keeping them around was resulting in some busy loops on timer events (i.e. retry after), making them unreliable, innacurate and CPU draining. they're now kept out whenever they're inactive.