This would include starting the configurable-http-proxy separately from JupyterHub, which means JupyterHub could be restarted without interrupting users who are currently using the system (though it would prevent users from logging in during the restart).
This also includes creating an ansible tag or something that makes it easy to just perform the tasks necessary for restarting JupyterHub.