dumb idea? Virtualized proxmox host in a cluster.

[manilx]

Hi

I have a 2 node cluster and have a Qdevice for Quorum. (Actually it is a VM on another hypervisor instead. Could use a Pie but...)
Anyway, as Qdevice is not the same as a full node (when you remove nodes etc) I had the idea to install another proxmox node as a VM in that hypervisor. Running fine (using nested virtualization), not using it for anything else obviously

Anything against using that "virtual node" as a 3rd node in my cluster instead of the qdevice???

Any input most welcomed.

 

[bbgeek17]

Anything against using that "virtual node" as a 3rd node in my cluster instead of the qdevice???

No, it makes no difference as long as its stable. If you use HA you will need to create a "group" of your "primary" cluster members allowed to take over VMs.

---

Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox

 

[manilx]

No, it makes no difference as long as its stable. If you use HA you will need to create a "group" of your "primary" cluster members allowed to take over VMs.

---

Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox

Yes, I already have HA and 2 groups defined.
I just wanted to replace a Qdevice with a proxmox host (I have another real one but don't need it and energy is expensive these days) as qdevice is inferior to a real host when removing hosts etc.

The proxmox VM is running stable (on xcp-ng as it is).

 

[Kingneutron]

Hi

I have a 2 node cluster and have a Qdevice for Quorum. (Actually it is a VM on another hypervisor instead. Could use a Pie but...)
Anyway, as Qdevice is not the same as a full node (when you remove nodes etc) I had the idea to install another proxmox node as a VM in that hypervisor. Running fine (using nested virtualization), not using it for anything else obviously

Anything against using that "virtual node" as a 3rd node in my cluster instead of the qdevice???

Any input most welcomed.

Now I'm not an expert, but from what I know clustering best practice recommends a dedicated network for corosync bc it's latency-sensitive. Depends on what hypervisor you're using - I wouldn't necessarily recommend Virtualbox as networking is slow on it. Also, does the virtual PMVE have enough resources to take over all node functions if one goes down?
Have you done any failover testing with this config?

 

[manilx]

As explained running it on xcpng. No failover needed, there will be no VM's running. I just need it as a replacement for the Qdevice.

 

[bbgeek17]

As explained running it on xcpng. No failover needed, there will be no VM's running. I just need it as a replacement for the Qdevice.

Do you also realize that instead of running a full PVE node with all the overhead, you can run a minimal installation of practically any Linux variant and still run the QDevice, conserving allocated resources?
Just a thought.

PS one of the benefits being - no need for extra traffic to replicate PMXCFS https://pve.proxmox.com/wiki/Proxmox_Cluster_File_System_(pmxcfs)

---

Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox

 

[manilx]

Do you also realize that instead of running a full PVE node with all the overhead, you can run a minimal installation of practically any Linux variant and still run the QDevice, conserving allocated resources?
Just a thought.

PS one of the benefits being - no need for extra traffic to replicate PMXCFS https://pve.proxmox.com/wiki/Proxmox_Cluster_File_System_(pmxcfs)

---

Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox

I *have* a qdevice serving quorum. That's not the issue.
BUT if you want to remove nodes etc it's better to have a full node as you have to delete the qdevice before, so you loose quorum on a 2 node cluster (if you want to take one out). That's the problem a virtualized node solves (or could).
See here: https://forum.proxmox.com/threads/c...le-disk-to-mirrored-zfs-lvm.83915/post-369425

Overhead is not an issue, lot's of CPU/RAM to spare. Energy cost cuts if the node does absolutely nothing else (no VM's).

 

[manilx]

Cluster is running fine after adding the "virtual node". Quorum test (shutting down hosts etc) without issue.