('/workspace/source/test/e2e/explainer/test_art_explainer.py', 48, 'Skipped: ODH does not support art explainer at the moment')('/workspace/source/test/e2e/predictor/test_grpc.py', 39, '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')test_case = TestCase(base_refs=['router-managed', 'workload-single-cpu', 'model-fb-opt-125m'], prompt='KServe is a', service_name=... {'name': 'model-fb-opt-125m-auth-invalid-20a12e51'}]}, 'status': None}, model_name='facebook/opt-125m') @pytest.mark.llminferenceservice @pytest.mark.auth @pytest.mark.parametrize( "test_case", [ pytest.param( TestCase( base_refs=[ "router-managed", "workload-single-cpu", "model-fb-opt-125m", ], prompt="KServe is a", service_name="auth-invalid-token-test", ), marks=[ pytest.mark.cluster_cpu, pytest.mark.cluster_single_node, ], id="auth-invalid-token", ), ], indirect=["test_case"], ids=generate_test_id, ) @log_execution def test_llm_auth_invalid_token_rejected(test_case: TestCase): # noqa: F811 """ Test that when auth is enabled: - Requests with MALFORMED tokens are rejected """ inject_k8s_proxy() kserve_client = KServeClient( config_file=os.environ.get("KUBECONFIG", "~/.kube/config"), client_configuration=client.Configuration(), ) service_name = test_case.llm_service.metadata.name sa_name = f"{service_name}-test-sa" ns = test_case.llm_service.metadata.namespace test_failed = False # Enable auth for this test if not test_case.llm_service.metadata.annotations: test_case.llm_service.metadata.annotations = {} test_case.llm_service.metadata.annotations[ "security.opendatahub.io/enable-auth" ] = "true" try: # Create LLMInferenceService create_llmisvc(kserve_client, test_case.llm_service) > wait_for_llm_isvc_ready( kserve_client, test_case.llm_service, test_case.wait_timeout ) llmisvc/test_llm_auth.py:440: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ args = (<kserve.api.kserve_client.KServeClient object at 0x7fc2591eca90>, {'api_version': 'serving.kserve.io/v1alpha1', 'kin...invali-efc9e473'}, {'name': 'model-fb-opt-125m-auth-invalid-20a12e51'}]}, 'status': None}, 900) kwargs = {}, func_name = 'wait_for_llm_isvc_ready' timestamp_start = '2026-07-28T17:30:30.884470', start_time = 1785259830.8848135 duration = 900.2377202510834, timestamp_end = '2026-07-28T17:45:31.122542' @functools.wraps(func) def wrapper(*args, **kwargs): func_name = func.__name__ timestamp_start = datetime.now().isoformat() logger.info( f"[{func_name}] [{timestamp_start}] start - args={args}, kwargs={kwargs}" ) start_time = time.time() try: > result = func(*args, **kwargs) llmisvc/logging.py:40: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ kserve_client = <kserve.api.kserve_client.KServeClient object at 0x7fc2591eca90> given = {'api_version': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceService', 'metadata': {'annotations': {'security....-auth-invali-efc9e473'}, {'name': 'model-fb-opt-125m-auth-invalid-20a12e51'}]}, 'status': None} timeout_seconds = 900 @log_execution def wait_for_llm_isvc_ready( kserve_client: KServeClient, given: V1alpha1LLMInferenceService, timeout_seconds: int = 900, ) -> str: def assert_llm_isvc_ready(): out = get_llmisvc( kserve_client, given.metadata.name, given.metadata.namespace, given.api_version.split("/")[1], ) if "status" not in out: raise AssertionError("No status found in LLM inference service") status = out["status"] if "conditions" not in status: raise AssertionError("No conditions found in status") expected_true_conditions = {"Ready", "WorkloadsReady", "RouterReady"} got_true_conditions = set() all_condition_types = set() conditions = status["conditions"] for condition in conditions: ctype = condition.get("type") all_condition_types.add(ctype) if condition.get("status") == "True": got_true_conditions.add(ctype) # When TokenizerReady is present, it must also be True if "TokenizerReady" in all_condition_types: expected_true_conditions.add("TokenizerReady") missing_conditions = expected_true_conditions - got_true_conditions if missing_conditions: raise AssertionError( f"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}" ) return True > return wait_for(assert_llm_isvc_ready, timeout=timeout_seconds, interval=1.0) llmisvc/test_llm_inference_service.py:1376: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ assertion_fn = <function wait_for_llm_isvc_ready.