{"created": 1783980686.8264694, "duration": 5992.918427228928, "exitcode": 2, "root": "/workspace/source/test/e2e", "environment": {}, "summary": {"passed": 25, "failed": 9, "error": 1, "total": 35, "collected": 42}, "collectors": [{"nodeid": "explainer/test_art_explainer.py", "outcome": "skipped", "result": [], "longrepr": "('/workspace/source/test/e2e/explainer/test_art_explainer.py', 38, 'Skipped: ODH does not support art explainer at the moment')"}, {"nodeid": "predictor/test_grpc.py", "outcome": "skipped", "result": [], "longrepr": "('/workspace/source/test/e2e/predictor/test_grpc.py', 35, 'Skipped: Not testable in ODH at the moment')"}, {"nodeid": "predictor/test_torchserve.py", "outcome": "skipped", "result": [], "longrepr": "('/workspace/source/test/e2e/predictor/test_torchserve.py', 34, 'Skipped: ODH does not support torchserve at the moment')"}], "tests": [{"nodeid": "llmisvc/test_gateway_section_name.py::test_gateway_section_name_propagation[cluster_single_node-cluster_cpu-with-section-name]", "lineno": 131, "outcome": "passed", "keywords": ["test_gateway_section_name_propagation[with-section-name]", "parametrize", "llmd_simulator", "cluster_single_node", "cluster_cpu", "llminferenceservice", "pytestmark", "with-section-name", "test_gateway_section_name.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.059462404999067076, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 8.084475783995003, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0006366980014718138, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-managed-scheduler-with-precise-prefix-cache-inline-config-workload-llmd-simulator-kvcache]", "lineno": 244, "outcome": "passed", "keywords": ["test_llm_inference_service[router-managed-scheduler-with-precise-prefix-cache-inline-config-workload-llmd-simulator-kvcache]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "llmd_simulator", "__wrapped__", "pytestmark", "router-managed-scheduler-with-precise-prefix-cache-inline-config-workload-llmd-simulator-kvcache", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 6.398342167994997, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 101.89368797199859, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0026791919954121113, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_gateway_section_name.py::test_gateway_section_name_propagation[cluster_single_node-cluster_cpu-without-section-name]", "lineno": 131, "outcome": "passed", "keywords": ["test_gateway_section_name_propagation[without-section-name]", "parametrize", "llmd_simulator", "cluster_single_node", "cluster_cpu", "llminferenceservice", "pytestmark", "without-section-name", "test_gateway_section_name.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.0343972139962716, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 16.657245200003672, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0006957619989407249, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_auth.py::test_llm_auth_enabled_requires_token[cluster_cpu-cluster_single_node-auth-enabled-default]", "lineno": 221, "outcome": "failed", "keywords": ["test_llm_auth_enabled_requires_token[auth-enabled-default]", "parametrize", "auth", "llminferenceservice", "cluster_cpu", "cluster_single_node", "__wrapped__", "pytestmark", "auth-enabled-default", "test_llm_auth.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.9547909350003465, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 903.5097690689945, "outcome": "failed", "crash": {"path": "/workspace/source/test/e2e/llmisvc/test_llm_inference_service.py", "lineno": 1199, "message": "AssertionError: Missing true conditions: {'Ready', 'WorkloadsReady'}, expected {'Ready', 'RouterReady', 'WorkloadsReady'}, got [{'lastTransitionTime': '2026-07-13T20:32:33Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-13T20:32:33Z', 'severity': 'Info', 'status': 'True', 'type': 'InferencePoolReady'}, {'lastTransitionTime': '2026-07-13T20:32:33Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'severity': 'Info', 'status': 'False', 'type': 'MainWorkloadReady'}, {'lastTransitionTime': '2026-07-13T20:32:20Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-13T20:32:33Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-13T20:32:56Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-13T20:32:56Z', 'severity': 'Info', 'status': 'True', 'type': 'SchedulerWorkloadReady'}, {'lastTransitionTime': '2026-07-13T20:32:33Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'WorkloadsReady'}]"}, "traceback": [{"path": "llmisvc/test_llm_auth.py", "lineno": 275, "message": ""}, {"path": "llmisvc/logging.py", "lineno": 40, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1204, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1215, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1199, "message": "AssertionError"}], "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python\n\ntest_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-enabled-89f54b63'}]},\n 'status': None}, model_name='facebook/opt-125m')\n\n @pytest.mark.llminferenceservice\n @pytest.mark.auth\n @pytest.mark.parametrize(\n \"test_case\",\n [\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"auth-enabled-test\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n ],\n id=\"auth-enabled-default\",\n ),\n ],\n indirect=[\"test_case\"],\n ids=generate_test_id,\n )\n @log_execution\n def test_llm_auth_enabled_requires_token(test_case: TestCase): # noqa: F811\n \"\"\"\n Test that when auth is enabled (default):\n - Requests WITH valid token succeed\n - Requests WITHOUT token are rejected (401/403)\n \"\"\"\n inject_k8s_proxy()\n \n kserve_client = KServeClient(\n config_file=os.environ.get(\"KUBECONFIG\", \"~/.kube/config\"),\n client_configuration=client.Configuration(),\n )\n \n service_name = test_case.llm_service.metadata.name\n sa_name = f\"{service_name}-test-sa\"\n test_failed = False\n \n # Enable auth for this test\n if not test_case.llm_service.metadata.annotations:\n test_case.llm_service.metadata.annotations = {}\n test_case.llm_service.metadata.annotations[\n \"security.opendatahub.io/enable-auth\"\n ] = \"true\"\n \n try:\n # Create LLMInferenceService\n create_llmisvc(kserve_client, test_case.llm_service)\n> wait_for_llm_isvc_ready(\n kserve_client, test_case.llm_service, test_case.wait_timeout\n )\n\nllmisvc/test_llm_auth.py:275: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nargs = (, {'api_version': 'serving.kserve.io/v1alpha1',\n 'kin...enable-a18fd8e2'},\n {'name': 'model-fb-opt-125m-auth-enabled-89f54b63'}]},\n 'status': None}, 900)\nkwargs = {}, func_name = 'wait_for_llm_isvc_ready'\ntimestamp_start = '2026-07-13T20:32:02.791845', start_time = 1783974722.7922158\nduration = 900.6645452976227, timestamp_end = '2026-07-13T20:47:03.456764'\n\n @functools.wraps(func)\n def wrapper(*args, **kwargs):\n func_name = func.__name__\n \n timestamp_start = datetime.now().isoformat()\n logger.info(\n f\"[{func_name}] [{timestamp_start}] start - args={args}, kwargs={kwargs}\"\n )\n start_time = time.time()\n \n try:\n> result = func(*args, **kwargs)\n\nllmisvc/logging.py:40: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nkserve_client = \ngiven = {'api_version': 'serving.kserve.io/v1alpha1',\n 'kind': 'LLMInferenceService',\n 'metadata': {'annotations': {'security....-auth-enable-a18fd8e2'},\n {'name': 'model-fb-opt-125m-auth-enabled-89f54b63'}]},\n 'status': None}\ntimeout_seconds = 900\n\n @log_execution\n def wait_for_llm_isvc_ready(\n kserve_client: KServeClient,\n given: V1alpha1LLMInferenceService,\n timeout_seconds: int = 900,\n ) -> str:\n def assert_llm_isvc_ready():\n out = get_llmisvc(\n kserve_client,\n given.metadata.name,\n given.metadata.namespace,\n given.api_version.split(\"/\")[1],\n )\n \n if \"status\" not in out:\n raise AssertionError(\"No status found in LLM inference service\")\n \n status = out[\"status\"]\n if \"conditions\" not in status:\n raise AssertionError(\"No conditions found in status\")\n \n expected_true_conditions = {\"Ready\", \"WorkloadsReady\", \"RouterReady\"}\n got_true_conditions = set()\n \n conditions = status[\"conditions\"]\n \n for condition in conditions:\n if condition.get(\"status\") == \"True\":\n got_true_conditions.add(condition.get(\"type\"))\n \n missing_conditions = expected_true_conditions - got_true_conditions\n if missing_conditions:\n raise AssertionError(\n f\"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}\"\n )\n return True\n \n> return wait_for(assert_llm_isvc_ready, timeout=timeout_seconds, interval=1.0)\n\nllmisvc/test_llm_inference_service.py:1204: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nassertion_fn = .assert_llm_isvc_ready at 0x7fb7b6036ca0>\ntimeout = 900, interval = 1.0\n\n def wait_for(\n assertion_fn: Callable[[], Any], timeout: float = 5.0, interval: float = 0.1\n ) -> Any:\n \"\"\"Wait for the assertion to succeed within timeout.\"\"\"\n deadline = time.time() + timeout\n last_msg = None\n while True:\n try:\n> return assertion_fn()\n\nllmisvc/test_llm_inference_service.py:1215: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\n def assert_llm_isvc_ready():\n out = get_llmisvc(\n kserve_client,\n given.metadata.name,\n given.metadata.namespace,\n given.api_version.split(\"/\")[1],\n )\n \n if \"status\" not in out:\n raise AssertionError(\"No status found in LLM inference service\")\n \n status = out[\"status\"]\n if \"conditions\" not in status:\n raise AssertionError(\"No conditions found in status\")\n \n expected_true_conditions = {\"Ready\", \"WorkloadsReady\", \"RouterReady\"}\n got_true_conditions = set()\n \n conditions = status[\"conditions\"]\n \n for condition in conditions:\n if condition.get(\"status\") == \"True\":\n got_true_conditions.add(condition.get(\"type\"))\n \n missing_conditions = expected_true_conditions - got_true_conditions\n if missing_conditions:\n> raise AssertionError(\n f\"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}\"\n )\nE AssertionError: Missing true conditions: {'Ready', 'WorkloadsReady'}, expected {'Ready', 'RouterReady', 'WorkloadsReady'}, got [{'lastTransitionTime': '2026-07-13T20:32:33Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-13T20:32:33Z', 'severity': 'Info', 'status': 'True', 'type': 'InferencePoolReady'}, {'lastTransitionTime': '2026-07-13T20:32:33Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'severity': 'Info', 'status': 'False', 'type': 'MainWorkloadReady'}, {'lastTransitionTime': '2026-07-13T20:32:20Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-13T20:32:33Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-13T20:32:56Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-13T20:32:56Z', 'severity': 'Info', 'status': 'True', 'type': 'SchedulerWorkloadReady'}, {'lastTransitionTime': '2026-07-13T20:32:33Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'WorkloadsReady'}]\n\nllmisvc/test_llm_inference_service.py:1199: AssertionError"}, "teardown": {"duration": 0.0029280969974934123, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-managed-workload-llmd-simulator0]", "lineno": 244, "outcome": "passed", "keywords": ["test_llm_inference_service[router-managed-workload-llmd-simulator0]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "llmd_simulator", "__wrapped__", "pytestmark", "router-managed-workload-llmd-simulator0", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.14000642099563265, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 66.38783556199633, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0024390960024902597, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-managed-workload-llmd-simulator1]", "lineno": 244, "outcome": "failed", "keywords": ["test_llm_inference_service[router-managed-workload-llmd-simulator1]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "llmd_simulator", "model_routing", "__wrapped__", "pytestmark", "router-managed-workload-llmd-simulator1", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.2274699569970835, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 1151.1598667669969, "outcome": "failed", "crash": {"path": "/workspace/source/test/e2e/llmisvc/test_llm_inference_service.py", "lineno": 1115, "message": "AssertionError: Service returned 503: inference error: ServiceUnavailable - failed to find endpoint candidates for serving the request"}, "traceback": [{"path": "llmisvc/test_llm_inference_service.py", "lineno": 816, "message": ""}, {"path": "llmisvc/logging.py", "lineno": 40, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1119, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1215, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1115, "message": "AssertionError"}], "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python\n\ntest_case = TestCase(base_refs=['router-managed', 'workload-llmd-simulator'], prompt='KServe is a', service_name='llmisvc-router-m... {'name': 'workload-llmd-simulator-llmisvc-8461fd55'}]},\n 'status': None}, model_name='facebook/opt-125m')\n\n @pytest.mark.llminferenceservice\n @pytest.mark.asyncio(loop_scope=\"session\")\n @pytest.mark.parametrize(\n \"test_case\",\n [\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-gateway-ref\",\n \"router-with-managed-route\",\n \"model-fb-opt-125m\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"custom-route-timeout-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"router-with-refs-test\",\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n routes=[ROUTER_ROUTES[0], ROUTER_ROUTES[1]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\"router-managed\", \"workload-pd-cpu\", \"model-fb-opt-125m\"],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"custom-route-timeout-pd-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"router-with-refs-pd-test\",\n response_assertion=assert_200_with_choices,\n expected_gateway=ROUTER_GATEWAYS[1],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[1]],\n routes=[ROUTER_ROUTES[2], ROUTER_ROUTES[3]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-dp-ep-gpu\",\n \"workload-dp-ep-prefill-gpu\",\n \"model-deepseek-v2-lite\",\n ],\n prompt=\"Delve into the multifaceted implications of a fully disaggregated cloud architecture, specifically \"\n \"where the compute plane (P) and the data plane (D) are independently deployed and managed for a \"\n \"geographically distributed, high-throughput, low-latency microservices ecosystem. Beyond the \"\n \"fundamental challenges of network latency and data consistency, elaborate on the advanced \"\n \"considerations and trade-offs inherent in such a setup: 1. Network Architecture and Protocols: \"\n \"How would the network fabric and underlying protocols (e.g., RDMA, custom transport layers) need to \"\n \"evolve to support optimal performance and minimize inter-plane communication overhead, especially for \"\n \"synchronous operations? Discuss the role of network programmability (e.g., SDN, P4) in