[
https://issues.apache.org/jira/browse/SPARK-32649?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17179385#comment-17179385
]
Cheng Su commented on SPARK-32649:
----------------------------------
[~cloud_fan] - sorry for ping as this is not major improvement. Could you let
me know if this is worth adding in upstream? The code change is pretty minor.
If yes, I can work on a PR for it. cc [~leanken] as well FYI thanks.
> Optimize BHJ/SHJ inner and semi join with empty hashed relation
> ---------------------------------------------------------------
>
> Key: SPARK-32649
> URL: https://issues.apache.org/jira/browse/SPARK-32649
> Project: Spark
> Issue Type: Sub-task
> Components: SQL
> Affects Versions: 3.1.0
> Reporter: Cheng Su
> Priority: Trivial
>
> With `EmptyHashedRelation` introduced in
> [https://github.com/apache/spark/pull/29389], it inspired me that there's a
> minor optimization we can apply to broadcast hash join and shuffled hash join
> if build side hashed relation is empty.
> If build side hashed relation is empty (i.e. build side is empty)
> 1.inner join: we don't need to execute stream side at all, just return an
> empty iterator -
> [https://github.com/apache/spark/blob/master/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/HashJoin.scala#L152]
> 2.semi join: we don't need to execute stream side at all, just return an
> empty iterator -
> [https://github.com/apache/spark/blob/master/sql/core/src/main/scala/org/apache/spark/sql/execution/joins/HashJoin.scala#L227]
> .
> This is not common that build side is empty, but in case it is, we can
> leverage it to not execute stream side at all for better query CPU/IO
> performance.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
