Re: BUG #15857: Parallel Hash Join makes join instead of exists

Thomas Munro <thomas.munro@gmail.com>

From: Thomas Munro <thomas.munro@gmail.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: krujkov@gmail.com, PostgreSQL mailing lists <pgsql-bugs@lists.postgresql.org>
Date: 2019-06-18T14:38:41Z
Lists: pgsql-bugs
On Wed, Jun 19, 2019 at 2:05 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> OK to push as far as I'm concerned.  I am not expecting to have to
> re-wrap given the lack of packagers complaints so far, and if we
> do re-wrap, I'd be OK with including this fix.

Done.

Vladimir, thanks for the report!  Unfortunately the fix is not
included in the out-of-schedule 11.4 release that's about to be made
(unless something triggers a re-wrap), but you should expect it in the
next regular release[1].  In the meantime you might be able to work
around the problem by using a different but equivalent join condition
so you get a semi-join, or by turning off enable_parallel_hash.  Even
with the fix you'll probably get better performance if you can
convince it to use a semi-join.

[1] https://www.postgresql.org/developer/roadmap/

-- 
Thomas Munro
https://enterprisedb.com



Commits

  1. Prevent Parallel Hash Join for JOIN_UNIQUE_INNER.