On Tue, 2022-10-18 at 10:55 +0900, Michael Paquier wrote:
> On Mon, Aug 22, 2022 at 05:16:58PM +0200, Michael Banck wrote:
> > The .backup files written to the archive (if archiving is on) are very
> > similar to the backup_label that's written/returned by
> > pg_stop_backup()/pg_backup_stop(), they just have a few extra lines
> > about the end of backup process that are missing from backup_label.
>
> Historically, there is "STOP WAL LOCATION" after "START WAL LOCATION",
> and "STOP TIME"/"STOP TIMELINE" at the end.
>
> > The parser in xlogrecovery.c however barfs on them because it does not
> > expect the additional STOP WAL LOCATION on line 2.
>
> Hm, no. I don't think that I'd want to expand the use of the backup
> history file in the context of recovery, so as we are free to add any
> extra information into it if necessary without impacting the
> compatibility of the recovery code. This file is primarily here for
> debugging, so I'd rather let it be used only for this purpose.
> Opinions of others are welcome, of course.
I tend to agree with you. It is easy to break PostgreSQL by manipulating
or removing "backup_label", and copying a file from the WAL archive and
renaming it to "backup_label" sounds like a footgun of the first order.
There is nothing that prevents you from copying the wrong file.
Such practices should not be encouraged.
Anybody who knows enough about PostgreSQL to be sure that what they are
doing is correct should be smart enough to know how to edit the copied file.
Yours,
Laurenz Albe