dynamically \"\n \"optimizing routing and traffic flow between P and D. 2. Advanced Data Consistency and Durability: \"\n \"Explore sophisticated data consistency models (e.g., causal consistency, strong eventual consistency) \"\n \"and their applicability in balancing performance and data integrity across a globally distributed data plane. \"\n \"Detail strategies for ensuring data durability and fault tolerance, including multi-region replication, \"\n \"intelligent partitioning, and recovery mechanisms in the event of partial or full plane failures. \"\n \"3. Dynamic Resource Orchestration and Cost Optimization: Analyze how an orchestration layer would intelligently \"\n \"manage the independent scaling of compute (P) and data (D) resources, considering fluctuating workloads, \"\n \"cost efficiency, and performance targets (e.g., using predictive analytics for resource provisioning). \"\n \"Discuss mechanisms for dynamically reallocating compute nodes to different data partitions based on \"\n \"workload patterns and data locality, potentially involving live migration strategies. \"\n \"4. Security and Compliance in a Distributed Landscape: Address the enhanced security perimeter \"\n \"challenges, including securing communication channels between P and D (encryption in transit, mutual TLS), \"\n \"fine-grained access control to data at rest and in motion, and identity management across disaggregated \"\n \"components. Discuss how such an architecture impacts compliance with regulatory frameworks (e.g., GDPR, HIPAA) \"\n \"concerning data sovereignty, privacy, and auditability. 5. Operational Complexity and Observability: \"\n \"Examine the increased complexity in monitoring, logging, and tracing across highly decoupled compute and \"\n \"data planes. What specialized tooling and practices (e.g., distributed tracing with OpenTelemetry, advanced AIOps) \"\n \"would be essential? How would incident response and troubleshooting differ in this disaggregated environment \"\n \"compared to traditional integrated systems? Consider the challenges of pinpointing root causes across \"\n \"independent failures. 6. Real-world Applicability and Future Trends: Identify specific industries \"\n \"or use cases (e.g., high-frequency trading, IoT edge processing, large language model inference) \"\n \"where the benefits of P/D disaggregation would strongly outweigh its complexities. \"\n \"Conclude by speculating on emerging technologies or paradigms (e.g., serverless compute functions \"\n \"directly interacting with object storage, in-memory disaggregation) that could further drive or \"\n \"transform P/D disaggregation in cloud computing.\",\n max_tokens=2000,\n ),\n marks=[\n pytest.mark.cluster_gpu,\n pytest.mark.cluster_nvidia,\n pytest.mark.cluster_nvidia_roce,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-no-scheduler\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"What is KServe?\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.no_scheduler,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"This test simulates DP+EP that can run on CPU, the idea is to test the LWS-based deployment, \"\n \"but without the resources requirements for DP+EP (GPUs and ROCe/IB).\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_multi_node],\n ),\n # Scheduler config tests\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-inline-config\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-inline-config-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Chat completions endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n model_name=\"Qwen/Qwen2.5-0.5B-Instruct\",\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-configmap-ref\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-configmap-ref-test\",\n before_test=[create_scheduler_configmap],\n after_test=[delete_scheduler_configmap],\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-replicas\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-ha-replicas-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-custom-template\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-custom-template-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Scheduler v0.6 \u2192 v0.7 migration tests.\n # Deploy v0.6-style configs and verify the controller migrates them\n # so the v0.7 scheduler boots successfully.\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-pd-config-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-pd-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-nonzero-threshold-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-threshold-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Precise prefix KV cache routing test\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-precise-prefix-cache-inline-config\",\n \"workload-llmd-simulator-kvcache\",\n ],\n prompt=\"KServe is a\",\n service_name=\"precise-prefix-cache-test\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Models endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=create_response_assertion(with_field=\"data\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/chat/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 LoRA adapter\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n model_name=f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/models (base + LoRA)\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=assert_models_contains(\n \"facebook/opt-125m\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n \"lora-adapter-1\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # PVC storage tests -- validate direct PVC volume mount with real vLLM serving\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-pd-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_multi_node,\n pytest.mark.pvc_storage,\n ],\n ),\n ],\n indirect=[\"test_case\"],\n ids=generate_test_id,\n )\n @log_execution\n def test_llm_inference_service(test_case: TestCase): # noqa: F811\n inject_k8s_proxy()\n \n kserve_client = KServeClient(\n config_file=os.environ.get(\"KUBECONFIG\", \"~/.kube/config\"),\n client_configuration=client.Configuration(),\n )\n \n service_name = test_case.llm_service.metadata.name\n if not test_case.llm_service.metadata.annotations:\n test_case.llm_service.metadata.annotations = {}\n \n test_case.llm_service.metadata.annotations[\n \"security.opendatahub.io/enable-auth\"\n ] = \"false\"\n prefix = test_case.log_prefix\n \n test_failed = False\n try:\n print(f\"{prefix} Creating LLMInferenceService {service_name}\")\n create_llmisvc(kserve_client, test_case.llm_service)\n print(f\"{prefix} Waiting for LLMInferenceService {service_name} to be ready\")\n wait_for_llm_isvc_ready(\n kserve_client, test_case.llm_service, test_case.wait_timeout\n )\n print(f\"{prefix} Waiting for model response from {service_name}\")\n> wait_for_model_response(\n kserve_client,\n test_case,\n test_case.wait_timeout,\n extra_headers=test_case.extra_headers,\n )\n\nllmisvc/test_llm_inference_service.py:816: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nargs = (, TestCase(base_refs=['router-managed', 'workload-llm... {'name': 'workload-llmd-simulator-llmisvc-8461fd55'}]},\n 'status': None}, model_name='facebook/opt-125m'), 900)\nkwargs = {'extra_headers': {'X-Gateway-Model-Name': 'publishers/kserve-ci-e2e-test/models/facebook/opt-125m'}}\nfunc_name = 'wait_for_model_response'\ntimestamp_start = '2026-07-13T20:35:18.591353', start_time = 1783974918.5917783\nduration = 1102.6349976062775, timestamp_end = '2026-07-13T20:53:41.226779'\n\n @functools.wraps(func)\n def wrapper(*args, **kwargs):\n func_name = func.