fix(serve): pass instance_type to _deploy once when creating an inference component (#6203)

Deploying a CustomOrchestrator in InferenceComponent mode through
ModelBuilder.deploy() always raised

    TypeError: _deploy() got multiple values for keyword argument 'instance_type'

In _deploy_for_ic the branch that creates a new inference component read
instance_type and initial_instance_count out of kwargs with kwargs.get,
passed them to self._deploy as explicit keyword arguments, and then spread
the unchanged **kwargs into the same call. kwargs.get leaves the key in
place, so Python saw each of those arguments twice. The only caller of
this branch, ModelBuilder.deploy, always supplies both keys, so the
failure was deterministic.

Consume the two keys with kwargs.pop instead so each is forwarded exactly
once. The defaults (self.instance_type and 1) are unchanged and kwargs is
a local dict, so the caller is unaffected. _deploy_core_endpoint reads the
values back through kwargs.get and receives the same data as before.

Add unit tests covering the create branch of _deploy_for_ic with explicit
instance arguments (fails with the TypeError before this change) and with
the default fallbacks.

Fixes #6199

Co-authored-by: lucasjia-aws <lucasjia@amazon.com>

Author: Om-singhaI

sagemaker-serve/src/sagemaker/serve/model_builder.py

@@ -4355,8 +4355,8 @@ def _deploy_for_ic(self, ic_data: Dict[str, Any], endpoint_name: str, **kwargs)
                 endpoint_type=EndpointType.INFERENCE_COMPONENT_BASED,
                 resources=resource_requirements,
                 inference_component_name=ic_name,
-                instance_type=kwargs.get("instance_type", self.instance_type),
-                initial_instance_count=kwargs.get("initial_instance_count", 1),
+                instance_type=kwargs.pop("instance_type", self.instance_type),
+                initial_instance_count=kwargs.pop("initial_instance_count", 1),
                 **kwargs,
             )
 

sagemaker-serve/tests/unit/test_model_builder_deploy.py

@@ -514,6 +514,74 @@ def test_does_ic_exist_false(self):
 
         self.assertFalse(result)
 
+    def test_deploy_for_ic_creates_new_ic_with_explicit_instance_kwargs(self):
+        """Test _deploy_for_ic forwards instance_type and initial_instance_count to _deploy once."""
+        builder = ModelBuilder(
+            model=Mock(),
+            role_arn="arn:aws:iam::123456789012:role/TestRole",
+            sagemaker_session=self.mock_session,
+        )
+        builder._does_ic_exist = Mock(return_value=False)
+        builder._deploy = Mock(return_value="endpoint")
+        built_model = Mock()
+        resource_requirements = ResourceRequirements(requests={"memory": 1024, "copies": 1})
+        ic_data = {
+            "Name": "test-ic",
+            "ResourceRequirements": resource_requirements,
+            "Model": built_model,
+        }
+
+        result = builder._deploy_for_ic(
+            ic_data=ic_data,
+            endpoint_name="test-endpoint",
+            container_timeout_in_seconds=600,
+            instance_type="ml.g5.xlarge",
+            initial_instance_count=2,
+        )
+
+        self.assertEqual(result, "endpoint")
+        builder._deploy.assert_called_once_with(
+            built_model=built_model,
+            endpoint_name="test-endpoint",
+            endpoint_type=EndpointType.INFERENCE_COMPONENT_BASED,
+            resources=resource_requirements,
+            inference_component_name="test-ic",
+            instance_type="ml.g5.xlarge",
+            initial_instance_count=2,
+            container_timeout_in_seconds=600,
+        )
+
+    def test_deploy_for_ic_creates_new_ic_with_default_instance_kwargs(self):
+        """Test _deploy_for_ic falls back to builder instance_type and one instance."""
+        builder = ModelBuilder(
+            model=Mock(),
+            role_arn="arn:aws:iam::123456789012:role/TestRole",
+            sagemaker_session=self.mock_session,
+        )
+        builder.instance_type = "ml.c5.xlarge"
+        builder._does_ic_exist = Mock(return_value=False)
+        builder._deploy = Mock(return_value="endpoint")
+        built_model = Mock()
+        resource_requirements = ResourceRequirements(requests={"memory": 1024, "copies": 1})
+        ic_data = {
+            "Name": "test-ic",
+            "ResourceRequirements": resource_requirements,
+            "Model": built_model,
+        }
+
+        result = builder._deploy_for_ic(ic_data=ic_data, endpoint_name="test-endpoint")
+
+        self.assertEqual(result, "endpoint")
+        builder._deploy.assert_called_once_with(
+            built_model=built_model,
+            endpoint_name="test-endpoint",
+            endpoint_type=EndpointType.INFERENCE_COMPONENT_BASED,
+            resources=resource_requirements,
+            inference_component_name="test-ic",
+            instance_type="ml.c5.xlarge",
+            initial_instance_count=1,
+        )
+
 
 class TestModelBuilderResetState(unittest.TestCase):
     """Test ModelBuilder _reset_build_state method."""