Is there some way to restrict which nodes within a cluster can host containers or VMs? I know that I can simply not place containers or VMs on various nodes, but I would prefer some way to restrict which nodes are shown in dialogs for hosting containers or VMs, such as migration targets.
Classify Cluster Node Roles?
- Thread starter jamin.collins
- Start date
Can you elaborate on this? I believe I've used them to some degree in the past, but I don't believe their use removed other non-HA member hosts from the migrate target dialog.
Setting HA on each VM seems quite tedious, is there either a scriptable or mass method of assignment?
NVM: found `ha-manager`
NVM: found `ha-manager`
Last edited:
So, HA seems to move the VMs to the desired hosts, but does not appear to alter (or restrict) the migration target drop down in the UI.
That's right but the HA mechanism would immedently remigrate the vm to a desired ha host group member if manually "wrong" migrated and even not back to "first" host when you eg. enabled maintenance mode on it (as reason for a manual migration).
You could assign the storage for the VMs only to the nodes where they should run. No storage, no images, no VMs.
My use case is (currently) to have two different classifications of proxmox hosts:
- CEPH OSD nodes
- Virt nodes
So, storage for the VMs is by definition the CEPH OSD nodes. However, I don't want to have any VMs running on the CEPH OSD nodes.
Interesting, that brings things one step closer. The UI now gives an error when I try to migrate a VM to one of the CEPH nodes.
Would still be better to not even list the invalid or undesired targets.
Would still be better to not even list the invalid or undesired targets.
The challenge is, I think, that the group is not "VM can be here" it is "VM can run here." If you shut down a VM you can move it outside the HA group. If you then start it, it will migrate off to a node that is in the group. So, if the VM is off, all the nodes are valid choices.
I know this thread is a bit old, but just had a power outage that resulted in some issues with this setup.
As a result of the power outage, all of my virt nodes went offline, while the CEPH nodes remained online and functional. As a result of this all the virtual machines seemed to migrate to the CEPH nodes, but since the backing storage isn't allowed on the CEPH nodes the virtual machines were all in an shutdown state. Not ideal, but also not truly an issue, yet.
Once I brought the virt nodes back online, I attempted to migrate the shutdown VMs back to the various virt nodes. Every attempt (UI or CLI) failed with an error that the CEPH node the virtual machine was on did not have access to the backing storage. Mind you, the virtual machines were all in a shutdown state, and I wasn't (AFAIK) asking in any way for them to be started, just migrated off the host where I didn't want them to run.
In the end, to migrate the hosts off the CEPH nodes, I had to enable the various backing storage pools on the CEPH nodes.
As a result of the power outage, all of my virt nodes went offline, while the CEPH nodes remained online and functional. As a result of this all the virtual machines seemed to migrate to the CEPH nodes, but since the backing storage isn't allowed on the CEPH nodes the virtual machines were all in an shutdown state. Not ideal, but also not truly an issue, yet.
Once I brought the virt nodes back online, I attempted to migrate the shutdown VMs back to the various virt nodes. Every attempt (UI or CLI) failed with an error that the CEPH node the virtual machine was on did not have access to the backing storage. Mind you, the virtual machines were all in a shutdown state, and I wasn't (AFAIK) asking in any way for them to be started, just migrated off the host where I didn't want them to run.
In the end, to migrate the hosts off the CEPH nodes, I had to enable the various backing storage pools on the CEPH nodes.
9.x has node affinity/negative affinity rules. I don’t recall offhand if it hides the node In dropdowns though. Mostly our migrations are for maintenance mode so automatic.