Repository navigation
A component with healthCheck type none always fails its deploy #96
Description
Found deploying Abecra staging (customer zero) with Cloudfall 0.6.0.
component.schema.json accepts healthCheck: {type: none} for a component that serves no HTTP (a queue worker), but engine/ansible/roles/cloudfall_deploy/tasks/main.yml ("Gate the release on its declared health check") always runs ansible.builtin.uri with the declared scheme, port and path. With type: none those are undefined, the task fails, and the first release of the component is stopped: "Release ... failed its health check and was stopped because no previous release exists".
Proposal
type: none: skip the HTTP gate; instead check the systemd unit isactiveand stays active for a short settle time (for example 10 s), so a worker that crashes on start still fails the deploy- A test in the deploy role for a component with
type: none
Acceptance
- A component with
healthCheck: {type: none}whose unit stays up deploys - One whose unit exits within the settle time fails the deploy and is rolled back
Workaround in use: the worker's health check points at the API of the same release.
Activity
shipmill-romamo commentedon Oct 20, 2026
Triage: implement.
Confirmed on main (348ed73):
component.schema.jsonacceptshealthCheck: {type: none}, but "Gate the release on its declared health check" inengine/ansible/roles/cloudfall_deploy/tasks/main.ymlalways runsansible.builtin.uriwithhealthCheck.scheme,portandpath, which anonecheck doesn't have. The task fails, the rescue runs, and a first release is stopped. The schema promises something the role breaks, so this is a bug fix.The plan:
type: none, skip the HTTP gate and check the systemd unit instead: it must beactiveright after the restart and stillactiveafter a fixed settle time (10 s), so a worker that crashes on start still fails and rolls backtype: httpkeeps today's tasks unchangednonecomponent whose unit stays up, and one whose unit exits inside the settle timeThe settle time is fixed, not a new field; a per-component setting can follow if 10 s turns out wrong for someone. I'll link the PR here.
shipmill-romamo commentedon Oct 20, 2026
The fix is in #107. For
healthCheck: {type: none}, the gate skips the HTTP request and checks the unit: it must beactiveright after the restart, stillactive10 s later, and itsNRestartscount can't go up in between. That way a worker that crash-loops underRestart=on-failurestill fails the gate. Any failure goes through the same rescue as an HTTP check.type: httpis unchanged.Trade-offs: every
type: nonedeploy now waits 10 s at the gate. The settle time is a role default (cloudfall_deploy_unit_settle_seconds), not a schema field.restart,rollbackand the health probe still use the shared gate with no settle wait. Nothing has run against a live systemd yet. This issue closes when #107 merges.shipmill-romamo commentedon Oct 20, 2026
Released in v0.8.0rc1 (pre-release).
uv tool install --prerelease allow cloudfall==0.8.0rc1A plain install gets it with the next stable release.
If you filed this: react 👍 on this comment once it works for you, or reply if it doesn't.