Re: Complex filters -> Bad row estimates -> bad query plan

From: Michael Lewis <mlewis(at)entrata(dot)com>
To: Mathieu Fenniak <mathieu(dot)fenniak(at)replicon(dot)com>
Cc: pgsql-general <pgsql-general(at)postgresql(dot)org>
Subject: Re: Complex filters -> Bad row estimates -> bad query plan
Date: 2019-08-21 18:08:27
Message-ID: CAHOFxGoQffz1bxmJi=R31f5xNiA4+rMOcUoQL8re43NwPVuFSA@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-general

If those conditions that are throwing off the stats are expected to be
minimally impactful/filtering few rows, then you can use the one
tried-and-true optimizer hint (aside from materialized CTEs, stylized
indexes, etc) --- OFFSET 0 at the end of a sub-query.

SELECT * FROM ( [your existing query without the sub-selects that are
complicated and produce bad estimates] OFFSET 0 ) WHERE [your other
conditions that don't produce good estimates]

If there is correlation between field1 and field2, you might also look at
CREATE STATISTICS assuming you are on PG 10 or 11.

Before I do any of that, I would try LEFT JOIN for Table3 and Table4 then
use the where conditon "AND 2 = COALESCE( Table3.Status, Table4.Status"
and see if the optimizer likes that option better.

In response to

Responses

Browse pgsql-general by date

  From Date Subject
Next Message Peter Wainaina 2019-08-21 18:17:40 Permission for not Django app to do Write and Read
Previous Message Rob Sargent 2019-08-21 17:54:14 Re: SELECT all the rows where id is children of other node.