keep workloads to their own zone depending on where they are. Hadoop achieve exactly that, I wonder if their solution could be used as an inspiration for solving this in kubernetes