Add a _placeholder_vector field when no vector field exists in the

patelchaitany · patelchaitany · commit 21896cef553b · 2026-06-25T17:54:59.000+05:30
schema, and populate it with a zero vector on writes. This is the recommended workaround per milvus-io/milvus#33853. Signed-off-by: Chaitany Patel <patelchaitany93@gmail.com>
diff --git a/sdk/python/feast/infra/online_stores/milvus_online_store/milvus.py b/sdk/python/feast/infra/online_stores/milvus_online_store/milvus.py
@@ -235,6 +235,18 @@ def _get_or_create_collection(
                                 max_length=field_max_length,
                             )
                         )
+            has_vector_field = any(
+                f.dtype in (DataType.FLOAT_VECTOR, DataType.BINARY_VECTOR)
+                for f in fields
+            )
+            if not has_vector_field:
+                fields.append(
+                    FieldSchema(
+                        name="_placeholder_vector",
+                        dtype=DataType.FLOAT_VECTOR,
+                        dim=1,
+                    )
+                )
             schema = CollectionSchema(
                 fields=fields, description="Feast feature view data"
             )
@@ -345,10 +357,13 @@ def online_write_batch(
                 "created_ts": created_ts_int,
             }
             single_entity_record.update(values_dict)
-            # Ensure all required fields exist, setting missing ones to empty strings
+            # Ensure all required fields exist, setting missing ones to defaults
             for field in required_fields:
                 if field not in single_entity_record:
-                    single_entity_record[field] = ""
+                    if field == "_placeholder_vector":
+                        single_entity_record[field] = [float("nan")]
+                    else:
+                        single_entity_record[field] = ""
             # Store only the latest event timestamp per entity
             if (
                 entity_key_str not in unique_entities
diff --git a/sdk/python/tests/unit/online_store/test_online_retrieval.py b/sdk/python/tests/unit/online_store/test_online_retrieval.py
@@ -1303,7 +1303,6 @@ def test_milvus_stored_writes_with_explode() -> None:
     )
 
     random.seed(42)
-    vector_length = 10
     runner = CliRunner()
     with runner.local_repo(
         example_repo_py=get_example_repo("example_rag_feature_repo.py"),
@@ -1357,10 +1356,10 @@ def milvus_explode_feature_view(inputs: dict[str, Any]):
                     "Document chunking example.",
                 ],
                 "vector": [
-                    [0.1] * vector_length,
-                    [0.2] * vector_length,
-                    [0.3] * vector_length,
-                    [0.4] * vector_length,
+                    [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+                    [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+                    [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+                    [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
                 ],
             }
             return output
@@ -1428,7 +1427,8 @@ def milvus_explode_feature_view(inputs: dict[str, Any]):
         )
 
         # Test vector search using Milvus
-        query_embedding = [0.1] * vector_length
+        # Query vector closest to doc_2/chunk-1 [0,0,1,0,...] then doc_1/chunk-2 [0,1,0,0,...]
+        query_embedding = [0.0, 0.3, 0.9, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]
 
         # First get Milvus client and search directly
         client = store._provider._online_store.client
