feat(inspect): Model deletion endpoint (#3128)

* feat(inspect): add endpoint to delete models and their artifacts

Signed-off-by: Dmitry Kalinin <dmitry.kalinin@intel.com>

* Fixed linter

Signed-off-by: Dmitry Kalinin <dmitry.kalinin@intel.com>

* Update application/backend/src/services/model_service.py

Co-authored-by: Max Xiang <maxx.rift@gmail.com>
Signed-off-by: Dmitry Kalinin <dmitry.kalinin@intel.com>

* Fixed comments

Signed-off-by: Dmitry Kalinin <dmitry.kalinin@intel.com>

* Fixed linter

Signed-off-by: Dmitry Kalinin <dmitry.kalinin@intel.com>

* Fixed comments

Signed-off-by: Dmitry Kalinin <dmitry.kalinin@intel.com>

* Fixed tests

Signed-off-by: Dmitry Kalinin <dmitry.kalinin@intel.com>

* Update application/backend/src/services/training_service.py

Co-authored-by: Max Xiang <maxx.rift@gmail.com>
Signed-off-by: Dmitry Kalinin <dmitry.kalinin@intel.com>

* Fixed tests

Signed-off-by: Dmitry Kalinin <dmitry.kalinin@intel.com>

* Fixed tests

Signed-off-by: Dmitry Kalinin <dmitry.kalinin@intel.com>

---------

Signed-off-by: Dmitry Kalinin <dmitry.kalinin@intel.com>
Co-authored-by: Max Xiang <maxx.rift@gmail.com>

Author: ActiveChooN

application/backend/src/api/endpoints/model_endpoints.py

@@ -68,3 +68,21 @@ async def predict(
         return await model_service.predict_image(model, image_bytes, request.app.state.active_models, device=device)
     except DeviceNotFoundError as e:
         raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=str(e))
+
+
+@model_router.delete(
+    "/{model_id}",
+    status_code=status.HTTP_204_NO_CONTENT,
+    responses={
+        status.HTTP_204_NO_CONTENT: {"description": "Model and exported artifacts successfully deleted"},
+        status.HTTP_400_BAD_REQUEST: {"description": "Invalid model ID"},
+        status.HTTP_404_NOT_FOUND: {"description": "Model not found"},
+    },
+)
+async def delete_model(
+    model_service: Annotated[ModelService, Depends(get_model_service)],
+    project_id: Annotated[UUID, Depends(get_project_id)],
+    model_id: Annotated[UUID, Depends(get_model_id)],
+) -> None:
+    """Delete a model and any exported artifacts."""
+    await model_service.delete_model(project_id=project_id, model_id=model_id)

application/backend/src/services/model_service.py

@@ -77,12 +77,22 @@ async def get_model_by_id(project_id: UUID, model_id: UUID) -> Model | None:
             repo = ModelRepository(session, project_id=project_id)
             return await repo.get_by_id(model_id)
 
-    @staticmethod
-    async def delete_model(project_id: UUID, model_id: UUID) -> None:
+    async def delete_model(self, project_id: UUID, model_id: UUID, delete_artifacts: bool = True) -> None:
+        if delete_artifacts:
+            model_binary_repo = ModelBinaryRepository(project_id=project_id, model_id=model_id)
+            try:
+                await model_binary_repo.delete_model_folder()
+            except FileNotFoundError:
+                logger.warning(
+                    "Model artifacts already absent on disk for model %s in project %s", model_id, project_id
+                )
+
         async with get_async_db_session_ctx() as session:
             repo = ModelRepository(session, project_id=project_id)
             return await repo.delete_by_id(model_id)
 
+        self.activate_model()
+
     @classmethod
     async def load_inference_model(cls, model: Model, device: str | None = None) -> OpenVINOInferencer:
         """Load a model for inference using the anomalib OpenVINO inferencer.

