Merge pull request #214 from roboflow/fix/improve_keyless_models_hand…

…ling

Fix problem with keyless access and Active Learning

inference/__init__.py

@@ -1,3 +1,3 @@
-from inference.core.interfaces.stream.stream import Stream
+from inference.core.interfaces.stream.stream import Stream  # isort:skip
 from inference.core.interfaces.stream.inference_pipeline import InferencePipeline
 from inference.models.utils import get_roboflow_model

inference/core/interfaces/stream/inference_pipeline.py

@@ -127,7 +127,9 @@ def init(
             active_learning_enabled (Optional[bool]): Flag to enable / disable Active Learning middleware (setting it
                 true does not guarantee any data to be collected, as data collection is controlled by Roboflow backend -
                 it just enables middleware intercepting predictions). If not given, env variable
-                `ACTIVE_LEARNING_ENABLED` will be used.
+                `ACTIVE_LEARNING_ENABLED` will be used. Please point out that Active Learning will be forcefully
+                disabled in a scenario when Roboflow API key is not given, as Roboflow account is required
+                for this feature to be operational.
 
         Other ENV variables involved in low-level configuration:
         * INFERENCE_PIPELINE_PREDICTIONS_QUEUE_SIZE - size of buffer for predictions that are ready for dispatching
@@ -170,6 +172,11 @@ def init(
                 f"with value: {ACTIVE_LEARNING_ENABLED}"
             )
             active_learning_enabled = ACTIVE_LEARNING_ENABLED
+        if api_key is None:
+            logger.info(
+                f"Roboflow API key not given - Active Learning is forced to be disabled."
+            )
+            active_learning_enabled = False
         if active_learning_enabled is True:
             active_learning_middleware = ThreadingActiveLearningMiddleware.init(
                 api_key=api_key,

inference/core/managers/active_learning.py

@@ -31,10 +31,10 @@ async def infer_from_request(
         self, model_id: str, request: InferenceRequest, **kwargs
     ) -> InferenceResponse:
         prediction = await super().infer_from_request(
-            model_id=model_id, request=request
+            model_id=model_id, request=request, **kwargs
         )
         active_learning_eligible = kwargs.get(ACTIVE_LEARNING_ELIGIBLE_PARAM, False)
-        if not active_learning_eligible:
+        if not active_learning_eligible or request.api_key is None:
             return prediction
         self.register(prediction=prediction, model_id=model_id, request=request)
         return prediction
@@ -108,11 +108,12 @@ class BackgroundTaskActiveLearningManager(ActiveLearningManager):
     async def infer_from_request(
         self, model_id: str, request: InferenceRequest, **kwargs
     ) -> InferenceResponse:
+        active_learning_eligible = kwargs.get(ACTIVE_LEARNING_ELIGIBLE_PARAM, False)
+        kwargs[ACTIVE_LEARNING_ELIGIBLE_PARAM] = False  # disabling AL in super-classes
         prediction = await super().infer_from_request(
-            model_id=model_id, request=request
+            model_id=model_id, request=request, **kwargs
         )
-        active_learning_eligible = kwargs.get(ACTIVE_LEARNING_ELIGIBLE_PARAM, False)
-        if not active_learning_eligible:
+        if not active_learning_eligible or request.api_key is None:
             return prediction
         if BACKGROUND_TASKS_PARAM not in kwargs:
             logger.warning(

inference/enterprise/stream_management/api/entities.py

@@ -51,7 +51,7 @@ class PipelineInitialisationRequest(BaseModel):
     sink_configuration: UDPSinkConfiguration = Field(
         description="Configuration of the sink."
     )
-    api_key: str = Field(description="Roboflow API key")
+    api_key: Optional[str] = Field(description="Roboflow API key", default=None)
     max_fps: Optional[Union[float, int]] = Field(
         description="Limit of FPS in video processing.", default=None
     )

inference/enterprise/stream_management/manager/inference_pipeline_manager.py

@@ -115,7 +115,7 @@ def _initialise_pipeline(self, request_id: str, payload: dict) -> None:
                 model_id=payload["model_id"],
                 video_reference=payload["video_reference"],
                 on_prediction=sink,
-                api_key=payload["api_key"],
+                api_key=payload.get("api_key"),
                 max_fps=payload.get("max_fps"),
                 watchdog=watchdog,
                 source_buffer_filling_strategy=source_buffer_filling_strategy,