__name__\n \n timestamp_start = datetime.now().isoformat()\n logger.info(\n f\"[{func_name}] [{timestamp_start}] start - args={args}, kwargs={kwargs}\"\n )\n start_time = time.time()\n \n try:\n> result = func(*args, **kwargs)\n\nllmisvc/logging.py:40: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nkserve_client = \ntest_case = TestCase(base_refs=['router-managed', 'workload-llmd-simulator'], prompt='KServe is a', service_name='llmisvc-router-m... {'name': 'workload-llmd-simulator-llmisvc-8461fd55'}]},\n 'status': None}, model_name='facebook/opt-125m')\ntimeout_seconds = 900\nextra_headers = {'X-Gateway-Model-Name': 'publishers/kserve-ci-e2e-test/models/facebook/opt-125m'}\n\n @log_execution\n def wait_for_model_response(\n kserve_client: KServeClient,\n test_case: TestCase, # noqa: F811\n timeout_seconds: int = 900,\n extra_headers: Optional[Dict[str, str]] = None,\n ) -> str:\n def get_successful_response():\n try:\n if test_case.url_getter:\n service_url = test_case.url_getter(kserve_client, test_case.llm_service)\n else:\n service_url = get_llm_service_url(kserve_client, test_case.llm_service)\n except Exception as e:\n raise AssertionError(f\"\u274c Failed to get service URL: {e}\") from e\n \n model_url = service_url + test_case.endpoint\n \n headers = {\"Content-Type\": \"application/json\"}\n if extra_headers:\n headers.update(extra_headers)\n \n if test_case.payload_formatter is not None:\n test_payload = test_case.payload_formatter(test_case)\n elif test_case.prompt is not None:\n test_payload = {\n \"model\": test_case.model_name\n if not extra_headers or MODEL_ROUTING_HEADER not in extra_headers\n else extra_headers[MODEL_ROUTING_HEADER],\n \"prompt\": test_case.prompt,\n \"max_tokens\": test_case.max_tokens,\n }\n else:\n test_payload = None\n \n logger.info(f\"Calling LLM service at {model_url} with payload {test_payload}\")\n try:\n if test_payload is not None:\n response = post_with_retry(\n model_url,\n headers=headers,\n json_data=test_payload,\n timeout=test_case.response_timeout,\n )\n else:\n response = get_with_retry(\n model_url,\n headers=headers,\n timeout=test_case.response_timeout,\n )\n except Exception as e:\n logger.error(f\"\u274c Failed to call model: {e}\")\n raise AssertionError(f\"\u274c Failed to call model: {e}\") from e\n \n logger.info(f\"Model response is {response.status_code}: {response.text[:500]}\")\n \n if 200 <= response.status_code < 300:\n return response\n raise AssertionError(\n f\"Service returned {response.status_code}: {response.text}\"\n )\n \n> response = wait_for(get_successful_response, timeout=timeout_seconds, interval=5.0)\n\nllmisvc/test_llm_inference_service.py:1119: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nassertion_fn = .get_successful_response at 0x7f8e716a72e0>\ntimeout = 900, interval = 5.0\n\n def wait_for(\n assertion_fn: Callable[[], Any], timeout: float = 5.0, interval: float = 0.1\n ) -> Any:\n \"\"\"Wait for the assertion to succeed within timeout.\"\"\"\n deadline = time.time() + timeout\n last_msg = None\n while True:\n try:\n> return assertion_fn()\n\nllmisvc/test_llm_inference_service.py:1215: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\n def get_successful_response():\n try:\n if test_case.url_getter:\n service_url = test_case.url_getter(kserve_client, test_case.llm_service)\n else:\n service_url = get_llm_service_url(kserve_client, test_case.llm_service)\n except Exception as e:\n raise AssertionError(f\"\u274c Failed to get service URL: {e}\") from e\n \n model_url = service_url + test_case.endpoint\n \n headers = {\"Content-Type\": \"application/json\"}\n if extra_headers:\n headers.update(extra_headers)\n \n if test_case.payload_formatter is not None:\n test_payload = test_case.payload_formatter(test_case)\n elif test_case.prompt is not None:\n test_payload = {\n \"model\": test_case.model_name\n if not extra_headers or MODEL_ROUTING_HEADER not in extra_headers\n else extra_headers[MODEL_ROUTING_HEADER],\n \"prompt\": test_case.prompt,\n \"max_tokens\": test_case.max_tokens,\n }\n else:\n test_payload = None\n \n logger.info(f\"Calling LLM service at {model_url} with payload {test_payload}\")\n try:\n if test_payload is not None:\n response = post_with_retry(\n model_url,\n headers=headers,\n json_data=test_payload,\n timeout=test_case.response_timeout,\n )\n else:\n response = get_with_retry(\n model_url,\n headers=headers,\n timeout=test_case.response_timeout,\n )\n except Exception as e:\n logger.error(f\"\u274c Failed to call model: {e}\")\n raise AssertionError(f\"\u274c Failed to call model: {e}\") from e\n \n logger.info(f\"Model response is {response.status_code}: {response.text[:500]}\")\n \n if 200 <= response.status_code < 300:\n return response\n> raise AssertionError(\n f\"Service returned {response.status_code}: {response.text}\"\n )\nE AssertionError: Service returned 503: inference error: ServiceUnavailable - failed to find endpoint candidates for serving the request\n\nllmisvc/test_llm_inference_service.py:1115: AssertionError"}, "teardown": {"duration": 0.0016204980056500062, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_auth.py::test_llm_auth_invalid_token_rejected[cluster_cpu-cluster_single_node-auth-invalid-token]", "lineno": 380, "outcome": "passed", "keywords": ["test_llm_auth_invalid_token_rejected[auth-invalid-token]", "parametrize", "auth", "llminferenceservice", "cluster_cpu", "cluster_single_node", "__wrapped__", "pytestmark", "auth-invalid-token", "test_llm_auth.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.19254248100332916, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 171.09302577200287, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0029024649993516505, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_auth.py::test_llm_auth_disabled_no_token_required[cluster_cpu-cluster_single_node-auth-disabled]", "lineno": 511, "outcome": "passed", "keywords": ["test_llm_auth_disabled_no_token_required[auth-disabled]", "parametrize", "auth", "llminferenceservice", "cluster_cpu", "cluster_single_node", "__wrapped__", "pytestmark", "auth-disabled", "test_llm_auth.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.15474095499666873, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 148.84028249199764, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.003040983996470459, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-with-gateway-ref-router-with-managed-route-model-fb-opt-125m-workload-llmd-simulator]", "lineno": 244, "outcome": "passed", "keywords": ["test_llm_inference_service[router-with-gateway-ref-router-with-managed-route-model-fb-opt-125m-workload-llmd-simulator]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "llmd_simulator", "custom_gateway", "__wrapped__", "pytestmark", "router-with-gateway-ref-router-with-managed-route-model-fb-opt-125m-workload-llmd-simulator", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.8073145279995515, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 93.09068209499674, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0019038159953197464, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-managed-workload-llmd-simulator2]", "lineno": 