[
https://issues.apache.org/jira/browse/SPARK-19894?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15904621#comment-15904621
]
Yuechen Chen commented on SPARK-19894:
--------------------------------------
https://github.com/apache/spark/pull/17238
> Tasks entirely assigned to one executor on Yarn-cluster mode for default-rack
> -----------------------------------------------------------------------------
>
> Key: SPARK-19894
> URL: https://issues.apache.org/jira/browse/SPARK-19894
> Project: Spark
> Issue Type: Bug
> Components: Scheduler, YARN
> Affects Versions: 2.1.0
> Environment: Yarn-cluster
> Reporter: Yuechen Chen
>
> In YARN-cluster mode, if driver has no rack information on two different
> hosts, these two hosts would both be recoginized as "/default-rack", which
> may cause some bugs.
> For example, if hosts of one executor and one external datasource are unknown
> by driver, this two hosts would be recoginized as the same rack
> "/default-rack", and then all tasks would be assigned to the executor.
> This bug would be avoided, if getRackForHost("unknown host") in YarnScheduler
> returns None, not Some("/default-rack").
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
