Currently the Rolling upgrades data retrieval is done using several tasks in parallel, where each task retrieves from the source cluster a subset of the data from one or more segments.

Each task should only retrieve data from segments that are confined to a single source server,  to avoid further remote calls inside the source server.