I have the same problem, and I think it is because the retry rule does not trigger when all pods are down. But I'm not completely sure about this.