To: Erik Jones <erik@myemma.com>
cc: pgsql-performance@postgresql.org
Subject: Re: More 8.2 client issues (Was: [Slow dump?) 
In-reply-to: <459BDC98.8030008@myemma.com> 
References: <459AD5A8.7030102@myemma.com> <22512.1167779361@sss.pgh.pa.us>
	<459BD61C.3030202@myemma.com> <11617.1167841479@sss.pgh.pa.us>
	<459BDC98.8030008@myemma.com>
Comments: In-reply-to Erik Jones <erik@myemma.com>
	message dated "Wed, 03 Jan 2007 10:40:56 -0600"
Date: Wed, 03 Jan 2007 11:56:20 -0500
Message-ID: <12515.1167843380@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Erik Jones <erik@myemma.com> writes:
> Tom Lane wrote:
>> I could see this taking an unreasonable amount of time if you had a huge
>> number of pg_class rows or a very long search_path --- is your database
>> at all out of the ordinary in those ways?
>> 
> Well, running "select count(*) from pg_class;" returns 524699 rows

Ouch.

> our search path is the default.  I'd also like to reiterate that \d 
> pg_class returns instantly when run from the 8.1.4 psql client connected 
> to the 8.2 db.

I think I know where the problem is: would you compare timing of

	select * from pg_class where c.relname ~ '^(pg_class)$';

	select * from pg_class where c.relname ~ '^pg_class$';

Recent versions of psql put parentheses into the regex pattern for
safety in case it's got "|", but I just realized that that probably
confuses the optimizer's check for an indexable regex :-(

However, this only explains slowdown in psql's \d commands, which
wasn't your original complaint ...

			regards, tom lane