Message-ID: <50B57440.1040802@2ndquadrant.com>
Date: Tue, 27 Nov 2012 18:17:36 -0800
From: Craig Ringer <craig@2ndquadrant.com>
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64;
	rv:15.0) Gecko/20120824 Thunderbird/15.0
MIME-Version: 1.0
To: Scott Marlowe <scott.marlowe@gmail.com>
CC: cedric@2ndquadrant.com, pgsql-performance@postgresql.org,
	Craig James <cjames@emolecules.com>,
	Kevin Grittner <kgrittn@mail.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>
Subject: Re: Hints (was Poor performance using CTE)
References: <20121121134232.68590@gmx.com>
	<CAFwQ8reE2vLynnE9zO_i7WhmktbiKaSYbBrCvJXCQ62feqBGZQ@mail.gmail.com>
	<201211231105.56380.cedric@2ndquadrant.com>
	<CAOR=d=1W8WqugTQBEe+kDUZbxBMXAfw413WzBBE1eOLD+A_mbw@mail.gmail.com>
In-Reply-To: 
 <CAOR=d=1W8WqugTQBEe+kDUZbxBMXAfw413WzBBE1eOLD+A_mbw@mail.gmail.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk
Sender: pgsql-performance-owner@postgresql.org

On 27/11/2012 3:42 PM, Scott Marlowe wrote:

> Here here!  PostgreSQL is well known for its extensibility and this is
> the perfect place for hints.
I agree with the sentiment and your concerns. However, this doesn't 
solve the CTE problem.

Some people are relying on the planner's inability to push conditions 
into / pull conditions out of CTEs, and otherwise re-arrange them. If 
support for optimising into eligible CTEs (ie CTE terms that contain 
only SELECT or VALUES and call no VOLATILE functions) then these 
applications will potentially encounter serious performance regressions.

Should this feature never be added to Pg, making it different and 
incompatible with other DBs that implement CTE optimisation, just 
because some people are using it for a hacky hint like OFFSET 0?

Should these applications just be broken by the update, with people told 
to add `OFFSET 0` or load some not-yet-existing hints module after 
reporting the performance issue to the list?

I don't think either of those are acceptable. Sooner or later somebody's 
going to want to add CTE optimisation, and I don't think that "you 
can't" or "great, we'll do it and break everything" are acceptable 
responses to any proposed patch someone might come up with to add that.

A GUC might be OK, as apps can always SET it before problem queries or 
not-yet-ported code. It'd probably reduce the rate at which people fixed 
their code considerably, though, going by past experience with 
standard_conforming_strings, etc, but it'd work.

--
Craig Ringer


-- 
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance