Currently to get the name of an index of a dask.dataframe, the dask must be computed. This is a small amount of metadata, which should change consistently. So we should be tracking it. This would touch a lot of dataframe creation code. And maybe give some idea about the difficulty of supporting MultiIndexs.