Re: Decoding of two-phase xacts missing from CREATE_REPLICATION_SLOT command

On Thu, Jun 10, 2021 at 4:13 AM Jeff Davis <pgsql(at)j-davis(dot)com> wrote:
>
> On Wed, 2021-06-09 at 17:27 +0530, Amit Kapila wrote:
> > 2. In the main patch [1], we do send two_phase option even during
> > START_REPLICATION for the very first time when the two_phase can be
> > enabled. There are reasons as described in the worker.c why we can't
> > enable it during CREATE_REPLICATION_SLOT.
>
> I'll have to catch up on the thread to digest that reasoning and how it
> applies to decoding vs. replication. But there don't seem to be changes
> to START_REPLICATION for twophase, so I don't quite follow your point.
>

I think it is because we pass there as an option as I have suggested
doing in the case of CREATE_REPLICATION_SLOT.

> Are you saying that we should not be able to create slots with twophase
> at all until the rest of the changes go in?
>

No, the slots will be created but two_phase option will be enabled
only after the initial tablesync is complete.

> > Now, if we want to do
> > protocol changes, I wonder why only do some changes and leave the
> > rest
> > for the next version?
>
> I started this thread because it's possible to create a slot a certain
> way using the SQL function create_logical_replication_slot(), but it's
> impossible over the replication protocol. That seems inconsistent to
> me.
>

Right, I understand that but on the protocol side, there are few more
things to be considered to allow subscribers to enable two_phase.
However, maybe, for now, we can do it just for create_replication_slot
and the start_replication stuff required for subscribers can be done
later. I was not completely sure if that is a good idea.

--
With Regards,
Amit Kapila.