Re: Replication & recovery_min_apply_delay

On 2019-Aug-02, Michael Paquier wrote:

> On Wed, Jul 31, 2019 at 04:43:26PM -0400, Alvaro Herrera wrote:
> > As for the test module, the one I submitted takes a lot of time to run
> > (well, 60s) and I don't think it's a good idea to include it as
> > something to be run all the time by every buildfarm member. I'm not
> > sure that there's a leaner way to test for this bug, though, but
> > certainly it'd be a good idea to ensure that this continues to work.
>
> Hmmm. Instead of that, wouldn't it be cleaner to maintain in the
> context of the startup process a marker similar to receivedUpto for
> the last LSN? The issue with this one is that it gets reset easily so
> we would lose track of it easily, and we need also to count with the
> case where a WAL receiver is not started. So I think that we should
> do that as a last replayed or received LSN if a WAL receiver is up and
> running, whichever is newer. Splitting the WAL receiver restart logic
> into a separate routine is a good idea in itself, the patch attempting
> to switch primary_conninfo to be reloadable could make use of that.

Konstantin, any interest in trying this?

--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services