<locals>.assert_llm_isvc_ready at 0x7fc258c50c20> timeout = 900, interval = 1.0 def wait_for( assertion_fn: Callable[[], Any], timeout: float = 5.0, interval: float = 0.1 ) -> Any: """Wait for the assertion to succeed within timeout.""" deadline = time.time() + timeout last_msg = None while True: try: > return assertion_fn() llmisvc/test_llm_inference_service.py:1387: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ def assert_llm_isvc_ready(): out = get_llmisvc( kserve_client, given.metadata.name, given.metadata.namespace, given.api_version.split("/")[1], ) if "status" not in out: raise AssertionError("No status found in LLM inference service") status = out["status"] if "conditions" not in status: raise AssertionError("No conditions found in status") expected_true_conditions = {"Ready", "WorkloadsReady", "RouterReady"} got_true_conditions = set() all_condition_types = set() conditions = status["conditions"] for condition in conditions: ctype = condition.get("type") all_condition_types.add(ctype) if condition.get("status") == "True": got_true_conditions.add(ctype) # When TokenizerReady is present, it must also be True if "TokenizerReady" in all_condition_types: expected_true_conditions.add("TokenizerReady") missing_conditions = expected_true_conditions - got_true_conditions if missing_conditions: > raise AssertionError( f"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}" ) E AssertionError: Missing true conditions: {'WorkloadsReady', 'Ready'}, expected {'WorkloadsReady', 'RouterReady', 'Ready'}, got [{'lastTransitionTime': '2026-07-28T17:31:04Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-28T17:31:04Z', 'severity': 'Info', 'status': 'True', 'type': 'InferencePoolReady'}, {'lastTransitionTime': '2026-07-28T17:31:04Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'severity': 'Info', 'status': 'False', 'type': 'MainWorkloadReady'}, {'lastTransitionTime': '2026-07-28T17:30:45Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-28T17:31:04Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-28T17:31:17Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-28T17:31:17Z', 'severity': 'Info', 'status': 'True', 'type': 'SchedulerWorkloadReady'}, {'lastTransitionTime': '2026-07-28T17:31:04Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'WorkloadsReady'}] llmisvc/test_llm_inference_service.py:1371: AssertionErrortest_case = TestCase(base_refs=['router-managed', 'workload-llmd-simulator-no-replicas', 'prometheus-scrape', 'scaling-hpa'], prom... {'name': 'scaling-hpa-autoscale-hpa-deplo-347a3180'}]}, 'status': None}, model_name='facebook/opt-125m') @pytest.mark.autoscaling_hpa @pytest.mark.parametrize( "test_case", [ pytest.param( TestCase( base_refs=[ "router-managed", "workload-llmd-simulator-no-replicas", "prometheus-scrape", "scaling-hpa", ], prompt="KServe is a", service_name="autoscale-hpa-deploy", ), marks=[ pytest.mark.cluster_cpu, pytest.mark.cluster_single_node, pytest.mark.llmd_simulator, ], ), ], indirect=["test_case"], ids=generate_test_id, ) @log_execution def test_llm_autoscaling_hpa_deployment(test_case: TestCase): """HPA + Deployment: HPA exists with WVA annotations; pods scale up under load.""" inject_k8s_proxy() kserve_client = _new_kserve_client() service_name = test_case.llm_service.metadata.name ns = test_case.namespace try: > _create_and_wait(kserve_client, test_case) llmisvc/test_llm_autoscaling_wva.py:542: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ kserve_client = <kserve.api.kserve_client.KServeClient object at 0x7fc258182f90> test_case = TestCase(base_refs=['router-managed', 'workload-llmd-simulator-no-replicas', 'prometheus-scrape', 'scaling-hpa'], prom... {'name': 'scaling-hpa-autoscale-hpa-deplo-347a3180'}]}, 'status': None}, model_name='facebook/opt-125m') def _create_and_wait(kserve_client, test_case): """Create LLMISVC and wait for it to be ready.""" create_llmisvc(kserve_client, test_case.llm_service) > wait_for_llm_isvc_ready( kserve_client, test_case.llm_service, test_case.wait_timeout ) llmisvc/test_llm_autoscaling_wva.py:482: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ args = (<kserve.api.kserve_client.KServeClient object at 0x7fc258182f90>, {'api_version': 'serving.kserve.io/v1alpha1', 'kin...e-hpa-4c186bcf'}, {'name': 'scaling-hpa-autoscale-hpa-deplo-347a3180'}]}, 'status': None}, 900) kwargs = {}, func_name = 'wait_for_llm_isvc_ready' timestamp_start = '2026-07-28T17:48:05.284533', start_time = 1785260885.2849283 duration = 900.7014257907867, timestamp_end = '2026-07-28T18:03:05.986374' @functools.wraps(func) def wrapper(*args, **kwargs): func_name = func.