Date: Sat, 01 Oct 2022 00:19:16 +0200
Subject: Re: disfavoring unparameterized nested loops
From: Benjamin Coutu <ben.coutu@zeyos.com>
To: Peter Geoghegan <pg@bowt.ie>
Cc: Tom Lane <tgl@sss.pgh.pa.us>,
	Bruce Momjian <bruce@momjian.us>,
	PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>,
	Robert Haas <robertmhaas@gmail.com>,
	David Rowley <dgrowleyml@gmail.com>
Message-ID: <bb1b14369bb1d8fc55ca@zeyos.com>
In-Reply-To: 
 <CAH2-Wz=DtkWYzJAC9B6TOUMnNZ4e=mQFDqZ12EVoK=9qkX60GA@mail.gmail.com>
References: 
 <CAH2-Wz=DtkWYzJAC9B6TOUMnNZ4e=mQFDqZ12EVoK=9qkX60GA@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain;
	charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Archived-At: 
 <https://www.postgresql.org/message-id/bb1b14369bb1d8fc55ca%40zeyos.com>
Precedence: bulk

> Sure, but the model isn't the problem here, really -- not to me. The
> problem is that the planner can in some cases choose a plan that is
> inherently unreasonable, at least in practical terms. You're talking
> about uncertainties. But I'm actually talking about the opposite thing
> -- certainty (albeit a limited kind of certainty that applies only to
> one narrow set of conditions).
=0AI absolutely agree and support your proposal to simply not generate thos=
e paths at all unless necessary.=0A=0A> For all I know you might be onto so=
mething. But it really seems
> independent to me.
=0AYeah, I=E2=80=98m sorry if I highjacked this thread for something relate=
d but technically different. I just wanted to expand on your proposal by ta=
king into account the join depth and also not just talking about unparametr=
ized nested loop joins. The research is very clear that the uncertainty is =
proportional to the join level, and that is the point I am trying to focus =
the discussion on.=0A=0AI really encourage everyone to read the VLDB paper.=
 BTW, the unnamed proprietary DBMSs in that paper are the 3 big ones from W=
ashington, California and NY, in that order.