On Wed, Dec 06, 2023 at 10:00:01AM +0300, Alexander Lakhin wrote:
> I agree with it. I had worried a bit about ReindexRelationConcurrently()
> becoming twofold for callers (it can leave the snapshot or pop it), but I
> couldn't find a way to hide this twofoldness inside without adding more
> complexity. On the other hand, ReindexRelationConcurrently() now satisfies
> EnsurePortalSnapshotExists() in all cases.
Thanks, applied that with a few more tests, covering a bit more than
the code path you've reported with a failure.
I was wondering if this should be backpatched, actually, but could not
make a case for it as we've never needed a snapshot after a reindex
until now, AFAIK.
--
Michael