__name__ timestamp_start = datetime.now().isoformat() logger.info( f"[{func_name}] [{timestamp_start}] start - args={args}, kwargs={kwargs}" ) start_time = time.time() try: > result = func(*args, **kwargs) llmisvc/logging.py:40: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ kserve_client = <kserve.api.kserve_client.KServeClient object at 0x7fc258182f90> given = {'api_version': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceService', 'metadata': {'annotations': {'security....toscale-hpa-4c186bcf'}, {'name': 'scaling-hpa-autoscale-hpa-deplo-347a3180'}]}, 'status': None} timeout_seconds = 900 @log_execution def wait_for_llm_isvc_ready( kserve_client: KServeClient, given: V1alpha1LLMInferenceService, timeout_seconds: int = 900, ) -> str: def assert_llm_isvc_ready(): out = get_llmisvc( kserve_client, given.metadata.name, given.metadata.namespace, given.api_version.split("/")[1], ) if "status" not in out: raise AssertionError("No status found in LLM inference service") status = out["status"] if "conditions" not in status: raise AssertionError("No conditions found in status") expected_true_conditions = {"Ready", "WorkloadsReady", "RouterReady"} got_true_conditions = set() all_condition_types = set() conditions = status["conditions"] for condition in conditions: ctype = condition.get("type") all_condition_types.add(ctype) if condition.get("status") == "True": got_true_conditions.add(ctype) # When TokenizerReady is present, it must also be True if "TokenizerReady" in all_condition_types: expected_true_conditions.add("TokenizerReady") missing_conditions = expected_true_conditions - got_true_conditions if missing_conditions: raise AssertionError( f"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}" ) return True > return wait_for(assert_llm_isvc_ready, timeout=timeout_seconds, interval=1.0) llmisvc/test_llm_inference_service.py:1376: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ assertion_fn = <function wait_for_llm_isvc_ready.<locals>.assert_llm_isvc_ready at 0x7fc253f751c0> timeout = 900, interval = 1.0 def wait_for( assertion_fn: Callable[[], Any], timeout: float = 5.0, interval: float = 0.1 ) -> Any: """Wait for the assertion to succeed within timeout.""" deadline = time.time() + timeout last_msg = None while True: try: > return assertion_fn() llmisvc/test_llm_inference_service.py:1387: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ def assert_llm_isvc_ready(): out = get_llmisvc( kserve_client, given.metadata.name, given.metadata.namespace, given.api_version.split("/")[1], ) if "status" not in out: raise AssertionError("No status found in LLM inference service") status = out["status"] if "conditions" not in status: raise AssertionError("No conditions found in status") expected_true_conditions = {"Ready", "WorkloadsReady", "RouterReady"} got_true_conditions = set() all_condition_types = set() conditions = status["conditions"] for condition in conditions: ctype = condition.get("type") all_condition_types.add(ctype) if condition.get("status") == "True": got_true_conditions.add(ctype) # When TokenizerReady is present, it must also be True if "TokenizerReady" in all_condition_types: expected_true_conditions.add("TokenizerReady") missing_conditions = expected_true_conditions - got_true_conditions if missing_conditions: > raise AssertionError( f"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}" ) E AssertionError: Missing true conditions: {'WorkloadsReady', 'Ready'}, expected {'WorkloadsReady', 'RouterReady', 'Ready'}, got [{'lastTransitionTime': '2026-07-28T17:48:45Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-28T17:48:45Z', 'severity': 'Info', 'status': 'True', 'type': 'InferencePoolReady'}, {'lastTransitionTime': '2026-07-28T17:48:45Z', 'severity': 'Info', 'status': 'True', 'type': 'MainWorkloadReady'}, {'lastTransitionTime': '2026-07-28T17:48:18Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-28T17:48:50Z', 'message': 'the HPA was unable to compute the replica count: unable to get external metric e2e-test-llm-autoscaling-hpa-deployment-9ba6f3f4/wva_desired_replicas/&LabelSelector{MatchLabels:map[string]string{variant_name: autoscale-hpa-deploy-kserve-hpa,},MatchExpressions:[]LabelSelectorRequirement{},}: unable to fetch metrics from external metrics API: scaledObject name is not specified', 'reason': 'FailedGetExternalMetric', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-28T17:48:50Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-28T17:48:45Z', 'message': 'the HPA was unable to compute the replica count: unable to get external metric e2e-test-llm-autoscaling-hpa-deployment-9ba6f3f4/wva_desired_replicas/&LabelSelector{MatchLabels:map[string]string{variant_name: autoscale-hpa-deploy-kserve-hpa,},MatchExpressions:[]LabelSelectorRequirement{},}: unable to fetch metrics from external metrics API: scaledObject name is not specified', 'reason': 'FailedGetExternalMetric', 'severity': 'Info', 'status': 'False', 'type': 'ScalingReady'}, {'lastTransitionTime': '2026-07-28T17:48:50Z', 