[
https://issues.apache.org/jira/browse/HIVE-17934?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16245252#comment-16245252
]
Zoltan Haindrich commented on HIVE-17934:
-----------------------------------------
[~ashutoshc] that sounds clearly bad :) so if there is a change like that in
the patch; that's by mistake.
usually the following pattern causes q.out changes:
* one of the TS / or parents were PARTIAL; and now the child is also PARTIAL
* one of the TS was scanning a partitioned table which had an empty partition -
lately I'm starting to question wether this should be done.
there are many stat changes in this patch; part of it is there because I've
uncovered them during this patch - this patch could benefit from those smaller
stat improvements (lesser q.out changes) - should I change the order; and try
to fix those first?
> Merging Statistics are promoted to COMPLETE (most of the time)
> --------------------------------------------------------------
>
> Key: HIVE-17934
> URL: https://issues.apache.org/jira/browse/HIVE-17934
> Project: Hive
> Issue Type: Bug
> Components: Statistics
> Reporter: Zoltan Haindrich
> Assignee: Zoltan Haindrich
> Attachments: HIVE-17934.01.patch, HIVE-17934.02.patch,
> HIVE-17934.03.patch, HIVE-17934.04.patch, HIVE-17934.05.patch,
> HIVE-17934.06wip01.patch
>
>
> in case multiple partition statistics are merged the STATS state is computed
> based on the datasize and rowcount;
> the merge may hide away non-existent stats in case there are other partition
> or operators which do contribute to the datasize and the rowcount.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
