JoshRosen commented on code in PR #43954:
URL: https://github.com/apache/spark/pull/43954#discussion_r1402597413
##########
core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala:
##########
@@ -2179,12 +2164,12 @@ private[spark] class DAGScheduler(
val message = s"Stage failed because barrier task $task finished
unsuccessfully.\n" +
failure.toErrorString
try {
- // killAllTaskAttempts will fail if a SchedulerBackend does not
implement killTask.
+ // cancelTasks will fail if a SchedulerBackend does not implement
killTask.
val reason = s"Task $task from barrier stage $failedStage
(${failedStage.name}) " +
"failed."
val job = jobIdToActiveJob.get(failedStage.firstJobId)
val shouldInterrupt = job.exists(j => shouldInterruptTaskThread(j))
- taskScheduler.killAllTaskAttempts(stageId, shouldInterrupt, reason)
Review Comment:
I think that there is a subtle difference here: `killAllTaskAttempts` only
kills tasks, whereas `cancelTasks` also calls `tsm.abort()` on the stage
attempts, which might enqueue a new `taskSetFailed` event for each task set and
I think that could have unintended side effects. Can you double-check whether
we think that change is okay?
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
