Gracefully degrading an HA cluster

D

dga

Member
Feb 11, 2023
3
0
6
Probably can't be done but...

Have a 3-node HA cluster. When a power failure occurs I'd like to shut down 2 of the nodes to conserve the UPS. Of course one node doesn't make a quorum. Is there any way to shut down 2 nodes (while maintaining normal HA functionality for other failure modes)? Or do I just accept that I can't do this, or at best can shut down 1 of the 3 nodes?

Thanks.
 
If you don't want to run the nodes all the time you are better off without a cluster. You could use the Proxmox Datacenter Manager for migrating vms between the nodes.
 
  • Like
Reactions: leesteken
Johannes S said:
If you don't want to run the nodes all the time you are better off without a cluster. You could use the Proxmox Datacenter Manager for migrating vms between the nodes.
Click to expand...
I wasn't aware of Data Center Manager. Thanks for the idea. I do have a use case for it that I'll explore with an off-site PVE server. Unfortunately I don't think it handles my original case - I do want to keep HA redundancy for those 3 nodes. And if that means the UPS will run the systems for a bit less time (and frankly the UPS should only be needed a couple times a year) that really isn't a major hardship.
 
dga said:
I wasn't aware of Data Center Manager. Thanks for the idea. I do have a use case for it that I'll explore with an off-site PVE server. Unfortunately I don't think it handles my original case - I do want to keep HA redundancy for those 3 nodes. And if that means the UPS will run the systems for a bit less time (and frankly the UPS should only be needed a couple times a year) that really isn't a major hardship.
Click to expand...
Another alternative would be pve-zsync: https://pve.proxmox.com/wiki/PVE-zsync You would still have to manually switch on the vm in case of a failure, but maybe this could be scripted? Another option would be to implement HA on the application layer (one example is Keepalived for pihole/dns https://forum.proxmox.com/threads/pi-hole-lxc-with-gravity-sync.109881/ Please note that in newer pihole versions gravity-sync doesn't work anymore, nebula-sync is one of the alternatives).

Or as third option: Reduce your cluster to two nodes and have the third node as stand-alone node. The third node will only be turned on if you need it and host the stuff where you don't need HA. Now of course you still need quorum for your cluster but this could be achieved with a low-power device (e.g. raspberry pi) or a docker container on your nas (if your NAS os happen to support it).
 
Johannes S said:
Or as third option: Reduce your cluster to two nodes and have the third node as stand-alone node. The third node will only be turned on if you need it and host the stuff where you don't need HA. Now of course you still need quorum for your cluster but this could be achieved with a low-power device (e.g. raspberry pi) or a docker container on your nas (if your NAS os happen to support it).
Click to expand...
Ahh.. I think this makes sense. I've got way more horsepower than I need with 3 nodes and 2 would do fine. And I've got a spare pi I used to use for the quorum and could again. If I do want the 3rd node then Data Center Manager could be a neat solution.

Thanks all!
 
You must log in or register to reply here.
Top Bottom