Manage nodes

Node writes (promote, demote, rm, update) must run on the Raft leader. Reads (ls, inspect) work on any manager.

sudo cellar node ls
sudo cellar node inspect <id>

# Role changes (target applies via heartbeat: re-issue cert + open/close Raft)
sudo cellar node promote <worker-id>
sudo cellar node demote <manager-id>

sudo cellar node update <id> --availability drain
sudo cellar node update <id> --label-add zone=a --label-rm zone
sudo cellar node update <id> --role manager   # same as promote

# Remove a down/orphaned node from cluster state (--force if still heartbeating)
sudo cellar node rm <id>
sudo cellar node rm --force <id>

Availability:

• pause — cordon only: no new placements; existing sandboxes stay.
• drain — cordon and reschedule running sandboxes onto other live nodes. Bind-mounted sandboxes stay until they stop or you remove them.

node rm / cellar leave delete the Raft node record (and Raft voter for managers). Running sandboxes still assigned to that node are rescheduled (bind mounts fail, same as a dead node). The remote daemon keeps its local identity until you run cellar leave there (or it observes removed on heartbeat and clears itself).