'severity': 'Info', 'status': 'True', 'type': 'SchedulerWorkloadReady'}, {'lastTransitionTime': '2026-07-28T17:48:45Z', 'message': 'the HPA was unable to compute the replica count: unable to get external metric e2e-test-llm-autoscaling-hpa-deployment-9ba6f3f4/wva_desired_replicas/&LabelSelector{MatchLabels:map[string]string{variant_name: autoscale-hpa-deploy-kserve-hpa,},MatchExpressions:[]LabelSelectorRequirement{},}: unable to fetch metrics from external metrics API: scaledObject name is not specified', 'reason': 'FailedGetExternalMetric', 'status': 'False', 'type': 'WorkloadsReady'}] llmisvc/test_llm_inference_service.py:1371: AssertionErrortest_case = TestCase(base_refs=['router-managed', 'workload-llmd-simulator-no-replicas', 'prometheus-scrape', 'scaling-keda'], pro... {'name': 'scaling-keda-autoscale-keda-dep-1ac84077'}]}, 'status': None}, model_name='facebook/opt-125m') @pytest.mark.autoscaling_keda @pytest.mark.parametrize( "test_case", [ pytest.param( TestCase( base_refs=[ "router-managed", "workload-llmd-simulator-no-replicas", "prometheus-scrape", "scaling-keda", ], prompt="KServe is a", service_name="autoscale-keda-deploy", ), marks=[ pytest.mark.cluster_cpu, pytest.mark.cluster_single_node, pytest.mark.llmd_simulator, ], ), ], indirect=["test_case"], ids=generate_test_id, ) @log_execution def test_llm_autoscaling_keda_deployment(test_case: TestCase): """KEDA + Deployment: ScaledObject exists with WVA annotations; no HPA; pods scale up under load.""" inject_k8s_proxy() kserve_client = _new_kserve_client() service_name = test_case.llm_service.metadata.name ns = test_case.namespace try: > _create_and_wait(kserve_client, test_case) llmisvc/test_llm_autoscaling_wva.py:606: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ kserve_client = <kserve.api.kserve_client.KServeClient object at 0x7fc258b2c6d0> test_case = TestCase(base_refs=['router-managed', 'workload-llmd-simulator-no-replicas', 'prometheus-scrape', 'scaling-keda'], pro... {'name': 'scaling-keda-autoscale-keda-dep-1ac84077'}]}, 'status': None}, model_name='facebook/opt-125m') def _create_and_wait(kserve_client, test_case): """Create LLMISVC and wait for it to be ready.""" create_llmisvc(kserve_client, test_case.llm_service) > wait_for_llm_isvc_ready( kserve_client, test_case.llm_service, test_case.wait_timeout ) llmisvc/test_llm_autoscaling_wva.py:482: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ args = (<kserve.api.kserve_client.KServeClient object at 0x7fc258b2c6d0>, {'api_version': 'serving.kserve.io/v1alpha1', 'kin...e-ked-101f2a9d'}, {'name': 'scaling-keda-autoscale-keda-dep-1ac84077'}]}, 'status': None}, 900) kwargs = {}, func_name = 'wait_for_llm_isvc_ready' timestamp_start = '2026-07-28T18:03:27.005951', start_time = 1785261807.0062244 duration = 900.011447429657, timestamp_end = '2026-07-28T18:18:27.017674' @functools.wraps(func) def wrapper(*args, **kwargs): func_name = func.__name__ timestamp_start = datetime.now().isoformat() logger.info( f"[{func_name}] [{timestamp_start}] start - args={args}, kwargs={kwargs}" ) start_time = time.time() try: > result = func(*args, **kwargs) llmisvc/logging.py:40: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ kserve_client = <kserve.api.kserve_client.KServeClient object at 0x7fc258b2c6d0> given = {'api_version': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceService', 'metadata': {'annotations': {'security....toscale-ked-101f2a9d'}, {'name': 'scaling-keda-autoscale-keda-dep-1ac84077'}]}, 'status': None} timeout_seconds = 900 @log_execution def wait_for_llm_isvc_ready( kserve_client: KServeClient, given: V1alpha1LLMInferenceService, timeout_seconds: int = 900, ) -> str: def assert_llm_isvc_ready(): out = get_llmisvc( kserve_client, given.metadata.name, given.metadata.namespace, given.api_version.split("/")[1], ) if "status" not in out: raise AssertionError("No status found in LLM inference service") status = out["status"] if "conditions" not in status: raise AssertionError("No conditions found in status") expected_true_conditions = {"Ready", "WorkloadsReady", "RouterReady"} got_true_conditions = set() all_condition_types = set() conditions = status["conditions"] for condition in conditions: ctype = condition.get("type") all_condition_types.add(ctype) if condition.get("status") == "True": got_true_conditions.add(ctype) # When TokenizerReady is present, it must also be True if "TokenizerReady" in all_condition_types: expected_true_conditions.add("TokenizerReady") missing_conditions = expected_true_conditions - got_true_conditions if missing_conditions: raise AssertionError( f"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}" ) return True > return wait_for(assert_llm_isvc_ready, timeout=timeout_seconds, interval=1.0) llmisvc/test_llm_inference_service.py:1376: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ assertion_fn = <function wait_for_llm_isvc_ready.