On 23/10/2026 15:42, David Steele wrote:
On 10/21/26 15:07, David Steele wrote:
Patch has been rebased.
A committer pointed out to me that part of this patch fixes a bug and
could be back-patched, so it should be separate.
Patch 01 now fixes only the possibility of a torn pg_control file.
Looks good to me.
It makes the copy out of memory to reduce the time the lock is held,
but it could alternately read the pg_control file in the lock as was
done by Thomas in c558e6fd92f. If the consensus is that the latter
approach is preferred I'm happy to make that change.
+0.5 for following the example of c558e6fd92f. This isn't
performance-sensitive so that aspect doesn't matter. But it seems a
little surprising if a backup might have more up-to-date values of some
fields than what the SQL functions show.
- Heikki
