[jira] [Commented] (IMPALA-10229) Analytic limit pushdown optimization can be applied incorrect when there are no analytic predicates

["Aman Sinha (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IMPALA-10229?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17210503#comment-17210503 ] 

Aman Sinha commented on IMPALA-10229:
-------------------------------------

Good catch !  I should have caught it during testing but it looks like the test results for the test with no filter condition is likely wrong. The prefix checking needs to be done for the outer order-by columns.  

> Analytic limit pushdown optimization can be applied incorrect when there are no analytic predicates
> ---------------------------------------------------------------------------------------------------
>
>                 Key: IMPALA-10229
>                 URL: https://issues.apache.org/jira/browse/IMPALA-10229
>             Project: IMPALA
>          Issue Type: Bug
>          Components: Frontend
>            Reporter: Tim Armstrong
>            Assignee: Tim Armstrong
>            Priority: Blocker
>              Labels: correctness
>
> {noformat}
> [localhost.EXAMPLE.COM:21050] default> select * from (select month, id, rank() over (partition by month order by id desc) rnk from functional_parquet.alltypes WHERE month >= 11) v order by month, id limit 3;
> +-------+------+-----+
> | month | id   | rnk |
> +-------+------+-----+
> | 11    | 6987 | 3   |
> | 11    | 6988 | 2   |
> | 11    | 6989 | 1   |
> +-------+------+-----+
> Fetched 3 row(s) in 4.16s
> {noformat}
> These are not the top 3 rows when ordering by month, id . Hive's result is correct:
> {noformat}
> +----------+-------+--------+
> | v.month  | v.id  | v.rnk  |
> +----------+-------+--------+
> | 11       | 3040  | 600    |
> | 11       | 3041  | 599    |
> | 11       | 3042  | 598    |
> +----------+-------+--------+
> {noformat}
> I think when there's no select predicates, that the ordering in the analytic sort needs to exactly match the TOP N sort ordering. I'm not sure if there are fixes needed for the case where there are select predicates.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscribe@impala.apache.org
For additional commands, e-mail: issues-all-help@impala.apache.org