[
https://issues.apache.org/jira/browse/DRILL-6755?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16757752#comment-16757752
]
Robert Hou commented on DRILL-6755:
-----------------------------------
Boaz suggested verifying this by joining with an empty file.
{noformat}
select count(*) from dfs.`/empty.json` E where E.l_orderkey in (select
L.l_orderkey from lineitem L);
{noformat}
I tested this with Drill 1.15. I had to turn off semijoins to get the desired
plan because if a semijoin is used, then the join is re-ordered so that the
empty file is on the build side (may be a bug).
I was able to verify that the hash join operator does not build a hash table
for this query.
> HashJoin should not build hash tables when probe side is empty.
> ---------------------------------------------------------------
>
> Key: DRILL-6755
> URL: https://issues.apache.org/jira/browse/DRILL-6755
> Project: Apache Drill
> Issue Type: Improvement
> Reporter: Timothy Farkas
> Assignee: Boaz Ben-Zvi
> Priority: Major
> Labels: ready-to-commit
> Fix For: 1.15.0
>
>
> Currently when doing an Inner or a Right join we still build hashtables when
> the probe side is empty. A performance optimization would be to not build
> them.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
