Register cudf.core.groupby.Grouper objects to dask grouper_dispatch

[brandon-b-miller]

This PR registers uses the (presumably shortly merged) dask Grouper dispatch to register cudf.core.groupby.Grouper objects to cudf.DataFrame objects. This should allow our own Grouper objects to be used in critical places in dask rather than pandas objects.

This solution is favorable IMO rather than changing cuDF to handle pandas grouper objects directly.

Xref dask/dask#9074

[brandon-b-miller]

@galipremsagar do I need to bump the dask version here to account for dask/dask#9074 ?

[galipremsagar]

@galipremsagar do I need to bump the dask version here to account for dask/dask#9074 ?

Ideally yes, but since there is no dask version released yet. We will have to move the dispatch importing and registering to the try/except block just below it. Similar to percentile_lookup dispatch. Like:

try:
    from dask.dataframe.dispatch import grouper_dispatch
    @grouper_dispatch.register((cudf.Series, cudf.DataFrame))
     def get_grouper_cudf(obj):
        return cudf.core.groupby.Grouper
except ImportError:
    pass

[brandon-b-miller]

tysm @galipremsagar !

[brandon-b-miller]

Is it worth mirroring the tests added in dask/dask@e531cd0 ?

[galipremsagar]

Is it worth mirroring the tests added in dask/dask@e531cd0 ?

Would it be easier to add an API call test that will go through the dispatch mechanisam? Instead of dispatch testing in dask/dask@e531cd0, since that is already taken care of in upstream.

[brandon-b-miller]

Is it worth mirroring the tests added in dask/dask@e531cd0 ?

Would it be easier to add an API call test that will go through the dispatch mechanisam? Instead of dispatch testing in dask/dask@e531cd0, since that is already taken care of in upstream.

In that case I think it's covered by #10889 , so maybe it's OK for now.

[galipremsagar]

Is it worth mirroring the tests added in dask/dask@e531cd0 ?

Would it be easier to add an API call test that will go through the dispatch mechanisam? Instead of dispatch testing in dask/dask@e531cd0, since that is already taken care of in upstream.

In that case I think it's covered by #10889 , so maybe it's OK for now.

Cool, then that sounds good to me.

[galipremsagar]

rerun tests

[galipremsagar]

@gpucibot merge

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (branch-22.08@efd2c39). Click here to learn what that means.
The diff coverage is n/a.

@@               Coverage Diff               @@
##             branch-22.08   #10838   +/-   ##
===============================================
  Coverage                ?   86.61%           
===============================================
  Files                   ?      144           
  Lines                   ?    23511           
  Branches                ?        0           
===============================================
  Hits                    ?    20364           
  Misses                  ?     3147           
  Partials                ?        0           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update efd2c39...5718428. Read the comment docs.

[galipremsagar]

@brandon-b-miller could you re-target this to 22.08 since we are pinning dask version to 2022.05.1 which anyway doesn't contain the dispatch?