On Tue, Jan 16, 2024 at 12:13 PM Hayato Kuroda (Fujitsu)
<kuroda.hayato@fujitsu.com> wrote:
>
> Dear Amit, Bharath,
>
> > This is a more strict check because it is possible that even if the
> > latest confirmed_flush location is not persisted there is no
> > meaningful decodable WAL between whatever the last confirmed_flush
> > location saved on disk and the shutdown_checkpoint record.
> > Kuroda-San/Vignesh, do you have any suggestion on this one?
>
> I think it should be as testcase explicitly. There are two reasons:
>
> * e0b2eed is a commit for backend codes, so it should be tested by src/test/*
> files. Each src/bin/XXX/*.pl files should test only their executable.
> * Assuming that the feature would be broken. In this case 003_logical_slots.pl
> would fail, but we do not have a way to recognize on the build farm.
> 038_save_logical_slots_shutdown.pl helps to distinguish the case.
+1 to keep 038_save_logical_slots_shutdown.pl as-is.
> Based on that, I think it is OK to add advance_wal() and comments, like Bharath's patch.
Thanks. I'll wait a while and then add it to CF to not lose it in the wild.
--
Bharath Rupireddy
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com