feast-dev · franciscojavierarceo · May 24, 2025 · May 24, 2025 · May 24, 2025 · May 24, 2025
@@ -56,6 +56,7 @@ The transformation workflow typically involves:
 3. **Chunking**: Split documents into smaller, semantically meaningful chunks
 4. **Embedding Generation**: Convert text chunks into vector embeddings
 5. **Storage**: Store embeddings and metadata in Feast's feature store
+
 ### Feature Transformation for LLMs
 
 Feast supports transformations that can be used to:

@@ -656,7 +656,7 @@ def _make_inferences(
     def _get_feature_views_to_materialize(
         self,
         feature_views: Optional[List[str]],
-    ) -> List[FeatureView]:
+    ) -> List[Union[FeatureView, OnDemandFeatureView]]:
         """
         Returns the list of feature views that should be materialized.
 
@@ -669,34 +669,53 @@ def _get_feature_views_to_materialize(
             FeatureViewNotFoundException: One of the specified feature views could not be found.
             ValueError: One of the specified feature views is not configured for materialization.
         """
-        feature_views_to_materialize: List[FeatureView] = []
+        feature_views_to_materialize: List[Union[FeatureView, OnDemandFeatureView]] = []
 
         if feature_views is None:
-            feature_views_to_materialize = utils._list_feature_views(
+            regular_feature_views = utils._list_feature_views(
                 self._registry, self.project, hide_dummy_entity=False
             )
-            feature_views_to_materialize = [
-                fv for fv in feature_views_to_materialize if fv.online
-            ]
+            feature_views_to_materialize.extend(
+                [fv for fv in regular_feature_views if fv.online]
+            )
             stream_feature_views_to_materialize = self._list_stream_feature_views(
                 hide_dummy_entity=False
             )
-            feature_views_to_materialize += [
-                sfv for sfv in stream_feature_views_to_materialize if sfv.online
-            ]
+            feature_views_to_materialize.extend(
+                [sfv for sfv in stream_feature_views_to_materialize if sfv.online]
+            )
+            on_demand_feature_views_to_materialize = self.list_on_demand_feature_views()
+            feature_views_to_materialize.extend(
+                [
+                    odfv
+                    for odfv in on_demand_feature_views_to_materialize
+                    if odfv.write_to_online_store
+                ]
+            )
         else:
             for name in feature_views:
+                feature_view: Union[FeatureView, OnDemandFeatureView]
                 try:
                     feature_view = self._get_feature_view(name, hide_dummy_entity=False)
                 except FeatureViewNotFoundException:
-                    feature_view = self._get_stream_feature_view(
-                        name, hide_dummy_entity=False
-                    )
+                    try:
+                        feature_view = self._get_stream_feature_view(
+                            name, hide_dummy_entity=False
+                        )
+                    except FeatureViewNotFoundException:
+                        feature_view = self.get_on_demand_feature_view(name)
 
-                if not feature_view.online:
+                if hasattr(feature_view, "online") and not feature_view.online:
                     raise ValueError(
                         f"FeatureView {feature_view.name} is not configured to be served online."
                     )
+                elif (
+                    hasattr(feature_view, "write_to_online_store")
+                    and not feature_view.write_to_online_store
+                ):
+                    raise ValueError(
+                        f"OnDemandFeatureView {feature_view.name} is not configured for write_to_online_store."
+                    )
                 feature_views_to_materialize.append(feature_view)
 
         return feature_views_to_materialize
@@ -866,7 +885,8 @@ def apply(
         views_to_update = [
             ob
             for ob in objects
-            if (
+            if
+            (
                 # BFVs are not handled separately from FVs right now.
                 (isinstance(ob, FeatureView) or isinstance(ob, BatchFeatureView))
                 and not isinstance(ob, StreamFeatureView)
@@ -1312,6 +1332,11 @@ def materialize_incremental(
         )
         # TODO paging large loads
         for feature_view in feature_views_to_materialize:
+            from feast.on_demand_feature_view import OnDemandFeatureView
+
+            if isinstance(feature_view, OnDemandFeatureView):
+                continue
+
             start_date = feature_view.most_recent_end_time
             if start_date is None:
                 if feature_view.ttl is None:
@@ -1352,12 +1377,13 @@ def tqdm_builder(length):
                 tqdm_builder=tqdm_builder,
             )
 
-            self._registry.apply_materialization(
-                feature_view,
-                self.project,
-                start_date,
-                end_date,
-            )
+            if not isinstance(feature_view, OnDemandFeatureView):
+                self._registry.apply_materialization(
+                    feature_view,
+                    self.project,
+                    start_date,
+                    end_date,
+                )
 
     def materialize(
         self,
@@ -1407,6 +1433,8 @@ def materialize(
         )
         # TODO paging large loads
         for feature_view in feature_views_to_materialize:
+            from feast.on_demand_feature_view import OnDemandFeatureView
+
             provider = self._get_provider()
             print(f"{Style.BRIGHT + Fore.GREEN}{feature_view.name}{Style.RESET_ALL}:")
 
