spark ContinuousExecution 源码

spark ContinuousExecution 代码

文件路径：/sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/continuous/ContinuousExecution.scala

/*
 * Licensed to the Apache Software Foundation (ASF) under one or more
 * contributor license agreements.  See the NOTICE file distributed with
 * this work for additional information regarding copyright ownership.
 * The ASF licenses this file to You under the Apache License, Version 2.0
 * (the "License"); you may not use this file except in compliance with
 * the License.  You may obtain a copy of the License at
 *
 *    http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

package org.apache.spark.sql.execution.streaming.continuous

import java.util.UUID
import java.util.concurrent.TimeUnit
import java.util.concurrent.atomic.AtomicReference
import java.util.function.UnaryOperator

import scala.collection.mutable.{Map => MutableMap}

import org.apache.spark.SparkEnv
import org.apache.spark.sql.SparkSession
import org.apache.spark.sql.catalyst.expressions.{CurrentDate, CurrentTimestampLike, LocalTimestamp}
import org.apache.spark.sql.catalyst.plans.logical.LogicalPlan
import org.apache.spark.sql.catalyst.streaming.{StreamingRelationV2, WriteToStream}
import org.apache.spark.sql.catalyst.trees.TreePattern.CURRENT_LIKE
import org.apache.spark.sql.connector.catalog.{SupportsRead, SupportsWrite, TableCapability}
import org.apache.spark.sql.connector.distributions.UnspecifiedDistribution
import org.apache.spark.sql.connector.read.streaming.{ContinuousStream, PartitionOffset, ReadLimit}
import org.apache.spark.sql.connector.write.{RequiresDistributionAndOrdering, Write}
import org.apache.spark.sql.errors.{QueryCompilationErrors, QueryExecutionErrors}
import org.apache.spark.sql.execution.SQLExecution
import org.apache.spark.sql.execution.datasources.v2.StreamingDataSourceV2Relation
import org.apache.spark.sql.execution.streaming._
import org.apache.spark.sql.streaming.Trigger
import org.apache.spark.util.Clock

class ContinuousExecution(
    sparkSession: SparkSession,
    trigger: Trigger,
    triggerClock: Clock,
    extraOptions: Map[String, String],
    plan: WriteToStream)
  extends StreamExecution(
    sparkSession, plan.name, plan.resolvedCheckpointLocation, plan.inputQuery, plan.sink,
    trigger, triggerClock, plan.outputMode, plan.deleteCheckpointOnStop) {

  @volatile protected var sources: Seq[ContinuousStream] = Seq()

  // For use only in test harnesses.
  private[sql] var currentEpochCoordinatorId: String = _

  // Throwable that caused the execution to fail
  private val failure: AtomicReference[Throwable] = new AtomicReference[Throwable](null)

  override val logicalPlan: WriteToContinuousDataSource = {
    val v2ToRelationMap = MutableMap[StreamingRelationV2, StreamingDataSourceV2Relation]()
    var nextSourceId = 0
    import org.apache.spark.sql.execution.datasources.v2.DataSourceV2Implicits._
    val _logicalPlan = analyzedPlan.transform {
      case s @ StreamingRelationV2(ds, sourceName, table: SupportsRead, options, output,
        catalog, identifier, _) =>
        val dsStr = if (ds.nonEmpty) s"[${ds.get}]" else ""
        if (!table.supports(TableCapability.CONTINUOUS_READ)) {
          throw QueryExecutionErrors.continuousProcessingUnsupportedByDataSourceError(sourceName)
        }

        v2ToRelationMap.getOrElseUpdate(s, {
          val metadataPath = s"$resolvedCheckpointRoot/sources/$nextSourceId"
          nextSourceId += 1
          logInfo(s"Reading table [$table] from DataSourceV2 named '$sourceName' $dsStr")
          // TODO: operator pushdown.
          val scan = table.newScanBuilder(options).build()
          val stream = scan.toContinuousStream(metadataPath)
          StreamingDataSourceV2Relation(output, scan, stream, catalog, identifier)
        })
    }

    sources = _logicalPlan.collect {
      case r: StreamingDataSourceV2Relation => r.stream.asInstanceOf[ContinuousStream]
    }
    uniqueSources = sources.distinct.map(s => s -> ReadLimit.allAvailable()).toMap

