('/workspace/source/test/e2e/explainer/test_art_explainer.py', 38, 'Skipped: ODH does not support art explainer at the moment')('/workspace/source/test/e2e/predictor/test_grpc.py', 35, 'Skipped: Not testable in ODH at the moment')('/workspace/source/test/e2e/predictor/test_torchserve.py', 34, 'Skipped: ODH does not support torchserve at the moment')file /workspace/source/test/e2e/batcher/test_raw_batcher.py, line 33 @pytest.mark.raw @pytest.mark.asyncio(scope="session") async def test_batcher_raw(rest_v1_client, network_layer): suffix = str(uuid.uuid4())[1:6] service_name = "isvc-raw-sklearn-batcher-" + suffix annotations = dict() annotations["serving.kserve.io/deploymentMode"] = "Standard" labels = dict() labels["networking.kserve.io/visibility"] = "exposed" predictor = V1beta1PredictorSpec( batcher=V1beta1Batcher( max_batch_size=32, max_latency=5000, ), min_replicas=1, sklearn=V1beta1SKLearnSpec( storage_uri="gs://kfserving-examples/models/sklearn/1.0/model", resources=V1ResourceRequirements( requests={"cpu": "50m", "memory": "128Mi"}, limits={"cpu": "100m", "memory": "256Mi"}, ), ), ) isvc = V1beta1InferenceService( api_version=constants.KSERVE_V1BETA1, kind=constants.KSERVE_KIND_INFERENCESERVICE, metadata=client.V1ObjectMeta( name=service_name, namespace=KSERVE_TEST_NAMESPACE, annotations=annotations, labels=labels, ), spec=V1beta1InferenceServiceSpec(predictor=predictor), ) kserve_client.create(isvc) try: kserve_client.wait_isvc_ready(service_name, namespace=KSERVE_TEST_NAMESPACE) except RuntimeError as e: print( kserve_client.api_instance.get_namespaced_custom_object( "serving.knative.dev", "v1", KSERVE_TEST_NAMESPACE, "services", service_name + "-predictor", ) ) pods = kserve_client.core_api.list_namespaced_pod( KSERVE_TEST_NAMESPACE, label_selector="serving.kserve.io/inferenceservice={}".format(service_name), ) for pod in pods.items: print(pod) raise e results = await predict_isvc( rest_v1_client, service_name, "./data/iris_batch_input.json", is_batch=True, network_layer=network_layer, ) assert all(x == results[0] for x in results) kserve_client.delete(service_name, KSERVE_TEST_NAMESPACE) file /workspace/source/test/e2e/common/gateway_proxy_istio.py, line 121 @pytest.fixture(autouse=True) def ensure_gateway_proxy_memory(test_case): E fixture 'test_case' not found > available fixtures: _session_event_loop, anyio_backend, anyio_backend_name, anyio_backend_options, assert_all_responses_were_requested, batcher/test_raw_batcher.py::<event_loop>, cache, capfd, capfdbinary, caplog, capsys, capsysbinary, configure_logger, cov, doctest_namespace, ensure_gateway_proxy_memory, event_loop, event_loop_policy, free_tcp_port, free_tcp_port_factory, free_udp_port, free_udp_port_factory, httpx_mock, include_metadata_in_junit_xml, json_metadata, metadata, monkeypatch, network_layer, no_cover, non_mocked_hosts, pytestconfig, record_property, record_testsuite_property, record_xml_attribute, recwarn, rest_v1_client, rest_v2_client, testrun_uid, tmp_path, tmp_path_factory, tmpdir, tmpdir_factory, unused_tcp_port, unused_tcp_port_factory, unused_udp_port, unused_udp_port_factory, worker_id > use 'pytest --fixtures [testpath]' for help on them. /workspace/source/test/e2e/common/gateway_proxy_istio.py:121/workspace/source/test/e2e/custom/test_custom_model_grpc.py:379: Not testable in ODH at the