Remove config override for vector dimensions

Author: saratpoluri

controller/src/controller/uuid_manager.py

@@ -37,13 +37,13 @@ def __init__(self, database=DEFAULT_DATABASE, reid_config_data=None):
     self.features_for_database_timestamps = {}  # Track when features were added
     self.quality_features = {}
     self.unique_id_count = 0
-    # vector_dimensions is optional; if absent, inferred from first observed embedding
+    # ReID embedding dimensions are inferred from the first observed embedding.
+    # Do not accept dimension overrides from reid_config_data.
     if reid_config_data is None:
       reid_config_data = {}
-    vector_dimensions = reid_config_data.get('vector_dimensions', None)
-    self._inferred_dimensions = vector_dimensions
+    self._inferred_dimensions = None
     self._dimensions_lock = threading.Lock()
-    self.reid_database = available_databases[database](dimensions=vector_dimensions)
+    self.reid_database = available_databases[database](dimensions=None)
     self.pool = concurrent.futures.ThreadPoolExecutor()
     self.similarity_query_times = collections.deque(
       maxlen=DEFAULT_MAX_SIMILARITY_QUERIES_TRACKED)
@@ -122,7 +122,7 @@ def _ensureReIDDimensions(self, embedding):
         log.info(f"Inferred ReID embedding dimensions from first observed vector: {dim}")
         try:
           self.reid_database.ensureSchema(dim)
-        except ValueError as err:
+        except (ValueError, RuntimeError) as err:
           log.error(f"ReID schema initialization failed: {err}")
           return False
         self._inferred_dimensions = dim

docs/user-guide/microservices/controller/Extended-ReID.md

@@ -158,21 +158,19 @@ controller/config/reid-config.json
 
 ### Configuration Parameters
 
-| Parameter                           | Type  | Default | Description                                                                                                                                                                                                                                                                                                                                                                                                                  |
-| ----------------------------------- | ----- | ------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `stale_feature_timeout_secs`        | float | 5.0     | How long (seconds) to accumulate features in memory before flushing to VDMS. Features older than this threshold are persisted to the database for long-term storage.                                                                                                                                                                                                                                                         |
-| `stale_feature_check_interval_secs` | float | 1.0     | How frequently (seconds) the background timer checks for stale features and flushes them to VDMS. More frequent checks ensure timely database updates.                                                                                                                                                                                                                                                                       |
-| `feature_accumulation_threshold`    | int   | 12      | Minimum number of quality features required before initiating a similarity query against the database. More features = higher statistical confidence in matching.                                                                                                                                                                                                                                                            |
-| `feature_slice_size`                | int   | 10      | When persisting features to VDMS, sample every Nth feature vector from the accumulated set to reduce database bloat. Example: slice_size=10 stores every 10th vector.                                                                                                                                                                                                                                                        |
-| `similarity_threshold`              | int   | 60      | Minimum similarity score (0-100) for a match to be considered valid. Higher values = stricter matching.                                                                                                                                                                                                                                                                                                                      |
-| `vector_dimensions`                 | int   | auto    | **Optional.** Override the embedding vector length used for the VDMS descriptor set schema. When absent (recommended), the controller infers the dimension automatically from the first received embedding. Set this only to validate that your producer always emits a specific length; a mismatch between the configured value and the received embedding causes the controller to reject all embeddings and log an error. |
+| Parameter                           | Type  | Default | Description                                                                                                                                                           |
+| ----------------------------------- | ----- | ------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `stale_feature_timeout_secs`        | float | 5.0     | How long (seconds) to accumulate features in memory before flushing to VDMS. Features older than this threshold are persisted to the database for long-term storage.  |
+| `stale_feature_check_interval_secs` | float | 1.0     | How frequently (seconds) the background timer checks for stale features and flushes them to VDMS. More frequent checks ensure timely database updates.                |
+| `feature_accumulation_threshold`    | int   | 12      | Minimum number of quality features required before initiating a similarity query against the database. More features = higher statistical confidence in matching.     |
+| `feature_slice_size`                | int   | 10      | When persisting features to VDMS, sample every Nth feature vector from the accumulated set to reduce database bloat. Example: slice_size=10 stores every 10th vector. |
+| `similarity_threshold`              | int   | 60      | Minimum similarity score (0-100) for a match to be considered valid. Higher values = stricter matching.                                                               |
 
