Can't groupby two or more keys

When attempting to groupby two or more keys, I get an attribute error. This does not occur when grouping by a single key
```python
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
in
----> 1 df.groupby(['path','time']).alt.mean().compute()

~/anaconda3/lib/python3.6/site-packages/dask/base.py in compute(self, **kwargs)
154 dask.base.compute
155 """
--> 156 (result,) = compute(self, traverse=False, **kwargs)
157 return result
158

~/anaconda3/lib/python3.6/site-packages/dask/base.py in compute(*args, **kwargs)
396 keys = [x.__dask_keys__() for x in collections]
397 postcomputes = [x.__dask_postcompute__() for x in collections]
--> 398 results = schedule(dsk, keys, **kwargs)
399 return repack([f(r, *a) for r, (f, a) in zip(results, postcomputes)])
400

~/anaconda3/lib/python3.6/site-packages/dask/threaded.py in get(dsk, result, cache, num_workers, pool, **kwargs)
74 results = get_async(pool.apply_async, len(pool._pool), dsk, result,
75 cache=cache, get_id=_thread_get_id,
---> 76 pack_exception=pack_exception, **kwargs)
77
78 # Cleanup pools associated to dead threads

~/anaconda3/lib/python3.6/site-packages/dask/local.py in get_async(apply_async, num_workers, dsk, result, cache, get_id, rerun_exceptions_locally, pack_exception, raise_exception, callbacks, dumps, loads, **kwargs)
465 finish_task(dsk, key, state, results, keyorder.get)
466 for f in posttask_cbs:
--> 467 f(key, res, dsk, state, worker_id)
468
469 while state['ready'] and len(state['running']) < num_workers:

~/anaconda3/lib/python3.6/site-packages/dask/cache.py in _posttask(self, key, value, dsk, state, id)
59 duration += max(self.durations.get(k, 0) for k in deps)
60 self.durations[key] = duration
---> 61 nb = self._nbytes(value) + overhead + sys.getsizeof(key) * 4
62 self.cache.put(key, value, cost=duration / nb / 1e9, nbytes=nb)
63

~/anaconda3/lib/python3.6/site-packages/cachey/nbytes.py in nbytes(o)
27
28 if name == 'pandas.core.series.Series':
---> 29 return _array(o._data.blocks[0].values) + _array(o.index._data)
30 elif name == 'pandas.core.frame.DataFrame':
31 return _array(o.index) + sum([_array(blk.values)

~/anaconda3/lib/python3.6/site-packages/cachey/nbytes.py in _array(x)
3
4 def _array(x):
----> 5 if x.dtype == 'O':
6 return sys.getsizeof('0'*100) * x.size
7 elif str(x.dtype) == 'category':

AttributeError: 'NoneType' object has no attribute 'dtype'
```
Is this a known problem? Or a bug?

Start with cachey/nbytes.py, especially the nbytes and _array entry points shown in the traceback, and reproduce the failure with df.groupby(['path','time']).alt.mean().compute(). Done means this grouped computation no longer raises the shown NoneType/dtype AttributeError; the payload names no repository test.

Written by the indexing model from the issue text.