From: Ronan Dunklau <ronan.dunklau@aiven.io>
To: pgsql-hackers@lists.postgresql.org, David Rowley <dgrowleyml@gmail.com>
Cc: Ranier Vilela <ranier.vf@gmail.com>
Subject: Re: Add proper planner support for ORDER BY / DISTINCT aggregates
Date: Thu, 15 Jul 2021 15:02:00 +0200
Message-ID: <2746808.Lq6AjvN2yG@aivenronan>
Organization: aiven
In-Reply-To: 
 <CAApHDvrqnL3U=H6KznCsqtROGTgVT0d3rT0p7zDbkbXb-Fy8kA@mail.gmail.com>
References: 
 <CAApHDvpHzfo92=R4W0+xVua3BUYCKMckWAmo-2t_KiXN-wYH=w@mail.gmail.com>
 <CAEudQAq-H2yNzCni1N_PW7PfrWr4jYMhzRAiQiGyAQ4vt4-sTg@mail.gmail.com>
 <CAApHDvrqnL3U=H6KznCsqtROGTgVT0d3rT0p7zDbkbXb-Fy8kA@mail.gmail.com>
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable
Content-Type: text/plain; charset="iso-8859-1"
Archived-At: 
 <https://www.postgresql.org/message-id/2746808.Lq6AjvN2yG%40aivenronan>
Precedence: bulk

Le mardi 13 juillet 2021, 06:44:12 CEST David Rowley a =E9crit :
> I've attached the updated patches.

The approach of building a pathkey for the first order by we find, then=20
appending to it as needed seems sensible but I'm a bit worried about users=
=20
starting to rely on this as an optimization. Even if we don't document it,=
=20
people may start to change the order of their target lists to "force" a=20
specific sort on the lower nodes. How confident are we that we won't change=
 this=20
or that we will be willing to break it ?

Generating all possible pathkeys and costing the resulting plans would be t=
oo=20
expensive, but maybe a more "stable" (and limited) approach would be fine, =
like=20
generating the pathkeys only if every ordered aggref shares the same prefix=
=2E I=20
don't think there would be any ambiguity here.

=2D-=20
Ronan Dunklau