[Debian-ha-maintainers] Debian HA documentation and version migration

[Valentin Vidić]

On Fri, Oct 04, 2019 at 03:21:19PM +0000, Michael Firth wrote:
> Apologies for mailing what seems to be a "maintainers" mailing list
> for Debian HA, but there doesn't seem to be a "users" list that I have
> been able to find.

Post Pacemaker users from RHEL, SUSE and Debian/Ubuntu are on
users at clusterlabs.org so you can also try that.

> Is there any more up-to-date information on running HA processes on
> Debian than the page at:
> 
> https://wiki.debian.org/Debian-HA
> 
> That, and the "Clusters from Scratch" page both seem to be only
> talking about Debian Jessie, which is now two releases behind the
> "stable" Debian release.
> 
> Also, some of the links on the top level "Debian-HA" page are now
> dead, for example the one to alioth.debian.org.

True, we need to clean that up but in general any upstream documentation
on Pacemaker should work.

> Most specifically, I have a pair of systems running Debian Stretch
> with HA pretty successfully.
> 
> What I would like to do, if possible, is to upgrade the machines to
> Buster at some point soon, however, it looks like the packages in
> Debian Buster require different configuration from what Jessie and
> Stretch needed, and I haven't yet found any good guidance on how to
> convert the configuration from one format to the other successfully.

Not sure what configuration you mean, but I think the main problem with
the upgrade is that Corosync 3 can't talk to older versions so in-place
upgrade of the cluster might not be possible. RHEL8 documentation
suggests you build a new cluster with Corosync 3 and migrate services
manually.

> Additionally, I'm not sure if will be possible to convert from one to
> the other without a massive amount of down time - it looks like there
> may be some interoperability between the versions of Pacemaker in
> Stretch and Buster, but I don't think it is possible to get the
> different versions of Corosync to communicate.

It should be possible to import the Pacemaker configuration from the
old version but you also need to check if the resources agents work
as expected in the new version. So again building a new cluster and
testing everything before migrating might be the safest option.

-- 
Valentin