[ClusterLabs] Clusterstack Dependency on SBD with systemd

[Klaus Wenninger]

Hi!

Wanted to pick up where we dropped the issue before Christmas.

Observations as a result of a startup-timeout issue led to
the awareness that the way we are starting SBD together with
the rest of the Clusterstack doesn't handle SBD startup-issues
properly.

SBD is integrated in the Clusterstack-startup using
dependencies & sequence via the SBD-unit-file:

  [Unit]
  Description=Shared-storage based fencing daemon
  Before=pacemaker.service
  After=systemd-modules-load.service iscsi.service
  PartOf=corosync.service
  RefuseManualStop=true
  RefuseManualStart=true

  [Install]
  RequiredBy=corosync.service

This is quite handy in the sense that if SBD is not present
nobody has to care about anything regarding SBD as it
would have to be if e.g. pacemaker directly required SBD - which
it actually does or should do in some way on clusters
using SBD.

What is working as desired with this setup is that we have
seamless startup/shutdown of SBD-service together with
corosync once SBD is enabled.

What does not work is that systemd doesn't wait/care
if SBD comes up properly before it starts pacemaker.

Just as a side-note:
We just have a startup-issue here as a positive startup
of SBD means that the watchdog (and you are strongly
advised to have a proper one with SBD) is engaged.
Thus any further problem SBD might face should reliably
lead to a reboot.

As a result of the discussion, and as there was no other
solution at sight that wouldn't have dependencies over
several packages, upstream-SBD merged the following
pull-request:

https://github.com/ClusterLabs/sbd/pull/39

Discussion here on the list and in this pull-request
showed that this approach of simply adding

[Unit]
Before=dlm.service   [Install]
  RequiredBy=pacemaker.service
  RequiredBy=dlm.service

to the SBD-unit-file solves the issue for the first
without having any impact on install/maintenance
procedures.
But it might not be the optimal longer-term solution.
Thus I'm writing this here as a platform for further
discussion.

And what showed up as possible caveat since the
discussion before Christmas, if you have SBD enabled
and you are doing an upgrade from a version without
the fix from this pull-request to a version with the
fix you would have to manually reenable the service
for the RequiredBy-links to be updated.

Regards,
Klaus