Re: GIN index not used

-----Original Message-----
From: Tom Lane [mailto:tgl(at)sss(dot)pgh(dot)pa(dot)us]
Sent: Friday, July 11, 2014 3:43 PM
To: Huang, Suya
Cc: Andreas Kretschmer; pgsql-performance(at)postgresql(dot)org
Subject: Re: [PERFORM] GIN index not used

"Huang, Suya" <Suya(dot)Huang(at)au(dot)experian(dot)com> writes:
> Just found out something here
> http://www.postgresql.org/message-id/17021.1234474178@sss.pgh.pa.us
> So I dropped the index and recreate it by specifying: using gin(terms_ts gin__int_ops) and the index works.

Oh, you're using contrib/intarray?

Pursuant to the thread you mention above, we removed intarray's <@ and @> operators (commit 65e758a4d3) but then reverted that (commit 156475a589) because of backwards-compatibility worries. It doesn't look like anything got done about it since then. Perhaps the extension upgrade infrastructure would offer a solution now.

> My PG version is 9.3.4, none-default planner settings:
> enable_mergejoin = off
> enable_nestloop = off

[ raised eyebrow... ] It's pretty hard to see how those would be a good idea. Not all problems are best solved by hash joins.

regards, tom lane

About the contrib/intarray, do I have other choices not using that one?

About the join, yeah, in our testing for DW-like queries, hash join does improved the performance greatly...

Thanks,
Suya