cugraph.dask.common.input_utils.get_distributed_data Drops Column Names

[alexbarghi-nv]

While investigating a bug involving concatenation of cudf series objects in cugraph.dask.uniform_neighbor_sample, another bug was found where after calling get_distributed_data, the column names of the returned dask future were removed. For some reason, calling merge instead of concat before get_distributed_data did not produce this bug.

s1 = ...
s2 = ...
df = dask_cudf.concat([s1, s2], axis=1)

ddf = get_distributed_data(ddf)
wait(ddf)
ddf = ddf.worker_to_parts

print(ddf[0][0].result().columns)

Expected:

["s1", "s2"]

Actual:

Int64Index([0, 1])

[rlratzel]

This should be combined with another story for removing get_distributed_data.

[rlratzel]

@alexbarghi-nv mentioned we're removing cugraph.dask.common.input_utils.get_distributed_data, which should resolve this issue.

[alexbarghi-nv]

Issue no longer relevant