<locals>.assert_llm_isvc_ready at 0x7fc253f75a80> timeout = 900, interval = 1.0 def wait_for( assertion_fn: Callable[[], Any], timeout: float = 5.0, interval: float = 0.1 ) -> Any: """Wait for the assertion to succeed within timeout.""" deadline = time.time() + timeout last_msg = None while True: try: > return assertion_fn() llmisvc/test_llm_inference_service.py:1387: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ def assert_llm_isvc_ready(): out = get_llmisvc( kserve_client, given.metadata.name, given.metadata.namespace, given.api_version.split("/")[1], ) if "status" not in out: raise AssertionError("No status found in LLM inference service") status = out["status"] if "conditions" not in status: raise AssertionError("No conditions found in status") expected_true_conditions = {"Ready", "WorkloadsReady", "RouterReady"} got_true_conditions = set() all_condition_types = set() conditions = status["conditions"] for condition in conditions: ctype = condition.get("type") all_condition_types.add(ctype) if condition.get("status") == "True": got_true_conditions.add(ctype) # When TokenizerReady is present, it must also be True if "TokenizerReady" in all_condition_types: expected_true_conditions.add("TokenizerReady") missing_conditions = expected_true_conditions - got_true_conditions if missing_conditions: > raise AssertionError( f"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}" ) E AssertionError: Missing true conditions: {'WorkloadsReady', 'Ready'}, expected {'WorkloadsReady', 'RouterReady', 'Ready'}, got [{'lastTransitionTime': '2026-07-28T18:03:55Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-28T18:03:55Z', 'severity': 'Info', 'status': 'True', 'type': 'InferencePoolReady'}, {'lastTransitionTime': '2026-07-28T18:03:55Z', 'severity': 'Info', 'status': 'True', 'type': 'MainWorkloadReady'}, {'lastTransitionTime': '2026-07-28T18:03:36Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-28T18:04:20Z', 'message': 'failed to ensure HPA is correctly created for ScaledObject: error parsing prometheus metadata: error parsing prometheus metadata: bearer token=<empty> is required when bearer auth is enabled', 'reason': 'ScaledObjectCheckFailed', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-28T18:04:20Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-28T18:03:55Z', 'message': 'failed to ensure HPA is correctly created for ScaledObject: error parsing prometheus metadata: error parsing prometheus metadata: bearer token=<empty> is required when bearer auth is enabled', 'reason': 'ScaledObjectCheckFailed', 'severity': 'Info', 'status': 'False', 'type': 'ScalingReady'}, {'lastTransitionTime': '2026-07-28T18:04:20Z', 'severity': 'Info', 'status': 'True', 'type': 'SchedulerWorkloadReady'}, {'lastTransitionTime': '2026-07-28T18:03:55Z', 'message': 'failed to ensure HPA is correctly created for ScaledObject: error parsing prometheus metadata: error parsing prometheus metadata: bearer token=<empty> is required when bearer auth is enabled', 'reason': 'ScaledObjectCheckFailed', 'status': 'False', 'type': 'WorkloadsReady'}] llmisvc/test_llm_inference_service.py:1371: AssertionErrortest_case = TestCase(base_refs=['router-managed', 'workload-llmd-simulator-lws', 'prometheus-scrape', 'scaling-hpa'], prompt='KSer... {'name': 'scaling-hpa-autoscale-hpa-lws-b344a3ff'}]}, 'status': None}, model_name='facebook/opt-125m') @pytest.mark.autoscaling_hpa @pytest.mark.parametrize( "test_case", [ pytest.param( TestCase( base_refs=[ "router-managed", "workload-llmd-simulator-lws", "prometheus-scrape", "scaling-hpa", ], prompt="KServe is a", service_name="autoscale-hpa-lws", ), marks=[ pytest.mark.cluster_cpu, pytest.mark.cluster_multi_node, pytest.mark.llmd_simulator, ], ), ], indirect=["test_case"], ids=generate_test_id, ) @log_execution def test_llm_autoscaling_hpa_lws(test_case: TestCase): """HPA + LWS: HPA exists with WVA annotations; pods scale under load.""" inject_k8s_proxy() kserve_client = _new_kserve_client() service_name = test_case.llm_service.metadata.name ns = test_case.namespace try: > _create_and_wait(kserve_client, test_case) llmisvc/test_llm_autoscaling_wva.py:670: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ kserve_client = <kserve.api.kserve_client.KServeClient object at 0x7fc259f65a50> test_case = TestCase(base_refs=['router-managed', 'workload-llmd-simulator-lws', 'prometheus-scrape', 