    // TODO (SPARK-27484): we should add the writing node before the plan is analyzed.
    val write = createWrite(plan.sink.asInstanceOf[SupportsWrite], extraOptions, _logicalPlan)

    if (hasDistributionRequirements(write) || hasOrderingRequirements(write)) {
      throw QueryCompilationErrors.writeDistributionAndOrderingNotSupportedInContinuousExecution()
    }

    val streamingWrite = write.toStreaming
    val customMetrics = write.supportedCustomMetrics.toSeq
    WriteToContinuousDataSource(streamingWrite, _logicalPlan, customMetrics)
  }

  private def hasDistributionRequirements(write: Write): Boolean = write match {
    case w: RequiresDistributionAndOrdering if w.requiredNumPartitions == 0 =>
      w.requiredDistribution match {
        case _: UnspecifiedDistribution =>
          false
        case _ =>
          true
      }
    case _ =>
      false
  }

  private def hasOrderingRequirements(write: Write): Boolean = write match {
    case w: RequiresDistributionAndOrdering if w.requiredOrdering.nonEmpty =>
      true
    case _ =>
      false
  }

  private val triggerExecutor = trigger match {
    case ContinuousTrigger(t) => ProcessingTimeExecutor(ProcessingTimeTrigger(t), triggerClock)
    case _ => throw new IllegalStateException(s"Unsupported type of trigger: $trigger")
  }

  override protected def runActivatedStream(sparkSessionForStream: SparkSession): Unit = {
    val stateUpdate = new UnaryOperator[State] {
      override def apply(s: State) = s match {
        // If we ended the query to reconfigure, reset the state to active.
        case RECONFIGURING => ACTIVE
        case _ => s
      }
    }

    do {
      runContinuous(sparkSessionForStream)
    } while (state.updateAndGet(stateUpdate) == ACTIVE)

    stopSources()
  }

  /**
   * Populate the start offsets to start the execution at the current offsets stored in the sink
   * (i.e. avoid reprocessing data that we have already processed). This function must be called
   * before any processing occurs and will populate the following fields:
   *  - currentBatchId
   *  - committedOffsets
   *  The basic structure of this method is as follows:
   *
   *  Identify (from the commit log) the latest epoch that has committed
   *  IF last epoch exists THEN
   *    Get end offsets for the epoch
   *    Set those offsets as the current commit progress
   *    Set the next epoch ID as the last + 1
   *    Return the end offsets of the last epoch as start for the next one
   *    DONE
   *  ELSE
   *    Start a new query log
   *  DONE
   */
  private def getStartOffsets(): OffsetSeq = {
    // Note that this will need a slight modification for exactly once. If ending offsets were
    // reported but not committed for any epochs, we must replay exactly to those offsets.
    // For at least once, we can just ignore those reports and risk duplicates.
    commitLog.getLatest() match {
      case Some((latestEpochId, _)) =>
        updateStatusMessage("Starting new streaming query " +
          s"and getting offsets from latest epoch $latestEpochId")
        val nextOffsets = offsetLog.get(latestEpochId).getOrElse {
          throw new IllegalStateException(
            s"Batch $latestEpochId was committed without end epoch offsets!")
        }
        committedOffsets = nextOffsets.toStreamProgress(sources)
        currentBatchId = latestEpochId + 1

        logDebug(s"Resuming at epoch $currentBatchId with committed offsets $committedOffsets")
        nextOffsets
      case None =>
        // We are starting this stream for the first time. Offsets are all None.
        updateStatusMessage("Starting new streaming query")
        logInfo(s"Starting new streaming query.")
        currentBatchId = 0
        OffsetSeq.fill(sources.map(_ => null): _*)
    }
  }

  /**
   * Do a continuous run.
   * @param sparkSessionForQuery Isolated [[SparkSession]] to run the continuous query with.
   */
  private def runContinuous(sparkSessionForQuery: SparkSession): Unit = {
    val offsets = getStartOffsets()

    if (currentBatchId > 0) {
      AcceptsLatestSeenOffsetHandler.setLatestSeenOffsetOnSources(Some(offsets), sources)
    }

    val withNewSources: LogicalPlan = logicalPlan transform {
      case relation: StreamingDataSourceV2Relation =>
        val loggedOffset = offsets.offsets(0)
        val realOffset = loggedOffset.map(off => relation.stream.deserializeOffset(off.json))
        val startOffset = realOffset.getOrElse(relation.stream.initialOffset)
        relation.copy(startOffset = Some(startOffset))
    }

    withNewSources.transformAllExpressionsWithPruning(_.containsPattern(CURRENT_LIKE)) {
      case (_: CurrentTimestampLike | _: CurrentDate | _: LocalTimestamp) =>
        throw new IllegalStateException("CurrentTimestamp, Now, CurrentDate and LocalTimestamp" +
          " not yet supported for continuous processing")
    }

