[ 
https://issues.apache.org/jira/browse/HIVE-14530?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15449405#comment-15449405
 ] 

Jesus Camacho Rodriguez commented on HIVE-14530:
------------------------------------------------

HIVE-13639 extended the logic to pull up predicates (and thus constants) 
through Union operators.

Originally, we were extracting them as a single disjunction of all the 
predicates in each input. For instance, for a Union with _n_ inputs:
_[pred1 OR pred2 OR ... predn]_ 

Instead, we factorize and identify common elements:
_[pred1, ..., predx, residualpred1 OR residualpred2 OR ... residualpredn]_
where _pred1, ..., predx_ are common elements found in every Union input, while 
_residualpred1 OR residualpred2 OR ... residualpredn_ are preds found only in 
input _1, 2, ..., n_, respectively.

This helped us identifying more expressions to pull up, and reducing the 
complexity of doing so. However, there was a bug that was making us pull up 
expressions even if they had not been found in every union input. This patch 
fixes that issue.

> Union All query returns incorrect results.
> ------------------------------------------
>
>                 Key: HIVE-14530
>                 URL: https://issues.apache.org/jira/browse/HIVE-14530
>             Project: Hive
>          Issue Type: Bug
>          Components: Query Planning
>    Affects Versions: 2.1.0
>         Environment: Hadoop 2.6
> Hive 2.1
>            Reporter: wenhe li
>            Assignee: Jesus Camacho Rodriguez
>         Attachments: HIVE-14530.01.patch
>
>
> create table dw_tmp.l_test1 (id bigint,val string,trans_date string) row 
> format delimited fields terminated by ' ' ;
> create table dw_tmp.l_test2 (id bigint,val string,trans_date string) row 
> format delimited fields terminated by ' ' ;  
> select * from dw_tmp.l_test1;
> 1       table_1      2016-08-11
> select * from dw_tmp.l_test2;
> 2       table_2      2016-08-11
> -- right like this
> select 
>     id,
>     'table_1' ,
>     trans_date
> from dw_tmp.l_test1
> union all
> select 
>     id,
>     val,
>     trans_date
> from dw_tmp.l_test2 ;
> 1       table_1     2016-08-11
> 2       table_2     2016-08-11
> -- incorrect
> select 
>     id,
>     999,
>     'table_1' ,
>     trans_date
> from dw_tmp.l_test1
> union all
> select 
>     id,
>     999,
>     val,
>     trans_date
> from dw_tmp.l_test2 ;
> 1       999     table_1     2016-08-11
> 2       999     table_1     2016-08-11     <-- here is wrong
> -- incorrect
> select 
>     id,
>     999,
>     666,
>     'table_1' ,
>     trans_date
> from dw_tmp.l_test1
> union all
> select 
>     id,
>     999,
>     666,
>     val,
>     trans_date
> from dw_tmp.l_test2 ;
> 1       999     666     table_1 2016-08-11
> 2       999     666     table_1 2016-08-11     <-- here is wrong
> -- right
> select 
>     id,
>     999,
>     'table_1' ,
>     trans_date,
>     '2016-11-11'
> from dw_tmp.l_test1
> union all
> select 
>     id,
>     999,
>     val,
>     trans_date,
>     trans_date
> from dw_tmp.l_test2 ;
> 1       999     table_1 2016-08-11      2016-11-11
> 2       999     table_2 2016-08-11      2016-08-11



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to