244, "outcome": "failed", "keywords": ["test_llm_inference_service[router-managed-workload-llmd-simulator2]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "llmd_simulator", "model_routing", "__wrapped__", "pytestmark", "router-managed-workload-llmd-simulator2", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.2569652429956477, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 1166.241355351005, "outcome": "failed", "crash": {"path": "/workspace/source/test/e2e/llmisvc/test_llm_inference_service.py", "lineno": 1115, "message": "AssertionError: Service returned 503: inference error: ServiceUnavailable - failed to find endpoint candidates for serving the request"}, "traceback": [{"path": "llmisvc/test_llm_inference_service.py", "lineno": 816, "message": ""}, {"path": "llmisvc/logging.py", "lineno": 40, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1119, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1215, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1115, "message": "AssertionError"}], "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python\n\ntest_case = TestCase(base_refs=['router-managed', 'workload-llmd-simulator'], prompt='What is KServe?', service_name='llmisvc-rout... {'name': 'workload-llmd-simulator-llmisvc-53a6ad30'}]},\n 'status': None}, model_name='facebook/opt-125m')\n\n @pytest.mark.llminferenceservice\n @pytest.mark.asyncio(loop_scope=\"session\")\n @pytest.mark.parametrize(\n \"test_case\",\n [\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-gateway-ref\",\n \"router-with-managed-route\",\n \"model-fb-opt-125m\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"custom-route-timeout-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"router-with-refs-test\",\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n routes=[ROUTER_ROUTES[0], ROUTER_ROUTES[1]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\"router-managed\", \"workload-pd-cpu\", \"model-fb-opt-125m\"],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"custom-route-timeout-pd-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"router-with-refs-pd-test\",\n response_assertion=assert_200_with_choices,\n expected_gateway=ROUTER_GATEWAYS[1],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[1]],\n routes=[ROUTER_ROUTES[2], ROUTER_ROUTES[3]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-dp-ep-gpu\",\n \"workload-dp-ep-prefill-gpu\",\n \"model-deepseek-v2-lite\",\n ],\n prompt=\"Delve into the multifaceted implications of a fully disaggregated cloud architecture, specifically \"\n \"where the compute plane (P) and the data plane (D) are independently deployed and managed for a \"\n \"geographically distributed, high-throughput, low-latency microservices ecosystem. Beyond the \"\n \"fundamental challenges of network latency and data consistency, elaborate on the advanced \"\n \"considerations and trade-offs inherent in such a setup: 1. Network Architecture and Protocols: \"\n \"How would the network fabric and underlying protocols (e.g., RDMA, custom transport layers) need to \"\n \"evolve to support optimal performance and minimize inter-plane communication overhead, especially for \"\n \"synchronous operations? Discuss the role of network programmability (e.g., SDN, P4) in dynamically \"\n \"optimizing routing and traffic flow between P and D. 2. Advanced Data Consistency and Durability: \"\n \"Explore sophisticated data consistency models (e.g., causal consistency, strong eventual consistency) \"\n \"and their applicability in balancing performance and data integrity across a globally distributed data plane. \"\n \"Detail strategies for ensuring data durability and fault tolerance, including multi-region replication, \"\n \"intelligent partitioning, and recovery mechanisms in the event of partial or full plane failures. \"\n \"3. Dynamic Resource Orchestration and Cost Optimization: Analyze how an orchestration layer would intelligently \"\n \"manage the independent scaling of compute (P) and data (D) resources, considering fluctuating workloads, \"\n \"cost efficiency, and performance targets (e.g., using predictive analytics for resource provisioning). \"\n \"Discuss mechanisms for dynamically reallocating compute nodes to different data partitions based on \"\n \"workload patterns and data locality, potentially involving live migration strategies. \"\n \"4. Security and Compliance in a Distributed Landscape: Address the enhanced security perimeter \"\n \"challenges, including securing communication channels between P and D (encryption in transit, mutual TLS), \"\n \"fine-grained access control to data at rest and in motion, and identity management across disaggregated \"\n \"components. Discuss how such an architecture impacts compliance with regulatory frameworks (e.g., GDPR, HIPAA) \"\n \"concerning data sovereignty, privacy, and auditability. 5. Operational Complexity and Observability: \"\n \"Examine the increased complexity in monitoring, logging, and tracing across highly decoupled compute and \"\n \"data planes. What specialized tooling and practices (e.g., distributed tracing with OpenTelemetry, advanced AIOps) \"\n \"would be essential? How would incident response and troubleshooting differ in this disaggregated environment \"\n \"compared to traditional integrated systems? Consider the challenges of pinpointing root causes across \"\n \"independent failures. 6. Real-world Applicability and Future Trends: Identify specific industries \"\n \"or use cases (e.g., high-frequency trading, IoT edge processing, large language model inference) \"\n \"where the benefits of P/D disaggregation would strongly outweigh its complexities. \"\n \"Conclude by speculating on emerging technologies or paradigms (e.g., serverless compute functions \"\n \"directly interacting with object storage, in-memory disaggregation) that could further drive or \"\n \"transform P/D disaggregation in cloud computing.\",\n max_tokens=2000,\n ),\n marks=[\n pytest.mark.cluster_gpu,\n pytest.mark.cluster_nvidia,\n pytest.mark.cluster_nvidia_roce,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-no-scheduler\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"What is KServe?\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.no_scheduler,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"This test simulates DP+EP that can run on CPU, the idea is to test the LWS-based deployment, \"\n \"but without the resources requirements for DP+EP (GPUs and ROCe/IB).\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_multi_node],\n ),\n # Scheduler config tests\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-inline-config\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-inline-config-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Chat completions endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n model_name=\"Qwen/Qwen2.5-0.5B-Instruct\",\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-configmap-ref\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-configmap-ref-test\",\n before_test=[create_scheduler_configmap],\n after_test=[delete_scheduler_configmap],\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-replicas\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-ha-replicas-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-custom-template\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-custom-template-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Scheduler v0.6 \u2192 v0.7 migration tests.\n # Deploy v0.6-style configs and verify the controller migrates them\n # so the v0.7 scheduler boots successfully.