * @param slideDuration  sliding interval of the window (i.e., the interval after which
   *                       the new DStream will generate RDDs); must be a multiple of this
   *                       DStream's batching interval
   * @param partitioner    partitioner for controlling the partitioning of each RDD in the new
   *                       DStream.
   */
  def groupByKeyAndWindow(
      windowDuration: Duration,
      slideDuration: Duration,
      partitioner: Partitioner
    ): DStream[(K, Seq[V])] = {
    val createCombiner = (v: Seq[V]) => new ArrayBuffer[V] ++= v
    val mergeValue = (buf: ArrayBuffer[V], v: Seq[V]) => buf ++= v
    val mergeCombiner = (buf1: ArrayBuffer[V], buf2: ArrayBuffer[V]) => buf1 ++= buf2
    self.groupByKey(partitioner)
        .window(windowDuration, slideDuration) // DStream.window会将当前的dstream封装成WindowedDStream，见下面的代码
        .combineByKey[ArrayBuffer[V]](createCombiner, mergeValue, mergeCombiner, partitioner)
        .asInstanceOf[DStream[(K, Seq[V])]]
  }

  /**
   * Return a new DStream in which each RDD contains all the elements in seen in a
   * sliding window of time over this DStream.
   * @param windowDuration width of the window; must be a multiple of this DStream's
   *                       batching interval
   * @param slideDuration  sliding interval of the window (i.e., the interval after which
   *                       the new DStream will generate RDDs); must be a multiple of this
   *                       DStream's batching interval
   */
  def window(windowDuration: Duration, slideDuration: Duration): DStream[T] = {
    new WindowedDStream(this, windowDuration, slideDuration)
  }

updateStateByKey

  /**
   * Return a new "state" DStream where the state for each key is updated by applying
   * the given function on the previous state of the key and the new values of each key.
   * [[org.apache.spark.Partitioner]] is used to control the partitioning of each RDD.
   * @param updateFunc State update function. If `this` function returns None, then
   *                   corresponding state key-value pair will be eliminated. Note, that
   *                   this function may generate a different a tuple with a different key
   *                   than the input key. It is up to the developer to decide whether to
   *                   remember the partitioner despite the key being changed.
   * @param partitioner Partitioner for controlling the partitioning of each RDD in the new
   *                    DStream
   * @param rememberPartitioner Whether to remember the paritioner object in the generated RDDs.
   * @tparam S State type
   */
  def updateStateByKey[S: ClassTag](
      updateFunc: (Iterator[(K, Seq[V], Option[S])]) => Iterator[(K, S)],
      partitioner: Partitioner,
      rememberPartitioner: Boolean
    ): DStream[(K, S)] = {
     new StateDStream(self, ssc.sc.clean(updateFunc), partitioner, rememberPartitioner)
  }

StateDStream
普通的DStream，都是直接从ParentRDD通过compute来得到当前的RDD
而StateDStream的特别之处，除了ParentRDD，还需要参考PreviousRDD，这个只存在在stream场景下，只有这个场景下，RDD之间才存在时间关系
PreviousRDD = getOrCompute(validTime - slideDuration)，即在DStream的generatedRDDs上前一个时间interval上的RDD
处理函数，val finalFunc = (iterator: Iterator[(K, (Seq[V], Seq[S]))]) => { }，需要3个参数，key，ParentRDD上的value，PreviousRDD上的value
处理函数需要考虑，当ParentRDD或PreviousRDD为空的情况

注意StateDStream，默认需要做persist和checkpoint

private[streaming]
class StateDStream[K: ClassTag, V: ClassTag, S: ClassTag](
    parent: DStream[(K, V)],
    updateFunc: (Iterator[(K, Seq[V], Option[S])]) => Iterator[(K, S)],
    partitioner: Partitioner,
    preservePartitioning: Boolean
  ) extends DStream[(K, S)](parent.ssc) {

  super.persist(StorageLevel.MEMORY_ONLY_SER) // RDD persist默认设为memory，因为后面的RDD需要用到

  override def dependencies = List(parent)

  override def slideDuration: Duration = parent.slideDuration

  override val mustCheckpoint = true  // 默认需要checkpoint，需要保持状态

  override def compute(validTime: Time): Option[RDD[(K, S)]] = {

    // Try to get the previous state RDD
    getOrCompute(validTime - slideDuration) match {

      case Some(prevStateRDD) => {    // If previous state RDD exists

        // Try to get the parent RDD
        parent.getOrCompute(validTime) match {  // 既有PreviousRDD，又有ParentRDD的case
          case Some(parentRDD) => {   // If parent RDD exists, then compute as usual

