Inspect startup and health-check expectations
Review deployment logs for process start, listening port, and the first health request. Confirm that the service binds to the host and port required by the platform, not only to a local interface or development default.
Check whether the endpoint is called before the application is ready and whether it depends on optional services. A liveness check indicates the process can continue; a readiness check may include dependencies required to serve traffic.
Fix the cause and verify recovery
Validate runtime variables and secrets without printing their values, then inspect database or external-service connectivity from the deployed environment. A local build cannot confirm production credentials or network routes.
Set startup grace periods based on observed initialization time, not to mask a hang. After a change, confirm the health check passes on a fresh deployment and genuine failures still remove the instance from service.