fix

Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>
franciscojavierarceo · Nov 26, 2025 · Nov 27, 2025 · Nov 28, 2025 · Nov 28, 2025 · Dec 4, 2025
commit d4adcd5ba0fea352762d8b0b3bc53663c0e9b12f
@@ -1387,11 +1387,13 @@ def get_historical_features(
         # EXCEPT for remote providers where the server needs to handle OnDemandFeatureViews
         provider_feature_refs = []
         transformation_view_names = [fv.name for fv, _ in unified_transformation_views]
-        odfv_names = [odfv.name for odfv in on_demand_feature_views]
 
         # Check if using remote offline store
         from feast.infra.offline_stores.remote import RemoteOfflineStoreConfig
-        is_remote_provider = isinstance(self.config.offline_store, RemoteOfflineStoreConfig)
+
+        is_remote_provider = isinstance(
+            self.config.offline_store, RemoteOfflineStoreConfig
+        )
 
         # For remote providers, send ALL feature references to the server
         # The server has access to the full registry and can handle OnDemandFeatureViews
@@ -2278,11 +2280,17 @@ def write_to_online_store(
 
         # If transform_on_write is not explicitly set, use auto-detection
         if transform_on_write is None:
-            feature_view = self._registry.get_any_feature_view(
+            fv_for_detection = self._registry.get_any_feature_view(
                 feature_view_name, self.project, allow_cache=allow_registry_cache
             )
             input_data = df if df is not None else inputs
-            transform_on_write = should_apply_transformation(feature_view, input_data)
+            # Only apply transformation detection for FeatureView or OnDemandFeatureView
+            if isinstance(fv_for_detection, (FeatureView, OnDemandFeatureView)):
+                transform_on_write = should_apply_transformation(
+                    fv_for_detection, input_data
+                )
+            else:
+                transform_on_write = True  # Default for other types
             if transform_on_write is None:
                 transform_on_write = True  # Default fallback
 
@@ -2305,8 +2313,14 @@ def write_to_online_store(
             if feature_column_names:
                 # For OnDemandFeatureViews with transform_on_write=False, skip feature column validation
                 # since the dataframe contains raw input data, not computed output features
-                missing_columns = [col for col in feature_column_names if col not in df.columns]
-                if missing_columns and isinstance(feature_view, OnDemandFeatureView) and not transform_on_write:
+                missing_columns = [
+                    col for col in feature_column_names if col not in df.columns
+                ]
+                if (
+                    missing_columns
+                    and isinstance(feature_view, OnDemandFeatureView)
+                    and not transform_on_write
+                ):
                     # Raw input data for OnDemandFeatureView - computed features not present yet, skip validation
                     pass
                 elif not missing_columns:

@@ -226,8 +226,10 @@ def __eq__(self, other):
 
         # Only compare bytecode if both UDFs have the same name and string representation
         # This makes serialization/deserialization more robust
-        if (self.udf.__name__ != other.udf.__name__ or
-            self.udf_string != other.udf_string):
+        if (
+            self.udf.__name__ != other.udf.__name__
+            or self.udf_string != other.udf_string
+        ):
             return False
 
         return True

@@ -794,9 +794,7 @@ def _augment_response_with_on_demand_transforms(
                 if initial_response_dict is None:
                     initial_response_dict = initial_response.to_dict()
                 # Always use transform_dict for OnDemandFeatureViews - it handles singleton mode properly
-                transformed_features_dict = odfv.transform_dict(
-                    initial_response_dict
-                )
+                transformed_features_dict = odfv.transform_dict(initial_response_dict)
                 transformed_features = transformed_features_dict
             elif mode in {"pandas", "substrait"}:
                 if initial_response_arrow is None:
@@ -1359,11 +1357,13 @@ def _get_feature_views_to_use(
 
     # Ensure OnDemandFeatureView source dependencies are included
     for odfv in od_fvs_to_use:
-        if hasattr(odfv, 'source_feature_view_projections'):
+        if hasattr(odfv, "source_feature_view_projections"):
             for source_fv_projection in odfv.source_feature_view_projections.values():
                 # Get the actual feature view from registry
                 try:
-                    source_fv = registry.get_any_feature_view(source_fv_projection.name, project, allow_cache)
+                    source_fv = registry.get_any_feature_view(
+                        source_fv_projection.name, project, allow_cache
+                    )
                     if source_fv and source_fv not in fvs_to_use:
                         fvs_to_use.append(source_fv)
                 except Exception:

@@ -60,13 +60,9 @@ def setUp(self):
         start_date = end_date - timedelta(days=15)
 
         driver_entities = [1001, 1002, 1003, 1004, 1005]
-        driver_df = create_driver_hourly_stats_df(
-            driver_entities, start_date, end_date
-        )
+        driver_df = create_driver_hourly_stats_df(driver_entities, start_date, end_date)
         driver_stats_path = os.path.join(data_dir, "driver_stats.parquet")
-        driver_df.to_parquet(
-            path=driver_stats_path, allow_truncated_timestamps=True
-        )
+        driver_df.to_parquet(path=driver_stats_path, allow_truncated_timestamps=True)
 
         driver = Entity(name="driver", join_keys=["driver_id"])
 