    reportTimeTaken("queryPlanning") {
      lastExecution = new IncrementalExecution(
        sparkSessionForQuery,
        withNewSources,
        outputMode,
        checkpointFile("state"),
        id,
        runId,
        currentBatchId,
        offsetSeqMetadata)
      lastExecution.executedPlan // Force the lazy generation of execution plan
    }

    val stream = withNewSources.collect {
      case relation: StreamingDataSourceV2Relation =>
        relation.stream.asInstanceOf[ContinuousStream]
    }.head

    sparkSessionForQuery.sparkContext.setLocalProperty(
      StreamExecution.IS_CONTINUOUS_PROCESSING, true.toString)
    sparkSessionForQuery.sparkContext.setLocalProperty(
      ContinuousExecution.START_EPOCH_KEY, currentBatchId.toString)
    // Add another random ID on top of the run ID, to distinguish epoch coordinators across
    // reconfigurations.
    val epochCoordinatorId = s"$runId--${UUID.randomUUID}"
    currentEpochCoordinatorId = epochCoordinatorId
    sparkSessionForQuery.sparkContext.setLocalProperty(
      ContinuousExecution.EPOCH_COORDINATOR_ID_KEY, epochCoordinatorId)

    // Use the parent Spark session for the endpoint since it's where this query ID is registered.
    val epochEndpoint = EpochCoordinatorRef.create(
      logicalPlan.write,
      stream,
      this,
      epochCoordinatorId,
      currentBatchId,
      sparkSession,
      SparkEnv.get)
    val epochUpdateThread = new Thread(new Runnable {
      override def run: Unit = {
        try {
          triggerExecutor.execute(() => {
            startTrigger()

            if (stream.needsReconfiguration && state.compareAndSet(ACTIVE, RECONFIGURING)) {
              if (queryExecutionThread.isAlive) {
                queryExecutionThread.interrupt()
              }
              false
            } else if (isActive) {
              currentBatchId = epochEndpoint.askSync[Long](IncrementAndGetEpoch)
              logInfo(s"New epoch $currentBatchId is starting.")
              true
            } else {
              false
            }
          })
        } catch {
          case _: InterruptedException =>
            // Cleanly stop the query.
        }
      }
    }, s"epoch update thread for $prettyIdString")

    try {
      epochUpdateThread.setDaemon(true)
      epochUpdateThread.start()

      updateStatusMessage("Running")
      reportTimeTaken("runContinuous") {
        SQLExecution.withNewExecutionId(lastExecution) {
          lastExecution.executedPlan.execute()
        }
      }

      val f = failure.get()
      if (f != null) {
        throw f
      }
    } catch {
      case t: Throwable if StreamExecution.isInterruptionException(t, sparkSession.sparkContext) &&
          state.get() == RECONFIGURING =>
        logInfo(s"Query $id ignoring exception from reconfiguring: $t")
        // interrupted by reconfiguration - swallow exception so we can restart the query
    } finally {
      // The above execution may finish before getting interrupted, for example, a Spark job having
      // 0 partitions will complete immediately. Then the interrupted status will sneak here.
      //
      // To handle this case, we do the two things here:
      //
      // 1. Clean up the resources in `queryExecutionThread.runUninterruptibly`. This may increase
      //    the waiting time of `stop` but should be minor because the operations here are very fast
      //    (just sending an RPC message in the same process and stopping a very simple thread).
      // 2. Clear the interrupted status at the end so that it won't impact the `runContinuous`
      //    call. We may clear the interrupted status set by `stop`, but it doesn't affect the query
      //    termination because `runActivatedStream` will check `state` and exit accordingly.
      queryExecutionThread.runUninterruptibly {
        try {
          epochEndpoint.askSync[Unit](StopContinuousExecutionWrites)
        } finally {
          SparkEnv.get.rpcEnv.stop(epochEndpoint)
          epochUpdateThread.interrupt()
          epochUpdateThread.join()
          // The following line must be the last line because it may fail if SparkContext is stopped
          sparkSession.sparkContext.cancelJobGroup(runId.toString)
        }
      }
      Thread.interrupted()
    }
  }

  /**
   * Report ending partition offsets for the given reader at the given epoch.
   */
  def addOffset(
      epoch: Long,
      stream: ContinuousStream,
      partitionOffsets: Seq[PartitionOffset]): Unit = {
    assert(sources.length == 1, "only one continuous source supported currently")

    val globalOffset = stream.mergeOffsets(partitionOffsets.toArray)
    val oldOffset = synchronized {
      offsetLog.add(epoch, OffsetSeq.fill(globalOffset))
      offsetLog.get(epoch - 1)
    }

