Cold Redundancy of a VM in a cluster of two nodes

[francesco.pernice]

Hello, I need help managing the redundancy of a VM.
At the moment I have 2 physical machines in Cluster MAIN and BACKUP (Proxmox 7.3)
both configured:

• in RAID1 ZFS (directly from the ProxMox installation).
• The two machines have two 1TB SSDs.

I know I can't have HA since I don't have 3 nodes but it's not a problem, I can wait before restarting the machine but it's important to have the data updated just before the service interruption.
As I said, my idea is to have 1 VM on main node that replicates every 5 minutes on the second one. Up to here I think I've arrived, in fact I can see the VM disk on the backup machine after replication.
My problem now is making sure that I can use this disk that is on the backup machine when the main physical node dies. Is there a "native" way of doing this without having to touch the system files too much? (I'm not a ZFS expert).

I tried to create a vm on the backup machine. Changing in the configuration file the disk with the replicated one, but it is not working.

Any advice/best practice?

I will bring new information in case of need

 

[fabian]

https://pve.proxmox.com/pve-docs/pve-admin-guide.html#_error_handling

 

[francesco.pernice]

Thanks, I've already read this section of the guide a good number of times. Yet there is something I don't understand, the guide says to change the expected votes for Quorum (and simultaneously tells you not to do it unless absolutely necessary). I wouldn't want to change those variables also because I'm afraid that when the Main machine is brought back to life this discrepancy might cause problems.

# pvecm expected 1

This line scares me, am i over-reacting? When the main comes back what should i do to avoid cluster problems?
It looks to me like i need a more simple solution, just use the replicated disk to start a new VM.

Thank you

 

[fabian]

ideally, you have more than two nodes (so that a single node going down still means the rest can establish quorum). if you only have two nodes and no tie-breaker in the form of a qdevice, you will need to temporarily disable regular quorum calculations (and yes, as you suspect and the docs call out, this is rather dangerous!) to recover and start the guest on your remaining node.

so, in order of preference:
- have more than two nodes
- if you can only have two nodes, setup a qdevice so you still have three votes
- if that is not an option (it very likely is! please re-evaluate your answers to the above two points ) temporarily set expected vote to 1 after carefully evaluating the situation, and reset it afterwards when it's no longer needed