\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-pd-config-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-pd-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-nonzero-threshold-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-threshold-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Precise prefix KV cache routing test\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-precise-prefix-cache-inline-config\",\n \"workload-llmd-simulator-kvcache\",\n ],\n prompt=\"KServe is a\",\n service_name=\"precise-prefix-cache-test\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Models endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=create_response_assertion(with_field=\"data\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/chat/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 LoRA adapter\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n model_name=f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/models (base + LoRA)\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=assert_models_contains(\n \"facebook/opt-125m\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n \"lora-adapter-1\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # PVC storage tests -- validate direct PVC volume mount with real vLLM serving\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-pd-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_multi_node,\n pytest.mark.pvc_storage,\n ],\n ),\n ],\n indirect=[\"test_case\"],\n ids=generate_test_id,\n )\n @log_execution\n def test_llm_inference_service(test_case: TestCase): # noqa: F811\n inject_k8s_proxy()\n \n kserve_client = KServeClient(\n config_file=os.environ.get(\"KUBECONFIG\", \"~/.kube/config\"),\n client_configuration=client.Configuration(),\n )\n \n service_name = test_case.llm_service.metadata.name\n if not test_case.llm_service.metadata.annotations:\n test_case.llm_service.metadata.annotations = {}\n \n test_case.llm_service.metadata.annotations[\n \"security.opendatahub.io/enable-auth\"\n ] = \"false\"\n prefix = test_case.log_prefix\n \n test_failed = False\n try:\n print(f\"{prefix} Creating LLMInferenceService {service_name}\")\n create_llmisvc(kserve_client, test_case.llm_service)\n print(f\"{prefix} Waiting for LLMInferenceService {service_name} to be ready\")\n wait_for_llm_isvc_ready(\n kserve_client, test_case.llm_service, test_case.wait_timeout\n )\n print(f\"{prefix} Waiting for model response from {service_name}\")\n> wait_for_model_response(\n kserve_client,\n test_case,\n test_case.wait_timeout,\n extra_headers=test_case.extra_headers,\n )\n\nllmisvc/test_llm_inference_service.py:816: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nargs = (, TestCase(base_refs=['router-managed', 'workload-llm... {'name': 'workload-llmd-simulator-llmisvc-53a6ad30'}]},\n 'status': None}, model_name='facebook/opt-125m'), 900)\nkwargs = {'extra_headers': {'X-Gateway-Model-Name': 'publishers/kserve-ci-e2e-test/models/facebook/opt-125m'}}\nfunc_name = 'wait_for_model_response'\ntimestamp_start = '2026-07-13T20:54:45.134725', start_time = 1783976085.1351664\nduration = 1102.556923866272, timestamp_end = '2026-07-13T21:13:07.692092'\n\n @functools.wraps(func)\n def wrapper(*args, **kwargs):\n func_name = func.__name__\n \n timestamp_start = datetime.now().isoformat()\n logger.info(\n f\"[{func_name}] [{timestamp_start}] start - args={args}, kwargs={kwargs}\"\n )\n start_time = time.time()\n \n try:\n> result = func(*args, **kwargs)\n\nllmisvc/logging.py:40: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nkserve_client = \ntest_case = TestCase(base_refs=['router-managed', 'workload-llmd-simulator'], prompt='What is KServe?', service_name='llmisvc-rout... {'name': 'workload-llmd-simulator-llmisvc-53a6ad30'}]},\n 'status': None}, model_name='facebook/opt-125m')\ntimeout_seconds = 900\nextra_headers = {'X-Gateway-Model-Name': 'publishers/kserve-ci-e2e-test/models/facebook/opt-125m'}\n\n @log_execution\n def wait_for_model_response(\n kserve_client: KServeClient,\n test_case: TestCase, # noqa: F811\n timeout_seconds: int = 900,\n extra_headers: Optional[Dict[str, str]] = None,\n ) -> str:\n def get_successful_response():\n try:\n if test_case.url_getter:\n service_url = test_case.url_getter(kserve_client, test_case.llm_service)\n else:\n service_url = get_llm_service_url(kserve_client, test_case.llm_service)\n except Exception as e:\n raise AssertionError(f\"\u274c Failed to get service URL: {e}\") from e\n \n model_url = service_url + test_case.endpoint\n \n headers = {\"Content-Type\": \"application/json\"}\n if extra_headers:\n headers.update(extra_headers)\n \n if test_case.payload_formatter is not None:\n test_payload = test_case.payload_formatter(test_case)\n elif test_case.prompt is not None:\n test_payload = {\n \"model\": test_case.model_name\n if not extra_headers or MODEL_ROUTING_HEADER not in extra_headers\n else extra_headers[MODEL_ROUTING_HEADER],\n \"prompt\": test_case.prompt,\n \"max_tokens\": test_case.max_tokens,\n }\n else:\n test_payload = None\n \n logger.info(f\"Calling LLM service at {model_url} with payload {test_payload}\")\n try:\n if test_payload is not None:\n response = post_with_retry(\n model_url,\n headers=headers,\n json_data=test_payload,\n timeout=test_case.response_timeout,\n )\n else:\n response = get_with_retry(\n model_url,\n headers=headers,\n timeout=test_case.response_timeout,\n )\n except Exception as e:\n logger.error(f\"\u274c Failed to call model: {e}\")\n raise AssertionError(f\"\u274c Failed to call model: {e}\") from e\n \n logger.info(f\"Model response is {response.status_code}: {response.text[:500]}\")\n \n if 200 <= response.status_code < 300:\n return response\n raise AssertionError(\n f\"Service returned {response.status_code}: {response.text}\"\n )\n \n> response = wait_for(get_successful_response, timeout=timeout_seconds, interval=5.0)\n\nllmisvc/test_llm_inference_service.py:1119: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nassertion_fn = .get_successful_response at 0x7f8e716045e0>\ntimeout = 900, interval = 5.0\n\n def wait_for(\n assertion_fn: Callable[[], Any], timeout: float = 5.0, interval: float = 0.1\n ) -> Any:\n \"\"\"Wait for the assertion to succeed within timeout.\"\"\"\n deadline = time.time() + timeout\n last_msg = None\n while True:\n try:\n> return assertion_fn()\n\nllmisvc/test_llm_inference_service.py:1215: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\n def get_successful_response():\n try:\n if test_case.url_getter:\n service_url = test_case.url_getter(kserve_client, test_case.llm_service)\n else:\n service_url = get_llm_service_url(kserve_client, test_case.llm_service)\n except Exception as e:\n raise AssertionError(f\"\u274c Failed to get service URL: {e}\") from e\n \n model_url = service_url + test_case.endpoint\n \n headers = {\"Content-Type\": \"application/json\"}\n if extra_headers:\n headers.update(extra_headers)\n \n if test_case.payload_formatter is not None:\n test_payload = test_case.payload_formatter(test_case)\n elif test_case.prompt is not None:\n test_payload = {\n \"model\": test_case.model_name\n if not extra_headers or MODEL_ROUTING_HEADER not in extra_headers\n else extra_headers[MODEL_ROUTING_HEADER],\n \"prompt\": test_case.prompt,\n \"max_tokens\": test_case.max_tokens,\n }\n else:\n test_payload = None\n \n logger.info(f\"Calling LLM service at {model_url} with payload {test_payload}\")\n