The cluster should be able to automatically re-balance itself, instead of being triggered manually.

This could be on a configured schedule (e.g. every 7 days) or perhaps on tolerances, such as a node is at 90% max-memory usage.