We recently found some interesting Starman feature.
Since it is a master/workers software there is, like in Apache, a way to
"gracefully" rebuild the workers without stopping master :
kill -HUP 123 (123 being the master PID).
Unlike stop/start of starman, the used workers will not be stopped until
there task is done. During that idle workers will indeed be killed and
This is useful when one wants to refresh Plack persistence on a
production running server.