@@ -1426,12 +1454,13 @@ def tqdm_builder(length):
                 tqdm_builder=tqdm_builder,
             )
 
-            self._registry.apply_materialization(
-                feature_view,
-                self.project,
-                start_date,
-                end_date,
-            )
+            if not isinstance(feature_view, OnDemandFeatureView):
+                self._registry.apply_materialization(
+                    feature_view,
+                    self.project,
+                    start_date,
+                    end_date,
+                )
 
     def _fvs_for_push_source_or_raise(
         self, push_source_name: str, allow_cache: bool
@@ -2033,9 +2062,9 @@ def retrieve_online_documents_v2(
             distance_metric: The distance metric to use for retrieval.
             query_string: The query string to retrieve the closest document features using keyword search (bm25).
         """
-        assert query is not None or query_string is not None, (
-            "Either query or query_string must be provided."
-        )
+        assert (
+            query is not None or query_string is not None
+        ), "Either query or query_string must be provided."
 
         (
             available_feature_views,
@@ -2348,9 +2377,9 @@ def write_logged_features(
         if not isinstance(source, FeatureService):
             raise ValueError("Only feature service is currently supported as a source")
 
-        assert source.logging_config is not None, (
-            "Feature service must be configured with logging config in order to use this functionality"
-        )
+        assert (
+            source.logging_config is not None
+        ), "Feature service must be configured with logging config in order to use this functionality"
 
         assert isinstance(logs, (pa.Table, Path))
 

@@ -196,9 +196,9 @@ def __init__(
             else:
                 features.append(field)
 
-        assert len([f for f in features if f.vector_index]) < 2, (
-            f"Only one vector feature is allowed per feature view. Please update {self.name}."
-        )
+        assert (
+            len([f for f in features if f.vector_index]) < 2
+        ), f"Only one vector feature is allowed per feature view. Please update {self.name}."
 
         # TODO(felixwang9817): Add more robust validation of features.
         cols = [field.name for field in schema]

@@ -187,9 +187,9 @@ def __init__(
         online_store: OnlineStore,
         **kwargs,
     ):
-        assert repo_config.offline_store.type == "snowflake.offline", (
-            "To use Snowflake Compute Engine, you must use Snowflake as an offline store."
-        )
+        assert (
+            repo_config.offline_store.type == "snowflake.offline"
+        ), "To use Snowflake Compute Engine, you must use Snowflake as an offline store."
 
         super().__init__(
             repo_config=repo_config,
@@ -210,11 +210,10 @@ def _materialize_one(
         project = task.project
         tqdm_builder = task.tqdm_builder if task.tqdm_builder else tqdm
 
-        assert isinstance(feature_view, BatchFeatureView) or isinstance(
-            feature_view, FeatureView
-        ), (
-            "Snowflake can only materialize FeatureView & BatchFeatureView feature view types."
-        )
+        assert (
+            isinstance(feature_view, BatchFeatureView)
+            or isinstance(feature_view, FeatureView)
+        ), "Snowflake can only materialize FeatureView & BatchFeatureView feature view types."
 
         entities = []
         for entity_name in feature_view.entities:

@@ -191,7 +191,9 @@ def evaluate_historical_retrieval():
             ):
                 # Make sure all event timestamp fields are tz-aware. We default tz-naive fields to UTC
                 entity_df_with_features[entity_df_event_timestamp_col] = (
-                    entity_df_with_features[entity_df_event_timestamp_col].apply(
+                    entity_df_with_features[
+                        entity_df_event_timestamp_col
+                    ].apply(
                         lambda x: x
                         if x.tzinfo is not None
                         else x.replace(tzinfo=timezone.utc)

@@ -326,9 +326,7 @@ def online_read(
         assert all(
             field in [f["name"] for f in collection["fields"]]
             for field in output_fields
-        ), (
-            f"field(s) [{[field for field in output_fields if field not in [f['name'] for f in collection['fields']]]}] not found in collection schema"
-        )
+        ), f"field(s) [{[field for field in output_fields if field not in [f['name'] for f in collection['fields']]]}] not found in collection schema"
         composite_entities = []
         for entity_key in entity_keys:
             entity_key_str = serialize_entity_key(
@@ -522,9 +520,7 @@ def retrieve_online_documents_v2(
         assert all(
             field in [f["name"] for f in collection["fields"]]
             for field in output_fields
-        ), (
-            f"field(s) [{[field for field in output_fields if field not in [f['name'] for f in collection['fields']]]}] not found in collection schema"
-        )
+        ), f"field(s) [{[field for field in output_fields if field not in [f['name'] for f in collection['fields']]]}] not found in collection schema"
 
