Not sure of a good test here given we already passed kwargs in tests and they were just getting dropped. Suggestions welcome.

While perhaps overkill, some version of my example from #3299 with the UnreliableStore target would be an effective test of store(..., retries=n).

For the student: how many retries with UnreliableStore and N partitions would it take to have a >99.9% chance of passing the test? :)

For the student: how many retries with UnreliableStore and N partitions would it take to have a >99.9% chance of passing the test? :)

For this reason, the UnreliableStore should probably be modified to just fail once and then succeed after, via some internal state memory.

We probably could create a custom get function and then pass some keyword argument(s) in and check for them in the custom get function. There might be something easier than this though.

We probably could create a custom get function and then pass some keyword argument(s) in and check for them in the custom get function. There might be something easier than this though.

That's what I would do (we've done similar things elsewhere). Something like:

def test_forwarded_kwargs()
    called = [False]
    def myget(*args, **kwargs):
        called[0] = True
        assert 'foo' in kwargs
        return dask.get(*args, **kwargs)
    store([source], [target], get=myget, foo='fake keyword')
    assert called[0]

Thanks. Went ahead and did exactly that.

Please let me know if there is anything else.