cloud-fan commented on a change in pull request #24110: [SPARK-25341][Core]
Support rolling back a shuffle map stage and re-generate the shuffle files
URL: https://github.com/apache/spark/pull/24110#discussion_r272449374
##########
File path:
core/src/main/scala/org/apache/spark/scheduler/TaskSchedulerImpl.scala
##########
@@ -254,15 +254,20 @@ private[spark] class TaskSchedulerImpl(
// that stage when an active TaskSetManager succeed.
stageIdToFinishedPartitions.getOrElseUpdate(taskSet.stageId, new BitSet)
val tsm = new TaskSetManager(this, taskSet, maxTaskFailures,
blacklistTrackerOpt)
- // TaskSet got submitted by DAGScheduler may have some already completed
- // tasks since DAGScheduler does not always know all the tasks that have
- // been completed by other tasksets when completing a stage, so we mark
- // those tasks as finished here to avoid launching duplicate tasks, while
- // holding the TaskSchedulerImpl lock.
- // See SPARK-25250 and `markPartitionCompletedInAllTaskSets()`
- stageIdToFinishedPartitions.get(taskSet.stageId).foreach {
- finishedPartitions =>
finishedPartitions.foreach(tsm.markPartitionCompleted(_, None))
+ // If this TaskSet belongs to a retrying indeterminate stage, we should
create
+ // a total new TaskSetManager.
+ if (!taskSet.properties.containsKey(SparkContext.IS_INDETERMINATE_STAGE)) {
Review comment:
I think we should add a new parameter in `TaskSet`.
----------------------------------------------------------------
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:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
