Out of interest, how are you handling persistence in your stack? Are the cluster machines homogeneous (i.e. all the same)? Can containers be deployed to any machine or must they be deployed to where the data was left?