Le mar. 8 janv., vers 08:13, Erika Knihti-Van Driessche exprimait :
> Hi,
Hey !
> I'm sure I'd regret creationg all the extra work for myself, but.. the
> customer insists that not all (at least production) applications may go
> down at the same time.
That's precisely the purpose of services or init script if you prefer.
For a customer, i manage 10 clusters x 5 environments, entirely with 1
service per cluster and 1 binary per environment. I also use ansible
to deploy and maintain streaming replication, logical migration,
monitoring, all based on services. It's easy with Debian packages,
because service creation is included.
> I have streaming replication in place also, and I think for minor upgrade
> it won't be a problem to first upgrade standby, resync replication, then
> failover primary to newer minor version standby (altho I did read that this
> is not supported but will likely work..).
It'll work.
> For major upgrade this isn't an option though, or I would have to
> implement Slony (I think)..
Why don't you use logical replication ? I recently migrate the 10x5
clusters from 9.6 to 10 with pglogical in 2 days of work. It was also
from Red Hat to Debian and i automate cluster per cluster with
services. Next step is to migrate from 10 to 11, also with logical
migration, by hand this time, because pglogical don't do so much and
due to the lack of documentation, i already wrote the missing parts.
> So, just creating separate services isn't really an option, because
> it won't help me with the upgrades.
It does as the major version is in the service name (i.e.
postgresql@<version>-<clustername>), you won't have to update it once
it's written.
Regards,
--
Stéphane KANSCHINE - https://www.hexack.fr./ - https://www.nuajik.io./
＠ stephane@hexack.fr
+33 6 64 31 72 52