inference_sdk/http/client.py

@@ -478,7 +478,7 @@ def clip_compare(
             )
         payload = self.__initialise_payload()
         payload["subject_type"] = subject_type
-        payload["prompt_type"] = subject_type
+        payload["prompt_type"] = prompt_type
         if subject_type == "image":
             encoded_image = load_static_inference_input(
                 inference_input=subject,

tests/inference/unit_tests/enterprise/stream_management/api/test_app.py

@@ -177,6 +177,44 @@ def test_initialise_pipeline_when_valid_payload_given(
     }, "CommandResponse must be serialised directly to JSON response"
 
 
+@mock.patch.object(app, "STREAM_MANAGER_CLIENT", new_callable=AsyncMock)
+def test_initialise_pipeline_when_valid_payload_given_without_api_key(
+    stream_manager_client: AsyncMock,
+) -> None:
+    # given
+    client = TestClient(app.app)
+    stream_manager_client.initialise_pipeline.return_value = CommandResponse(
+        status="success",
+        context=CommandContext(request_id="my_request", pipeline_id="my_pipeline"),
+    )
+
+    # when
+    response = client.post(
+        "/initialise",
+        json={
+            "model_id": "some/1",
+            "video_reference": "rtsp://some:543",
+            "sink_configuration": {
+                "type": "udp_sink",
+                "host": "127.0.0.1",
+                "port": 9090,
+            },
+            "model_configuration": {"type": "object-detection"},
+            "active_learning_enabled": True,
+        },
+    )
+
+    # then
+    assert response.status_code == 200, "Status code for success must be 200"
+    assert response.json() == {
+        "status": "success",
+        "context": {
+            "request_id": "my_request",
+            "pipeline_id": "my_pipeline",
+        },
+    }, "CommandResponse must be serialised directly to JSON response"
+
+
 @mock.patch.object(app, "STREAM_MANAGER_CLIENT", new_callable=AsyncMock)
 def test_pause_pipeline_when_successful_response_expected(
     stream_manager_client: AsyncMock,

tests/inference/unit_tests/enterprise/stream_management/api/test_stream_manager_client.py

@@ -506,6 +506,52 @@ async def test_stream_manager_client_can_successfully_initialise_pipeline(
     )
 
 
+@pytest.mark.asyncio
+@mock.patch.object(stream_manager_client, "establish_socket_connection")
+async def test_stream_manager_client_can_successfully_initialise_pipeline_without_api_key(
+    establish_socket_connection_mock: AsyncMock,
+) -> None:
+    # given
+    reader = assembly_socket_reader(
+        message={
+            "request_id": "my_request",
+            "pipeline_id": "new_pipeline",
+            "response": {"status": "success"},
+        },
+        header_size=4,
+    )
+    writer = DummyStreamWriter()
+    establish_socket_connection_mock.return_value = (reader, writer)
+    initialisation_request = PipelineInitialisationRequest(
+        model_id="some/1",
+        video_reference="rtsp://some:543",
+        sink_configuration=UDPSinkConfiguration(
+            type="udp_sink",
+            host="127.0.0.1",
+            port=9090,
+        ),
+        model_configuration=ObjectDetectionModelConfiguration(type="object_detection"),
+    )
+    client = StreamManagerClient.init(
+        host="127.0.0.1",
+        port=7070,
+        operations_timeout=1.0,
+        header_size=4,
+        buffer_size=16438,
+    )
+
+    # when
+    result = await client.initialise_pipeline(
+        initialisation_request=initialisation_request
+    )
+
+    # then
+    assert result == CommandResponse(
+        status="success",
+        context=CommandContext(request_id="my_request", pipeline_id="new_pipeline"),
+    )
+
+
 @pytest.mark.asyncio
 @mock.patch.object(stream_manager_client, "establish_socket_connection")
 async def test_stream_manager_client_can_successfully_terminate_pipeline(

tests/inference/unit_tests/enterprise/stream_management/manager/test_inference_pipeline_manager.py

@@ -60,7 +60,7 @@ def test_inference_pipeline_manager_when_init_pipeline_operation_is_requested(
 
