During the Rolling Update all Pods (new and old) are put in the same cluster. For some use cases (e.g. when an app deployed on WF is not backwards compatible and uses cache) we need to separate those two clusters.