To: Mario Weilguni <mweilguni@sime.com>
cc: pgsql-performance@postgresql.org
Subject: Re: Very slow INFORMATION_SCHEMA 
In-reply-to: <481EEFD4.3060203@sime.com> 
References: <481B82AE.8050607@gmail.com> <26118.1209766858@sss.pgh.pa.us>
	<481EEFD4.3060203@sime.com>
Comments: In-reply-to Mario Weilguni <mweilguni@sime.com>
	message dated "Mon, 05 May 2008 13:30:28 +0200"
Date: Mon, 05 May 2008 20:56:26 -0400
Message-ID: <5792.1210035386@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Mario Weilguni <mweilguni@sime.com> writes:
> I can confirm this for a quite larger result set (4020 rows) for a DB 
> with 410 tables and a lot of foreign key constraints.
> ...
> This is Postgresql 8.2.4, on a Dual-Core XEON 3.6GHz. With nested_loops 
> off, I get a very fast response (330ms).

FWIW, it looks like 8.3 is significantly smarter about this example
--- it's able to push the toplevel conditions on CONSTRAINT_SCHEMA
and CONSTRAINT_TYPE down inside the UNION, where 8.2 fails to do so.

Which is not to say that there's not more left to do on optimizing
the information_schema views.  In this particular case, for example,
I wonder why the UNION in INFORMATION_SCHEMA.TABLE_CONSTRAINTS isn't a
UNION ALL.  There's probably a lot more such micro-optimizations that
could be done if anyone was motivated to look at it.

			regards, tom lane