Re: Changing the state of data checksums in a running cluster

Andres Freund <andres@anarazel.de>

From: Andres Freund <andres@anarazel.de>

To: Andreas Karlsson <andreas@proxel.se>

Cc: Tomas Vondra <tomas@vondra.me>, Daniel Gustafsson <daniel@yesql.se>, Bernd Helmle <mailings@oopsware.de>, Michael Paquier <michael@paquier.xyz>, Michael Banck <mbanck@gmx.net>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>

Date: 2025-11-21T20:28:40Z

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 →

Use correct datatype for PID
- 0ca1b3010597 19 (unreleased) landed
Improve comments in online checksums code
- cd857dec0e0a 19 (unreleased) landed
Fix checksum state transition during promotion
- 5fee7cab1b87 19 (unreleased) landed
Fix regex searching for page verification failures in tests
- 486b9a9b9eb4 19 (unreleased) landed
Apply data-checksum worker throttling parameters
- 9a39056c418c 19 (unreleased) landed
Skip WAL for unlogged main fork during online checksum enable
- 2018bd616790 19 (unreleased) landed
Revert "Get rid of WALBufMappingLock"
- c13070a27b63 19 (unreleased) cited
Get rid of WALBufMappingLock
- bc22dc0e0ddc 18.0 cited
Improve grammar of options for command arrays in TAP tests
- ce1b0f9da03e 18.0 cited

Hi,

On 2025-11-21 01:44:31 +0100, Andreas Karlsson wrote:
> On 11/20/25 11:34 AM, Tomas Vondra wrote:
> > On 11/19/25 22:03, Andreas Karlsson wrote:
> > > I have been following these discussions but not read the patch in detail.
> > > 
> > > This patch makes me worried especially with the new issues recently
> > > uncovered. This was already a quite big patch and to fix these issues it
> > > will likely have to become even bigger and given how this would become a
> > > very rarely stressed code paths I wonder if we can actually ever become
> > > confident that the patch works in all edge cases.
> > > 
> > > Something like this need to be easy to understand for us to have any
> > > hope at all to be comfortable in the correctness. Can we actually do that?
> > > 
> > 
> > How's this different from any other complex patch? We get more familiar
> > with the problem during review, identify issues, improve the patch to
> > address them. And then again and again.
> 
> The difference I see is in how rarely anyone actually switches checksum
> state in a production database, especially now that we enabled them by
> default. A complex and rarely stressed code path is a minefield.

FWIW, I think this is actually a good feature build the infrastructure for
features (i.e. dynamically reconfiguring the cluster while running) like this,
precisely because it isn't *constantly* used.

Greetings,

Andres Freund