4 ms·
Also, for homelabbing, there's a magic incantation that allows you to configure Proxmox away from the default High Availability cluster mode. By default, the H
by alias_neo 3y ago
Also, for homelabbing, there's a magic incantation that allows you to configure Proxmox away from the default High Availability cluster mode.
By default, the HA mode insists that all nodes are up or others won't come up, making this tweak (it may be contained in the OP scripts, I don't recall) allows you to retain many of the benefits of having Proxmox machines "connected", without requiring you treat them like a HA cluster.
In my case, I have a node I shutdown when it's not used, I use it for specific occasional tasks, and then I have nodes I want up all of the time.
With the non-HA tweaks, you can still do things like centrally manage, migrate VMs and containers between nodes etc, without the limitations of it wanting them all up and available all of the time.
- johntash 3y agoWould this work for having proxmox modes across multiple datacenters? IIRC it was sensitive to latency I don't really use the HA features of proxmox that often, mostly just for migrating vms across hosts. But I had been thinking of re-installing some of my remote servers with proxmox and hooking them up to the same cluster so that I could potentially shut off all of the servers in my basement and run everything from a remote location, then migrate the vms back later.
- j45 3y agoOP may know better, but in part it may depend on the ms latency between the sites?
- alias_neo 3y agoI've not noticed anything latency-sensitivty, perhaps the Ceph feature? I'm not using that though. There's also the issue of quorum of it's only a 2-node cluster, though I've personally not had any issues with it, despite repeatedly restarting the different nodes. I see no reason why you couldn't do this to allow you to move your VMs from home to elsewhere and back again as you just mentioned.
- johntash 3y agoceph is latency sensitive, but that should only matter for the servers running ceph and not (as much) the clients using it. I thought quorum feature of proxmox was also latency sensitive, but this is the only thing I found: https://pve.proxmox.com/wiki/Cluster_Manager#pvecm_cluster_network_requirements https://pve.proxmox.com/wiki/Cluster_Manager#pvecm_cluster_n... Maybe this doesn't matter if not using HA? I'll read more about it and maybe test it out
- alias_neo 3y agoYeah I've not had any issues. I have nodes they are offline for months when not in use, it doesn't seem to have any effect of you're not using HA. Just my experience though, YMMV.