feat: Add feature view versioning support to FAISS online store

When enable_online_feature_view_versioning is enabled, FAISS indices are namespaced by versioned table keys (e.g. project_driver_stats_v2) so multiple feature view versions can coexist in memory. Reuses the shared compute_table_id() from helpers.py for consistency with PostgreSQL and MySQL stores. Signed-off-by: yassinnouh21 <yassinnouh21@gmail.com>
feast-dev · ntkathole · Apr 11, 2026 · Apr 10, 2026 · Apr 10, 2026 · 5123bed4762406e6bdc2ab1810d11fff4c2b6896
commit 5123bed4762406e6bdc2ab1810d11fff4c2b6896
@@ -142,7 +142,7 @@ class VersionedOnlineReadNotSupported(FeastError):
     def __init__(self, store_name: str, version: int):
         super().__init__(
             f"Versioned feature reads (@v{version}) are not yet supported by {store_name}. "
-            f"Currently only SQLite, PostgreSQL, and MySQL support version-qualified feature references. "
+            f"Currently only SQLite, PostgreSQL, MySQL, and FAISS support version-qualified feature references. "
         )
 
 

@@ -8,6 +8,7 @@
 
 from feast import Entity, FeatureView, RepoConfig
 from feast.infra.key_encoding_utils import serialize_entity_key
+from feast.infra.online_stores.helpers import compute_table_id
 from feast.infra.online_stores.online_store import OnlineStore
 from feast.protos.feast.types.EntityKey_pb2 import EntityKey as EntityKeyProto
 from feast.protos.feast.types.Value_pb2 import Value as ValueProto
@@ -43,16 +44,21 @@ def teardown(self):
         self.entity_keys = {}
 
 
+def _table_id(project: str, table: FeatureView, enable_versioning: bool = False) -> str:
+    return compute_table_id(project, table, enable_versioning)
+
+
 class FaissOnlineStore(OnlineStore):
-    _index: Optional[faiss.IndexIVFFlat] = None
-    _in_memory_store: InMemoryStore = InMemoryStore()
-    _config: Optional[FaissOnlineStoreConfig] = None
     _logger: logging.Logger = logging.getLogger(__name__)
 
-    def _get_index(self, config: RepoConfig) -> faiss.IndexIVFFlat:
-        if self._index is None or self._config is None:
-            raise ValueError("Index is not initialized")
-        return self._index
+    def __init__(self):
+        super().__init__()
+        self._indices: Dict[str, faiss.IndexIVFFlat] = {}
+        self._in_memory_stores: Dict[str, InMemoryStore] = {}
+        self._config: Optional[FaissOnlineStoreConfig] = None
+
+    def _get_index(self, table_key: str) -> Optional[faiss.IndexIVFFlat]:
+        return self._indices.get(table_key)
 
     def update(
         self,
@@ -63,32 +69,45 @@ def update(
         entities_to_keep: Sequence[Entity],
         partial: bool,
     ):
-        feature_views = tables_to_keep
-        if not feature_views:
-            return
-
-        feature_names = [f.name for f in feature_views[0].features]
-        dimension = len(feature_names)
-
         self._config = FaissOnlineStoreConfig(**config.online_store.dict())
-        if self._index is None or not partial:
-            quantizer = faiss.IndexFlatL2(dimension)
-            self._index = faiss.IndexIVFFlat(quantizer, dimension, self._config.nlist)
-            self._index.train(
-                np.random.rand(self._config.nlist * 100, dimension).astype(np.float32)
-            )
-            self._in_memory_store = InMemoryStore()
+        versioning = config.registry.enable_online_feature_view_versioning
+
+        for table in tables_to_delete:
+            table_key = _table_id(config.project, table, versioning)
+            self._indices.pop(table_key, None)
+            self._in_memory_stores.pop(table_key, None)
+
+        for table in tables_to_keep:
+            table_key = _table_id(config.project, table, versioning)
+            feature_names = [f.name for f in table.features]
+            dimension = len(feature_names)
+
+            if table_key not in self._indices or not partial:
+                quantizer = faiss.IndexFlatL2(dimension)
+                index = faiss.IndexIVFFlat(quantizer, dimension, self._config.nlist)
+                index.train(
+                    np.random.rand(self._config.nlist * 100, dimension).astype(
+                        np.float32
+                    )
+                )
+                self._indices[table_key] = index
+                self._in_memory_stores[table_key] = InMemoryStore()
 
