Re: Planner join order regression from PG 15 to PG 16+: 70ms -> 1440ms (self-contained reproducer included)

From: Andrei Lepikhov <lepihov(at)gmail(dot)com>
To: Mauro Gatti <mauro(dot)gt70(at)gmail(dot)com>, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
Cc: pgsql-performance(at)lists(dot)postgresql(dot)org
Subject: Re: Planner join order regression from PG 15 to PG 16+: 70ms -> 1440ms (self-contained reproducer included)
Date: 2026-03-06 09:05:55
Message-ID: a026f7c2-6439-4a7e-b9fb-e88fc0ccd2fd@gmail.com
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgsql-performance

On 6/3/26 09:24, Mauro Gatti wrote:
> In any case, I don’t think there are other workable solutions without
> changing the code.
The gist of the issue is hidden [anti]correlations between columns.
I like the idea of grouping columns into a custom type when an object in
the data model is defined by several properties. This approach could
help reduce the number of clauses, hide correlations within an object,
and enable the use of histogram statistics.

I'm not sure if this idea is ready to use yet, or if Postgres has
everything needed to make it work well - just note.

--
regards, Andrei Lepikhov,
pgEdge

In response to

Browse pgsql-performance by date

  From Date Subject
Next Message Pavel Stehule 2026-03-06 09:06:42 Re: proposal: schema variables
Previous Message Mauro Gatti 2026-03-06 08:24:03 Re: Planner join order regression from PG 15 to PG 16+: 70ms -> 1440ms (self-contained reproducer included)