Health checks
Both endpoints are unauthenticated and answered by every role, including workers (which serve nothing else).
Liveness: GET /api/health
Section titled “Liveness: GET /api/health”The process is up.
{ "ok": true, "role": "api" }The necodoc-api image’s Docker HEALTHCHECK calls this, as do the chart’s startup and liveness probes.
Readiness: GET /api/health/ready
Section titled “Readiness: GET /api/health/ready”200 when the node can do its job, 503 otherwise.
{ "ready": true, "role": "api", "database": true, "queue": true, "storage": { "driver": "s3", "ok": true }, "workersOnline": 2}database:select 1succeeds.queue: pg-boss has started on this node.storage: the output storage round-trips (see Storage). On failure it carries anerrormessage. The check times out after 5 seconds, so a hung mount fails the probe instead of hanging it.workersOnline: workers with a recent heartbeat. Informational only: an API node with no workers is still ready (it can accept async work), but nothing renders. Alert on0.
Point load balancers and Kubernetes readiness probes at this endpoint. Use a timeout above 5 seconds.
Behind the app container both endpoints are at the same paths, e.g. https://necodoc.example.com/api/health/ready.