James Szalay

ALT A screenshot from https://ngrok.com/blog/probes that reads: “You might notice that when you set the /live endpoint to 503, an odd sequence of events happen. The liveness probe fails. The kubelet restarts pod-a’s container. Before the startup probe has a chance to fire, a liveness probe fires and fails because the container is still in startup. The kubelet restarts the container again, before it has a chance to pass its startup probe! This is a real Kubernetes bug that I found during the making of this post. Liveness probes shouldn’t be firing before the startup probe has succeeded. I reproduced this in k3s, minikube, and kind against the latest Kubernetes at time of writing v1.36.2. The bug appears to have been introduced in v1.35.0. I’ve filed an issue about it and SIG Node have accepted it and marked it priority/important-soon.”