deploy-vps fails a healthy rollout when the post-deploy prune collides #89

New issue

Closed

opened 2026-09-16 18:53:15 +00:00 by ahmad · 0 comments

ahmad commented

2026-09-16 18:53:15 +00:00

Owner

Context

deploy-vps ends its remote script with an image prune after the health wait. On a host where two deploys overlap, the second prune is refused by the daemon with a prune operation is already running, the script exits non-zero, and the action reports Deployment of <version> failed for a rollout it had just declared healthy. Seen on amtronics/agent-bus v0.3.0 (run 17, 2026-09-16): agent-bus-1 is healthy after 122s. followed by the prune error, a failed job, a false tracking issue and a red Telegram message, while the container served the new version throughout.

Scope

In: the prune is housekeeping, not part of the rollout — its failure is logged and ignored, and a deployment's verdict is the health wait's alone.
Out: pruning policy (what and how old), the health wait itself.

Acceptance criteria

Given a rollout whose containers report healthy, when the post-deploy prune fails for any reason, then the action still reports the deployment as succeeded and the prune error appears in the log.
Given a rollout whose containers do not report healthy in time, when the script ends, then the action still reports the deployment as failed, as today.

### Context `deploy-vps` ends its remote script with an image prune after the health wait. On a host where two deploys overlap, the second prune is refused by the daemon with `a prune operation is already running`, the script exits non-zero, and the action reports `Deployment of <version> failed` for a rollout it had just declared healthy. Seen on `amtronics/agent-bus` v0.3.0 (run 17, 2026-09-16): `agent-bus-1 is healthy after 122s.` followed by the prune error, a failed job, a false tracking issue and a red Telegram message, while the container served the new version throughout. ### Scope - In: the prune is housekeeping, not part of the rollout — its failure is logged and ignored, and a deployment's verdict is the health wait's alone. - Out: pruning policy (what and how old), the health wait itself. ### Acceptance criteria - [ ] Given a rollout whose containers report healthy, when the post-deploy prune fails for any reason, then the action still reports the deployment as succeeded and the prune error appears in the log. - [ ] Given a rollout whose containers do not report healthy in time, when the script ends, then the action still reports the deployment as failed, as today.

ahmad added the

labels

2026-09-16 18:53:15 +00:00

ahmad referenced this issue from a commit

2026-09-16 18:56:19 +00:00

fix: Keep a colliding prune from failing a healthy rollout

ahmad added

state/in-review