MIR-906

Deploy should reset crash cooldown for the app

Done public

phinze Opened Mar 27, 2026 Updated Mar 27, 2026

Add `miren app restart` and reset crash cooldown on deploy

Bug

Deploying a new version of an app does not reset the crash cooldown. If an app is in crash backoff (e.g. from a server restart), a fresh miren deploy creates new sandboxes but they can't boot because the cooldown timer is still active.

coordinator.httpingress sandbox died early while acquiring lease │ error: "failed to request pool capacity: sandbox died while booting: application in crash cooldown until 2026-03-27T15:38:27Z (consecutive crashes: 6)" app: app/meet

Expected

A deploy is an explicit operator action saying "run this new version now." It should reset the crash counter and cooldown timer, allowing the new version to boot immediately.

Context

Discovered during MIR-890. After rolling Garden back from a broken dev build, apps were stuck in crash cooldown. Redeploying meet created a new version with new sandboxes, but they still couldn't start due to the cooldown from the old version's crashes.

See also MIR-905 (miren app restart) for the manual reset path.