So much this. Dask does not remove the need to babysit workers and partitions, and that's the real difference between pandas and spark.