[ClusterLabs] DRBD + VDO HowTo?

[Andrei Borzenkov]

On Tue, May 18, 2021 at 11:43 AM Eric Robinson <eric.robinson at psmnv.com> wrote:
> > But really, the most simple would be to use systemd service. Then you do
> > not really need to monitor anything. Resource is assumed to be active when
> > service is started. That is enough to quickly get it going.
> >
>
> That was the first thing I tried. The systemd service does not work because it wants to stop and start all vdo devices, but mine are on different nodes.

Now, that's becoming ridiculous. Nobody said "use whatever upstream
ships" because upstream probably never intended it to be used in a
cluster environment. But if you are going to write your own LSB/OCF
script anyway, you can just as well write your own service. Which is
much simpler at this point

> Also, the assumption that the resource is active does not seem to be safe to make. I've been doing a lot of additional testing, and I think the reason why systemd, ocf, and lsb scripts have all failed me is because Pacemaker is not honoring the order constraints. IT keeps trying to start the vdo device before promoting drbd on the target node. I will check it some more and confirm.

Pacemaker is correctly honoring your constraints. Your constraints do
not forbid starting VDO on demoted instances. See another e-mail.