A running process is not necessarily ready to serve useful traffic.

A liveness check asks whether the process is still functioning enough that restarting it could help. A readiness check asks whether the instance should currently receive new work.

An application may be live but not ready while starting, warming a cache, waiting for a required dependency or draining before shutdown.

During a graceful shutdown, an instance should normally stop accepting new work, allow suitable in-flight operations to finish within a bounded period, then terminate. Killing it immediately can turn routine deployment into request failures or partial work.

Health checks should test the property their consumer needs. A liveness probe that fails whenever a remote dependency is briefly unavailable can create restart loops without repairing that dependency.

Lifecycle signals let deployment infrastructure distinguish “process exists” from “instance can safely receive work.”