Re: speed up a logical replica setup

On Wed, Jul 3, 2024 at 9:21 AM Hayato Kuroda (Fujitsu)
<kuroda(dot)hayato(at)fujitsu(dot)com> wrote:
>
> Dear Amit,
>
> > IIUC, the problem is that the consistent_lsn value returned by
> > setup_publisher() is the "end +1" location of the required LSN whereas
> > the recovery_target_lsn used in wait_for_end_recovery() expects the
> > LSN value to be "start" location of required LSN.
>
> Yeah, right. It is same as my understanding.
>
> > This sounds like an ugly hack to me and don't know if we can use it.
>
> I also think it is hacky, but I could not find better solutions.
>
> > The ideal way to fix this is to get the start_lsn from the
> > create_logical_slot functionality or have some parameter like
> > recover_target_end_lsn but I don't know if this is a good time to
> > extend such a functionality.
>
> I felt that such approach might be used for HEAD, but not suitable for PG17.
> Alternative approach I came up with was to insert a tuple while waiting the
> promotion. It can generate a WAL record so that standby can finish after the
> application. But I'm not sure how do we do and it seems to lead an additional
> timing issue. Also, this does not improve the behavior of the command - normal
> user may have to wait some time by the command.
>

BTW, I think the time required by standby to reach a consistent state
after startup is any way unpredictable. For example, if we consider
that in real-world scenarios between the time we have stopped standby
and restarted it, there could be many transactions on primary that
need to be replicated before we reach recover_target_lsn.

I don't think adding additional (dummy) WAL records is a good solution
but it is better to hear from others.

> Do you have any other idea?
>

The other idea could be that we use the minimum restart_lsn of all the
slots created by this tool as a consistent_lsn. We can probably get
that value by using pg_get_replication_slots() but this idea needs
further evaluation as to whether it will lead to a consistent
subscriber.

--
With Regards,
Amit Kapila.