Cordon, drain and maintenance

Action	Effect	Command
Cordon	No new instances on the node. Running instances keep running.	stratumctl nodes cordon n-abc123
Uncordon	The node takes new instances again.	stratumctl nodes uncordon n-abc123
Drain	Cordon, then move every instance off the node by the gentlest method it qualifies for.	stratumctl nodes drain n-abc123
Maintenance	Takes the node out of service for host work. Every instance on it goes to error.	stratumctl nodes maintenance n-abc123

Console: the node’s action menu under Admin › Metal.

Drain

See what a drain would do before starting it:

stratumctl nodes evacuation-plan n-abc123

Each instance is classified:

Class	When	What happens
live	Running, root volume on shared storage	Live-migrated, no downtime.
rehome	Stopped, on shared storage	Only its home node changes.
cold	Root volume on the node’s own LVM-thin pool	Stopped and moved with its disk.
blocked	No room elsewhere, or stop-protected and unable to move live	Not moved.

A drain that would stop a running instance refuses and names it. --allow-disruption permits the stop:

stratumctl nodes drain n-abc123 --allow-disruption

On a cluster

Drain a node, then reboot or shut it down, to keep its instances running through host work. Instances on Ceph move live; instances on a node-local pool are stopped, and drain refuses unless you pass --allow-disruption.

On a single node

There is no other node to move instances to, so drain and maintenance do not help with host work on a single node. Instead:

• A reboot (systemctl reboot) shuts instances down cleanly and starts them again afterwards.
• A full stop for hardware work: stratumctl admin cluster shutdown. See Power and shutdown.

cordon is still useful to stop new launches while preparing.