'scaling-hpa'], prompt='KSer... {'name': 'scaling-hpa-autoscale-hpa-lws-b344a3ff'}]}, 'status': None}, model_name='facebook/opt-125m') def _create_and_wait(kserve_client, test_case): """Create LLMISVC and wait for it to be ready.""" create_llmisvc(kserve_client, test_case.llm_service) > wait_for_llm_isvc_ready( kserve_client, test_case.llm_service, test_case.wait_timeout ) llmisvc/test_llm_autoscaling_wva.py:482: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ args = (<kserve.api.kserve_client.KServeClient object at 0x7fc259f65a50>, {'api_version': 'serving.kserve.io/v1alpha1', 'kin...ale-hpa-b29acdba'}, {'name': 'scaling-hpa-autoscale-hpa-lws-b344a3ff'}]}, 'status': None}, 900) kwargs = {}, func_name = 'wait_for_llm_isvc_ready' timestamp_start = '2026-07-28T18:18:48.789897', start_time = 1785262728.7902377 duration = 900.5734322071075, timestamp_end = '2026-07-28T18:33:49.363672' @functools.wraps(func) def wrapper(*args, **kwargs): func_name = func.__name__ timestamp_start = datetime.now().isoformat() logger.info( f"[{func_name}] [{timestamp_start}] start - args={args}, kwargs={kwargs}" ) start_time = time.time() try: > result = func(*args, **kwargs) llmisvc/logging.py:40: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ kserve_client = <kserve.api.kserve_client.KServeClient object at 0x7fc259f65a50> given = {'api_version': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceService', 'metadata': {'annotations': {'security....autoscale-hpa-b29acdba'}, {'name': 'scaling-hpa-autoscale-hpa-lws-b344a3ff'}]}, 'status': None} timeout_seconds = 900 @log_execution def wait_for_llm_isvc_ready( kserve_client: KServeClient, given: V1alpha1LLMInferenceService, timeout_seconds: int = 900, ) -> str: def assert_llm_isvc_ready(): out = get_llmisvc( kserve_client, given.metadata.name, given.metadata.namespace, given.api_version.split("/")[1], ) if "status" not in out: raise AssertionError("No status found in LLM inference service") status = out["status"] if "conditions" not in status: raise AssertionError("No conditions found in status") expected_true_conditions = {"Ready", "WorkloadsReady", "RouterReady"} got_true_conditions = set() all_condition_types = set() conditions = status["conditions"] for condition in conditions: ctype = condition.get("type") all_condition_types.add(ctype) if condition.get("status") == "True": got_true_conditions.add(ctype) # When TokenizerReady is present, it must also be True if "TokenizerReady" in all_condition_types: expected_true_conditions.add("TokenizerReady") missing_conditions = expected_true_conditions - got_true_conditions if missing_conditions: raise AssertionError( f"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}" ) return True > return wait_for(assert_llm_isvc_ready, timeout=timeout_seconds, interval=1.0) llmisvc/test_llm_inference_service.py:1376: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ assertion_fn = <function wait_for_llm_isvc_ready.<locals>.assert_llm_isvc_ready at 0x7fc253f76700> timeout = 900, interval = 1.0 def wait_for( assertion_fn: Callable[[], Any], timeout: float = 5.0, interval: float = 0.1 ) -> Any: """Wait for the assertion to succeed within timeout.""" deadline = time.time() + timeout last_msg = None while True: try: > return assertion_fn() llmisvc/test_llm_inference_service.py:1387: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ def assert_llm_isvc_ready(): out = get_llmisvc( kserve_client, given.metadata.name, given.metadata.namespace, given.api_version.split("/")[1], ) if "status" not in out: raise AssertionError("No status found in LLM inference service") status = out["status"] if "conditions" not in status: raise AssertionError("No conditions found in status") expected_true_conditions = {"Ready", "WorkloadsReady", "RouterReady"} got_true_conditions = set() all_condition_types = set() conditions = status["conditions"] for condition in conditions: ctype = condition.get("type") all_condition_types.add(ctype) if condition.get("status") == "True": got_true_conditions.add(ctype) # When TokenizerReady is present, it must also be True if "TokenizerReady" in all_condition_types: expected_true_conditions.add("TokenizerReady") missing_conditions = expected_true_conditions - got_true_conditions if missing_conditions: > raise AssertionError( f"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}" ) E AssertionError: Missing true conditions: {'WorkloadsReady', 'Ready'}, expected {'WorkloadsReady', 'RouterReady', 'Ready'}, got [{'lastTransitionTime': '2026-07-28T18:19:20Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-28T18:19:20Z', 'severity': 'Info', 'status': 'True', 'type': 