Re: New standby_slot_names GUC in PG 17

Hi,

On Tue, Jun 25, 2024 at 10:24:41AM +0530, Amit Kapila wrote:
> On Tue, Jun 25, 2024 at 8:20 AM Masahiko Sawada <sawada(dot)mshk(at)gmail(dot)com> wrote:
> >
> > On Tue, Jun 25, 2024 at 11:21 AM Zhijie Hou (Fujitsu)
> > <houzj(dot)fnst(at)fujitsu(dot)com> wrote:
> > >
> > > I agree the current name seems too generic and the suggested ' synchronized_standby_slots '
> > > is better than the current one.
> > >
> > > Some other ideas could be:
> > >
> > > synchronize_slots_on_standbys: it indicates that the standbys that enabled
> > > slot sync should be listed in this GUC.
> > >
> > > logical_replication_wait_slots: it means the logical replication(logical
> > > Walsender process) will wait for these slots to advance the confirm flush
> > > lsn before proceeding.
> >
> > I feel that the name that has some connection to "logical replication"
> > also sounds good. Let me add some ideas:
> >
> > - logical_replication_synchronous_standby_slots (might be too long)
> > - logical_replication_synchronous_slots
> >
>
> I see your point about keeping logical_replication in the name but
> that could also lead one to think that this list can contain logical
> slots.

Agree, and we may add the same functionality for physical replication slots
in the future too (it has been discussed in the thread [1]). So I don't think
"logical" should be part of the name.

> OTOH, there is some value in keeping '_standby_' in the name as
> that is more closely associated with physical standby's and this list
> contains physical slots corresponding to physical standby's. So, my
> preference is in order as follows: synchronized_standby_slots,
> wait_for_standby_slots, logical_replication_wait_slots,
> logical_replication_synchronous_slots, and
> logical_replication_synchronous_standby_slots.
>

I like the idea of having "synchronize[d]" in the name as it makes think of
the feature it is linked to [2]. The slots mentioned in this parameter are
linked to the "primary_slot_name" parameter on the standby, so what about?

synchronized_primary_slot_names

It makes clear it is somehow linked to "primary_slot_name" and that we want them
to be in sync.

So I'd vote for (in that order);

synchronized_primary_slot_names, synchronized_standby_slots

[1]: https://www.postgresql.org/message-id/bb437218-73bc-34c3-b8fb-8c1be4ddaec9%40enterprisedb.com
[2]: https://git.postgresql.org/gitweb/?p=postgresql.git;a=commit;h=93db6cbda037f1be9544932bd9a785dabf3ff712

Regards,

--
Bertrand Drouvot
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com