Journal / Operations
Health checks that tell the truth
A process can be alive while the application is unusable. A useful health model separates liveness, readiness and dependency checks.
The underlying problem
Liveness answers whether the process should be restarted. Readiness answers whether it can receive traffic. Dependency checks describe whether downstream services are available. Mixing these into one probe can create restart loops during an ordinary database outage.
A practical approach
Start with a lightweight liveness endpoint that does not require external dependencies. Make readiness reflect essential prerequisites, and expose richer diagnostics through protected observability endpoints. Alert on sustained user-visible failures, not on every transient probe miss.