Cluster node "standby" mode?

[esi_y]

Does PVE have anything (could be series of steps, but something deterministic, not "it depends") that will do the equivalent of RHEL's pcs cluster standby?

I understand I can do ha-manager crm-command node-maintenance enable to have whatever HA migrate elsewhere and I can migrate non-HA VMs manually (it's scriptable to move them later back too), but how do I prevent e.g. another user to create new resources manually on that node during that period?

Obviously, I do not want to remove the node from the cluster, it will be back from standby later on. Any better idea than shutting down the pve* services and leave others in limbo?

 

[LnxBil]

Any better idea than shutting down the pve* services and leave others in limbo?

We recently needed to shut down a node to save heat, so similar but same .... but other than that ... no idea.

Yet I have an immediate "counter question" for you: why would you want a standby node in a PVE cluster? I can answer the question for other type of clusters and environments, yet not for PVE. Can you elaborate?

 

[esi_y]

We recently needed to shut down a node to save heat, so similar but same .... but other than that ... no idea.

Yet I have an immediate "counter question" for you: why would you want a standby node in a PVE cluster? I can answer the question for other type of clusters and environments, yet not for PVE. Can you elaborate?

I do not think PVE-kind-of cluster is special in this, the HA command is there for a reason too. You are doing any sort of maintenance and really do not want anyone else to be attempting to start up new resources on that very node for that limited period. Note I do not need pveproxy off, they might be just fine connected to the GUI through the node just not be creating anything there (nor migrating to). Yes they might use other nodes, but how do they know which one is now "standby"?

 

[LnxBil]

I do not think PVE-kind-of cluster is special in this, the HA command is there for a reason too. You are doing any sort of maintenance and really do not want anyone else to be attempting to start up new resources on that very node for that limited period.

Okay, but maintenance is not a "valid" standby in my book. When I think about standby, I mean "on and can take work, but does not do anything now". It's a waste of energy and hardware cost ... unless things like Oracle 10-day-rule in licensing kicks in and a socket license cost is at 15k. I thought you meant something like that.

 

[esi_y]

Okay, but maintenance is not a "valid" standby in my book. When I think about standby, I mean "on and can take work, but does not do anything now". It's a waste of energy and hardware cost ... unless things like Oracle 10-day-rule in licensing kicks in and a socket license cost is at 15k. I thought you meant something like that.

I see. I have no strong opinion what it should be called, but used the shortest possible description and put it into quotes as it is essentially pacemaker vocabulary.

There's also "maintenance-mode" for the whole cluster there, but the "standby" is single node only. It's neither online nor offline, that's about it I guess. Interestingly it e.g. still votes in the quorum, it just does not allow new resources to be started there and (as expected) survives in that status through a power cycle. It's essentially what PVE supports for HA resources, I just wonder how come it is HA-feature only and no one needed this so far? I really thought it is there somewhere but I just cannot find it.