non-HA cluster quorum and votes

[M@ario]

Hi all!

Today I had to reboot my 8 node non-HA cluster because of a storage migration. When I turned on the nodes one-by-one the cluster did not get quorate on the first 4 nodes and the VMs on these nodes did not start. After the 5th node the cluster got the quorum and the VMs on node 5 to 8 started normally. I had to start the VMs on 1-4 manually.
I think this is because the "expected votes" value in the cluster configuration is set to 8. AFAIK that means the cluster does not become quorate until at least half of the nodes (4) are online.

Question: Is it safe, in a non-HA cluster, to set expected votes permanently to 1?
This should make the cluster always quorate and so the VMs are always started automatically.

 

[udo]

Hi all!

Today I had to reboot my 8 node non-HA cluster because of a storage migration. When I turned on the nodes one-by-one the cluster did not get quorate on the first 4 nodes and the VMs on these nodes did not start. After the 5th node the cluster got the quorum and the VMs on node 5 to 8 started normally. I had to start the VMs on 1-4 manually.
I think this is because the "expected votes" value in the cluster configuration is set to 8. AFAIK that means the cluster does not become quorate until at least half of the nodes (4) are online.

Hi!

Question: Is it safe, in a non-HA cluster, to set expected votes permanently to 1?
...

No!!
If your Networkconnection between parts of your cluster lost, all nodes have quorum and all VMs can run.
Say, if you have trouble with node 3 because of a network issue and you shutdown the node and start the VMs from node 3 on other nodes (shared storage on anothe NIC) all is fine. But if you now start node 3 again, the node get quorum (but don't see the rest of the qluster and don't know that the VMs allready run on different nodes) and start all VMs which are autostarted - best thing to get data corruption inside the VM-hdd...

If you restartet your cluster node by node you don't have trouble with the quorum and autostart.

Udo

 

[M@ario]

Hi!

No!!
If your Networkconnection between parts of your cluster lost, all nodes have quorum and all VMs can run.
Say, if you have trouble with node 3 because of a network issue and you shutdown the node and start the VMs from node 3 on other nodes (shared storage on anothe NIC) all is fine. But if you now start node 3 again, the node get quorum (but don't see the rest of the qluster and don't know that the VMs allready run on different nodes) and start all VMs which are autostarted - best thing to get data corruption inside the VM-hdd...

I'm talking of a NON-HA cluster (no rgmanager running)! You can't simply start a VM on another node with this type of cluster (it's like the "cluster" in 1.x). All you can do (and all I want) is live migration between cluster nodes. So this should not be a problem with "expected votes=1".

If you restartet your cluster node by node you don't have trouble with the quorum and autostart.

I had to take the shared storage offline, so I needed to stop the whole cluster first.

 

[dietmar]

So this should not be a problem with "expected votes=1".

Quorum is also used for other things (pmxcfs), so please do not set that!