In #9406 @rjzamora did some nice work in implementing a shuffle-based groupby/agg operation for dask dataframe. However, it is currently only implemented for groupby(...).agg({...}), and some of the simpler aggregation functions which might benefit from it are un-implemented. This includes groupby(...).min, `groupby(...).sum and similar:

So, the following snippet would use a shuffle-based approach:

import dask.datasets

ddf = dask.datasets.timeseries()
ddf.groupby("id").agg({"x": "mean"}, split_out=2)

While the following semantically identical one would not:

import dask.datasets

ddf = dask.datasets.timeseries()
ddf.groupby("id").x.mean(split_out=2)

It should be relatively straightforward to retrofit the simpler aggregate functions to also take advantage of the new shuffle-base approach where appropriate.