drill-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Deneche A. Hakim (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (DRILL-4449) Wrong results when using metadata cache with specific set of queries
Date Tue, 08 Mar 2016 15:58:40 GMT

    [ https://issues.apache.org/jira/browse/DRILL-4449?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15185131#comment-15185131
] 

Deneche A. Hakim commented on DRILL-4449:
-----------------------------------------

a clarification here. To have a reproduction the table must be partitioned and referenced
in both inner queries with different filters. The filters need to trigger a parquet partition
pruning and leave more than one file after the pruning.

> Wrong results when using metadata cache with specific set of queries
> --------------------------------------------------------------------
>
>                 Key: DRILL-4449
>                 URL: https://issues.apache.org/jira/browse/DRILL-4449
>             Project: Apache Drill
>          Issue Type: Bug
>          Components: Storage - Parquet
>    Affects Versions: 1.5.0
>            Reporter: Deneche A. Hakim
>            Assignee: Deneche A. Hakim
>            Priority: Critical
>             Fix For: 1.6.0
>
>
> We are still working on a reproduction but when we have a query similar to this one:
> {noformat}
> with q1 as (
> select a.field
> from `table` a
> where <some condition that causes the table to be pruned>
> group by a.field
> having ...
> )
> , q2 as (
> select a.field
> from `table` a
> where <some other pruning condition>
> group by a.field
> )
> select * from (
> select count(*) as cnt from q1
> union all
> select count(*) as cnt from q2
> );
> {noformat}
> The table is partitioned and both sub queries will force a parquet pruning on the table.
Because we share the parquet metadata object in ParquetGroupScan, the second query end up
being "over pruned" and we get wrong results.
> The plan doesn't show the problem.



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

Mime
View raw message