 ### Embedding Dimension Inference
 
-The controller automatically infers the ReID embedding dimension from the first vector it receives at runtime, removing the need to keep `vector_dimensions` in configuration synchronized with the model:
+The controller automatically infers the ReID embedding dimension from the first vector it receives at runtime:
 
-- **Auto-inference (default)**: On the first decoded embedding the controller reads the vector length from the payload, creates the VDMS descriptor set schema with that dimension, and locks that dimension for the process lifetime. All subsequent embeddings are validated against that inferred length; mismatches are discarded with a warning.
-- **Optional override** (`vector_dimensions` in config): Set this when you want the controller to reject embeddings whose length does not match a specific value. Useful in deployments where the embedding model is fixed and any deviation should be treated as a misconfiguration. A mismatch triggers an error and no vectors are stored for that run.
+- **Runtime inference only**: On the first decoded embedding the controller reads the vector length from the payload, creates the VDMS descriptor set schema with that dimension, and locks that dimension for the process lifetime. All subsequent embeddings are validated against that inferred length; mismatches are discarded with a warning.
 - **Switching ReID models**: Because the dimension is locked after the first embedding, switching to a model with a different output length requires restarting the controller. The VDMS descriptor set must also be recreated if the stored dimension differs (VDMS does not support in-place schema migration).
 - **Base64 compatibility**: The controller decodes base64 embeddings using the payload byte length by default. Producers can also include an optional `embedding_dimensions` field alongside `embedding_vector`; if provided, it must match the packed float count.
 
@@ -191,7 +189,7 @@ python scene_controller.py \
 **Current Implementation Note**:
 
 - `stale_feature_timeout_secs`, `stale_feature_check_interval_secs`, `feature_accumulation_threshold`, `feature_slice_size`, and `similarity_threshold` are fully implemented
-- `vector_dimensions` is optional; omit it to let the controller infer dimensions automatically from the first received embedding
+- ReID embedding dimensions are inferred at runtime from the first received embedding; there is no configuration override for dimension
 - All semantic metadata attributes are currently used for TIER 1 filtering. Selective metadata filtering is planned for Phase 2.
 
 ### Tuning Recommendations

tests/sscape_tests/uuid_manager/test_uuid_manager.py

@@ -18,36 +18,6 @@
 class TestUUIDManagerInitialization:
   """Test UUIDManager initialization and basic setup."""
 
-  def test_initialization_applies_vector_dimensions_from_reid_config(self):
-    """Verify explicit vector_dimensions is forwarded to the selected DB constructor."""
-    mock_db_instance = MagicMock()
-    captured_kwargs = {}
-
-    def fake_db_constructor(**kwargs):
-      captured_kwargs.update(kwargs)
-      return mock_db_instance
-
-    with patch.dict(UUIDManager.__init__.__globals__['available_databases'], {'VDMS': fake_db_constructor}):
-      manager = UUIDManager(database="VDMS", reid_config_data={"vector_dimensions": 512})
-
-    assert captured_kwargs.get('dimensions') == 512
-    assert manager.reid_database is mock_db_instance
-
-  def test_initialization_without_vector_dimensions_passes_none_to_database(self):
-    """Verify that omitting vector_dimensions passes None to the DB constructor (auto-infer)."""
-    mock_db_instance = MagicMock()
-    captured_kwargs = {}
-
-    def fake_db_constructor(**kwargs):
-      captured_kwargs.update(kwargs)
-      return mock_db_instance
-
-    with patch.dict(UUIDManager.__init__.__globals__['available_databases'], {'VDMS': fake_db_constructor}):
-      manager = UUIDManager(database="VDMS", reid_config_data={})
-
-    assert captured_kwargs.get('dimensions') is None, "Should pass None when vector_dimensions absent"
-    assert manager._inferred_dimensions is None, "Should start with no inferred dimensions"
-
   @patch('controller.uuid_manager.VDMSDatabase')
   def test_initialization_with_default_database(self, mock_vdms_class):
     """Verify UUIDManager initializes with default VDMS database."""
@@ -386,7 +356,11 @@ def test_assign_id_does_not_increment_counter_when_reid_present(self, mock_vdms_
     mock_vdms_instance = MagicMock()
     mock_vdms_class.return_value = mock_vdms_instance
 
