FIX-#0000: make sure pickling is zero-copy for Ray

Signed-off-by: Anatoly Myachev <[email protected]>

Author: anmyachev

modin/core/dataframe/pandas/partitioning/partition_manager.py

@@ -805,7 +805,9 @@ def split_pandas_df_into_partitions(
         parts = [
             [
                 update_bar(
-                    put_func(col_part.iloc[i : i + row_chunksize]),
+                    put_func(
+                        col_part.iloc[i : i + row_chunksize].copy()
+                    ),  # `copy()` to fix zero-copy pickling
                 )
                 for col_part in col_parts
             ]

modin/test/storage_formats/pandas/test_internals.py

@@ -1511,3 +1511,24 @@ def assert_materialized(obj):
 
     assert call_queue == reconstructed_queue
     assert_everything_materialized(reconstructed_queue)
+
+
+@pytest.mark.skipif(Engine.get() != "Ray", reason="Ray specific")
+def test_zero_copy_pickling():
+    import ray
+
+    df = pd.DataFrame(np.zeros((100, 100)))
+    part = ray.get(df._query_compiler._modin_frame._partitions[0][0]._data)
+
+    try:
+        part.values[0, 0] = 10
+    except ValueError as err:
+        breakpoint()
+        if "assignment destination is read-only" in str(err):
+            pass
+        else:
+            # Unexpected exception
+            raise err
+    else:
+        # The exception must be thrown
+        raise RuntimeError("not zero copy pickling")