opschain admin server node
Included from Basic tier
The per-node operations verbs for high-availability estates: durably drain
ONE server node so it can be killed and rolled (quiesce), see which nodes
are held (list), and bring a node back with an explicit operator act
(resume). Nodes are addressed by their STABLE server name — the
server_name config key (or OPSCHAIN_SERVER_NAME), defaulting to the
node's hostname — never by the per-boot replica id.
A quiesced node drains its own participation while the rest of the estate
carries on: readiness flips to 503 (the load balancer routes around it),
job offers stop, its scheduler shards, leadership and membership are handed
back to peers, and its in-flight run drives detach for peers to adopt. The
node is then safe to kill — kill -9 included — because every run is
resumable from the store. The posture is durable: a binary restarted under
it boots HELD (healthy and inspectable, taking no work) until resume.
There is deliberately no timer or automatic rejoin — resume is always a
human act.
This is the belt-and-braces binary-roll procedure: quiesce the node, kill it, replace the binary, inspect the held boot, resume — node by node across the fleet, with running changes riding through on their peers.
| Subcommand | Purpose |
|---|---|
list | Show the quiesced nodes, and the answering node's own name and held state — opschain docs cli/admin-server-node-list. |
quiesce <server-name> | Durably drain one node for a kill-and-roll — opschain docs cli/admin-server-node-quiesce. |
resume <server-name> | The explicit human resume a held node waits for — opschain docs cli/admin-server-node-resume. |
See also
opschain docs cli/admin-server— the fleet-wide dispatch quiesce (every node at once, runners still parked).opschain docs cli/admin-platform— the estate-wide system quiesce and the emergency halt tier.