 @pytest.mark.timeout(30)
 @mock.patch.object(inference_pipeline_manager.InferencePipeline, "init")
-def test_inference_pipeline_manager_when_init_pipeline_operation_is_requested_but_invalid_payload_sent(
+def test_inference_pipeline_manager_when_init_pipeline_operation_is_requested_without_api_key(
     pipeline_init_mock: MagicMock,
 ) -> None:
     # given
@@ -70,7 +70,7 @@ def test_inference_pipeline_manager_when_init_pipeline_operation_is_requested_bu
         command_queue=command_queue, responses_queue=responses_queue
     )
     init_payload = assembly_valid_init_payload()
-    del init_payload["model_configuration"]
+    del init_payload["api_key"]
 
     # when
     command_queue.put(("1", init_payload))
@@ -82,24 +82,18 @@ def test_inference_pipeline_manager_when_init_pipeline_operation_is_requested_bu
     status_2 = responses_queue.get()
 
     # then
-    assert (
-        status_1[0] == "1"
-    ), "First request should be reported in responses_queue at first"
-    assert (
-        status_1[1]["status"] == OperationStatus.FAILURE
-    ), "Init operation should fail"
-    assert (
-        status_1[1]["error_type"] == ErrorType.INVALID_PAYLOAD
-    ), "Invalid Payload error is expected"
+    assert status_1 == (
+        "1",
+        {"status": OperationStatus.SUCCESS},
+    ), "Initialisation operation must succeed"
     assert status_2 == (
         "2",
         {"status": OperationStatus.SUCCESS},
-    ), "Termination of pipeline must happen"
-
+    ), "Termination operation must succeed"
 
 @pytest.mark.timeout(30)
 @mock.patch.object(inference_pipeline_manager.InferencePipeline, "init")
-def test_inference_pipeline_manager_when_init_pipeline_operation_is_requested_but_api_key_not_given(
+def test_inference_pipeline_manager_when_init_pipeline_operation_is_requested_but_invalid_payload_sent(
     pipeline_init_mock: MagicMock,
 ) -> None:
     # given
@@ -109,7 +103,7 @@ def test_inference_pipeline_manager_when_init_pipeline_operation_is_requested_bu
         command_queue=command_queue, responses_queue=responses_queue
     )
     init_payload = assembly_valid_init_payload()
-    del init_payload["api_key"]
+    del init_payload["model_configuration"]
 
     # when
     command_queue.put(("1", init_payload))

tests/inference_sdk/unit_tests/http/test_client.py

@@ -1561,6 +1561,47 @@ def test_clip_compare_when_both_prompt_and_subject_are_texts(
     }, "Request must contain API key, subject and prompt types as text, exact values of subject and list of prompt values"
 
 
+@mock.patch.object(client, "load_static_inference_input")
+def test_clip_compare_when_mixed_input_is_given(
+    load_static_inference_input_mock: MagicMock,
+    requests_mock: Mocker,
+) -> None:
+    # given
+    api_url = "http://some.com"
+    http_client = InferenceHTTPClient(api_key="my-api-key", api_url=api_url)
+    load_static_inference_input_mock.side_effect = [
+        [("base64_image_1", 0.5)]
+    ]
+    requests_mock.post(
+        f"{api_url}/clip/compare",
+        json={
+            "frame_id": None,
+            "time": 0.1435863340011565,
+            "similarity": [0.8963012099266052, 0.8830886483192444],
+        },
+    )
+
+    # when
+    result = http_client.clip_compare(
+        subject="/some/image.jpg",
+        prompt=["dog", "house"],
+    )
+
+    # then
+    assert result == {
+        "frame_id": None,
+        "time": 0.1435863340011565,
+        "similarity": [0.8963012099266052, 0.8830886483192444],
+    }, "Result must match the value returned by HTTP endpoint"
+    assert requests_mock.request_history[0].json() == {
+        "api_key": "my-api-key",
+        "subject": {"type": "base64", "value": "base64_image_1"},
+        "prompt": ["dog", "house"],
+        "prompt_type": "text",
+        "subject_type": "image",
+    }, "Request must contain API key, subject and prompt types as text, exact values of subject and list of prompt values"
+
+
 @mock.patch.object(client, "load_static_inference_input")
 def test_clip_compare_when_both_prompt_and_subject_are_images(
     load_static_inference_input_mock: MagicMock,