Currently the whole file is read when creating the dataset, this is no good. See also: https://github.com/pydata/xarray/issues/790