momentfile /workspace/source/test/e2e/graph/test_inference_graph.py, line 944 @pytest.mark.raw @pytest.mark.asyncio(scope="session") async def test_inference_graph_raw_mode(rest_v1_client, network_layer): logger.info("Starting test test_inference_graph_raw_mode") suffix = str(uuid.uuid4())[1:6] sklearn_name = "isvc-sklearn-graph-raw-" + suffix xgb_name = "isvc-xgboost-graph-raw-" + suffix graph_name = "model-chainer-raw-" + suffix annotations = dict() annotations["serving.kserve.io/deploymentMode"] = "Standard" labels = dict() labels["networking.kserve.io/visibility"] = "exposed" sklearn_predictor = V1beta1PredictorSpec( min_replicas=1, sklearn=V1beta1SKLearnSpec( storage_uri="gs://kfserving-examples/models/sklearn/1.0/model", resources=V1ResourceRequirements( requests={"cpu": "50m", "memory": "128Mi"}, limits={"cpu": "100m", "memory": "256Mi"}, ), ), ) sklearn_isvc = V1beta1InferenceService( api_version=constants.KSERVE_V1BETA1, kind=constants.KSERVE_KIND_INFERENCESERVICE, metadata=client.V1ObjectMeta( name=sklearn_name, namespace=KSERVE_TEST_NAMESPACE, annotations=annotations, labels=labels, ), spec=V1beta1InferenceServiceSpec(predictor=sklearn_predictor), ) xgb_predictor = V1beta1PredictorSpec( min_replicas=1, xgboost=V1beta1XGBoostSpec( storage_uri="gs://kfserving-examples/models/xgboost/1.5/model", resources=V1ResourceRequirements( requests={"cpu": "50m", "memory": "128Mi"}, limits={"cpu": "100m", "memory": "256Mi"}, ), ), ) xgb_isvc = V1beta1InferenceService( api_version=constants.KSERVE_V1BETA1, kind=constants.KSERVE_KIND_INFERENCESERVICE, metadata=client.V1ObjectMeta( name=xgb_name, namespace=KSERVE_TEST_NAMESPACE, annotations=annotations, labels=labels, ), spec=V1beta1InferenceServiceSpec(predictor=xgb_predictor), ) nodes = { "root": V1alpha1InferenceRouter( router_type="Sequence", steps=[ V1alpha1InferenceStep( service_name=sklearn_name, ), V1alpha1InferenceStep( service_name=xgb_name, data="$request", ), ], ) } graph_spec = V1alpha1InferenceGraphSpec( nodes=nodes, ) ig = V1alpha1InferenceGraph( api_version=constants.KSERVE_V1ALPHA1, kind=constants.KSERVE_KIND_INFERENCEGRAPH, metadata=client.V1ObjectMeta( name=graph_name, namespace=KSERVE_TEST_NAMESPACE, annotations=annotations ), spec=graph_spec, ) kserve_client = KServeClient( config_file=os.environ.get("KUBECONFIG", "~/.kube/config") ) kserve_client.create(sklearn_isvc) kserve_client.create(xgb_isvc) kserve_client.wait_isvc_ready(sklearn_name, namespace=KSERVE_TEST_NAMESPACE) kserve_client.wait_isvc_ready(xgb_name, namespace=KSERVE_TEST_NAMESPACE) kserve_client.create_inference_graph(ig) kserve_client.wait_ig_ready(graph_name, namespace=KSERVE_TEST_NAMESPACE) # Below checks are raw deployment specific. They ensure raw k8s resources created instead of knative resources dep = kserve_client.app_api.read_namespaced_deployment( graph_name, namespace=KSERVE_TEST_NAMESPACE ) if not dep: raise RuntimeError( "Deployment doesn't exist for InferenceGraph {} in raw deployment mode".format( graph_name ) ) svc = kserve_client.core_api.read_namespaced_service( graph_name, namespace=KSERVE_TEST_NAMESPACE ) if not svc: raise RuntimeError( "Service doesn't exist for InferenceGraph {} in raw deployment mode".format( graph_name ) ) try: knativeroute = kserve_client.api_instance.get_namespaced_custom_object( "serving.knative.dev", "v1", KSERVE_TEST_NAMESPACE, "routes", graph_name ) if knativeroute: raise RuntimeError( "Knative route resource shouldn't exist for InferenceGraph {}".format( graph_name ) + "in raw deployment mode" ) except client.rest.ApiException: logger.info("Expected error in finding knative route in raw deployment mode") try: knativesvc = kserve_client.api_instance.get_namespaced_custom_object( "serving.knative.dev", "v1", KSERVE_TEST_NAMESPACE, "services", graph_name ) if knativesvc: raise RuntimeError( "Knative resources shouldn't exist for InferenceGraph {} ".format( graph_name ) + "in raw deployment mode" ) except client.rest.ApiException: logger.info("Expected error in finding knative service in raw deployment mode") # TODO Fix this when we enable ALB creation for IG raw deployment mode. This is required for traffic ingress # for this predict api call to work # # res = await predict_ig( # rest_v1_client, # graph_name, # os.path.join(IG_TEST_RESOURCES_BASE_LOCATION, "iris_input.json"), # network_layer=network_layer, # ) # assert res["predictions"] == [1, 1] kserve_client.delete_inference_graph(graph_name, KSERVE_TEST_NAMESPACE) kserve_client.delete(sklearn_name, KSERVE_TEST_NAMESPACE) kserve_client.delete(xgb_name, KSERVE_TEST_NAMESPACE) file /workspace/source/test/e2e/common/gateway_proxy_istio.py, line 121 @pytest.fixture(autouse=True) def ensure_gateway_proxy_memory(test_case): E fixture 'test_case' not found > available fixtures: _session_event_loop, anyio_backend, anyio_backend_name, anyio_backend_options, assert_all_responses_were_requested, cache, capfd, capfdbinary, caplog, capsys, capsysbinary, configure_logger, cov, doctest_namespace, ensure_gateway_proxy_memory, event_loop, event_loop_policy, free_tcp_port, free_tcp_port_factory, free_udp_port, free_udp_port_factory, graph/__init__.py::<event_loop>, graph/test_inference_graph.py::<event_loop>, httpx_mock, include_metadata_in_junit_xml, json_metadata, metadata, monkeypatch, network_layer, no_cover, non_mocked_hosts, pytestconfig, record_property, record_testsuite_property, record_xml_attribute, recwarn, rest_v1_client, rest_v2_client, testrun_uid, tmp_path, tmp_path_factory, tmpdir, tmpdir_factory, unused_tcp_port, unused_tcp_port_factory, unused_udp_port, unused_udp_port_factory, worker_id > use 'pytest --fixtures [testpath]' for help on them. /workspace/source/test/e2e/common/gateway_proxy_istio.py:121file /workspace/source/test/e2e/graph/test_inference_graph.py, line 1104 @pytest.mark.raw @pytest.mark.asyncio(scope="session") async def test_inference_graph_raw_mode_with_hpa(rest_v1_client, network_layer): logger.info("Starting test test_inference_graph_raw_mode_with_hpa") suffix = str(uuid.uuid4())[1:6] sklearn_name = "isvc-sklearn-graph-raw-hpa-" + suffix xgb_name = "isvc-xgboost-graph-raw-hpa-" + suffix graph_name = "model-chainer-raw-hpa-" + suffix annotations = dict() annotations["serving.kserve.io/deploymentMode"] = "Standard" # annotations["serving.kserve.io/max-scale"] = '5' # annotations["serving.kserve.io/metric"] = 'rps' # annotations["serving.kserve.io/min-scale"] = '2' # annotations["serving.kserve.io/target"] = '30' labels = dict() labels["networking.kserve.io/visibility"] = "exposed" sklearn_predictor = V1beta1PredictorSpec( min_replicas=1, sklearn=V1beta1SKLearnSpec( storage_uri="gs://kfserving-examples/models/sklearn/1.0/model", resources=V1ResourceRequirements( requests={"cpu": "50m", "memory": "128Mi"}, limits={"cpu": "100m", "memory": "256Mi"}, ), ), ) sklearn_isvc = V1beta1InferenceService( api_version=constants.KSERVE_V1BETA1, kind=constants.KSERVE_KIND_INFERENCESERVICE, metadata=client.V1ObjectMeta( name=sklearn_name, namespace=KSERVE_TEST_NAMESPACE, annotations=annotations, labels=labels, ), spec=V1beta1InferenceServiceSpec(predictor=sklearn_predictor), ) xgb_predictor = V1beta1PredictorSpec( min_replicas=1, xgboost=V1beta1XGBoostSpec( storage_uri="gs://kfserving-examples/models/xgboost/1.5/model", resources=V1ResourceRequirements( requests={"cpu": "50m", "memory": "128Mi"}, limits={"cpu": "100m", "memory": "256Mi"}, ), ), ) xgb_isvc = V1beta1InferenceService( api_version=constants.KSERVE_V1BETA1, kind=constants.KSERVE_KIND_INFERENCESERVICE, metadata=client.V1ObjectMeta( name=xgb_name, namespace=KSERVE_TEST_NAMESPACE, annotations=annotations, labels=labels, ), spec=V1beta1InferenceServiceSpec(predictor=xgb_predictor), ) nodes = { "root": V1alpha1InferenceRouter( router_type="Sequence", steps=[ V1alpha1InferenceStep( service_name=sklearn_name, ), V1alpha1InferenceStep( service_name=xgb_name, data="$request", ), ], ) } graph_spec = V1alpha1InferenceGraphSpec( nodes=nodes, ) ig = V1alpha1InferenceGraph( api_version=constants.KSERVE_V1ALPHA1, kind=constants.KSERVE_KIND_INFERENCEGRAPH, metadata=client.V1ObjectMeta( name=graph_name, namespace=KSERVE_TEST_NAMESPACE, annotations=annotations ), spec=graph_spec, ) kserve_client = KServeClient( config_file=os.environ.get("KUBECONFIG", "~/.kube/config") ) kserve_client.create(sklearn_isvc) kserve_client.create(xgb_isvc) kserve_client.wait_isvc_ready(sklearn_name, namespace=KSERVE_TEST_NAMESPACE) kserve_client.wait_isvc_ready(xgb_name, namespace=KSERVE_TEST_NAMESPACE) kserve_client.create_inference_graph(ig) kserve_client.wait_ig_ready(graph_name, namespace=KSERVE_TEST_NAMESPACE) # Below checks are raw deployment specific. They ensure raw k8s resources created instead of knative resources dep = kserve_client.app_api.read_namespaced_deployment( graph_name, namespace=KSERVE_TEST_NAMESPACE ) if not dep: raise RuntimeError( "Deployment doesn't exist for InferenceGraph {} in raw deployment mode".format( graph_name ) ) svc = kserve_client.core_api.read_namespaced_service( graph_name, namespace=KSERVE_TEST_NAMESPACE ) if not svc: raise RuntimeError( "Service doesn't exist for InferenceGraph {} in raw deployment mode".format( graph_name ) ) # hpa = kserve_client.hpa_v2_api.read_namespaced_horizontal_pod_autoscaler(graph_name, # namespace=KSERVE_TEST_NAMESPACE) # if not hpa: # raise RuntimeError("HPA doesn't exist for InferenceGraph {} in raw deployment mode".format(graph_name)) try: knativeroute = kserve_client.api_instance.get_namespaced_custom_object( "serving.knative.dev", "v1", KSERVE_TEST_NAMESPACE, "routes", graph_name ) if knativeroute: raise RuntimeError( "Knative route resource shouldn't exist for InferenceGraph {} ".format( graph_name ) + "in raw deployment mode" ) except client.rest.ApiException: logger.info("Expected error in finding knative route in raw deployment mode") try: knativesvc = kserve_client.api_instance.get_namespaced_custom_object( "serving.knative.dev", "v1", KSERVE_TEST_NAMESPACE, "services", graph_name ) if knativesvc: raise