-    manager = UUIDManager()
+    def fake_db_constructor(**kwargs):
+      return mock_vdms_instance
+
+    with patch.dict(UUIDManager.__init__.__globals__['available_databases'], {'VDMS': fake_db_constructor}):
+      manager = UUIDManager()
     initial_count = manager.unique_id_count
 
     obj = MagicMock()
@@ -428,7 +402,11 @@ def test_assign_id_gathers_quality_features_for_new_tracker(self, mock_vdms_clas
     mock_vdms_instance = MagicMock()
     mock_vdms_class.return_value = mock_vdms_instance
 
-    manager = UUIDManager()
+    def fake_db_constructor(**kwargs):
+      return mock_vdms_instance
+
+    with patch.dict(UUIDManager.__init__.__globals__['available_databases'], {'VDMS': fake_db_constructor}):
+      manager = UUIDManager()
 
     obj = MagicMock()
     obj.rv_id = "new_tracker_with_features"
@@ -472,7 +450,11 @@ def test_assign_id_does_not_submit_query_without_sufficient_features(self, mock_
     mock_vdms_instance = MagicMock()
     mock_vdms_class.return_value = mock_vdms_instance
 
-    manager = UUIDManager()
+    def fake_db_constructor(**kwargs):
+      return mock_vdms_instance
+
+    with patch.dict(UUIDManager.__init__.__globals__['available_databases'], {'VDMS': fake_db_constructor}):
+      manager = UUIDManager()
     manager.pool = MagicMock()
 
     obj = MagicMock()
@@ -495,7 +477,11 @@ def test_assign_id_submits_query_with_sufficient_features(self, mock_vdms_class)
     mock_vdms_instance = MagicMock()
     mock_vdms_class.return_value = mock_vdms_instance
 
-    manager = UUIDManager()
+    def fake_db_constructor(**kwargs):
+      return mock_vdms_instance
+
+    with patch.dict(UUIDManager.__init__.__globals__['available_databases'], {'VDMS': fake_db_constructor}):
+      manager = UUIDManager()
     manager.pool = MagicMock()
 
     obj = MagicMock()
@@ -669,21 +655,6 @@ def test_reject_embedding_with_inconsistent_dimension(self):
     assert result is False, "Should reject embedding with different dimension"
     assert manager._inferred_dimensions == 256, "Locked dimension should remain unchanged"
 
-  def test_explicit_config_dimension_used_as_initial_inferred(self):
-    """Verify that a configured vector_dimensions seeds _inferred_dimensions."""
-    manager, _ = self._make_manager_with_mock_db(reid_config_data={"vector_dimensions": 512})
-
-    assert manager._inferred_dimensions == 512, "Configured dimension should seed inferred value"
-
-  def test_reject_embedding_mismatching_configured_dimension(self):
-    """Verify embedding is rejected when its dimension differs from an explicitly configured one."""
-    manager, _ = self._make_manager_with_mock_db(reid_config_data={"vector_dimensions": 512})
-    wrong = np.arange(256, dtype=np.float32)
-
-    result = manager._ensureReIDDimensions(wrong)
-
-    assert result is False, "Should reject embedding whose dimension mismatches configured value"
-
   def test_ensure_schema_error_causes_false_return(self):
     """Verify False is returned and dimension remains unset when ensureSchema raises."""
     mock_db = MagicMock()