Message-ID: <50AD0EA0.7070009@joeconway.com>
Date: Wed, 21 Nov 2012 09:25:52 -0800
From: Joe Conway <mail@joeconway.com>
User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US;
	rv:1.8.0.12) Gecko/20071019 Fedora/1.5.0.12-3.fc6 pango-text
	Thunderbird/1.5.0.12 Mnenhy/0.7.5.666
MIME-Version: 1.0
To: Heikki Linnakangas <hlinnakangas@vmware.com>
CC: Kevin Grittner <kgrittn@mail.com>, Craig Ringer <craig@2ndQuadrant.com>,
	Craig James <cjames@emolecules.com>, Tom Lane <tgl@sss.pgh.pa.us>,
	Claudio Freire <klaussfreire@gmail.com>,
	Merlin Moncure <mmoncure@gmail.com>,
	David Greco <David_Greco@harte-hanks.com>, pgsql-performance@postgresql.org
Subject: Re: Hints (was Poor performance using CTE)
References: <20121121134232.68590@gmx.com> <50ACFBBD.30204@vmware.com>
In-Reply-To: <50ACFBBD.30204@vmware.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit

On 11/21/2012 08:05 AM, Heikki Linnakangas wrote:
> Rather than telling the planner what to do or not to do, I'd much rather
> have hints that give the planner more information about the tables and
> quals involved in the query. A typical source of bad plans is when the
> planner gets its cost estimates wrong. So rather than telling the
> planner to use a nested loop join for "a INNER JOIN b ON a.id = b.id",
> the user could tell the planner that there are only 10 rows that match
> the "a.id = b.id" qual. That gives the planner the information it needs
> to choose the right plan on its own. That kind of hints would be much
> less implementation specific and much more likely to still be useful, or
> at least not outright counter-productive, in a future version with a
> smarter planner.
> 
> You could also attach that kind of hints to tables and columns, which
> would be more portable and nicer than decorating all queries.

I like this idea, but also think that if we have a syntax to allow
hints, it would be nice to have a simple way to ignore all hints (yes, I
suppose I'm suggesting yet another GUC). That way after sprinkling your
SQL with hints, you could easily periodically (e.g. after a Postgres
upgrade) test what would happen if the hints were removed.

Joe
-- 
Joe Conway
credativ LLC: http://www.credativ.us
Linux, PostgreSQL, and general Open Source
Training, Service, Consulting, & 24x7 Support