[
https://issues.apache.org/jira/browse/SPARK-52508?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18021454#comment-18021454
]
Aparna Garg commented on SPARK-52508:
-------------------------------------
User 'EnricoMi' has created a pull request for this issue:
https://github.com/apache/spark/pull/51200
> Read from fallback storage should consider replication delay
> ------------------------------------------------------------
>
> Key: SPARK-52508
> URL: https://issues.apache.org/jira/browse/SPARK-52508
> Project: Spark
> Issue Type: Sub-task
> Components: Kubernetes
> Affects Versions: 4.1.0
> Reporter: Enrico Minack
> Priority: Major
>
> Using the storage decommissioning feature on Kubernetes with a distributed
> filesystem as the fallback storage might run into the situation where an
> executor cannot see the shuffle data on the distributed filesystem that has
> just been written by the decommissioned executor. This is caused by some
> replication delay. Given the dependent executor knows the location of the
> shuffle data is the fallback storage, it can defer reading on a
> {{FileNotFoundException}}.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