            // Define the function for the mapPartition operation on cogrouped RDD;
            // first map the cogrouped tuple to tuples of required type,
            // and then apply the update function
            val updateFuncLocal = updateFunc
            val finalFunc = (iterator: Iterator[(K, (Seq[V], Seq[S]))]) => {
              val i = iterator.map(t => {
                (t._1, t._2._1, t._2._2.headOption)
              })
              updateFuncLocal(i)
            }
            val cogroupedRDD = parentRDD.cogroup(prevStateRDD, partitioner) //`(k, a) cogroup (k, b)` produces k -> Seq(ArrayBuffer as, ArrayBuffer bs)
            val stateRDD = cogroupedRDD.mapPartitions(finalFunc, preservePartitioning)
            Some(stateRDD)
          }
          case None => {    // If parent RDD does not exist，ParentRDD不存在

            // Re-apply the update function to the old state RDD
            val updateFuncLocal = updateFunc
            val finalFunc = (iterator: Iterator[(K, S)]) => {
              val i = iterator.map(t => (t._1, Seq[V](), Option(t._2))) // 直接把ParentRDD置空，Seq[V]()
              updateFuncLocal(i)
            }
            val stateRDD = prevStateRDD.mapPartitions(finalFunc, preservePartitioning)
            Some(stateRDD)
          }
        }
      }

      case None => {    // If previous session RDD does not exist (first input data)

        // Try to get the parent RDD
        parent.getOrCompute(validTime) match {
          case Some(parentRDD) => {   // If parent RDD exists, then compute as usual，PreviousRDD为空的case，说明是第一个state RDD

            // Define the function for the mapPartition operation on grouped RDD;
            // first map the grouped tuple to tuples of required type,
            // and then apply the update function
            val updateFuncLocal = updateFunc
            val finalFunc = (iterator: Iterator[(K, Seq[V])]) => {
              updateFuncLocal(iterator.map(tuple => (tuple._1, tuple._2, None))) // 把PreviousRDD置为None
            }

            val groupedRDD = parentRDD.groupByKey(partitioner)
            val sessionRDD = groupedRDD.mapPartitions(finalFunc, preservePartitioning)
            //logDebug("Generating state RDD for time " + validTime + " (first)")
            Some(sessionRDD)
          }
          case None => { // If parent RDD does not exist, then nothing to do!，previous和parent都没有，当然啥也做不了
            //logDebug("Not generating state RDD (no previous state, no parent)")
            None
          }
        }
      }
    }
  }
}

TransformedDStream
首先这是个比较通用的operation，可以通过自定义的transformFunc，将一组parentRDDs计算出当前的RDD
需要注意的是，这些parentRDDs必须在同一个streamContext下，并且有相同的slideDuration
在DStream接口中，可以提供transform和transformWith两种，参考下面源码

private[streaming]
class TransformedDStream[U: ClassTag] (
    parents: Seq[DStream[_]],
    transformFunc: (Seq[RDD[_]], Time) => RDD[U]
  ) extends DStream[U](parents.head.ssc) {

  require(parents.length > 0, "List of DStreams to transform is empty")
  require(parents.map(_.ssc).distinct.size == 1, "Some of the DStreams have different contexts")
  require(parents.map(_.slideDuration).distinct.size == 1,
    "Some of the DStreams have different slide durations")

  override def dependencies = parents.toList

  override def slideDuration: Duration = parents.head.slideDuration

  override def compute(validTime: Time): Option[RDD[U]] = {
    val parentRDDs = parents.map(_.getOrCompute(validTime).orNull).toSeq
    Some(transformFunc(parentRDDs, validTime))
  }
}

  /**
   * Return a new DStream in which each RDD is generated by applying a function
   * on each RDD of 'this' DStream.
   */
  def transform[U: ClassTag](transformFunc: (RDD[T], Time) => RDD[U]): DStream[U] = {
    val cleanedF = context.sparkContext.clean(transformFunc)
    val realTransformFunc =  (rdds: Seq[RDD[_]], time: Time) => {
      assert(rdds.length == 1)
      cleanedF(rdds.head.asInstanceOf[RDD[T]], time)
    }
    new TransformedDStream[U](Seq(this), realTransformFunc) // this，单个RDD
  }
  /**
   * Return a new DStream in which each RDD is generated by applying a function
   * on each RDD of 'this' DStream and 'other' DStream.
   */
  def transformWith[U: ClassTag, V: ClassTag](
      other: DStream[U], transformFunc: (RDD[T], RDD[U], Time) => RDD[V]
    ): DStream[V] = {
    val cleanedF = ssc.sparkContext.clean(transformFunc)
    val realTransformFunc = (rdds: Seq[RDD[_]], time: Time) => {
      assert(rdds.length == 2)
      val rdd1 = rdds(0).asInstanceOf[RDD[T]]
      val rdd2 = rdds(1).asInstanceOf[RDD[U]]
      cleanedF(rdd1, rdd2, time)
    }
    new TransformedDStream[V](Seq(this, other), realTransformFunc) // this and other，多个RDDs
  }