On 24-May-20 15:48, Stephen Frost wrote:
> That really shouldn't be possible. I'm very curious as to exactly what
> happened that resulted in your primary/replica being 'out of sync', as
> you say.
Hi Stephen,
Actually this was more a hypothetical question to find a solution in
case some day one of our standby clusters goes out of sync and we have
to rebuild it having a very big database.
With proper WAL archiving this shouldn't happen but we wanted to be
prepared for this scenario just in case.
We did some tests measuring IO and traffic and are very happy with the
results. We will definitely be adding pgBackRest to our toolchain.
Regarding my initial question, I still believe that the world deserves a
simple direct pg_basebackup replacement even when putting an additional
"repo host" in the middle is a better idea in the long term.
As you said, all the pieces are there and it would be quite easy to
write a new "pg_basebackup_delta" script that could be executed on the
standby host to:
1) setup a pgBackRest repo on the primary host (via SSH)
2) create a backup on the primary host (via SSH)
3) do a delta restore on the standby
Even when the repository on the primary host is only created temporarily
(and require double storage, resources, etc), it may still be worth
considering the traffic that can be saved by doing a delta restore on a
standby host in a different region, right?
Thanks and congratulations for the good work.