[ML][Pipelines] fix: avoid raising OSError for invalid paths in LocalPathField deserialization

sdk/ml/azure-ai-ml/azure/ai/ml/_schema/core/fields.py

@@ -131,14 +131,20 @@ def _resolve_path(self, value) -> Path:
         """Resolve path to absolute path based on base_path in context.
         Will resolve the path if it's already an absolute path.
         """
-        result = Path(value)
-        base_path = Path(self.context[BASE_PATH_CONTEXT_KEY])
-        if not result.is_absolute():
-            result = base_path / result
         try:
-            return result.resolve()
+            result = Path(value)
+            base_path = Path(self.context[BASE_PATH_CONTEXT_KEY])
+            if not result.is_absolute():
+                result = base_path / result
+
+            # for non-path string like "azureml:/xxx", OSError can be raised in either
+            # resolve() or is_dir() or is_file()
+            result = result.resolve()
+            if (self._allow_dir and result.is_dir()) or (self._allow_file and result.is_file()):
+                return result
         except OSError:
             raise self.make_error("invalid_path")
+        raise self.make_error("path_not_exist", path=result.as_posix(), allow_type=self.allowed_path_type)
 
     @property
     def allowed_path_type(self) -> str:
@@ -154,16 +160,12 @@ def _validate(self, value):
 
         if value is None:
             return
-        path = self._resolve_path(value)
-        if (self._allow_dir and path.is_dir()) or (self._allow_file and path.is_file()):
-            return
-        raise self.make_error("path_not_exist", path=path.as_posix(), allow_type=self.allowed_path_type)
+        self._resolve_path(value)
 
     def _serialize(self, value, attr, obj, **kwargs) -> typing.Optional[str]:
         # do not block serializing None even if required or not allow_none.
         if value is None:
             return None
-        self._validate(value)
         # always dump path as absolute path in string as base_path will be dropped after serialization
         return super(LocalPathField, self)._serialize(self._resolve_path(value).as_posix(), attr, obj, **kwargs)
 

sdk/ml/azure-ai-ml/azure/ai/ml/operations/_online_deployment_operations.py

@@ -162,7 +162,7 @@ def begin_create_or_update(
                 operation_config=self._operation_config,
             )
             if deployment.data_collector:
-                self._register_collection_data_assets(deployment= deployment)
+                self._register_collection_data_assets(deployment=deployment)
 
             upload_dependencies(deployment, orchestrators)
             try:
@@ -351,16 +351,13 @@ def _register_collection_data_assets(self, deployment: OnlineDeployment) -> None
         for collection in deployment.data_collector.collections:
             data_name = deployment.endpoint_name + "-" + deployment.name + "-" + collection
             data_object = Data(
-                name = data_name,
-                path = deployment.data_collector.destination.path
+                name=data_name,
+                path=deployment.data_collector.destination.path
                 if deployment.data_collector.destination and deployment.data_collector.destination.path
                 else DEFAULT_MDC_PATH,
-                is_anonymous= True
-        )
+                is_anonymous=True,
+            )
             result = self._all_operations._all_operations[AzureMLResourceType.DATA].create_or_update(data_object)
             deployment.data_collector.collections[collection].data = DataAsset(
-                data_id = result.id,
-                path = result.path,
-                name = result.name,
-                version = result.version
-                )
+                data_id=result.id, path=result.path, name=result.name, version=result.version
+            )

sdk/ml/azure-ai-ml/tests/component/unittests/test_component_schema.py

@@ -328,6 +328,21 @@ def test_dump_with_non_existent_base_path(self):
         component_entity._base_path = "/non/existent/path"
         component_entity._to_dict()
 
+    def test_arm_code_from_rest_object(self):
+        arm_code = (
+            "azureml:/subscriptions/xxx/resourceGroups/xxx/providers/Microsoft.MachineLearningServices/"
+            "workspaces/zzz/codes/90b33c11-365d-4ee4-aaa1-224a042deb41/versions/1"
+        )
+        yaml_path = "./tests/test_configs/components/helloworld_component.yml"
+        yaml_component = load_component(yaml_path)
+
+        from azure.ai.ml.entities import Component
+
+        rest_object = yaml_component._to_rest_object()
+        rest_object.properties.component_spec["code"] = arm_code
+        component = Component._from_rest_object(rest_object)
+        assert component.code == arm_code[8:]
+
 
 @pytest.mark.timeout(_COMPONENT_TIMEOUT_SECOND)
 @pytest.mark.unittest