True for long running services, but not necessarily ephemeral workloads. Those need a system to schedule them or else once they complete they just stop running.

True for long running services, but not necessarily ephemeral workloads. Those need a system to schedule them or else once they complete they just stop running. Also, as long running services crash or reboot naturally they need to be rescheduled or else your cluster slowly dies, and as your cluster size increases the mtbf decreases and the need to reschedule workloads continually increases.