[
https://issues.apache.org/jira/browse/HIVE-10456?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14509487#comment-14509487
]
Gunther Hagleitner commented on HIVE-10456:
-------------------------------------------
This doesn't seem right - the clear method you use leaves a broken partition
behind right? You clean up some stuff but don't nuke the whole container. I
think the logic should be: if it has any spilled partitions, throw away the
whole container and make sure it's not in the cache (shouldn't be). If there
are no partitions spilled leave alone for reuse.
> Grace Hash Join should not load spilled partitions on abort
> -----------------------------------------------------------
>
> Key: HIVE-10456
> URL: https://issues.apache.org/jira/browse/HIVE-10456
> Project: Hive
> Issue Type: Bug
> Affects Versions: 1.2.0
> Reporter: Prasanth Jayachandran
> Assignee: Prasanth Jayachandran
> Attachments: HIVE-10456.1.patch
>
>
> Grace Hash Join loads the spilled partitions to complete the join in
> closeOp(). This should not happen when closeOp with abort is invoked. Instead
> it should clean up all the spilled data.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
