[jira] [Commented] (HIVE-3891) physical optimizer changes for auto sort-merge join

["Namit Jain (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-3891?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13573240#comment-13573240 ] 

Namit Jain commented on HIVE-3891:
----------------------------------

After HIVE-3403, an auto sortmerge join can be performed.
However, consider the following scenario:
T1 - big table: 3 partitions : big partitions
T2 - small table: 1 partition
both T1 and T2 are sorted/bucketed on the same key.
If a join is being performed between T1 and T2, with HIVE-3403, T2 would be picked up as the table on which the mapper would run.
Ideally, if T2 is so small that it can fit in memory, a better approach would be map-join. 
The physical optimizer should change the sort-merge join task into a backup task, with all tables being candidates for map-join
(similar to auto convert mapjoin).
                
> physical optimizer changes for auto sort-merge join
> ---------------------------------------------------
>
>                 Key: HIVE-3891
>                 URL: https://issues.apache.org/jira/browse/HIVE-3891
>             Project: Hive
>          Issue Type: Bug
>            Reporter: Namit Jain
>            Assignee: Namit Jain
>


--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira