[jira] [Commented] (HIVE-12165) wrong result when hive.optimize.sampling.orderby=true with some aggregate functions

["Aihua Xu (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-12165?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15134435#comment-15134435 ] 

Aihua Xu commented on HIVE-12165:
---------------------------------

Internally when the query is compiled, an order by is added and we are trying to do a parallel order by for this case. Ideally maybe we can do another reduce to merge the result. But that probably is big change. I will force it to use 1 reducer for this case just the same as hive.optimize.sampling.orderby=false right now.

> wrong result when hive.optimize.sampling.orderby=true with some aggregate functions
> -----------------------------------------------------------------------------------
>
>                 Key: HIVE-12165
>                 URL: https://issues.apache.org/jira/browse/HIVE-12165
>             Project: Hive
>          Issue Type: Bug
>         Environment: hortonworks  2.3
>            Reporter: ErwanMAS
>            Assignee: Aihua Xu
>            Priority: Critical
>
> This simple query give wrong result , when , i use the parallel order .
> {noformat}
> select count(*) , count(distinct dummyint ) , min(dummyint),max(dummyint) from foobar_1M ;
> {noformat}
> Current wrong result :
> {noformat}
> c0	c1	c2	c3
> 32740	32740	0	163695
> 113172	113172	163700	729555
> 54088	54088	729560	999995
> {noformat}
> Right result :
> {noformat}
> c0	c1	c2	c3
> 1000000	1000000	0	999999
> {noformat}
> The sql script for my test 
> {noformat}
> drop table foobar_1 ;
> create table foobar_1 ( dummyint int  , dummystr string ) ;
> insert into table foobar_1 select count(*),'dummy 0'  from foobar_1 ;
> drop table foobar_1M ;
> create table foobar_1M ( dummyint bigint  , dummystr string ) ;
> insert overwrite table foobar_1M
>        select val_int  , concat('dummy ',val_int) from
>              ( select ((((((d_1*10)+d_2)*10+d_3)*10+d_4)*10+d_5)*10+d_6) as val_int from foobar_1
>                  lateral view outer explode(split("0,1,2,3,4,5,6,7,8,9",",")) tbl_1 as d_1
>                  lateral view outer explode(split("0,1,2,3,4,5,6,7,8,9",",")) tbl_2 as d_2
>                  lateral view outer explode(split("0,1,2,3,4,5,6,7,8,9",",")) tbl_3 as d_3
>                  lateral view outer explode(split("0,1,2,3,4,5,6,7,8,9",",")) tbl_4 as d_4
>                  lateral view outer explode(split("0,1,2,3,4,5,6,7,8,9",",")) tbl_5 as d_5
>                  lateral view outer explode(split("0,1,2,3,4,5,6,7,8,9",",")) tbl_6 as d_6  ) as f                 ;
> set hive.optimize.sampling.orderby.number=10000;
> set hive.optimize.sampling.orderby.percent=0.1f;
> set mapreduce.job.reduces=3 ;
> set hive.optimize.sampling.orderby=false;
> select count(*) , count(distinct dummyint ) , min(dummyint),max(dummyint) from foobar_1M ;
> set hive.optimize.sampling.orderby=true;
> select count(*) , count(distinct dummyint ) , min(dummyint),max(dummyint) from foobar_1M ;
> {noformat}



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