RE: Changing the setting of wal_sender_timeout per standby

From: Kyotaro HORIGUCHI [mailto:horiguchi.kyotaro@lab.ntt.co.jp]
> At Fri, 14 Sep 2018 18:22:33 +0900, Masahiko Sawada <sawada.mshk@gmail.com>
> wrote in
> <CAD21AoBr2Y=N4iH8+6m5ara2GWdKE6ZrzWaqjZux6ErZ9pyAxQ@mail.gmail.com>
> > On Thu, Sep 13, 2018 at 12:32 PM, Michael Paquier <michael@paquier.xyz>
> wrote:
> > > On Thu, Sep 13, 2018 at 01:14:12AM +0000, Tsunakawa, Takayuki wrote:
> > >> Some customer wants to change the setting per standby, i.e., a shorter
> > >> timeout for a standby in the same region to enable faster detection
> > >> failure and failover, and a longer timeout for a standby in the remote
> > >> region (for disaster recovery) to avoid mis-judging its health.
> > >
> > > This argument is sensible.
> > >
> > >> The current PGC_HUP allows to change the setting by editing
> > >> postgresql.conf or ALTER SYSTEM and then sending SIGHUP to a specific
> > >> walsender.  But that's not easy to use.  The user has to do it upon
> > >> every switchover and failover.
> > >>
> > >> With PGC_BACKEND, the user would be able to tune the timeout as follows:
> > >>
> > >> [recovery.conf]
> > >> primary_conninfo = '... options=''-c wal_sender_timeout=60000'' ...'
> > >>
> > >> With PGC_USERSET, the user would be able to use different user
> > >> accounts for each standby, and tune the setting as follows:
> > >>
> > >> ALTER USER repluser_remote SET wal_sender_timeout = 60000;
> > >
> > > It seems to me that switching to PGC_BACKENDwould cover already all
> the
> > > use-cases you are mentioning, as at the end one would just want to
> > > adjust the WAL sender timeout on a connection basis depending on the
> > > geographical location of the receiver and the latency between primary
> > > and standby.
> >
> > +1 for PGC_BACKEND. It looks enough for most use cases.
> 
> +1, and we need a means to see the actual value, in
> pg_stat_replication?

Thanks, the patch attached.  I'll add this to the next CF shortly.  As Michael said, I think viewing the configured value would be a separate feature.


Regards
Takayuki Tsunakawa