[
https://issues.apache.org/jira/browse/DRILL-3327?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14598533#comment-14598533
]
Aman Sinha commented on DRILL-3327:
-----------------------------------
Yes, the problem could potentially manifest if a single partition spans
multiple batches. If you are making some code changes, you might want to test
with smaller direct memory (maybe 1-2GB) such that you get multiple batches
output from the Sort. Also, it is useful to test with partition-by where the
pb column has a single value:
{code}
create table tt1 as select 5 as const, c_integer from j1 ;
select row_number() over (partition by const) from tt1 order by 1;
{code}
> row_number function returns incorrect result when only order by clause is
> specified
> -----------------------------------------------------------------------------------
>
> Key: DRILL-3327
> URL: https://issues.apache.org/jira/browse/DRILL-3327
> Project: Apache Drill
> Issue Type: Bug
> Components: Query Planning & Optimization
> Affects Versions: 1.0.0
> Reporter: Victoria Markman
> Assignee: Aman Sinha
> Priority: Critical
> Labels: window_function
> Fix For: 1.1.0
>
> Attachments:
> 0001-DRILL-3327-If-no-partition-by-is-specified-read-all-.patch, j1.tar,
> row_number_wrong_result.tar
>
>
> Queries returning wrong result:
> {code}
> select c_integer, row_number() over(order by c_integer) from j1 order by 2;
> select c_integer, row_number() over(order by c_integer) from j1 order by 2
> desc;
> select c_integer, row_number() over(order by c_integer desc) from j1 order by
> 2;
> select c_integer, row_number() over(order by c_integer desc) from j1 order by
> 2;
> select c_integer, row_number() over(order by c_integer nulls first) from j1
> order by c_integer nulls last, row_number() over(order by c_integer nulls
> first);
> select c_integer, row_number() over(order by c_integer nulls last) from j1
> order by c_integer nulls first, row_number() over(order by c_integer nulls
> last);
> {code}
> In attached tar file (row_number_wrong_result.tar) you will find:
> 1. *.sql - file that contains query
> 2. *.res - expected result (generated from Postgres)
> 3. *.out - result returned by drill
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