@@ -133,7 +129,7 @@ def test_view(inputs: dict[str, Any]) -> dict[str, Any]:
         )
 
     def tearDown(self):
-        if hasattr(self, 'temp_dir'):
+        if hasattr(self, "temp_dir"):
             shutil.rmtree(self.temp_dir, ignore_errors=True)
 
     def test_online_retrieval(self):

@@ -64,13 +64,9 @@ def setUp(self):
         start_date = end_date - timedelta(days=15)
 
         driver_entities = [1001, 1002, 1003, 1004, 1005]
-        driver_df = create_driver_hourly_stats_df(
-            driver_entities, start_date, end_date
-        )
+        driver_df = create_driver_hourly_stats_df(driver_entities, start_date, end_date)
         driver_stats_path = os.path.join(data_dir, "driver_stats.parquet")
-        driver_df.to_parquet(
-            path=driver_stats_path, allow_truncated_timestamps=True
-        )
+        driver_df.to_parquet(path=driver_stats_path, allow_truncated_timestamps=True)
 
         driver = Entity(
             name="driver", join_keys=["driver_id"], value_type=ValueType.INT64
@@ -123,9 +119,7 @@ def setUp(self):
         )
         def pandas_view(inputs: pd.DataFrame) -> pd.DataFrame:
             df = pd.DataFrame()
-            df["conv_rate_plus_acc_pandas"] = (
-                inputs["conv_rate"] + inputs["acc_rate"]
-            )
+            df["conv_rate_plus_acc_pandas"] = inputs["conv_rate"] + inputs["acc_rate"]
             return df
 
         @on_demand_feature_view(
@@ -136,9 +130,7 @@ def pandas_view(inputs: pd.DataFrame) -> pd.DataFrame:
         def python_view(inputs: dict[str, Any]) -> dict[str, Any]:
             output: dict[str, Any] = {
                 "conv_rate_plus_acc_python": conv_rate + acc_rate
-                for conv_rate, acc_rate in zip(
-                    inputs["conv_rate"], inputs["acc_rate"]
-                )
+                for conv_rate, acc_rate in zip(inputs["conv_rate"], inputs["acc_rate"])
             }
             return output
 
@@ -244,8 +236,9 @@ def python_stored_writes_feature_view(
         assert len(self.store.list_stream_feature_views()) == 0
 
     def tearDown(self):
-        if hasattr(self, 'temp_dir'):
+        if hasattr(self, "temp_dir"):
             import shutil
+
             shutil.rmtree(self.temp_dir, ignore_errors=True)
 
     def test_setup(self):
@@ -424,13 +417,9 @@ def setUp(self):
         start_date = end_date - timedelta(days=15)
 
         driver_entities = [1001, 1002, 1003, 1004, 1005]
-        driver_df = create_driver_hourly_stats_df(
-            driver_entities, start_date, end_date
-        )
+        driver_df = create_driver_hourly_stats_df(driver_entities, start_date, end_date)
         driver_stats_path = os.path.join(data_dir, "driver_stats.parquet")
-        driver_df.to_parquet(
-            path=driver_stats_path, allow_truncated_timestamps=True
-        )
+        driver_df.to_parquet(path=driver_stats_path, allow_truncated_timestamps=True)
 
         driver = Entity(name="driver", join_keys=["driver_id"])
 
@@ -648,9 +637,7 @@ def pandas_view(features_df: pd.DataFrame) -> pd.DataFrame:
             "val_to_add_2",
         ]
         resp_online_missing_entity = self.store.get_online_features(
-            entity_rows=[
-                {"driver_id": 1234567890, "val_to_add": 0, "val_to_add_2": 1}
-            ],
+            entity_rows=[{"driver_id": 1234567890, "val_to_add": 0, "val_to_add_2": 1}],
             features=[
                 "driver_hourly_stats:conv_rate",
                 "driver_hourly_stats:acc_rate",
@@ -693,8 +680,9 @@ def pandas_view(features_df: pd.DataFrame) -> pd.DataFrame:
         assert sorted(resp_online.columns) != sorted(resp_offline.columns)
 
     def tearDown(self):
-        if hasattr(self, 'temp_dir'):
+        if hasattr(self, "temp_dir"):
             import shutil
+
             shutil.rmtree(self.temp_dir, ignore_errors=True)
 
     def test_setup(self):
@@ -1451,7 +1439,7 @@ def python_stored_writes_feature_view_explode_singleton(
 
             assert odfv_applied.entities == [chunk.name, document.name]
 
-                # Note here that after apply() is called, the entity_columns are populated with the join_key
+            # Note here that after apply() is called, the entity_columns are populated with the join_key
             assert odfv_applied.entity_columns[1].name == chunk.join_key
             assert odfv_applied.entity_columns[0].name == document.join_key