application/backend/src/services/training_service.py

@@ -109,13 +109,8 @@ async def _run_training_job(cls, job: Job, job_service: JobService) -> Model | N
                 job_id=job.id, status=JobStatus.FAILED, message=f"Failed with exception: {str(e)}"
             )
             if model.export_path:
-                logger.warning("Deleting partially created model with id: %s", model.id)
-                await cls._cleanup_partial_model(
-                    job=job,
-                    model=model,
-                    delete_model_record=True,
-                    model_service=model_service,
-                )
+                logger.warning(f"Deleting partially created model with id: {model.id}")
+                await model_service.delete_model(project_id=project_id, model_id=model.id, delete_artifacts=True)
             raise e
         finally:
             logger.debug("Syncing progress with db stopped")
@@ -218,27 +213,10 @@ async def _handle_job_cancellation(job_service: JobService, job: Job, model: Mod
             status=JobStatus.CANCELED,
             message="Training cancelled by user",
         )
-        await TrainingService._cleanup_partial_model(job=job, model=model, delete_model_record=False)
-
-    @staticmethod
-    async def _cleanup_partial_model(
-        *,
-        job: Job,
-        model: Model,
-        delete_model_record: bool,
-        model_service: ModelService | None = None,
-    ) -> None:
-        """Remove partially exported artifacts and optionally delete model record."""
-        if not model.export_path:
-            return
 
         model_binary_repo = ModelBinaryRepository(project_id=job.project_id, model_id=model.id)
         await model_binary_repo.delete_model_folder()
 
-        if delete_model_record:
-            service = model_service or ModelService()
-            await service.delete_model(project_id=job.project_id, model_id=model.id)
-
     @staticmethod
     def _compute_export_size(path: str | None) -> int | None:
         if path is None:

application/backend/tests/unit/endpoints/test_models.py

@@ -10,7 +10,8 @@
 from api.dependencies import get_model_service
 from main import app
 from pydantic_models import Model, ModelList
-from services import ModelService
+from services import ModelService, ResourceNotFoundError
+from services.exceptions import ResourceType
 
 
 @pytest.fixture
@@ -51,3 +52,24 @@ def test_get_models(fxt_client, fxt_model_service, fxt_model, fxt_project):
     assert len(response.json()["models"]) == 1
     assert response.json()["models"][0]["size"] == 1024
     fxt_model_service.get_model_list.assert_called_once_with(project_id=project_id)
+
+
+def test_delete_model_success(fxt_client, fxt_model_service, fxt_project):
+    project_id = fxt_project.id
+    model_id = uuid4()
+
+    response = fxt_client.delete(f"/api/projects/{project_id}/models/{model_id}")
+
+    assert response.status_code == status.HTTP_204_NO_CONTENT
+    fxt_model_service.delete_model.assert_called_once_with(project_id=project_id, model_id=model_id)
+
+
+def test_delete_model_not_found(fxt_client, fxt_model_service, fxt_project):
+    project_id = fxt_project.id
+    model_id = uuid4()
+    fxt_model_service.delete_model.side_effect = ResourceNotFoundError(ResourceType.MODEL, str(model_id))
+
+    response = fxt_client.delete(f"/api/projects/{project_id}/models/{model_id}")
+
+    assert response.status_code == status.HTTP_404_NOT_FOUND
+    assert "not found" in response.json()["detail"].lower()

application/backend/tests/unit/services/test_model_service.py

@@ -128,7 +128,9 @@ def test_delete_model(self, fxt_model_service, fxt_model_repository, fxt_model,
         with patch("services.model_service.ModelRepository") as mock_repo_class:
             mock_repo_class.return_value = fxt_model_repository
 
-            asyncio.run(fxt_model_service.delete_model(fxt_project.id, fxt_model.id))
+            asyncio.run(
+                fxt_model_service.delete_model(fxt_model_service, fxt_project.id, fxt_model.id, delete_artifacts=False)
+            )
 
         fxt_model_repository.delete_by_id.assert_called_once_with(fxt_model.id)
 

application/backend/tests/unit/services/test_training_service.py

@@ -203,7 +203,6 @@ def mock_train_model(cls, model, synchronization_parameters: ProgressSyncParams,
                 asyncio.run(TrainingService.train_pending_job())
 
             # Verify cleanup was called
-            fxt_mock_binary_repos[0].return_value.delete_model_folder.assert_called_once()
             fxt_mock_model_service.delete_model.assert_called_once()
             call_args = fxt_mock_model_service.delete_model.call_args
             assert call_args[1]["project_id"] == fxt_job.project_id
@@ -255,6 +254,7 @@ def test_train_pending_job_cancelled(
         fxt_mock_job_service_class,
         fxt_mock_model_service_class,
         fxt_mock_job_service,
+        fxt_mock_binary_repos,
     ):
         """Training should mark job as cancelled when cancellation flag is set."""
         fxt_job.payload = {"model_name": "padim"}