Horton

Re: BUG: Postgres 14 + vacuum_defer_cleanup_age + FOR UPDATE + UPDATE

Andres Freund <andres@anarazel.de>

View thread View on postgr.es
From: Andres Freund <andres@anarazel.de>
To: Peter Geoghegan <pg@bowt.ie>
Cc: Heikki Linnakangas <hlinnaka@iki.fi>, Andrey Borodin <amborodin86@gmail.com>, Michail Nikolaev <michail.nikolaev@gmail.com>, Thomas Munro <thomas.munro@gmail.com>, Matthias van de Meent <boekewurm+postgres@gmail.com>, PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Date: 2023-02-06T21:02:05Z
Lists: pgsql-hackers 
Hi,

On 2023-02-04 11:10:55 -0800, Peter Geoghegan wrote:
> On Sat, Feb 4, 2023 at 2:57 AM Andres Freund <andres@anarazel.de> wrote:
> > Is there a good way to make breakage in the page recycling mechanism
> > visible with gist? I guess to see corruption, I'd have to halt a scan
> > before a page is visited with gdb, then cause the page to be recycled
> > prematurely in another session, then unblock the first? Which'd then
> > visit that page, thinking it to be in a different part of the tree than
> > it actually is?
> 
> Yes. This bug is similar to an ancient nbtree bug fixed back in 2012,
> by commit d3abbbeb.
> 
> > which clearly doesn't seem right.
> >
> > I just can't quite judge how bad that is.
> 
> It's really hard to judge, even if you're an expert. We're talking
> about a fairly chaotic scenario. My guess is that there is a very
> small chance of a very unpleasant scenario if you have a GiST index
> that has regular page deletions, and if you use
> vacuum_defer_cleanup_age. It's likely that most GiST indexes never
> have any page deletions due to the workload characteristics.

Thanks.


Sounds like a problem here is too hard to repro. I mostly wanted to know how
to be more confident about a fix working correctly. There's no tests for the
whole page recycling behaviour, afaics, so it's a bit scary to change things
around.

I didn't quite feel confident pushing a fix for this just before a minor
release, so I'll push once the minor releases are tagged. A quite minimal fix
to GetFullRecentGlobalXmin() in 12-13 (returning FirstNormalTransactionId if
epoch == 0 and RecentGlobalXmin > nextxid_xid), and the slightly larger fix in
14+.

Greetings,

Andres Freund

Commits

  1. Fix incorrect TAP test ordering

  2. pg_amcheck: Minor test speedups

  3. amcheck: Fix FullTransactionIdFromXidAndCtx() for xids before epoch 0

  4. amcheck: Fix ordering bug in update_cached_xid_range()

  5. Fix corruption due to vacuum_defer_cleanup_age underflowing 64bit xids

  6. Add hardening to catch invalid TIDs in indexes.

  7. Use full 64-bit XID for checking if a deleted GiST page is old enough.

  8. Avoid early reuse of btree pages, causing incorrect query results.