venkata91 commented on a change in pull request #30691:
URL: https://github.com/apache/spark/pull/30691#discussion_r646974281
##########
File path: core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala
##########
@@ -2000,6 +2025,149 @@ private[spark] class DAGScheduler(
}
}
+ /**
+ * Schedules shuffle merge finalize.
+ */
+ private[scheduler] def scheduleShuffleMergeFinalize(stage: ShuffleMapStage):
Unit = {
+ logInfo(("%s (%s) scheduled for finalizing" +
+ " shuffle merge in %s s").format(stage, stage.name,
shuffleMergeFinalizeWaitSec))
+ shuffleMergeFinalizeScheduler.schedule(
+ new Runnable {
+ override def run(): Unit = finalizeShuffleMerge(stage)
+ },
+ shuffleMergeFinalizeWaitSec,
+ TimeUnit.SECONDS
+ )
+ }
+
+ /**
+ * DAGScheduler notifies all the remote shuffle services chosen to serve
shuffle merge request for
+ * the given shuffle map stage to finalize the shuffle merge process for
this shuffle. This is
+ * invoked in a separate thread to reduce the impact on the DAGScheduler
main thread, as the
+ * scheduler might need to talk to 1000s of shuffle services to finalize
shuffle merge.
+ */
+ private[scheduler] def finalizeShuffleMerge(stage: ShuffleMapStage): Unit = {
+ logInfo("%s (%s) finalizing the shuffle merge".format(stage, stage.name))
+ externalShuffleClient.foreach { shuffleClient =>
+ val shuffleId = stage.shuffleDep.shuffleId
+ val numMergers = stage.shuffleDep.getMergerLocs.length
+ val numResponses = new AtomicInteger()
+ val results = (0 until numMergers).map(_ =>
SettableFuture.create[Boolean]())
+ val timedOut = new AtomicBoolean()
Review comment:
After having offline discussion, we simplified it further just by
posting `ShuffleMergeFinalized` only in the finally block that way it
guarantees the event itself would get posted only once.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
