Health probes

Pod	State	Startup	Liveness	Readiness
web-1	Ready	S passed	L 0/3	R 0/3
web-2	Ready	S passed	L 0/3	R 0/3
web-3	Ready	S passed	L 0/3	R 0/3

Crate lamp: lit while the pod is Ready and in the Service's endpointsPlungers S, L, R: the startup, liveness and readiness probes, one tap per periodRed: a failed probe, and the tower's count of consecutive failures (L and R need 3, S allows 10)The kubelet runs each probe once per period. A failing readiness probe (Overload web-2) only takes the pod out of the Service: it keeps running and comes back when the probe passes. A failing liveness probe (Hang web-1) means the container is stuck, so after three failures the kubelet kills and restarts it. A slow starter (web-3) is the trap: without a startup probe, liveness starts checking after its initial delay and kills the app before it finishes starting, again and again, with CrashLoopBackOff growing the wait. With a startup probe on, liveness and readiness hold off until the app answers, up to ten failures.

Next: Requests, limits & QoS →