try:\n if test_payload is not None:\n response = post_with_retry(\n model_url,\n headers=headers,\n json_data=test_payload,\n timeout=test_case.response_timeout,\n )\n else:\n response = get_with_retry(\n model_url,\n headers=headers,\n timeout=test_case.response_timeout,\n )\n except Exception as e:\n logger.error(f\"\u274c Failed to call model: {e}\")\n raise AssertionError(f\"\u274c Failed to call model: {e}\") from e\n \n logger.info(f\"Model response is {response.status_code}: {response.text[:500]}\")\n \n if 200 <= response.status_code < 300:\n return response\n> raise AssertionError(\n f\"Service returned {response.status_code}: {response.text}\"\n )\nE AssertionError: Service returned 503: inference error: ServiceUnavailable - failed to find endpoint candidates for serving the request\n\nllmisvc/test_llm_inference_service.py:1115: AssertionError"}, "teardown": {"duration": 0.001919557005749084, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-managed-workload-single-cpu-model-fb-opt-125m]", "lineno": 244, "outcome": "passed", "keywords": ["test_llm_inference_service[router-managed-workload-single-cpu-model-fb-opt-125m]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "__wrapped__", "pytestmark", "router-managed-workload-single-cpu-model-fb-opt-125m", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.39690002700081095, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 166.7140256739949, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.003495750999718439, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-custom-route-timeout-scheduler-managed-workload-single-cpu-model-fb-opt-125m]", "lineno": 244, "outcome": "failed", "keywords": ["test_llm_inference_service[router-custom-route-timeout-scheduler-managed-workload-single-cpu-model-fb-opt-125m]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "__wrapped__", "pytestmark", "router-custom-route-timeout-scheduler-managed-workload-single-cpu-model-fb-opt-125m", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.19910014199558645, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 902.2552381480054, "outcome": "failed", "crash": {"path": "/workspace/source/test/e2e/llmisvc/test_llm_inference_service.py", "lineno": 1199, "message": "AssertionError: Missing true conditions: {'Ready', 'WorkloadsReady'}, expected {'Ready', 'RouterReady', 'WorkloadsReady'}, got [{'lastTransitionTime': '2026-07-13T20:57:09Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-13T20:57:09Z', 'severity': 'Info', 'status': 'True', 'type': 'InferencePoolReady'}, {'lastTransitionTime': '2026-07-13T20:57:09Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'severity': 'Info', 'status': 'False', 'type': 'MainWorkloadReady'}, {'lastTransitionTime': '2026-07-13T20:57:01Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-13T20:57:09Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-13T20:57:33Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-13T20:57:33Z', 'severity': 'Info', 'status': 'True', 'type': 'SchedulerWorkloadReady'}, {'lastTransitionTime': '2026-07-13T20:57:09Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'WorkloadsReady'}]"}, "traceback": [{"path": "llmisvc/test_llm_inference_service.py", "lineno": 812, "message": ""}, {"path": "llmisvc/logging.py", "lineno": 40, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1204, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1215, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1199, "message": "AssertionError"}], "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python\n\ntest_case = TestCase(base_refs=['router-custom-route-timeout', 'scheduler-managed', 'workload-single-cpu', 'model-fb-opt-125m'], p... {'name': 'model-fb-opt-125m-custom-route-928a8601'}]},\n 'status': None}, model_name='facebook/opt-125m')\n\n @pytest.mark.llminferenceservice\n @pytest.mark.asyncio(loop_scope=\"session\")\n @pytest.mark.parametrize(\n \"test_case\",\n [\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-gateway-ref\",\n \"router-with-managed-route\",\n \"model-fb-opt-125m\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"custom-route-timeout-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"router-with-refs-test\",\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n routes=[ROUTER_ROUTES[0], ROUTER_ROUTES[1]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\"router-managed\", \"workload-pd-cpu\", \"model-fb-opt-125m\"],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"custom-route-timeout-pd-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"router-with-refs-pd-test\",\n response_assertion=assert_200_with_choices,\n expected_gateway=ROUTER_GATEWAYS[1],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[1]],\n routes=[ROUTER_ROUTES[2], ROUTER_ROUTES[3]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-dp-ep-gpu\",\n \"workload-dp-ep-prefill-gpu\",\n \"model-deepseek-v2-lite\",\n ],\n prompt=\"Delve into the multifaceted implications of a fully disaggregated cloud architecture, specifically \"\n \"where the compute plane (P) and the data plane (D) are independently deployed and managed for a \"\n \"geographically distributed, high-throughput, low-latency microservices ecosystem. Beyond the \"\n \"fundamental challenges of network latency and data consistency, elaborate on the advanced \"\n \"considerations and trade-offs inherent in such a setup: 1. Network Architecture and Protocols: \"\n \"How would the network fabric and underlying protocols (e.g., RDMA, custom transport layers) need to \"\n \"evolve to support optimal performance and minimize inter-plane communication overhead, especially for \"\n \"synchronous operations? Discuss the role of network programmability (e.g., SDN, P4) in dynamically \"\n \"optimizing routing and traffic flow between P and D. 2. Advanced Data Consistency and Durability: \"\n \"Explore sophisticated data consistency models (e.g., causal consistency, strong eventual consistency) \"\n \"and their applicability in balancing performance and data integrity across a globally distributed data plane. \"\n \"Detail strategies for ensuring data durability and fault tolerance, including multi-region replication, \"\n \"intelligent partitioning, and recovery mechanisms in the event of partial or full plane failures. \"\n \"3. Dynamic Resource Orchestration and Cost Optimization: Analyze how an orchestration layer would intelligently \"\n \"manage the independent scaling of compute (P) and data (D) resources, considering fluctuating workloads, \"\n \"cost efficiency, and performance targets (e.g., using predictive analytics for resource provisioning). \"\n \"Discuss mechanisms for dynamically reallocating compute nodes to different data partitions based on \"\n \"workload patterns and data locality, potentially involving live migration strategies. \"\n \"4. Security and Compliance in a Distributed Landscape: Address the enhanced security perimeter \"\n \"challenges, including securing communication channels between P and D (encryption in transit, mutual TLS), \"\n \"fine-grained access control to data at rest and in motion, and identity management across disaggregated \"\n \"components. Discuss how such an architecture impacts compliance with regulatory frameworks (e.g., GDPR, HIPAA) \"\n \"concerning data sovereignty, privacy, and auditability. 