2 node cluster setup advice

[SeppeM]

I’m experimenting with setting up a converged Proxmox cluster and only have two nodes. Adding a third is not an option at the moment, unless I use a much weaker leftover server.

I am looking for any advice from people whom might have ran a two-node cluster in production?
On the other end would I perhaps be better of adding a third weak node and not just have quorum device?


For context: we already have a separate storage cluster, so these nodes will be dedicated to compute.

 

[UdoB]

On the other end would I perhaps be better of adding a third weak node and not just have quorum device?

That decision is completely up to you

You seem to have some other equipment. I would check if a small VM or a container could be placed there - to implement the Quorum Device.

On the other hand a physical third server, even if small and old, lifts some restrictions and offers some additional options when (not: if) one of the two main nodes fail.

Note also that while the QD ist just and only a third vote, a "real" third server needs to get the same subscription license as the two main nodes.

 

[mr44er]

Adding a third is not an option at the moment, unless I use a much weaker leftover server.

A weak server is better than nothing, you really want quorum (for that it is good enough). 3 nodes (regardless if one is slow) are always better than 2.

 

[SeppeM]

A weak server is better than nothing, you really want quorum (for that it is good enough). 3 nodes (regardless if one is slow) are always better than 2.

I have plenty systems I can use for quorum but indeed as you and @UdoB have said I will go trough our inventory here and try to create a 3rd actual server which will take over some VM's if 1 of the main 2 end up dying. If i'm not wrong I do have to try and recreate stuff like network interfaces to exist on all 3 servers equally (name wise that is) in case when vm's get moved or they won't be able to start due to lack of a correct interface to bind to.

 

[mr44er]

3rd actual server which will take over some VM's if 1 of the main 2 end up dying

The main purpose of 3rd node is just to avoid split-brain situation and halt everything. VMs could/should hop between the stronger node1 and 2 in case of a failure, the third node therefore ensures the avoidance of split-brain with quorum. If this node is slower but still has enough resources to take over VMs in the event of a failure, all the better. In each of these cases, it's better than two nodes. Another option would be four nodes: your two fast ones and two older ones. You could configure it so that if the fast node 1 or 2 dies, node 3 takes over 50% of the VMs, and node 4 also takes over the remaining 50%.

If i'm not wrong I do have to try and recreate stuff like network interfaces to exist on all 3 servers equally (name wise that is) in case when vm's get moved or they won't be able to start due to lack of a correct interface to bind to.

Yes, the real NIC-names are not important, only the vmbridges have to be the same and in the same numbering scheme. Not required but ideal also the same bandwidth.