From: Tom Lane <tgl@sss.pgh.pa.us>
To: Timothy Garnett <tgarnett@panjiva.com>
cc: pgsql-performance@lists.postgresql.org
Subject: Re: Is it possible to specify minimum number of rows planner should
 consider?
In-reply-to: 
 <CAPcyiQ2Nbw+caY0y+Gywf+58dmVkpuqPSqJuDLcoL90DbnwA7w@mail.gmail.com>
References: 
 <CAPcyiQ2Nbw+caY0y+Gywf+58dmVkpuqPSqJuDLcoL90DbnwA7w@mail.gmail.com>
Comments: In-reply-to Timothy Garnett <tgarnett@panjiva.com>
	message dated "Mon, 28 Sep 2020 16:12:37 -0400"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <408271.1601327211.1@sss.pgh.pa.us>
Date: Mon, 28 Sep 2020 17:06:51 -0400
Message-ID: <408272.1601327211@sss.pgh.pa.us>
Precedence: bulk

Timothy Garnett <tgarnett@panjiva.com> writes:
> Is there some way to tell the planner that unless it's guaranteed by a
> constraint or some such it shouldn't guess that the selectivity of a
> filter/anti-join is 1 row (e.g. minimum to consider is 2 rows unless it's
> guaranteed to be 1 row) or somehow otherwise make it more conservative
> around the worst case possibilities.

There's been some discussion in that area, but it's a hard problem
to solve in general, and especially so if you'd like to not break
a ton of queries that work nicely today.

			regards, tom lane