rune restart / stop

rune restart

Replaces every instance with a fresh one at the current spec. The server stamps a new template generation in a single write and the reconciler swaps the instances, so the desired scale never dips through zero. Restarting a stopped service starts it at its last non-zero scale.

rune restart api
rune restart api -d            # detach — exit after the request is sent
rune restart api --timeout=15m

Use after rotating a secret, updating a configmap, or anything else that requires a fresh process. Mounted secrets/configs do not hot-reload.

Recovering instances stuck in create

An instance that never got a container — an unpullable image or a rejected registry credential exhausting every create attempt — ends up frozen in Stalled. The reconciler deliberately stops retrying it automatically.

rune restart re-arms those instances: it clears the create-attempt counter and the retry backoff so the reconciler tries again on its next tick. The re-arm is best-effort and runs per instance after the generation restamp — if one fails it is logged, not reported, so a stuck instance can survive a restart. Fix the underlying cause first (renew the credential, push the missing tag), then:

rune restart api

If a Stalled instance is still carrying the old template generation when the wait runs, restart gives up promptly rather than burning the full --timeout — that instance holds the slot and the reconciler will not retry it, so the replacement can never be created:

restart cannot proceed: api-7c2e (Stalled) never got a container and remain
stalled, so the replacement cannot be created. Either the instance exhausted
its create attempts again (check `rune describe service api -n default` for
the reason — an unpullable image or rejected registry credential is the usual
cause), or this server predates the restart re-arm fix, in which case
`rune scale api 0 -n default && rune scale api 1 -n default` clears the slot

Almost always the underlying problem is still there: read the reason from rune describe service api, fix it, then rune restart again. The message also names an older-server case it can't rule out from the client side — on a server without the re-arm behaviour, rune scale api 0 && rune scale api 1 is the only way to clear the slot.

An instance that re-stalls on the new generation is a different case: it counts as a failed replacement, so the wait runs to its timeout instead of exiting early.

rune stop

Scales to 0 but keeps the spec. Restore later with rune scale <name> N or rune cast.

rune stop api
rune stop api -d   # detach

To remove the spec entirely, use rune delete.

See also

• Scale & restart
• rune scale