If you are running a cluster that uses kube-proxy for service IP to endpoint mapping, then it is expected that kube-proxy will be restarted shortly after it exits by a monitoring process (e.g. systemd, monit, supervisord, etc). In fact, in later versions of Kubernetes, kube-proxy runs as a privileged container and the kubelet ensures that it stays running. Since it is expected to be restarted quickly, cleaning up iptables would just cause them to be modified unnecessarily each time kube-proxy was restarted.

If you don't want kube-proxy to manage iptables for you, then you can decide not to run it in your cluster at all, or to manually clean up the iptables rules after you stop it.