We have a fairly large cluster now and several hundred applications running. Some of these applications are business critical - ie, they go down and we start to loose money while some of the applications are simple intern toys or hackathon projects.
We would be interested in potentially setting a "priority" level on applications - a simple weight of which applications get offer priority in the case of partial cluster outage or if a slave goes down and tasks need to be rescheduled.
We could do this with roles, but right now marathon only supports one additional role beyond the default '*' role
Another problem is preemption. If a subset of machines go down, we want to kill the super-low priority offers on other machines if room isn't available.