Daemon stuck on an old deployment: promotions and scale-to-0 are accepted but never executed

TraderMigs · 2026-11-11 07:54:58.538963+00

Plan and setup Hobby team, 1 project, 1 daemon (voice-worker), region iad1, 1 instance, 1 GiB, built from a Dockerfile. "Auto-assign custom production domains" is off, so production deployments are staged and then promoted. Daemon quota is not exhausted (2 h 55 m of 5 Active CPU-hours, 139 of 420 GB-hours). What happens Since about 2026-11-11 03:30 UTC the daemon controller accepts changes for this daemon but does not carry them out. The running instance keeps working normally, on the deployment that was promoted the day before. 1. Three production promotions did not roll the daemon (all times UTC, 2026-11-11) dpl_BunEVZ5FCYBoHgiXoii8ZCAgGF3u promoted 04:44:41 (created through the API) dpl_6h1v2dpYX8gsVhRaghVZrYk67Dsb promoted 05:43:20 (created through the API) dpl_EGzkA29tmoxZYotNYGFJUJX38UkN promoted 06:22:52 (created by a git push) Each one became the current production deployment (readySubstate PROMOTED, all aliases completed) and each lists the daemon under Resources with its own image. The daemon stayed on the previous deployment. 2. A scale-down was accepted but not executed 06:47:23: Scale dialog, iad1 set from 1 to 0. Response: "Daemon scaled". The daemon record then showed instances 0 in all regions, configuredInstances iad1 = 1, desired 0, status "reconciling". The instance was still alive and working more than 20 minutes later. 07:09:50: "Clear scaling overrides". This worked as a configuration change: the record returned to instances iad1 = 1, desired 1, status "ready". Ten minutes later the same old instance is still the one running, on the old deployment. Daemon record (GET /v1/projects/{id}/daemons), unchanged through all of the above deploymentId dpl_7BD9MhzwZeaWM9X6GqtackLi7Bu4 (promoted 2026-11-10 16:13:30) image daemon.voice-worker@sha256:8e3025faee9e... instance dmni_01m733x9qjk48a76spvvjcjexb, startedAt 2026-11-10T16:15:56.986Z, alive true instance lastCheckedAt 2026-11-11T03:30:52.581Z, which has not advanced since daemon updatedAt 2026-11-10T16:15:56.966Z What worked before 2026-11-06: a promotion rolled the daemon within 2 seconds. 2026-11-10: a promotion rolled the daemon within 2 minutes 28 seconds. The documentation says new instances start "right away" when a deployment becomes the production deployment. Not tried Restarting the instance, redeploying with a renamed daemon, or switching to direct production deployments. Questions Is the daemon controller stuck for this project, and can it be reset on Vercel's side? Is there a supported way to make the daemon reconcile to the current production deployment? Is there anything on the Hobby plan that pauses daemon rollouts without stopping the running instance?

Plan and setup

Hobby team, 1 project, 1 daemon (voice-worker), region iad1, 1 instance, 1 GiB, built from a Dockerfile.
"Auto-assign custom production domains" is off, so production deployments are staged and then promoted.
Daemon quota is not exhausted (2 h 55 m of 5 Active CPU-hours, 139 of 420 GB-hours).

What happens

Since about 2026-11-11 03:30 UTC the daemon controller accepts changes for this daemon but does not carry them out. The running instance keeps working normally, on the deployment that was promoted the day before.

1. Three production promotions did not roll the daemon (all times UTC, 2026-11-11)

dpl_BunEVZ5FCYBoHgiXoii8ZCAgGF3u promoted 04:44:41 (created through the API)
dpl_6h1v2dpYX8gsVhRaghVZrYk67Dsb promoted 05:43:20 (created through the API)
dpl_EGzkA29tmoxZYotNYGFJUJX38UkN promoted 06:22:52 (created by a git push)

Each one became the current production deployment (readySubstate PROMOTED, all aliases completed) and each lists the daemon under Resources with its own image. The daemon stayed on the previous deployment.

2. A scale-down was accepted but not executed

06:47:23: Scale dialog, iad1 set from 1 to 0. Response: "Daemon scaled".
The daemon record then showed instances 0 in all regions, configuredInstances iad1 = 1, desired 0, status "reconciling". The instance was still alive and working more than 20 minutes later.
07:09:50: "Clear scaling overrides". This worked as a configuration change: the record returned to instances iad1 = 1, desired 1, status "ready". Ten minutes later the same old instance is still the one running, on the old deployment.

Daemon record (GET /v1/projects/{id}/daemons), unchanged through all of the above

deploymentId dpl_7BD9MhzwZeaWM9X6GqtackLi7Bu4 (promoted 2026-11-10 16:13:30)
image daemon.voice-worker@sha256:8e3025faee9e...
instance dmni_01m733x9qjk48a76spvvjcjexb, startedAt 2026-11-10T16:15:56.986Z, alive true
instance lastCheckedAt 2026-11-11T03:30:52.581Z, which has not advanced since
daemon updatedAt 2026-11-10T16:15:56.966Z

What worked before

2026-11-06: a promotion rolled the daemon within 2 seconds.
2026-11-10: a promotion rolled the daemon within 2 minutes 28 seconds.

The documentation says new instances start "right away" when a deployment becomes the production deployment.

Not tried

Restarting the instance, redeploying with a renamed daemon, or switching to direct production deployments.

Questions

Is the daemon controller stuck for this project, and can it be reset on Vercel's side?
Is there a supported way to make the daemon reconcile to the current production deployment?
Is there anything on the Hobby plan that pauses daemon rollouts without stopping the running instance?