-        self._in_memory_store.update(feature_names, {})
+            self._in_memory_stores[table_key].update(feature_names, {})
 
     def teardown(
         self,
         config: RepoConfig,
         tables: Sequence[FeatureView],
         entities: Sequence[Entity],
     ):
-        self._index = None
-        self._in_memory_store.teardown()
+        versioning = config.registry.enable_online_feature_view_versioning
+        for table in tables:
+            table_key = _table_id(config.project, table, versioning)
+            self._indices.pop(table_key, None)
+            store = self._in_memory_stores.pop(table_key, None)
+            if store is not None:
+                store.teardown()
 
     def online_read(
         self,
@@ -97,23 +116,28 @@ def online_read(
         entity_keys: List[EntityKeyProto],
         requested_features: Optional[List[str]] = None,
     ) -> List[Tuple[Optional[datetime], Optional[Dict[str, ValueProto]]]]:
-        if self._index is None:
+        versioning = config.registry.enable_online_feature_view_versioning
+        table_key = _table_id(config.project, table, versioning)
+        index = self._get_index(table_key)
+        in_memory_store = self._in_memory_stores.get(table_key)
+
+        if index is None or in_memory_store is None:
             return [(None, None)] * len(entity_keys)
 
         results: List[Tuple[Optional[datetime], Optional[Dict[str, Any]]]] = []
         for entity_key in entity_keys:
             serialized_key = serialize_entity_key(
                 entity_key, config.entity_key_serialization_version
             ).hex()
-            idx = self._in_memory_store.entity_keys.get(serialized_key, -1)
+            idx = in_memory_store.entity_keys.get(serialized_key, -1)
             if idx == -1:
                 results.append((None, None))
             else:
-                feature_vector = self._index.reconstruct(int(idx))
+                feature_vector = index.reconstruct(int(idx))
                 feature_dict = {
                     name: ValueProto(double_val=value)
                     for name, value in zip(
-                        self._in_memory_store.feature_names, feature_vector
+                        in_memory_store.feature_names, feature_vector
                     )
                 }
                 results.append((None, feature_dict))
@@ -128,8 +152,16 @@ def online_write_batch(
         ],
         progress: Optional[Callable[[int], Any]],
     ) -> None:
-        if self._index is None:
-            self._logger.warning("Index is not initialized. Skipping write operation.")
+        versioning = config.registry.enable_online_feature_view_versioning
+        table_key = _table_id(config.project, table, versioning)
+        index = self._get_index(table_key)
+        in_memory_store = self._in_memory_stores.get(table_key)
+
+        if index is None or in_memory_store is None:
+            self._logger.warning(
+                "Index for table '%s' is not initialized. Skipping write operation.",
+                table_key,
+            )
             return
 
         feature_vectors = []
@@ -142,7 +174,7 @@ def online_write_batch(
             feature_vector = np.array(
                 [
                     feature_dict[name].double_val
-                    for name in self._in_memory_store.feature_names
+                    for name in in_memory_store.feature_names
                 ],
                 dtype=np.float32,
             )
@@ -153,21 +185,17 @@ def online_write_batch(
         feature_vectors_array = np.array(feature_vectors)
 
         existing_indices = [
-            self._in_memory_store.entity_keys.get(sk, -1) for sk in serialized_keys
+            in_memory_store.entity_keys.get(sk, -1) for sk in serialized_keys
         ]
         mask = np.array(existing_indices) != -1
         if np.any(mask):
-            self._index.remove_ids(
-                np.array([idx for idx in existing_indices if idx != -1])
-            )
+            index.remove_ids(np.array([idx for idx in existing_indices if idx != -1]))
 
-        new_indices = np.arange(
-            self._index.ntotal, self._index.ntotal + len(feature_vectors_array)
-        )
-        self._index.add(feature_vectors_array)
+        new_indices = np.arange(index.ntotal, index.ntotal + len(feature_vectors_array))
+        index.add(feature_vectors_array)
 
         for sk, idx in zip(serialized_keys, new_indices):
-            self._in_memory_store.entity_keys[sk] = idx
+            in_memory_store.entity_keys[sk] = idx
 
         if progress:
             progress(len(data))
@@ -189,12 +217,16 @@ def retrieve_online_documents(
             Optional[ValueProto],
         ]
     ]:
-        if self._index is None:
+        versioning = config.registry.enable_online_feature_view_versioning
+        table_key = _table_id(config.project, table, versioning)
+        index = self._get_index(table_key)
+
+        if index is None:
             self._logger.warning("Index is not initialized. Returning empty result.")
             return []
 
         query_vector = np.array(embedding, dtype=np.float32).reshape(1, -1)
-        distances, indices = self._index.search(query_vector, top_k)
+        distances, indices = index.search(query_vector, top_k)
 
         results: List[
             Tuple[
@@ -209,7 +241,7 @@ def retrieve_online_documents(
             if idx == -1:
                 continue
 
-            feature_vector = self._index.reconstruct(int(idx))
+            feature_vector = index.reconstruct(int(idx))
 
             timestamp = Timestamp()
             timestamp.GetCurrentTime()
@@ -237,5 +269,4 @@ async def online_read_async(
         entity_keys: List[EntityKeyProto],
         requested_features: Optional[List[str]] = None,
     ) -> List[Tuple[Optional[datetime], Optional[Dict[str, ValueProto]]]]:
-        # Implement async read if needed
         raise NotImplementedError("Async read is not implemented for FaissOnlineStore")
@@ -274,6 +274,12 @@ def _check_versioned_read_support(self, grouped_refs):
             supported_types.append(PostgreSQLOnlineStore)
         except ImportError:
             pass
+        try:
+            from feast.infra.online_stores.faiss_online_store import FaissOnlineStore
+
+            supported_types.append(FaissOnlineStore)
+        except ImportError:
+            pass
 
         if isinstance(self, tuple(supported_types)):
             return