Re: Logical Decoding Failover

On Sun, Aug 7, 2016 at 9:29 PM, Colin Morelli <colin(dot)morelli(at)gmail(dot)com>
wrote:

> Venkata,
>
> Thanks for the reply. Unfortunately something like PgPool still won't
> create the replication slots on all hosts, and record the LSN in a way that
> is reusable on the secondary.
>

Yes, thats correct, pgPool does not have anything to do with replication
slots. That is something which you need to manually configure.

> This sort of puts logical decoding at odds with HA, correct? In the case
> of master failover, there's no way to: a) know where in the stream you read
> to, or b) convert that to something that can be understood by a replica
> that's about to be promoted.
>

Can you please clarify, what did you exactly mean here ? are you referring
to cascading standbys or something ? Failover is something which
application has to do, to reconnect to the promoted standby. As far as
logical decoding is concerned, you need to have a replication slot
configured for the new master to ensure all the WALs which are not
replicated to standby are retained. After promotion, the standby database
becomes a standalone instance and the replication must be re-enabled. You
can automate the standby promotion using tools like pgpool-II or pacemaker.

I am trying to differentiate standby-promotion and failover mechanism here.

Are there any scheduled improvements here either for 9.6 or 9.7?
>

I am not 100% sure, if there is anything developed from an failover
mechanism perspective.

Regards,
Venkata B N

Fujitsu Australia