Horton

Re: Damage control for planner's get_actual_variable_endpoint() runaway

Robert Haas <robertmhaas@gmail.com>

View thread View on postgr.es
From: Robert Haas <robertmhaas@gmail.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Andres Freund <andres@anarazel.de>, Jakub Wartak <jakub.wartak@enterprisedb.com>, Simon Riggs <simon.riggs@enterprisedb.com>, Alvaro Herrera <alvherre@alvh.no-ip.org>, PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Date: 2022-11-21T17:39:18Z
Lists: pgsql-hackers 
On Mon, Nov 21, 2022 at 12:38 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Andres Freund <andres@anarazel.de> writes:
> > This can't quite be right - isn't this only applying the limit if we found a
> > visible tuple?
>
> What it's restricting is the number of heap page fetches, which
> might be good enough.  We don't have a lot of visibility here
> into how many index pages were scanned before returning the next
> not-dead index entry, so I'm not sure how hard it'd be to do better.

Oh. That's really sad. Because I think the whole problem here is that
the number of dead index entries can be huge.

-- 
Robert Haas
EDB: http://www.enterprisedb.com

Commits

  1. YA attempt at taming worst-case behavior of get_actual_variable_range.

  2. Improve performance of get_actual_variable_range with recently-dead tuples.

  3. Use SnapshotDirty rather than an active snapshot to probe index endpoints.