From: dwsteele (@dwsteele) <noreply+dwsteele@github.com>
To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>
Subject: Re: [pgbackrest/pgbackrest] issue #2913: Proposal: Checkpoint saves in resume feature override for S3 object lock
Date: Mon, 09 Nov 2026 19:11:49 +0000
Message-ID: <gh-pgbackrest-pgbackrest-2913-c6748969086@github.com> (raw)
In-Reply-To: <gh-pgbackrest-pgbackrest-2913@github.com>
References: <gh-pgbackrest-pgbackrest-2913@github.com>
I think we'd be open to this change but not as written (yes, I peeked!). The problem is that it takes a very S3-centric approach to the problem which does not honor our storage abstraction. To make it work you'd need to create a generic parameter to pass to storageOpenWriteP(), something like `tmp`, defined as a file that is not intended to last longer than the current operation. Then that gets pushed down to the driver and the driver can decide what to do with it. What you did inside the S3 driver looks fine at a glance, but how the flag gets there is important.
Further, if possible, I'd like to make this a repo-storage-* option that works with GCS, Azure, and S3 (using different mode lists). That would not be needed for a first draft, though, and it might not even be possible to make it work sensibly.
So, yes -- I think the idea has merit and I'm happy to look at a patch that works roughly as described above.