pcs
pcs copied to clipboard
ClusterLabs
pcs cluster disable/start/stop should accept a --corosync-only and --pacemaker-only parameter
It would be useful to be able to only stop pacemaker or corosync with the pcs cluster disable/start/stop commands
Well you could just stop pacemaker and continue to have corosync quorum information up.
It feels wrong to install pacemaker on an arbitrator-only node. Also, corosync is A LOT more tolerant regarding different corosync versions inside the cluster. But an outdated pacemaker installation on the arbitrator f***s the whole cluster up majorly, as I just experienced again.
@ChristianCiach I do not get what you are asking about. If an arbitrator-only node is not a part of any cluster, it does not matter what pacemaker version is installed there. If such arbitrator node is a part of a cluster, why install pacemaker version different than what other nodes in the cluster have in the first place?
why install pacemaker version different than what other nodes in the cluster have in the first place?
Exactly this was the issue. The arbitrator node has just been "forgotten" to be updated. Yes, this was a human error, but as pacemaker is not necessary on an arbitrator-only node, I would prefer to not install pacemaker at all on this node, just to minimize the chance of those possible issues.
Also, pacemaker checks for the existance of the OCF scripts, even when this node is set to "standby". It feels dirty to install pacemaker + our custom OCF scripts just for pcs to work.
@ChristianCiach: You must NOT run pacemaker as a cluster node on arbitrator! It is only installed as a dependency for PCS but you should not be starting it at all. The whole point of an arbitrator is that it has to be independent from any cluster.
But wouldn't pacemaker be started by pcs when doing a "pcs cluster start"?
@ChristianCiach: It wouldn't as long as there's no cluster defined on arbitrator. If your arbitrator service is installed on a cluster node, you should either remove the node from cluster or use another machine.
I don't really get what you are saying. AFAIK, the arbitrator would become part of the cluster as soon as I configure corosync through pcs. Maybe I am wrong on this part. I will investigate this. Maybe it would be better for me to not install pcs on the arbitrator at all, but then I would need to configure corosync manually on this node.
@ChristianCiach: Just to be sure, are we talking about booth arbitrator, right?
@ChristianCiach: Ah, confusion of terms then. You cannot actually use a third node like that without running full corosync+pacemaker stack. You either need to keep the third node as a full cluster node or take a look at corosync-qdevice which does exactly what you wanted (you'll need to use pcs-0.9.158 or higher).
@lyon667 Awesome! I didn't know I could use corosync-qdevice without a full corosync installation! This seems promising!
Actually, I may even use Booth instead. When we created the cluster around 3 to 4 years ago, Booth was at version 0.1.0 or 0.2.0. That wasn't really trustworthy and I wasn't even able to get it working. Maybe I will give it another shot. Thanks for your help this far!