Hi,
On 2023-11-20 11:24:25 -0500, Robert Haas wrote:
> I do also think it is worth considering how this proposal interacts
> with the proposal to remove backup_label. If that proposal goes
> through, then this proposal is obsolete, I believe.
I think it's the opposite, if anything. Today you can at least tell there was
use of a backup_label by looking for backup_label.old and you can verify
fairly easily in a restore script that backup_label is present. If we "just"
use pg_control, neither of those is as easy. I.e. log messages would be more
important, not less. Depending on how things work out, we might need to
reformulate and/or move them a bit, but that doesn't seem like a big deal.
> But if this is a good idea, does that mean that's not a good idea? Or would
> we try to make the pg_control which that patch would drop in place have some
> internal difference which we could use to drive a similar log message?
I think we absolutely have to. If there's no way to tell whether an "external"
pg_backup_start/stop() procedure actually used the proper pg_control, it'd
make the situation substantially worse compared to today's, already bad,
situation.
Greetings,
Andres Freund