'InferencePoolReady'}, {'lastTransitionTime': '2026-07-28T18:19:07Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-28T18:19:37Z', 'message': 'the HPA was unable to compute the replica count: unable to get external metric e2e-test-llm-autoscaling-hpa-lws-d4cbcfd2/wva_desired_replicas/&LabelSelector{MatchLabels:map[string]string{variant_name: autoscale-hpa-lws-kserve-hpa,},MatchExpressions:[]LabelSelectorRequirement{},}: unable to fetch metrics from external metrics API: scaledObject name is not specified', 'reason': 'FailedGetExternalMetric', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-28T18:19:37Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-28T18:19:21Z', 'message': 'the HPA was unable to compute the replica count: unable to get external metric e2e-test-llm-autoscaling-hpa-lws-d4cbcfd2/wva_desired_replicas/&LabelSelector{MatchLabels:map[string]string{variant_name: autoscale-hpa-lws-kserve-hpa,},MatchExpressions:[]LabelSelectorRequirement{},}: unable to fetch metrics from external metrics API: scaledObject name is not specified', 'reason': 'FailedGetExternalMetric', 'severity': 'Info', 'status': 'False', 'type': 'ScalingReady'}, {'lastTransitionTime': '2026-07-28T18:19:37Z', 'severity': 'Info', 'status': 'True', 'type': 'SchedulerWorkloadReady'}, {'lastTransitionTime': '2026-07-28T18:19:20Z', 'severity': 'Info', 'status': 'True', 'type': 'WorkerWorkloadReady'}, {'lastTransitionTime': '2026-07-28T18:19:21Z', 'message': 'the HPA was unable to compute the replica count: unable to get external metric e2e-test-llm-autoscaling-hpa-lws-d4cbcfd2/wva_desired_replicas/&LabelSelector{MatchLabels:map[string]string{variant_name: autoscale-hpa-lws-kserve-hpa,},MatchExpressions:[]LabelSelectorRequirement{},}: unable to fetch metrics from external metrics API: scaledObject name is not specified', 'reason': 'FailedGetExternalMetric', 'status': 'False', 'type': 'WorkloadsReady'}] llmisvc/test_llm_inference_service.py:1371: AssertionErrortest_case = TestCase(base_refs=['router-managed', 'workload-llmd-simulator-lws', 'prometheus-scrape', 'scaling-keda'], prompt='KSe... {'name': 'scaling-keda-autoscale-keda-lws-1337f511'}]}, 'status': None}, model_name='facebook/opt-125m') @pytest.mark.autoscaling_keda @pytest.mark.parametrize( "test_case", [ pytest.param( TestCase( base_refs=[ "router-managed", "workload-llmd-simulator-lws", "prometheus-scrape", "scaling-keda", ], prompt="KServe is a", service_name="autoscale-keda-lws", ), marks=[ pytest.mark.cluster_cpu, pytest.mark.cluster_multi_node, pytest.mark.llmd_simulator, ], ), ], indirect=["test_case"], ids=generate_test_id, ) @log_execution def test_llm_autoscaling_keda_lws(test_case: TestCase): """KEDA + LWS: ScaledObject exists with WVA annotations; pods scale under load.""" inject_k8s_proxy() kserve_client = _new_kserve_client() service_name = test_case.llm_service.metadata.name ns = test_case.namespace try: > _create_and_wait(kserve_client, test_case) llmisvc/test_llm_autoscaling_wva.py:728: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ kserve_client = <kserve.api.kserve_client.KServeClient object at 0x7fc258161310> test_case = TestCase(base_refs=['router-managed', 'workload-llmd-simulator-lws', 'prometheus-scrape', 'scaling-keda'], prompt='KSe... {'name': 'scaling-keda-autoscale-keda-lws-1337f511'}]}, 'status': None}, model_name='facebook/opt-125m') def _create_and_wait(kserve_client, test_case): """Create LLMISVC and wait for it to be ready.""" create_llmisvc(kserve_client, test_case.llm_service) > wait_for_llm_isvc_ready( kserve_client, test_case.llm_service, test_case.wait_timeout ) llmisvc/test_llm_autoscaling_wva.py:482: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ args = (<kserve.api.kserve_client.KServeClient object at 0x7fc258161310>, {'api_version': 'serving.kserve.io/v1alpha1', 'kin...e-ked-231d315d'}, {'name': 'scaling-keda-autoscale-keda-lws-1337f511'}]}, 'status': None}, 900) kwargs = {}, func_name = 'wait_for_llm_isvc_ready' timestamp_start = '2026-07-28T18:34:15.068504', start_time = 1785263655.0688283 duration = 900.7306854724884, timestamp_end = '2026-07-28T18:49:15.799516' @functools.wraps(func) def wrapper(*args, **kwargs): func_name = func.