         # Find the vector search field if we need it
         ann_search_field = None

@@ -460,9 +460,9 @@ def retrieve_online_documents_v2(
             where the first item is the event timestamp for the row, and the second item is a dict of feature
             name to embeddings.
         """
-        assert embedding is not None or query_string is not None, (
-            "Either embedding or query_string must be specified"
-        )
+        assert (
+            embedding is not None or query_string is not None
+        ), "Either embedding or query_string must be specified"
         raise NotImplementedError(
             f"Online store {self.__class__.__name__} does not support online retrieval"
         )

@@ -73,9 +73,9 @@ def _get_client(self, config: RepoConfig) -> QdrantClient:
         if self._client:
             return self._client
         online_store_config = config.online_store
-        assert isinstance(online_store_config, QdrantOnlineStoreConfig), (
-            "Invalid type for online store config"
-        )
+        assert isinstance(
+            online_store_config, QdrantOnlineStoreConfig
+        ), "Invalid type for online store config"
 
         assert online_store_config.similarity and (
             online_store_config.similarity.lower() in DISTANCE_MAPPING

@@ -790,12 +790,12 @@ def _get_vector_field(table: FeatureView) -> str:
     vector_fields: List[Field] = [
         f for f in table.features if getattr(f, "vector_index", None)
     ]
-    assert len(vector_fields) > 0, (
-        f"No vector field found, please update feature view = {table.name} to declare a vector field"
-    )
-    assert len(vector_fields) < 2, (
-        "Only one vector field is supported, please update feature view = {table.name} to declare one vector field"
-    )
+    assert (
+        len(vector_fields) > 0
+    ), f"No vector field found, please update feature view = {table.name} to declare a vector field"
+    assert (
+        len(vector_fields) < 2
+    ), "Only one vector field is supported, please update feature view = {table.name} to declare one vector field"
     vector_field: str = vector_fields[0].name
     return vector_field
 

@@ -16,7 +16,7 @@
 import pyarrow as pa
 from tqdm import tqdm
 
-from feast import OnDemandFeatureView, importer
+from feast import importer
 from feast.base_feature_view import BaseFeatureView
 from feast.batch_feature_view import BatchFeatureView
 from feast.data_source import DataSource
@@ -38,6 +38,7 @@
 from feast.infra.provider import Provider
 from feast.infra.registry.base_registry import BaseRegistry
 from feast.infra.supported_async_methods import ProviderAsyncMethods
+from feast.on_demand_feature_view import OnDemandFeatureView
 from feast.online_response import OnlineResponse
 from feast.protos.feast.core.Registry_pb2 import Registry as RegistryProto
 from feast.protos.feast.types.EntityKey_pb2 import EntityKey as EntityKeyProto
@@ -420,13 +421,22 @@ def ingest_df_to_offline_store(self, feature_view: FeatureView, table: pa.Table)
     def materialize_single_feature_view(
         self,
         config: RepoConfig,
-        feature_view: FeatureView,
+        feature_view: Union[FeatureView, OnDemandFeatureView],
         start_date: datetime,
         end_date: datetime,
         registry: BaseRegistry,
         project: str,
         tqdm_builder: Callable[[int], tqdm],
     ) -> None:
+        from feast.on_demand_feature_view import OnDemandFeatureView
+
+        if isinstance(feature_view, OnDemandFeatureView):
+            if not feature_view.write_to_online_store:
+                raise ValueError(
+                    f"OnDemandFeatureView {feature_view.name} does not have write_to_online_store enabled"
+                )
+            return
+
         assert (
             isinstance(feature_view, BatchFeatureView)
             or isinstance(feature_view, StreamFeatureView)
@@ -496,9 +506,9 @@ def write_feature_service_logs(
         config: RepoConfig,
         registry: BaseRegistry,
     ):
-        assert feature_service.logging_config is not None, (
-            "Logging should be configured for the feature service before calling this function"
-        )
+        assert (
+            feature_service.logging_config is not None
+        ), "Logging should be configured for the feature service before calling this function"
 
         self.offline_store.write_logged_features(
             config=config,
@@ -516,9 +526,9 @@ def retrieve_feature_service_logs(
         config: RepoConfig,
         registry: BaseRegistry,
     ) -> RetrievalJob:
-        assert feature_service.logging_config is not None, (
-            "Logging should be configured for the feature service before calling this function"
-        )
+        assert (
+            feature_service.logging_config is not None
+        ), "Logging should be configured for the feature service before calling this function"
 
         logging_source = FeatureServiceLoggingSource(feature_service, config.project)
         schema = logging_source.get_schema(registry)

@@ -217,7 +217,7 @@ def ingest_df_to_offline_store(
     def materialize_single_feature_view(
         self,
         config: RepoConfig,
-        feature_view: FeatureView,
+        feature_view: Union[FeatureView, OnDemandFeatureView],
         start_date: datetime,
         end_date: datetime,
         registry: BaseRegistry,