Hello, I am trying to mask out one node in my cluster from migrations of vms from crashed hosts. I have a stretched ceph cluster with two data centers of two nodes each. I have a fifth node that acts as a quorum member only and does not have any storage but acts has a witness node for the cluster. I do not want any vms to migrate to this node. I have masked out the storage and disable cluster services on the node. When I perform a fail over vms will move to this node and go into a HA state of Freeze. This is a running Proxmox VE 9.1.7 so I cannot use HA groups to restrict this.
Problem with blocking node from HA VM migrations
- Thread starter flannery65
- Start date
Does that fifth node need to be a full PVE node? As a QDevice only it would vote but have no PVE services...
I am thinking that might be the way to go, however how would I configure that as a ceph witness node so that my storage would still avoid split brain when one data center goes down? Or is it just simply a "fifth" vote what is automatically recognized by the storage cluster?
QDevice works for the cluster. You could just have 3 Ceph monitors, which is all Proxmox recommends: https://pve.proxmox.com/wiki/Deploy_Hyper-Converged_Ceph_Cluster#pve_ceph_monitors
My test environment is a stretched cluster like yours. The witness node is a full cluster member.
I use a HA Node Affinity rule to prevent unwanted VMs from running on the witness node. The rule includes all the hosts except for the witness and is set in strict mode. The VMs are assigned to this rule.
I just tested it by moving 6 VMs onto one node and then crashed it. The VMs only restarted on the hosts assigned to the group.
I use a HA Node Affinity rule to prevent unwanted VMs from running on the witness node. The rule includes all the hosts except for the witness and is set in strict mode. The VMs are assigned to this rule.
I just tested it by moving 6 VMs onto one node and then crashed it. The VMs only restarted on the hosts assigned to the group.