[
https://issues.apache.org/jira/browse/SPARK-4020?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14177678#comment-14177678
]
Apache Spark commented on SPARK-4020:
-------------------------------------
User 'andrewor14' has created a pull request for this issue:
https://github.com/apache/spark/pull/2865
> Failed executor not properly removed if it has not run tasks
> ------------------------------------------------------------
>
> Key: SPARK-4020
> URL: https://issues.apache.org/jira/browse/SPARK-4020
> Project: Spark
> Issue Type: Bug
> Reporter: Andrew Or
> Assignee: Andrew Or
>
> If an executor has not run tasks, then the DAG scheduler won't remove the
> block manager that belongs to this executor. This is because we don't add the
> executor to `activeExecutorIds` in `TaskSchedulerImpl` until we have
> successfully scheduled a task on the executor.
> The motivation for fixing this is to avoid having the
> `BlockManagerMasterActor` to remove the associated block manager through a
> timeout when it can just do so right away.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
