test(serve): Skip hang-prone endpoint integ tests in PR check

sagemaker-serve/pyproject.toml

@@ -74,7 +74,7 @@ python_classes = ["Test*"]
 python_functions = ["test_*"]
 addopts = "-v --tb=short"
 markers = [
-    "skip_in_pr_check: mark a test that is excluded from PR check runs. Long-running or hang-prone tests that would otherwise push the run past the CodeBuild timeout; they run in a dedicated scheduled CI run instead.",
+    "slow_test: mark a test that is excluded from PR check runs. Long-running or hang-prone tests that would otherwise push the run past the CodeBuild timeout; they run in a dedicated scheduled CI run instead.",
 ]
 
 [tool.black]

sagemaker-serve/tests/integ/test_ai_inference_recommender_enhancements_integration.py

@@ -95,7 +95,6 @@ def test_list_benchmarks_and_recommendations_plumbing():
     logger.info("Non-matching filters correctly returned empty lists.")
 
 
-@pytest.mark.slow_test
 @pytest.mark.gpu_intensive
 def test_recommendation_deploy_best_and_compare_e2e():
     """Full flow across all three enhancements, sharing one rec job + endpoint:

sagemaker-serve/tests/integ/test_ai_inference_recommender_integration.py

@@ -71,7 +71,6 @@ def _build_jumpstart_model_builder(role_arn):
 
 
 @pytest.mark.slow_test
-@pytest.mark.skip_in_pr_check
 def test_benchmark_workflow_end_to_end():
     """Deploy a JumpStart endpoint, run a benchmark against it, parse the result."""
     logger.info("Starting AI inference recommender benchmark integration test...")
@@ -127,7 +126,6 @@ def test_benchmark_workflow_end_to_end():
         )
 
 
-@pytest.mark.slow_test
 @pytest.mark.gpu_intensive
 def test_recommendation_workflow_end_to_end():
     """Run an AI recommendation via generate_deployment_recommendations and deploy the top recommendation."""

sagemaker-serve/tests/integ/test_ai_inference_recommender_sdkt_ic_integration.py

@@ -63,7 +63,6 @@ def _src(uri):
     ]
 
 
-@pytest.mark.slow_test
 @pytest.mark.gpu_intensive
 def test_deploy_sdkt_model_as_inference_component():
     """A model carrying SD/KT AdditionalModelDataSources deploys as an

sagemaker-serve/tests/integ/test_in_process_integration.py

@@ -50,7 +50,6 @@ def invoke(self, input_object, model):
         return {"result": result, "operation": f"multiply by {factor}"}
 
 
-@pytest.mark.slow_test
 def test_in_process_build_deploy_invoke_cleanup():
     """Integration test for In-Process mode build, deploy, invoke, and cleanup workflow"""
     logger.info("Starting In-Process integration test...")

sagemaker-serve/tests/integ/test_jumpstart_deploy_parity.py

@@ -26,7 +26,6 @@
 MODEL_NAME_PREFIX = "js-netiso-test"
 
 
-@pytest.mark.slow_test
 def test_jumpstart_build_enables_network_isolation():
     """Integration test verifying JumpStart models are built with EnableNetworkIsolation.
 
@@ -71,7 +70,6 @@ def test_jumpstart_build_enables_network_isolation():
 VOLUME_SIZE_INSTANCE_TYPE = "ml.inf2.xlarge"
 
 
-@pytest.mark.slow_test
 def test_jumpstart_build_sets_volume_size():
     """Integration test verifying volume_size from model specs is propagated.
 

sagemaker-serve/tests/integ/test_jumpstart_integration.py

@@ -32,7 +32,6 @@
 SERVE_SAGEMAKER_ENDPOINT_TIMEOUT = 15
 
 
-@pytest.mark.slow_test
 @pytest.mark.gpu_intensive
 def test_jumpstart_build_deploy_invoke_cleanup():
     """Integration test for JumpStart model build, deploy, invoke, and cleanup workflow"""

sagemaker-serve/tests/integ/test_jumpstart_vllm_integration.py

@@ -27,7 +27,6 @@
 MODEL_NAME_PREFIX = "js-vllm-test-model"
 
 
-@pytest.mark.slow_test
 def test_jumpstart_vllm_build():
     """Integration test for JumpStart model using vLLM container image.
 

sagemaker-serve/tests/integ/test_model_customization_deployment.py

@@ -118,7 +118,7 @@ def test_build_from_training_job(self, training_job_name, sagemaker_session):
         assert model_builder.image_uri is not None
         assert model_builder.instance_type is not None
 
-    @pytest.mark.skip_in_pr_check
+    @pytest.mark.slow_test
     def test_deploy_from_training_job(self, training_job_name, sagemaker_session):
         """Deploy, reuse, invoke, and clean up one training-job endpoint."""
         test_id = uuid.uuid4().hex
@@ -325,6 +325,7 @@ def test_build_from_model_package(self, model_package_arn, sagemaker_session):
         assert model is not None
         assert model.model_arn is not None
 