RuntimeError( "Knative resources shouldn't exist for InferenceGraph {} ".format( graph_name ) + "in raw deployment mode" ) except client.rest.ApiException: logger.info("Expected error in finding knative route in raw deployment mode") # TODO Fix this when we enable ALB creation for IG raw deployment mode. This is required for traffic ingress # for this predict api call to work # # res = await predict_ig( # rest_v1_client, # graph_name, # os.path.join(IG_TEST_RESOURCES_BASE_LOCATION, "iris_input.json"), # network_layer=network_layer, # ) # assert res["predictions"] == [1, 1] kserve_client.delete_inference_graph(graph_name, KSERVE_TEST_NAMESPACE) kserve_client.delete(sklearn_name, KSERVE_TEST_NAMESPACE) kserve_client.delete(xgb_name, KSERVE_TEST_NAMESPACE) file /workspace/source/test/e2e/common/gateway_proxy_istio.py, line 121 @pytest.fixture(autouse=True) def ensure_gateway_proxy_memory(test_case): E fixture 'test_case' not found > available fixtures: _session_event_loop, anyio_backend, anyio_backend_name, anyio_backend_options, assert_all_responses_were_requested, cache, capfd, capfdbinary, caplog, capsys, capsysbinary, configure_logger, cov, doctest_namespace, ensure_gateway_proxy_memory, event_loop, event_loop_policy, free_tcp_port, free_tcp_port_factory, free_udp_port, free_udp_port_factory, graph/__init__.py::<event_loop>, graph/test_inference_graph.py::<event_loop>, httpx_mock, include_metadata_in_junit_xml, json_metadata, metadata, monkeypatch, network_layer, no_cover, non_mocked_hosts, pytestconfig, record_property, record_testsuite_property, record_xml_attribute, recwarn, rest_v1_client, rest_v2_client, testrun_uid, tmp_path, tmp_path_factory, tmpdir, tmpdir_factory, unused_tcp_port, unused_tcp_port_factory, unused_udp_port, unused_udp_port_factory, worker_id > use 'pytest --fixtures [testpath]' for help on them. /workspace/source/test/e2e/common/gateway_proxy_istio.py:121file /workspace/source/test/e2e/logger/test_raw_logger.py, line 40 @pytest.mark.raw @pytest.mark.asyncio(scope="session") async def test_kserve_logger(rest_v1_client, network_layer): suffix = str(uuid.uuid4())[1:6] msg_dumper = "message-dumper-raw-" + suffix before(msg_dumper) service_name = "isvc-logger-raw-" + suffix predictor = V1beta1PredictorSpec( min_replicas=1, logger=V1beta1LoggerSpec( mode="all", url="http://" + msg_dumper + "-predictor" + "." + KSERVE_TEST_NAMESPACE + ".svc.cluster.local", ), sklearn=V1beta1SKLearnSpec( storage_uri="gs://kfserving-examples/models/sklearn/1.0/model", resources=V1ResourceRequirements( requests={"cpu": "10m", "memory": "128Mi"}, limits={"cpu": "100m", "memory": "256Mi"}, ), ), ) await base_test(msg_dumper, service_name, predictor, rest_v1_client, network_layer) file /workspace/source/test/e2e/common/gateway_proxy_istio.py, line 121 @pytest.fixture(autouse=True) def ensure_gateway_proxy_memory(test_case): E fixture 'test_case' not found > available fixtures: _session_event_loop, anyio_backend, anyio_backend_name, anyio_backend_options, assert_all_responses_were_requested, cache, capfd, capfdbinary, caplog, capsys, capsysbinary, configure_logger, cov, doctest_namespace, ensure_gateway_proxy_memory, event_loop, event_loop_policy, free_tcp_port, free_tcp_port_factory, free_udp_port, free_udp_port_factory, httpx_mock, include_metadata_in_junit_xml, json_metadata, logger/test_raw_logger.py::<event_loop>, metadata, monkeypatch, network_layer, no_cover, non_mocked_hosts, pytestconfig, record_property, record_testsuite_property, record_xml_attribute, recwarn, rest_v1_client, rest_v2_client, testrun_uid, tmp_path, tmp_path_factory, tmpdir, tmpdir_factory, unused_tcp_port, unused_tcp_port_factory, unused_udp_port, unused_udp_port_factory, worker_id > use 'pytest --fixtures [testpath]' for help on them. /workspace/source/test/e2e/common/gateway_proxy_istio.py:121file /workspace/source/test/e2e/predictor/test_autoscaling.py, line 206 @pytest.mark.raw @pytest.mark.asyncio(scope="session") async def test_sklearn_scale_raw(rest_v1_client, network_layer): suffix = str(uuid.uuid4())[1:6] service_name = "isvc-sklearn-scale-raw-" + suffix predictor = V1beta1PredictorSpec( min_replicas=1, scale_metric="cpu", scale_target=50, sklearn=V1beta1SKLearnSpec( storage_uri=MODEL, resources=V1ResourceRequirements( requests={"cpu": "25m", "memory": "128Mi"}, limits={"cpu": "50m", "memory": "256Mi"}, ), ), ) annotations = {"serving.kserve.io/deploymentMode": "Standard"} labels = dict() labels["networking.kserve.io/visibility"] = "exposed" isvc = V1beta1InferenceService( api_version=constants.KSERVE_V1BETA1, kind=constants.KSERVE_KIND_INFERENCESERVICE, metadata=client.V1ObjectMeta( name=service_name, namespace=KSERVE_TEST_NAMESPACE, annotations=annotations, labels=labels, ), spec=V1beta1InferenceServiceSpec(predictor=predictor), ) kserve_client = KServeClient( config_file=os.environ.get("KUBECONFIG", "~/.kube/config") ) kserve_client.create(isvc) kserve_client.wait_isvc_ready(service_name, namespace=KSERVE_TEST_NAMESPACE) api_instance = kserve_client.api_instance hpa_resp = api_instance.list_namespaced_custom_object( group="autoscaling", version="v1", namespace=KSERVE_TEST_NAMESPACE, label_selector=f"serving.kserve.io/inferenceservice={service_name}", plural="horizontalpodautoscalers", ) assert hpa_resp["items"][0]["spec"]["targetCPUUtilizationPercentage"] == 50 res = await predict_isvc( rest_v1_client, service_name, INPUT, network_layer=network_layer ) assert res["predictions"] == [1, 1] kserve_client.delete(service_name, KSERVE_TEST_NAMESPACE) file /workspace/source/test/e2e/common/gateway_proxy_istio.py, line 121 @pytest.fixture(autouse=True) def ensure_gateway_proxy_memory(test_case): E fixture 'test_case' not found > available fixtures: _cleanup_orphaned_predictor_isvcs, _session_event_loop, anyio_backend, anyio_backend_name, anyio_backend_options, assert_all_responses_were_requested, cache, capfd, capfdbinary, caplog, capsys, capsysbinary, configure_logger, cov, doctest_namespace, ensure_gateway_proxy_memory, event_loop, event_loop_policy, free_tcp_port, free_tcp_port_factory, free_udp_port, free_udp_port_factory, httpx_mock, include_metadata_in_junit_xml, json_metadata, metadata, monkeypatch, network_layer, no_cover, non_mocked_hosts, predictor/test_autoscaling.py::<event_loop>, pytestconfig, record_property, record_testsuite_property, record_xml_attribute, recwarn, rest_v1_client, rest_v2_client, testrun_uid, tmp_path, tmp_path_factory, tmpdir, tmpdir_factory, unused_tcp_port, unused_tcp_port_factory, unused_udp_port, unused_udp_port_factory, worker_id > use 'pytest --fixtures [testpath]' for help on them. /workspace/source/test/e2e/common/gateway_proxy_istio.py:121