An interactive walkthrough of Kubernetes probes surfaces a liveness probe regression accepted by SIG Node

Sources: ngrok blog, HN discussion
Summary: The post reports that liveness probes fire before the startup probe has succeeded, so a container killed by a liveness failure can be killed again during restart before it ever gets a chance to start. It states the behaviour was reproduced in k3s, minikube, and kind against v1.36.2, traced to v1.35.0, filed as an upstream issue, and that SIG Node accepted it and marked it priority/important-soon. The post also documents the ordinary probe failure modes, including a liveness probe that checks a database turning one database blip into cluster-wide CrashLoopBackOff.
Why it matters: Anyone running v1.35.0 or later with both a startup probe and a liveness probe configured can hit premature restarts.
Follow-up: Track an upstream fix for the liveness and startup probe ordering.