__name__ timestamp_start = datetime.now().isoformat() logger.info( f"[{func_name}] [{timestamp_start}] start - args={args}, kwargs={kwargs}" ) start_time = time.time() try: > result = func(*args, **kwargs) llmisvc/logging.py:40: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ kserve_client = <kserve.api.kserve_client.KServeClient object at 0x7fc258161310> given = {'api_version': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceService', 'metadata': {'annotations': {'security....toscale-ked-231d315d'}, {'name': 'scaling-keda-autoscale-keda-lws-1337f511'}]}, 'status': None} timeout_seconds = 900 @log_execution def wait_for_llm_isvc_ready( kserve_client: KServeClient, given: V1alpha1LLMInferenceService, timeout_seconds: int = 900, ) -> str: def assert_llm_isvc_ready(): out = get_llmisvc( kserve_client, given.metadata.name, given.metadata.namespace, given.api_version.split("/")[1], ) if "status" not in out: raise AssertionError("No status found in LLM inference service") status = out["status"] if "conditions" not in status: raise AssertionError("No conditions found in status") expected_true_conditions = {"Ready", "WorkloadsReady", "RouterReady"} got_true_conditions = set() all_condition_types = set() conditions = status["conditions"] for condition in conditions: ctype = condition.get("type") all_condition_types.add(ctype) if condition.get("status") == "True": got_true_conditions.add(ctype) # When TokenizerReady is present, it must also be True if "TokenizerReady" in all_condition_types: expected_true_conditions.add("TokenizerReady") missing_conditions = expected_true_conditions - got_true_conditions if missing_conditions: raise AssertionError( f"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}" ) return True > return wait_for(assert_llm_isvc_ready, timeout=timeout_seconds, interval=1.0) llmisvc/test_llm_inference_service.py:1376: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ assertion_fn = <function wait_for_llm_isvc_ready.<locals>.assert_llm_isvc_ready at 0x7fc253f774c0> timeout = 900, interval = 1.0 def wait_for( assertion_fn: Callable[[], Any], timeout: float = 5.0, interval: float = 0.1 ) -> Any: """Wait for the assertion to succeed within timeout.""" deadline = time.time() + timeout last_msg = None while True: try: > return assertion_fn() llmisvc/test_llm_inference_service.py:1387: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ def assert_llm_isvc_ready(): out = get_llmisvc( kserve_client, given.metadata.name, given.metadata.namespace, given.api_version.split("/")[1], ) if "status" not in out: raise AssertionError("No status found in LLM inference service") status = out["status"] if "conditions" not in status: raise AssertionError("No conditions found in status") expected_true_conditions = {"Ready", "WorkloadsReady", "RouterReady"} got_true_conditions = set() all_condition_types = set() conditions = status["conditions"] for condition in conditions: ctype = condition.get("type") all_condition_types.add(ctype) if condition.get("status") == "True": got_true_conditions.add(ctype) # When TokenizerReady is present, it must also be True if "TokenizerReady" in all_condition_types: expected_true_conditions.add("TokenizerReady") missing_conditions = expected_true_conditions - got_true_conditions if missing_conditions: > raise AssertionError( f"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}" ) E AssertionError: Missing true conditions: {'WorkloadsReady', 'Ready'}, expected {'WorkloadsReady', 'RouterReady', 'Ready'}, got [{'lastTransitionTime': '2026-07-28T18:34:31Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-28T18:34:31Z', 'severity': 'Info', 'status': 'True', 'type': 'InferencePoolReady'}, {'lastTransitionTime': '2026-07-28T18:34:26Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-28T18:34:57Z', 'message': 'failed to ensure HPA is correctly created for ScaledObject: error parsing prometheus metadata: error parsing prometheus metadata: bearer token=<empty> is required when bearer auth is enabled', 'reason': 'ScaledObjectCheckFailed', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-28T18:34:57Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-28T18:34:31Z', 'message': 'failed to ensure HPA is correctly created for ScaledObject: error parsing prometheus metadata: error parsing prometheus metadata: bearer token=<empty> is required when bearer auth is enabled', 'reason': 'ScaledObjectCheckFailed', 'severity': 'Info', 'status': 'False', 'type': 'ScalingReady'}, {'lastTransitionTime': '2026-07-28T18:34:57Z', 'severity': 'Info', 'status': 'True', 'type': 'SchedulerWorkloadReady'}, {'lastTransitionTime': '2026-07-28T18:34:35Z', 'severity': 'Info', 'status': 'True', 'type': 'WorkerWorkloadReady'}, {'lastTransitionTime': '2026-07-28T18:34:35Z', 'message': 'failed to ensure HPA is correctly created for ScaledObject: error parsing prometheus metadata: error parsing prometheus metadata: bearer token=<empty> is required when bearer auth is enabled', 'reason': 'ScaledObjectCheckFailed', 'status': 'False', 'type': 'WorkloadsReady'}] llmisvc/test_llm_inference_service.py:1371: AssertionError