On Mon, Sep 27, 2021 at 11:20 AM Ajin Cherian <itsajin@gmail.com> wrote:
>
> On Sat, Sep 25, 2021 at 4:28 AM tushar <tushar.ahuja@enterprisedb.com> wrote:
> >
> > On 9/24/21 10:36 PM, Robert Haas wrote:
> > > Here's v9, fixing the issue reported by Fujii Masao.
> >
> > Please refer this scenario where publication on v14RC1 and subscription
> > on HEAD (w/patch)
> >
> > --create a subscription with parameter two_phase=1 on HEAD
> >
> > postgres=# CREATE SUBSCRIPTION r1015 CONNECTION 'dbname=postgres
> > host=localhost port=5454' PUBLICATION p WITH (two_phase=1);
> > NOTICE: created replication slot "r1015" on publisher
> > CREATE SUBSCRIPTION
> > postgres=#
> >
> > --check on 14RC1
> >
> > postgres=# select two_phase from pg_replication_slots where
> > slot_name='r105';
> > two_phase
> > -----------
> > f
> > (1 row)
> >
> > so are we silently ignoring this parameter as it is not supported on
> > v14RC/HEAD ? and if yes then why not we just throw an message like
> > ERROR: unrecognized subscription parameter: "two_phase"
> >
> > --
>
> There is usually a time lag between a subscription created with two_phase on and
> the slot on the publisher enabling two_phase. It only happens after a
> tablesync is completed and
> the apply worker is restarted. There are logs which indicate that this
> has happened. If you could share the
> logs (on publisher and subscriber) when this happens, I could have a look.
>
And in case you do see a problem, I request you create a seperate
thread for this. I didn't want to derail this patch.
regards,
Ajin Cherian
Fujitsu Australia
