drbd-graceful-shutdown.service interfering with Pacemaker

[Akemi Yagi]

On Mon, Jan 27, 2025 at 4:00 AM Lars Ellenberg
<lars.ellenberg at linbit.com> wrote:
>
> On Thu, Jan 23, 2025 at 12:56:11PM -0800, Reid Wahl wrote:
> > Hi, we at Red Hat received a user bug report today involving drbd and Pacemaker.
> >
> > When shutting down the system (e.g., with `shutdown -r now`),
> > Pacemaker hangs when trying to stop the cluster resource that manages
> > drbd. If the user manually stops Pacemaker before shutdown (e.g.,
> > `systemctl stop pacemaker`), the problem does not occur.
>
> So apparently we forgot to add a "Before=pacemaker.service"
> or even better "Before=resource-agents-deps.target"
> to the graceful shutdown service.
>
> If you don't need it in your environment, you can still "mask" it.
>
> I have some changes queued up for our systemd thingies,
> I'll work that in as well.
>
> BTW, stopping pacemaker explicitly first should be a generic recommendation
> and has proven useful in many cases.
> Because all sorts of missing dependencies for all sorts of things, which,
> because of the racy nature, only "sometimes" are stopped "too early".
>
> And because, whenever there is a problem during shutdown of pacemaker,
> for whatever reason, and you already initiated the system shutdown,
> you have very limited means to "help it along", and sometimes need to
> wait out all of the timeouts and escalation timeouts.
>
> Thank you for the report.
>
>     Lars

Hi,

If you run RHEL (or its rebuild), you can get the drbd-utils package
that has the proposed fix. It is currently in the elrepo-testing
repository.

EL8:
https://elrepo.org/linux/testing/el8/x86_64/RPMS/drbd90-utils-9.30.0-2.el8.elrepo.x86_64.rpm

EL9:
https://elrepo.org/linux/testing/el9/x86_64/RPMS/drbd9x-utils-9.30.0-2.el9.elrepo.x86_64.rpm

Akemi