PR #93 changes health probe behavior, but some important invariants and ordering assumptions are only implicit in the code.
Requested follow-up:
- add a short comment near
/healthz/startup explaining that it is only safe because the HTTP listener binds after Database::new().await and migrations complete
- add a short comment in the shutdown/readiness path explaining that DB checks are intentionally skipped during shutdown and the outer
Shutting down response wins
- add a short comment noting that Cloud Run does not use readiness probes for traffic routing, so
/healthz/ready is mainly useful for smoke tests and Kubernetes-style consumers
Context:
This is non-blocking for PR #93 but worth documenting so the current behavior does not depend on silent assumptions.
PR #93 changes health probe behavior, but some important invariants and ordering assumptions are only implicit in the code.
Requested follow-up:
/healthz/startupexplaining that it is only safe because the HTTP listener binds afterDatabase::new().awaitand migrations completeShutting downresponse wins/healthz/readyis mainly useful for smoke tests and Kubernetes-style consumersContext:
This is non-blocking for PR #93 but worth documenting so the current behavior does not depend on silent assumptions.