Nested Proxmox for HA, is it possible?

[Proxfan2025]

Hi,

I will have several critical VMs running as DDNS or OpenVPN or SSH bastion servers. They are not heavy workloads but are not allowed to stop. However, my company has only one physical server for budget reasons.

So, I fancy that I could set up 3 nested Proxmox as guests, to only run above light-weight servers as members of Ceph HA cluster. Other VMs that are not critical but performance-oriented, would be set up directly on a physical, 1st tier Proxmox server. Would it be possible?

I attached 2 diagrams that explained this strategy. The 3 nodes nested Proxmox will be scaled out to 4 nodes if the 2nd physical server is added. I'd appreciate it if someone give some advice to tell me whether this is a ridiculous or valid idea.

 

[LnxBil]

I will have several critical VMs running a [... ] my company has only one physical server for budget reasons.

That states everything and the answer is NO.

Ask the people which decided the budget what 4 hours of downtime cost (e.g. with hardware support of 4 hours recovery time - which is also expensive) and then they should decide again. I they're also buy cheap without a good support contract, ask them what 24h or more of total downtime cost. Normally, they give you enough money.

Always ask the tough questions and most companies nowadays have to conform to e.g. ISO9001, so they're supposed to have redundancy in place.

So, I fancy that I could set up 3 nested Proxmox as guests, to only run above light-weight servers as members of Ceph HA cluster. Other VMs that are not critical but performance-oriented, would be set up directly on a physical, 1st tier Proxmox server. Would it be possible?

Sure, but useless. This setup would imply that PVE is not stable, so that the OS will crash a lot, which is not the case. Normally, the hardware is the problem. Depending on what type of server this is, a failed PSU can kill everything. Nothing to prevent it (besides redundant PSU). If you have that, regard a mainboard failure, pdu, a network card failure or any kind of other SPOF and you're done. RAID controller firmware error (had that one before) ... there are a million things than can go wrong including a cosmic ray hit.

Guideline: If it has to be one server, go with redundant everything, e.g.

• two power supplies
• at least 2 disks (better more)
• a lot of ECC memory with at least enough that one dimm can fail and the rest can run everything
• network setup including switchES

 

[Proxfan2025]

Hi LnxCar,
Sorry for my delayed reply. Thanks a lot for detailed comments!

You have points... yes, if business servers are down, it needs overwhelming costs for restoration. I talked to my boss & customer to reconsider the budgetary lines.

Also, I completely agree that PVE is very stable unless hardware issues occur, or some operation mistake take place. Overall, I will make it more simple in design to protect redundancy effectively.

Thanks & have a good day.

 

[LnxBil]

I talked to my boss & customer to reconsider the budgetary lines.

Yes, just make it clear to them and get everything in writing so that you can show it to them in the case of an error.