[
https://issues.apache.org/jira/browse/SPARK-48870?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated SPARK-48870:
-----------------------------------
Labels: pull-request-available (was: )
> spark.excludeOnFailure.killExcludedExecutors does not take effect in Spark on
> YARN mode.
> ----------------------------------------------------------------------------------------
>
> Key: SPARK-48870
> URL: https://issues.apache.org/jira/browse/SPARK-48870
> Project: Spark
> Issue Type: Bug
> Components: Spark Core, YARN
> Affects Versions: 3.4.3
> Reporter: JacobZheng
> Priority: Major
> Labels: pull-request-available
>
> In one of my Spark jobs, I enabled
> *spark.excludeOnFailure.killExcludedExecutors=true* and
> *spark.excludeOnFailure=true*, but I found that the excluded executor was not
> killed. The logs contain the following error message:
> {code:java}
> Not attempting to kill executors on excluded host cdh-001 since allocation
> client is not defined.
> {code}
> I checked the relevant code and found that the reason is that the
> *healthTracker* is initialized before the *sc.schedulerBackend*, causing
> *sc.schedulerBackend* to be null when *healthTracker* is initialized.
> {code:scala}
> private def maybeCreateHealthTracker(sc: SparkContext): Option[HealthTracker]
> = {
> if (HealthTracker.isExcludeOnFailureEnabled(sc.conf)) {
> val executorAllocClient: Option[ExecutorAllocationClient] =
> sc.schedulerBackend match {
> case b: ExecutorAllocationClient => Some(b)
> case _ => None
> }
> Some(new HealthTracker(sc, executorAllocClient))
> } else {
> None
> }
> } {code}
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]