+    @pytest.mark.slow_test
     def test_deploy_from_model_package(
         self, model_package_arn, cleanup_endpoints, sagemaker_session
     ):

sagemaker-serve/tests/integ/test_optimize_integration.py

@@ -39,7 +39,7 @@
 DJL_LMI_VERSION = "0.31.0"
 
 
-@pytest.mark.skip_in_pr_check
+@pytest.mark.slow_test
 def test_optimize_build_deploy_invoke_cleanup():
     """Integration test for Optimize workflow"""
     logger.info("Starting Optimize integration test...")

sagemaker-serve/tests/integ/test_passthrough_source_code_repack_integration.py

@@ -68,7 +68,6 @@ def _tar_members(s3_client, s3_uri):
     return tarfile.open(fileobj=io.BytesIO(body), mode="r:gz").getnames()
 
 
-@pytest.mark.slow_test
 def test_build_repacks_source_code_into_artifact():
     """build() with image_uri + model artifact + source_code repacks code/ into
     the model.tar.gz. No deploy - runs in seconds."""

sagemaker-serve/tests/integ/test_private_hub_artifact_resolution.py

@@ -154,7 +154,6 @@ def execution_role():
     return f"arn:aws:iam::{account_id}:role/Admin"
 
 
-@pytest.mark.slow_test
 def test_from_jumpstart_config_derives_hub_arn(private_hub, sagemaker_session):
     """Verify from_jumpstart_config correctly derives hub_arn from hub_name."""
     js_config = JumpStartConfig(
@@ -176,7 +175,6 @@ def test_from_jumpstart_config_derives_hub_arn(private_hub, sagemaker_session):
     logger.info("hub_arn correctly derived: %s", mb.hub_arn)
 
 
-@pytest.mark.slow_test
 def test_build_resolves_artifacts_via_private_hub(private_hub, execution_role, sagemaker_session):
     """Verify build() resolves model data through the private hub."""
     js_config = JumpStartConfig(
@@ -388,7 +386,6 @@ def _deploy_and_assert_hub_access_config(
                 logger.warning("Cleanup failed for %s: %s", kwargs, e)
 
 
-@pytest.mark.slow_test
 def test_deploy_with_no_s3_execution_role(private_hub, no_s3_execution_role, sagemaker_session):
     """E2E: deploy from a private hub with an execution role that has ZERO
     S3 permissions. Passes only when the SDK attaches HubAccessConfig to
@@ -406,7 +403,6 @@ def test_deploy_with_no_s3_execution_role(private_hub, no_s3_execution_role, sag
     )
 
 
-@pytest.mark.slow_test
 def test_deploy_with_aliased_hub_content_name(
     private_hub, aliased_model_reference, no_s3_execution_role, sagemaker_session
 ):

sagemaker-serve/tests/integ/test_train_inference_e2e_integration.py

@@ -34,7 +34,6 @@
 TRAINING_JOB_PREFIX = "e2e-v3-pytorch"
 
 
-@pytest.mark.slow_test
 def test_train_inference_e2e_build_deploy_invoke_cleanup():
     """Integration test for Train-Inference E2E workflow"""
     logger.info("Starting Train-Inference E2E integration test...")

sagemaker-serve/tests/integ/test_triton_integration.py

@@ -44,7 +44,6 @@ def forward(self, x):
         return torch.softmax(self.linear(x), dim=1)
 
 
-@pytest.mark.slow_test
 def test_triton_build_deploy_invoke_cleanup():
     """Integration test for Triton model build, deploy, invoke, and cleanup workflow"""
     logger.info("Starting Triton integration test...")

sagemaker-serve/tox.ini

@@ -62,14 +62,13 @@ markers =
     canary_quick
     cron
     local_mode
-    slow_test
+    slow_test: mark a test that is excluded from PR check runs. Long-running or hang-prone tests that would otherwise push the run past the CodeBuild timeout; they run in a dedicated scheduled CI run instead.
     release
     image_uris_unit_test
     timeout: mark a test as a timeout.
     gpu_intensive: mark a test as GPU resource intensive (runs on scheduled CI, not PR checks).
     us_east_1: mark a test that requires us-east-1 test account credentials (784379639078).
     import_model: mark a test that creates a Bedrock model import job. Concurrent model import jobs are capped at 1 by a non-raisable Bedrock service quota, so these run serially in a dedicated scheduled CI run, not in PR checks.
-    skip_in_pr_check: mark a test that is excluded from PR check runs. Long-running or hang-prone tests that would otherwise push the run past the CodeBuild timeout; they run in a dedicated scheduled CI run instead.
 
 [testenv]
 setenv =