In #7580 we removed the ability for get_worker to return non-task worker functions to return the worker. This is a confusing statement -- essentially for the following is no longer supported

from dask.distributed import Client, get_worker

def main():
    c = Client()
    def foo():
        worker = get_worker()
        print(worker)

    c.run(foo)

if __name__ == "__main__":
    main()

client.run runs functions on each worker but there is no task so get_worker returns None. This is functionality RAPIDS folks have been using like the following:

https://github.com/rapidsai/raft/blob/05d899b36b76545d2439dbe47e4659d644ced227/python/raft-dask/raft_dask/common/comms.py#L410-L414

We are fixing this by relying on the dask_worker arg for client.run functions:

If your function takes an input argument named dask_worker then that variable will be populated with the worker itself.

We have a fix for this in Dask-CUDA and working on one for RAFT. I think we should include a statement like the following to warn legacy users of the change get_worker:

If you need to get the worker in a non-task function like client.run please use dask_worker in the function argument (xref: client.run)