[
https://issues.apache.org/jira/browse/SPARK-27348?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16812657#comment-16812657
]
Shixiong Zhu commented on SPARK-27348:
--------------------------------------
[~sandeep.katta2007] I cannot reproduce this locally. Ideally, when we decide
to remove an executor, we should remove it from all places rather than counting
on a TCP disconnect event which may not happen sometimes.
> HeartbeatReceiver doesn't remove lost executors from
> CoarseGrainedSchedulerBackend
> ----------------------------------------------------------------------------------
>
> Key: SPARK-27348
> URL: https://issues.apache.org/jira/browse/SPARK-27348
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 2.4.0
> Reporter: Shixiong Zhu
> Priority: Major
>
> When a heartbeat timeout happens in HeartbeatReceiver, it doesn't remove lost
> executors from CoarseGrainedSchedulerBackend. When a connection of an
> executor is not gracefully shut down, CoarseGrainedSchedulerBackend may not
> receive a disconnect event. In this case, CoarseGrainedSchedulerBackend still
> thinks a lost executor is still alive. CoarseGrainedSchedulerBackend may ask
> TaskScheduler to run tasks on this lost executor. This task will never finish
> and the job will hang forever.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