    // If offset hasn't changed since last epoch, there's been no new data.
    if (oldOffset.contains(OffsetSeq.fill(globalOffset))) {
      noNewData = true
    }

    awaitProgressLock.lock()
    try {
      awaitProgressLockCondition.signalAll()
    } finally {
      awaitProgressLock.unlock()
    }
  }

  /**
   * Mark the specified epoch as committed. All readers must have reported end offsets for the epoch
   * before this is called.
   */
  def commit(epoch: Long): Unit = {
    updateStatusMessage(s"Committing epoch $epoch")

    assert(sources.length == 1, "only one continuous source supported currently")
    assert(offsetLog.get(epoch).isDefined, s"offset for epoch $epoch not reported before commit")

    synchronized {
      // Record offsets before updating `committedOffsets`
      recordTriggerOffsets(from = committedOffsets, to = availableOffsets, latest = latestOffsets)
      if (queryExecutionThread.isAlive) {
        commitLog.add(epoch, CommitMetadata())
        val offset =
          sources(0).deserializeOffset(offsetLog.get(epoch).get.offsets(0).get.json)
        committedOffsets ++= Seq(sources(0) -> offset)
        sources(0).commit(offset)
      } else {
        return
      }
    }

    // Since currentBatchId increases independently in cp mode, the current committed epoch may
    // be far behind currentBatchId. It is not safe to discard the metadata with thresholdBatchId
    // computed based on currentBatchId. As minLogEntriesToMaintain is used to keep the minimum
    // number of batches that must be retained and made recoverable, so we should keep the
    // specified number of metadata that have been committed.
    if (minLogEntriesToMaintain <= epoch) {
      purge(epoch + 1 - minLogEntriesToMaintain)
    }

    awaitProgressLock.lock()
    try {
      awaitProgressLockCondition.signalAll()
    } finally {
      awaitProgressLock.unlock()
    }
  }

  /**
   * Blocks the current thread until execution has committed at or after the specified epoch.
   */
  private[sql] def awaitEpoch(epoch: Long): Unit = {
    def notDone = {
      val latestCommit = commitLog.getLatest()
      latestCommit match {
        case Some((latestEpoch, _)) =>
          latestEpoch < epoch
        case None => true
      }
    }

    while (notDone) {
      awaitProgressLock.lock()
      try {
        awaitProgressLockCondition.await(100, TimeUnit.MILLISECONDS)
        if (streamDeathCause != null) {
          throw streamDeathCause
        }
      } finally {
        awaitProgressLock.unlock()
      }
    }
  }

  /**
   * Stores error and stops the query execution thread to terminate the query in new thread.
   */
  def stopInNewThread(error: Throwable): Unit = {
    if (failure.compareAndSet(null, error)) {
      logError(s"Query $prettyIdString received exception $error")
      stopInNewThread()
    }
  }

  /**
   * Stops the query execution thread to terminate the query in new thread.
   */
  private def stopInNewThread(): Unit = {
    new Thread("stop-continuous-execution") {
      setDaemon(true)

      override def run(): Unit = {
        try {
          ContinuousExecution.this.stop()
        } catch {
          case e: Throwable =>
            logError(e.getMessage, e)
            throw e
        }
      }
    }.start()
  }

  /**
   * Stops the query execution thread to terminate the query.
   */
  override def stop(): Unit = {
    // Set the state to TERMINATED so that the batching thread knows that it was interrupted
    // intentionally
    state.set(TERMINATED)
    if (queryExecutionThread.isAlive) {
      // The query execution thread will clean itself up in the finally clause of runContinuous.
      // We just need to interrupt the long running job.
      interruptAndAwaitExecutionThreadTermination()
    }
    logInfo(s"Query $prettyIdString was stopped")
  }
}

object ContinuousExecution {
  val START_EPOCH_KEY = "__continuous_start_epoch"
  val EPOCH_COORDINATOR_ID_KEY = "__epoch_coordinator_id"
}

相关信息

spark 源码目录

相关文章

spark ContinuousDataSourceRDD 源码

spark ContinuousQueuedDataReader 源码

spark ContinuousRateStreamSource 源码

spark ContinuousTaskRetryException 源码

spark ContinuousTextSocketSource 源码

spark ContinuousWriteRDD 源码

spark EpochCoordinator 源码

spark EpochTracker 源码

spark WriteToContinuousDataSource 源码

spark WriteToContinuousDataSourceExec 源码