feast-dev · Vperiodt · Mar 14, 2026 · Mar 15, 2026 · Mar 15, 2026 · Mar 16, 2026
diff --git a/.secrets.baseline b/.secrets.baseline
@@ -1485,14 +1485,14 @@
         "filename": "sdk/python/tests/universal/feature_repos/repo_configuration.py",
         "hashed_secret": "d90e76ef629fb00c95f4e84fec29fbda111e2392",
         "is_verified": false,
-        "line_number": 466
+        "line_number": 475
       },
       {
         "type": "Secret Keyword",
         "filename": "sdk/python/tests/universal/feature_repos/repo_configuration.py",
         "hashed_secret": "5baa61e4c9b93f3f0682250b6cf8331b7ee68fd8",
         "is_verified": false,
-        "line_number": 468
+        "line_number": 477
       }
     ],
     "sdk/python/tests/universal/feature_repos/universal/data_sources/file.py": [
@@ -1564,5 +1564,5 @@
       }
     ]
   },
-  "generated_at": "2026-07-30T09:40:48Z"
+  "generated_at": "2026-07-30T13:06:49Z"
 }
@@ -46,6 +46,7 @@
     opt_float,
 )
 from feast.repo_config import FeastConfigBaseModel, RepoConfig
+from feast.utils import compute_non_entity_date_range
 
 
 def _read_data_source(data_source: DataSource, repo_path: str) -> Table:
@@ -495,6 +496,59 @@ class DuckDBOfflineStoreConfig(FeastConfigBaseModel):
     staging_location_endpoint_override: Optional[str] = None
 
 
+def _build_entity_df_from_sources(
+    config: RepoConfig,
+    feature_views: List[FeatureView],
+    start_date: datetime,
+    end_date: datetime,
+    data_source_reader: Callable[[DataSource, str], Table],
+) -> pd.DataFrame:
+    """Build a synthetic entity DataFrame for non-entity mode.
+
+    Scans each feature view's source within the date range to discover all unique
+    entity key combinations, then creates a single row per entity with
+    event_timestamp = end_date so the PIT join returns the latest feature values.
+    """
+    entity_dfs: List[pd.DataFrame] = []
+    repo_path = str(config.repo_path) if config.repo_path else ""
+
+    for fv in feature_views:
+        source = fv.batch_source
+        assert source is not None, f"Feature view '{fv.name}' has no batch_source"
+        table = data_source_reader(source, repo_path)
+        ts_col = source.timestamp_field
+
+        if fv.projection.join_key_map:
+            join_keys = list(fv.projection.join_key_map.values())
+        elif fv.entity_columns:
+            join_keys = [e.name for e in fv.entity_columns]
+        else:
+            join_keys = list(fv.entities) if fv.entities else []
+
+        filtered = table.filter(
+            (table[ts_col] >= ibis.literal(start_date))
+            & (table[ts_col] <= ibis.literal(end_date))
+        )
+
+        if join_keys:
+            sub_df = filtered.select(join_keys).execute().drop_duplicates()
+        else:
+            sub_df = pd.DataFrame(index=[0])
+
+        entity_dfs.append(sub_df)
+
+    if not entity_dfs:
+        return pd.DataFrame(columns=["event_timestamp"])
+
+    combined = pd.concat(entity_dfs, ignore_index=True).drop_duplicates()
+    combined["event_timestamp"] = (
+        pd.Timestamp(end_date).tz_convert("UTC")
+        if pd.Timestamp(end_date).tzinfo
+        else pd.Timestamp(end_date, tz="UTC")
+    )
+    return combined
+
+
 class DuckDBOfflineStore(OfflineStore):
     @staticmethod
     def pull_latest_from_table_or_query(
@@ -527,11 +581,30 @@ def get_historical_features(
         config: RepoConfig,
         feature_views: List[FeatureView],
         feature_refs: List[str],
-        entity_df: Union[pd.DataFrame, str],
+        entity_df: Optional[Union[pd.DataFrame, str]],
         registry: BaseRegistry,
         project: str,
         full_feature_names: bool = False,
+        start_date: Optional[datetime] = None,
+        end_date: Optional[datetime] = None,
     ) -> RetrievalJob:
+        non_entity_mode = entity_df is None
+
+        if non_entity_mode:
+            start_date, end_date = compute_non_entity_date_range(
+                feature_views,
+                start_date=start_date,
+                end_date=end_date,
+            )
+
+            entity_df = _build_entity_df_from_sources(
+                config=config,
+                feature_views=feature_views,
+                start_date=start_date,
+                end_date=end_date,
+                data_source_reader=_read_data_source,
+            )
+
         return get_historical_features_ibis(
             config=config,
             feature_views=feature_views,

@@ -11,8 +11,7 @@
 
 
 @pytest.mark.integration
-@pytest.mark.universal_offline_stores
-@pytest.mark.ray_offline_stores_only
+@pytest.mark.universal_offline_stores(only=["file", "duckdb"])
 def test_non_entity_mode_basic(environment, universal_data_sources):
     """Test historical features retrieval without entity_df (non-entity mode).
 

@@ -731,7 +731,7 @@ def test_historical_features_field_mapping(
 
 
 @pytest.mark.integration
-@pytest.mark.universal_offline_stores(only=["file"])
+@pytest.mark.universal_offline_stores(only=["file", "duckdb"])
 def test_historical_features_non_entity_retrieval(environment):
     """Test get_historical_features with entity_df=None using start_date/end_date.