On Mon, Dec 9, 2024 at 6:43 AM Joe Wildish wrote:
Overall, your solution seems okay, but:
> a fix has gone in to PG17 that sorts this problem.
>
> However, we can't go to 17 yet, so need a solution for 15 and 16.
Honestly, this would seem like a really, really strong reason to push for
v17.
We maintain c.50 logical replicas. Typically the producer version is 12 or 13,
and the subscriber version is 14. We intend to upgrade the subscribers to 15
using pg_upgrade. However, we ran into an unexpected problem with that
approach. I couldn't find much being mentioned about it on the web,
