Drbd/pacemaker integration question

[Devin]

@Anton

handlers {
      fence-peer "/usr/lib/drbd/crm-fence-peer.9.sh";
      after-resync-target "/usr/lib/drbd/crm-unfence-peer.9.sh";
    }

This is likely the problem right here. This was how we implemented fencing handlers previously. By abusing the after-resync-target handler. However, with recent DRBD versions we no longer do this. You will want to instead use the unfence-peer handler. See the user’s guide here for an example: https://linbit.com/drbd-user-guide/drbd-guide-9_0-en/#s-pacemaker-fencing-cib

@Anton

resource 'ha-nfs-clone' (id: drbd-fence-by-handler-ha-nfs-ha-nfs-clone)
    Rules:
      Rule: role=Promoted score=-INFINITY (id: drbd-fence-by-handler-ha-nfs-rule-ha-nfs-clone)
        Expression: #uname ne memverge2 (id: drbd-fence-by-handler-ha-nfs-expr-28-ha-nfs-clone)

That is the “strange location constraint” I was talking about in my previous post. This basically says “never promote on a node where the uname is not equal to memverg2”. You can tell this was set by the fencing handler via the name: drbd-fence-by-handler...
This would perfectly explain the behavior you observe. This should be removed automatically by the unfence-peer handler, but you currently don’t have one configured.