I propose that a Mesos scheduler should optionally implement an request in order for Mesos to determine its health. Health could be expressed in terms of metrics mean percentile values e.g.: mean75, mean95, mean98, mean99, mean999. The values are open to interpretation and are domain specific, but could represent queue latency, free memory and so forth. Multiple metrics could be returned.
DC/OS packaging could allow for the expression of the health endpoint URI. A formulae could also be expressed that determines what represents green, amber and red conditions using the metric data and its mean percentiles.
DC/OS packaging could also allow web-assets to be declared (html fragments and js) that are responsible for rendering a framework's status panel. Spark, Cassandra and other frameworks would all then be able to render their state in a domain specific manner. I would recommend constraining the styling such that css should not be required and the rendering can then conform to the style that is already present.