Kubernetes Probes as Failure Detectors: Semantics, Timing, and Cascading Risk Kubernetes probes are distributed-systems failure detectors with different control effects. A readiness failure removes a Pod from Service endpoints; a liveness failure asks the kubelet to restart the container. The distinction matters because detection is necessarily imperfect: aggressive thresholds reduce detection latency but increase false positives under transient load. Readiness protects traffic; ...