Does restarting kubelet stop all nodes?

Question

If I run systemctl restart kubelet will it impact the other running nodes? Will it stop the cluster? Can you foresee any impact? Any help would be appreciated!

Answer 1

Before answering, small disclaimer: restart is not due to multiple potentially breaking configuration changes to kubelet and kubelet is indeed restarted not crashed due to misconfiguration. Answer is aimed at scenario of simple restart of kubelet (with maybe small and non-breaking configuration change)

will it impact the other running nodes?

Just a restart as such should not be an issue (providing it is really restart as defined in disclaimer above). There is difference if you restart kubelet on master or worker node. During restart on master as long as all system pods are running uninterrupted all should be well but if during kubelet downtime any system pod needs restart as well you are in trouble until kubelet gets operational again... For worker node (if you didn’t change default) Kubernetes waits 5 min for node to get back to ready state (kubelet gets operational after restart). Again supposing that pods are live and well during that time - if any of them fails liveness probe it will be restarted on another node but it will not be communicated back to node in question until kubelet is back online (and docker will continue to run it until then)...

Will it stop the cluster?

Again, if on worker then - no, if on master then no IF system pods (api, controller, scheduler, DNS, proxy...) continue to run uninterrupted on master during restart.

Can you foresee any impact?

If after restart kubelet crashes and you are operating on master and then any system pod crashes while kubelet is down you are in for a trouble (depending on what crashed).

Make sure that you didn’t introduce any breaking kubelet config changes that will break kubelet on node you are restarting, especially for master node...

To conclude: simple restart should not be an issue, make sure that your config is correct, kubelet restart is proper and monitor system pods during restart if restarting on master.