Health and Readiness Endpoints

Split liveness and readiness for this service. A load balancer should not send traffic to a process that cannot reach its database, and a restart should not happen just because the database blipped if the process is fine. SERVICE Name: [NAME] Dependencies that mean "cannot serve": [DATABASE, QUEUE, NONE] Dependencies that are degraded but alive: [EMAIL, SEARCH] Framework: [FRAMEWORK] DELIVER 1) GET /health — process is up, no dependency calls, 200. This is what the restart policy hits. 2) GET /ready — checks dependencies with a short timeout, 200 or 503, body lists which check failed without leaking host strings that are secrets. 3) Timeouts and caching: do not open a new connection storm; reuse the pool and cache the ready result for one second. 4) The Kubernetes or compose probe config, with the right path on the right probe. 5) A failure mode note: a migration in progress, a drained instance, a disk full. Do not put a secret in the health body. Do not require auth on the probe if the probe cannot send it; bind the endpoint carefully instead and say how.