mridulm commented on code in PR #42950:
URL: https://github.com/apache/spark/pull/42950#discussion_r1334896106
##########
core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala:
##########
@@ -1903,13 +1903,20 @@ private[spark] class DAGScheduler(
case smt: ShuffleMapTask =>
val shuffleStage = stage.asInstanceOf[ShuffleMapStage]
- shuffleStage.pendingPartitions -= task.partitionId
+ // Ignore task completion for old attempt of indeterminate stage
+ val ignoreIndeterminate = stage.isIndeterminate &&
+ task.stageAttemptId < stage.latestInfo.attemptNumber()
+ if (!ignoreIndeterminate) {
+ shuffleStage.pendingPartitions -= task.partitionId
+ }
val status = event.result.asInstanceOf[MapStatus]
val execId = status.location.executorId
logDebug("ShuffleMapTask finished on " + execId)
if (executorFailureEpoch.contains(execId) &&
smt.epoch <= executorFailureEpoch(execId)) {
logInfo(s"Ignoring possibly bogus $smt completion from executor
$execId")
Review Comment:
So this is kind of funny - take a look at what the above replaced @cloud-fan
:
https://github.com/apache/spark/pull/16620/files#diff-85de35b2e85646ed499c545a3be1cd3ffd525a88aae835a9c621f877eebadcb6R1183
:-)
Both of these actually do not account for DETERMINATE/INDETERMINATE changes
we made subsequently.
IMO, for INDETERMINATE stages, we should ignore task completion events from
previous attempts - since we have already cancelled the state.
Having said that, I have not thought through the nuances here.
+CC @jiangxb1987 as well
--
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.
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
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
