[
https://issues.apache.org/jira/browse/HIVE-20585?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16788578#comment-16788578
]
Zoltan Haindrich commented on HIVE-20585:
-----------------------------------------
resetting the col stats cache is just bad...it will go south in production but
will pass in the tests;
the cache is incorrectly returning bad results - that's the real issue here; it
could also cause cbo off - so we should fix the incorrect caching.
please restore the original title I think the summary should reflect what the
issue is; and not what the actual action item in the patch.
> Reset aggregate column stat cache before each q test
> ----------------------------------------------------
>
> Key: HIVE-20585
> URL: https://issues.apache.org/jira/browse/HIVE-20585
> Project: Hive
> Issue Type: Bug
> Components: Tests
> Reporter: Zoltan Haindrich
> Assignee: Miklos Gergely
> Priority: Major
> Attachments: HIVE-20585.01.patch, HIVE-20585.02.patch
>
>
> If column stats are fetched(HIVE-17084) then list_bucket_dml_4.q's output is
> flactuating; sometimes it has column stats COMPLETE; sometimes just partial.
> Running locally produces COMPLETE.
> running together with join33.q,list_bucket_dml_4.q cause it to degrade to
> PARTIAL
> The issue is that the aggregate column stat cache may store statistics for
> common tables in the q tests like src, src1, srcpart etc, which may affect
> wether the aggregate stats received would be complete or partial.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
