From: Tom Lane <tgl@sss.pgh.pa.us>
To: Jeremy Altavilla <jeremyaltavilla@gmail.com>
cc: pgsql-performance@postgresql.org
Subject: Re: Analyze results in more expensive query plan
In-reply-to: 
 <CAG3Nut_TY+cn_THAW3PKE+cNUod8HbaLu+tADrT4vEGcvUWCRg@mail.gmail.com>
References: 
 <CAG3Nut_TY+cn_THAW3PKE+cNUod8HbaLu+tADrT4vEGcvUWCRg@mail.gmail.com>
Comments: In-reply-to Jeremy Altavilla <jeremyaltavilla@gmail.com>
	message dated "Thu, 16 May 2019 16:03:18 -0400"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <6604.1558100135.1@sss.pgh.pa.us>
Date: Fri, 17 May 2019 09:35:35 -0400
Message-ID: <6605.1558100135@sss.pgh.pa.us>
Precedence: bulk

Jeremy Altavilla <jeremyaltavilla@gmail.com> writes:
> We have several select statements whose performance is greatly improved by
> deleting some stats from pg_statistic.

You might have better results by setting up some "extended stats" for
the combination of bag_type columns that this query depends on.  Per your
description, there's a fair amount of cross-column correlation, which
the planner will not expect without some extended stats to tell it so.

https://www.postgresql.org/docs/10/planner-stats.html#PLANNER-STATS-EXTENDED

			regards, tom lane