Re: pg15b2: large objects lost on upgrade

Peter Geoghegan <pg@bowt.ie>

From: Peter Geoghegan <pg@bowt.ie>

To: Robert Haas <robertmhaas@gmail.com>

Cc: Tom Lane <tgl@sss.pgh.pa.us>, "Jonathan S. Katz" <jkatz@postgresql.org>, Andres Freund <andres@anarazel.de>, Noah Misch <noah@leadboat.com>, Andrew Dunstan <andrew@dunslane.net>, Bruce Momjian <bruce@momjian.us>, Michael Paquier <michael@paquier.xyz>, Justin Pryzby <pryzby@telsasoft.com>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>, Shruthi Gowda <gowdashru@gmail.com>

Date: 2022-08-04T18:04:14Z

Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

Revert recent changes to 002_pg_upgrade.pl.
- 6f7e7d0c482d 15.0 landed
- 87e22f675fd8 16.0 landed
Revise test case added in 43746996399541ecb5c7b188725a5f097c15ceae.
- d92f2bc0dae3 15.0 landed
- 212bdc0cbc32 16.0 landed
Use TRUNCATE to preserve relfilenode for pg_largeobject + index.
- bbe08b8869bd 16.0 landed
- 4ab5dae9472c 15.0 landed
Preserve relfilenode of pg_largeobject and its index across pg_upgrade.
- a2996478c32d 15.0 landed
- d498e052b4b8 16.0 landed
Have VACUUM warn on relfrozenxid "in the future".
- e83ebfe6d767 15.0 cited
Set relfrozenxid to oldest extant XID seen by VACUUM.
- 0b018fabaaba 15.0 cited
pg_upgrade: Preserve relfilenodes and tablespace OIDs.
- 9a974cbcba00 15.0 cited
Perform a lot more sanity checks when freezing tuples.
- 699bf7d05c68 11.0 cited

On Thu, Aug 4, 2022 at 9:44 AM Robert Haas <robertmhaas@gmail.com> wrote:
> But if you don't want to do that, and you also don't want to have
> random failures, the only alternatives are weakening the check and
> removing the test. It's kind of hard to say which is better, but I'm
> inclined to think that if we just weaken the test we're going to think
> we've got coverage for this kind of problem when we really don't.

Perhaps amcheck's verify_heapam() function can be used here. What
could be better than exhaustively verifying that the relfrozenxid (and
relminmxid) invariants hold for every single tuple in the table? Those
are the exact conditions that we care about, as far as
relfrozenxid/relminmxid goes.

My sense is that that has a much better chance of detecting a real bug
at some point. This approach is arguably an example of property-based
testing.

-- 
Peter Geoghegan