[ 
https://issues.apache.org/jira/browse/HIVE-19326?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16475802#comment-16475802
 ] 

Zoltan Haindrich commented on HIVE-19326:
-----------------------------------------

reviewboard: https://reviews.apache.org/r/67126/
[~ashutoshc] it's not possible to get rid of tezIndexForUnion because operator 
ids are being reused in {{from ( SEL ... UNION ALL SEL ... ) insert  into ... 
}} queries 

I think HIVE-19237 should probably restore consistency in cloneOperatorTree as 
well; and then it would be okay to just remove that index

> stats auto gather: incorrect aggregation during UNION queries (may lead to 
> incorrect results)
> ---------------------------------------------------------------------------------------------
>
>                 Key: HIVE-19326
>                 URL: https://issues.apache.org/jira/browse/HIVE-19326
>             Project: Hive
>          Issue Type: Bug
>          Components: Statistics
>            Reporter: Sergey Shelukhin
>            Assignee: Zoltan Haindrich
>            Priority: Critical
>         Attachments: HIVE-19326.01wip01.patch, HIVE-19326.02.patch, 
> HIVE-19326.03.patch, HIVE-19326.04.patch
>
>
> Found when investigating the results change after converting tables to MM, 
> turns out the MM result is correct but the current one is not.
> The test ends like so:
> {noformat}
> desc formatted small_alltypesorc_a;
> ANALYZE TABLE small_alltypesorc_a COMPUTE STATISTICS;
> desc formatted small_alltypesorc_a;
> insert into table small_alltypesorc_a select * from small_alltypesorc1a;
> desc formatted small_alltypesorc_a;
> {noformat}
> The results from the descs in the golden file are:
> {noformat}
>       COLUMN_STATS_ACCURATE   {\"BASIC_STATS\":\"true\"}
>       numFiles                1                   
>       numRows                 5                               
> ...
>       COLUMN_STATS_ACCURATE   {\"BASIC_STATS\":\"true\"}
>       numFiles                1                   
>       numRows                 15                                
> ...
>       COLUMN_STATS_ACCURATE   {\"BASIC_STATS\":\"true\"}
>       numFiles                2                   
>       numRows                 20                              
> {noformat}
> Note the result change after analyze - the original nomRows is inaccurate, 
> but  BASIC_STATS is set to true.
> I am assuming with metadata only optimization this can produce incorrect 
> results.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to