tgravescs commented on a change in pull request #27636:
[SPARK-30873][CORE][YARN]Handling Node Decommissioning for Yarn cluster manger
in Spark
URL: https://github.com/apache/spark/pull/27636#discussion_r389722064
##########
File path: core/src/main/scala/org/apache/spark/scheduler/TaskSetManager.scala
##########
@@ -798,6 +799,23 @@ private[spark] class TaskSetManager(
if (fetchFailed.bmAddress != null) {
blacklistTracker.foreach(_.updateBlacklistForFetchFailure(
fetchFailed.bmAddress.host, fetchFailed.bmAddress.executorId))
+
+ // Do account fetch failure exception raised by decommissioned
+ // node against stage failure. Here the logic is to specify,
+ // if the task failed due to fetchFailed of decommission nodes than
+ // don't count towards the stageFailure. countTowardsStageFailures
+ // variable of TaskEndReason, that can be used in DAG scheduler to
account
+ // fetch failure while checking the stage abort
+ decommissionTracker match {
+ case Some(decommissionTracker) =>
+ if
(decommissionTracker.isNodeDecommissioned(fetchFailed.bmAddress.host)) {
+ logInfo(s"Do not count fetch failure from decommissioned" +
+ s" node ${fetchFailed.bmAddress.host}")
+ fetchFailed.countTowardsStageFailures = false
Review comment:
We have a lot of logic in DAGScheduler doing account and just assuming this
is decommission. If we know its being decommissioned perhaps we should add
something to indicate it so we aren't guessing
----------------------------------------------------------------
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]
