[
https://issues.apache.org/jira/browse/SPARK-18820?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15740754#comment-15740754
]
jin xing commented on SPARK-18820:
----------------------------------
[~lins05]
Thanks a lot for your comment : )
In our company's cluster, we find lots of the NullPointerException described
above;
Checking source code, I found CoarseGrainedSchedulerBackend will
executorDataMap first, then reply "RegisteredExecutor";
After updating executorDataMap, the new joined executor may be sent
"LaunchTask", which will result in "LaunchTask" arrives before than
"RegisteredExecutor";
How do you think about this?
> Driver may send "LaunchTask" before executor receive "RegisteredExecutor"
> -------------------------------------------------------------------------
>
> Key: SPARK-18820
> URL: https://issues.apache.org/jira/browse/SPARK-18820
> Project: Spark
> Issue Type: Bug
> Components: Scheduler
> Affects Versions: 1.6.3
> Environment: spark-1.6.3
> Reporter: jin xing
>
> CoarseGrainedSchedulerBackend will update executorDataMap after receiving
> "RegisterExecutor", thus task scheduler may assign tasks on to this executor;
> If LaunchTask arrives at CoarseGrainedExecutorBackend before
> RegisteredExecutor, it will result in NullPointerException and executor
> backend will exit;
> Is it a bug? If so can I make a pr? I think driver should send "LaunchTask"
> after "RegisteredExecutor" is already received.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
