add 2 qdevice
- Thread starter backbeard
- Start date
Why?
Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox
Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox
I have a kind of weird setup currently, it's a temporary setup until I have time to do it properly.
1 Main node with a virtualised firewall and other essential services, 2 other proxmox nodes that aren't always online to save energy and other reasons. I have also a qdevice setup on a cloud vm that connect home using a vpn. If the other 2 nodes go down it's all good as I have the qdevice on the cloud vm, however, if the vpn link is served I cannot start the virtualised firewall as the qdevice is unreachable hence no quorum. So, for a short time, I want to add a lxc container with a qdevice, just to be sure the VMs on the main node can start and provide connectivity to the remove qdevice.
I know isn't a pretty setup, but that's the setup I want for now.
Any help?
1 Main node with a virtualised firewall and other essential services, 2 other proxmox nodes that aren't always online to save energy and other reasons. I have also a qdevice setup on a cloud vm that connect home using a vpn. If the other 2 nodes go down it's all good as I have the qdevice on the cloud vm, however, if the vpn link is served I cannot start the virtualised firewall as the qdevice is unreachable hence no quorum. So, for a short time, I want to add a lxc container with a qdevice, just to be sure the VMs on the main node can start and provide connectivity to the remove qdevice.
I know isn't a pretty setup, but that's the setup I want for now.
Any help?
That's not how a QDevice works and not how Proxmox is designed to work. However, you can give a node more votes (and make it a single point of failure) if you really want to. I don't know the command/configuration but you can probably find it by searching the forum.
EDIT: Why run a cluster at all? Maybe separate nodes with remote-migrate might fit your use-case better?
EDIT: Why run a cluster at all? Maybe separate nodes with remote-migrate might fit your use-case better?
You would be well advised to move your Qdevice to a local raspberry pi, or even a Virtualbox VM running on your desktop - as long as it's on 24/7.
Clearly you have not tested a DR situation, you don't want a Qdevice in the cloud somewhere and unreachable if things go sideways.
And as leesteken mentioned, you should seriously consider unclustering if most of your infrastructure is powered off -- cluster is designed for HA and failover. You can run stuff on separate nodes easily with backup and restore, and manage different nodes from separate tabs in the browser.
Clearly you have not tested a DR situation, you don't want a Qdevice in the cloud somewhere and unreachable if things go sideways.
And as leesteken mentioned, you should seriously consider unclustering if most of your infrastructure is powered off -- cluster is designed for HA and failover. You can run stuff on separate nodes easily with backup and restore, and manage different nodes from separate tabs in the browser.
I totally agree, uncluster my setup will be a better solution, however, this would require some downtime that I can't plan right now.
My question is if we can actually have 2 qdevices, is that actually supported?
Such a configuration would not make technical sense, therefor it has not been tested, and that means it's not supported.
Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox
My understanding is that you put the 2nd qdevice with 2 votes, and the other with 1 vote hoping for an odd number with your 2 node cluster.
quorum expect = 5
I think people on the forum get pedantic around the even numbered quorums, due to the situ where 1 node+1qdevice are in an island opposed to the other node and +1qdevice. Which I expect will happen every now and then, but since my 3 nodes are hopefully still in contact I will be fine.
I could be wrong. Maybe the votes should 1 and a 3 vote? for the qdevices (in 2 node clustrer) but that makes 6.
When my workstation reboots or goes down, it takes 2 votes with it so expect = 3 afterward.
Or if my 1 vote qdevice goes down, the cluster behaves with 2 nodes and 2 votes from one qdevice