Lawrence Mitchell
Lawrence Mitchell
with `export UCX_TLS=^cuda_ipc`: ``` $ dask-scheduler --protocol ucx --scheduler-file foo.json & $ for i in $(seq 0 7); do CUDA_VISIBLE_DEVICES=$i dask-cuda-worker --scheduler-file foo.json --protocol ucx &; done $ python local_cudf_merge.py...
> What we would do in that case is to call `numba.cuda.current_context(devnum=N)`, where `N` is the index for the device the worker should use. However, it may not be that...
I made some progress (with the goal of using cuda-python), unfortunately that exposed some bugs in cuda-python which are fixed (but not yet in a version that we can use...
/ok to test
I would suggest that we introduce this change in behaviour by adding an [appropriate deprecation](https://docs.rapids.ai/api/cudf/stable/developer_guide/contributing_guide/#deprecating-and-removing-code), we can then fix up any internal usages and remove in a subsequent release.
> @wence- > > > I would suggest that we introduce this change in behaviour by adding an [appropriate deprecation](https://docs.rapids.ai/api/cudf/stable/developer_guide/contributing_guide/#deprecating-and-removing-code) > > this isn't a change in behavior. this is...
> I got a similar issue with `python=3.12`. Python downgrade helped. @VitaliyaIoffe, I think you were affected by https://github.com/rapidsai/cudf/issues/15989, we will hopefully have a fix for that soon.
/ok to test
/ok to tset
> /ok to test @aaronmondal you don't have the permissions to do that