[jira] [Commented] (DRILL-7141) Hash-Join (and Agg) should always spill to disk the least used partition

["Boaz Ben-Zvi (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/DRILL-7141?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16967208#comment-16967208 ] 

Boaz Ben-Zvi commented on DRILL-7141:
-------------------------------------

This enhancement requires the use of good statistics (on the probe-side), as spilling decisions (i.e., which partitions) happen during the build phase.

 

> Hash-Join (and Agg) should always spill to disk the least used partition
> ------------------------------------------------------------------------
>
>                 Key: DRILL-7141
>                 URL: https://issues.apache.org/jira/browse/DRILL-7141
>             Project: Apache Drill
>          Issue Type: Improvement
>          Components: Execution - Relational Operators
>    Affects Versions: 1.15.0
>            Reporter: Kunal Khatua
>            Assignee: Boaz Ben-Zvi
>            Priority: Major
>
> When the probe-side data for a hash join is skewed, it is preferable to have the corresponding partition on the build side to be in memory. 
> Currently, with the spill-to-disk feature, the partition selected for spilling to disk is done at random. This means that a highly skewed probe-side data would also spill for lack of a corresponding hash table partition in memory. 



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