We delegate it to the dataset so a pandas dataset will copy locally, a spark or dask dataset will work on that cluster