Fix null partitions

distributed/shuffle/_arrow.py

@@ -79,7 +79,9 @@ def list_of_buffers_to_table(data: list[bytes]) -> pa.Table:
     """Convert a list of arrow buffers and a schema to an Arrow Table"""
     import pyarrow as pa
 
-    return pa.concat_tables(deserialize_table(buffer) for buffer in data)
+    return pa.concat_tables(
+        (deserialize_table(buffer) for buffer in data), promote=True
+    )
 
 
 def serialize_table(table: pa.Table) -> bytes:

distributed/shuffle/tests/test_shuffle.py

@@ -18,6 +18,7 @@
 from distributed.shuffle._core import ShuffleId, ShuffleRun, barrier_key
 from distributed.worker import Status
 
+np = pytest.importorskip("numpy")
 pd = pytest.importorskip("pandas")
 dd = pytest.importorskip("dask.dataframe")
 
@@ -2057,6 +2058,28 @@ async def test_handle_null_partitions_p2p_shuffling(c, s, *workers):
     await check_scheduler_cleanup(s)
 
 
+@gen_cluster(client=True)
+async def test_handle_null_partitions_p2p_shuffling_2(c, s, a, b):
+    def make_partition(i):
+        """Return null column for one partition"""
+        if i % 2 == 1:
+            return pd.DataFrame({"a": np.random.random(10), "b": [None] * 10})
+        return pd.DataFrame({"a": np.random.random(10), "b": np.random.random(10)})
+
+    ddf = dd.from_map(make_partition, range(50))
+    out = ddf.shuffle(on="a", shuffle="p2p", ignore_index=True)
+    result, expected = c.compute([ddf, out])
+    del out
+    result = await result
+    expected = await expected
+    dd.assert_eq(result, expected)
+    del result
+
+    await check_worker_cleanup(a)
+    await check_worker_cleanup(b)
+    await check_scheduler_cleanup(s)
+
+
 @gen_cluster(client=True)
 async def test_set_index_p2p(c, s, *workers):
     df = pd.DataFrame({"a": [1, 2, 3, 4, 5, 6, 7, 8], "b": 1})