Build a livenessProbe, readinessProbe, or startupProbe spec with any handler (httpGet, tcpSocket, exec, grpc). See the exact time to detect failure that the kubelet will use — periodSeconds × failureThreshold + timeoutSeconds — plus a check-timing timeline so you can tell at a glance when the pod restarts (liveness), leaves the Service endpoints (readiness), or fails to start (startup).
liveness and startup. readiness accepts >1.
terminationGracePeriodSeconds on a probe overrides the pod's for the SIGKILL after a liveness failure — useful when the app needs quicker draining than the rest of the pod.
>=200 and <400. Redirects are followed as normal responses (302 → success). host defaults to the pod IP; a header Host: foo is preferred over the field.httpGet when possible.grpc.health.v1.Health/Check RPC. Requires --feature-gates=GRPCContainerProbe=true before 1.24 and the app to implement it.terminationGracePeriodSeconds on the probe if you want liveness-triggered kills to have a different grace than pod-triggered ones. Feature is GA in 1.27.