5. Operational Complexity and Observability: \"\n \"Examine the increased complexity in monitoring, logging, and tracing across highly decoupled compute and \"\n \"data planes. What specialized tooling and practices (e.g., distributed tracing with OpenTelemetry, advanced AIOps) \"\n \"would be essential? How would incident response and troubleshooting differ in this disaggregated environment \"\n \"compared to traditional integrated systems? Consider the challenges of pinpointing root causes across \"\n \"independent failures. 6. Real-world Applicability and Future Trends: Identify specific industries \"\n \"or use cases (e.g., high-frequency trading, IoT edge processing, large language model inference) \"\n \"where the benefits of P/D disaggregation would strongly outweigh its complexities. \"\n \"Conclude by speculating on emerging technologies or paradigms (e.g., serverless compute functions \"\n \"directly interacting with object storage, in-memory disaggregation) that could further drive or \"\n \"transform P/D disaggregation in cloud computing.\",\n max_tokens=2000,\n ),\n marks=[\n pytest.mark.cluster_gpu,\n pytest.mark.cluster_nvidia,\n pytest.mark.cluster_nvidia_roce,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-no-scheduler\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"What is KServe?\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.no_scheduler,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"This test simulates DP+EP that can run on CPU, the idea is to test the LWS-based deployment, \"\n \"but without the resources requirements for DP+EP (GPUs and ROCe/IB).\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_multi_node],\n ),\n # Scheduler config tests\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-inline-config\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-inline-config-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Chat completions endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n model_name=\"Qwen/Qwen2.5-0.5B-Instruct\",\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-configmap-ref\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-configmap-ref-test\",\n before_test=[create_scheduler_configmap],\n after_test=[delete_scheduler_configmap],\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-replicas\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-ha-replicas-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-custom-template\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-custom-template-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Scheduler v0.6 \u2192 v0.7 migration tests.\n # Deploy v0.6-style configs and verify the controller migrates them\n # so the v0.7 scheduler boots successfully.\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-pd-config-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-pd-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-nonzero-threshold-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-threshold-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Precise prefix KV cache routing test\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-precise-prefix-cache-inline-config\",\n \"workload-llmd-simulator-kvcache\",\n ],\n prompt=\"KServe is a\",\n service_name=\"precise-prefix-cache-test\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Models endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=create_response_assertion(with_field=\"data\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/chat/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 LoRA adapter\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n model_name=f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/models (base + LoRA)\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=assert_models_contains(\n \"facebook/opt-125m\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n \"lora-adapter-1\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # PVC storage tests -- validate direct PVC volume mount with real vLLM serving\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-pd-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_multi_node,\n pytest.mark.pvc_storage,\n ],\n ),\n ],\n indirect=[\"test_case\"],\n ids=generate_test_id,\n )\n @log_execution\n def test_llm_inference_service(test_case: TestCase): # noqa: F811\n inject_k8s_proxy()\n \n kserve_client = KServeClient(\n config_file=os.environ.get(\"KUBECONFIG\", \"~/.kube/config\"),\n client_configuration=client.Configuration(),\n )\n \n service_name = test_case.llm_service.metadata.name\n if not test_case.llm_service.metadata.annotations:\n test_case.llm_service.metadata.annotations = {}\n \n test_case.llm_service.metadata.annotations[\n \"security.opendatahub.io/enable-auth\"\n ] = \"false\"\n prefix = test_case.log_prefix\n \n test_failed = False\n try:\n print(f\"{prefix} Creating LLMInferenceService {service_name}\")\n create_llmisvc(kserve_client, test_case.llm_service)\n print(f\"{prefix} Waiting for LLMInferenceService {service_name} to be ready\")\n> wait_for_llm_isvc_ready(\n kserve_client, test_case.llm_service, test_case.wait_timeout\n )\n\nllmisvc/test_llm_inference_service.py:812: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nargs = (, {'api_version': 'serving.kserve.io/v1alpha1',\n 'kin...m-rout-4c4b9f6e'},\n {'name': 'model-fb-opt-125m-custom-route-928a8601'}]},\n 'status': None}, 900)\nkwargs = {}, func_name = 'wait_for_llm_isvc_ready'\ntimestamp_start = '2026-07-13T20:56:47.463858', start_time = 1783976207.464244\nduration = 900.760556936264, timestamp_end = '2026-07-13T21:11:48.224804'\n\n @functools.wraps(func)\n def wrapper(*args, **kwargs):\n func_name = func.__name__\n \n timestamp_start = datetime.now().isoformat()\n logger.info(\n f\"[{func_name}] [{timestamp_start}] start - args={args}, kwargs={kwargs}\"\n )\n start_time = time.time()\n \n try:\n> result = func(*args, **kwargs)\n\nllmisvc/logging.py:40: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nkserve_client = \ngiven = {'api_version': 'serving.kserve.io/v1alpha1',\n 'kind': 'LLMInferenceService',\n 'metadata': {'annotations': {'security....-custom-rout-4c4b9f6e'},\n {'name': 'model-fb-opt-125m-custom-route-928a8601'}]},\n 'status': None}\ntimeout_seconds = 900\n\n @log_execution\n def wait_for_llm_isvc_ready(\n kserve_client: KServeClient,\n given: V1alpha1LLMInferenceService,\n timeout_seconds: int = 900,\n ) -> str:\n def assert_llm_isvc_ready():\n out = get_llmisvc(\n kserve_client,\n given.metadata.name,\n given.metadata.namespace,\n given.api_version.split(\"/\")[1],\n )\n \n if \"status\" not in out:\n raise AssertionError(\"No status found in LLM inference service\")\n \n status = out[\"status\"]\n if \"conditions\" not in status:\n raise AssertionError(\"No conditions found in status\")\n \n expected_true_conditions = {\"Ready\", \"WorkloadsReady\", \"RouterReady\"}\n got_true_conditions = set()\n \n conditions = status[\"conditions\"]\n \n for condition in conditions:\n if condition.get(\"status\") == \"True\":\n got_true_conditions.add(condition.get(\"type\"))\n \n missing_conditions = expected_true_conditions - got_true_conditions\n if missing_conditions:\n raise AssertionError(\n f\"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}\"\n )\n return True\n \n> return wait_for(assert_llm_isvc_ready, timeout=timeout_seconds, interval=1.0)\n\nllmisvc/test_llm_inference_service.py:1204: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nassertion_fn = .assert_llm_isvc_ready at 0x7fb7b5e7d080>\ntimeout = 900, interval = 1.0\n\n def wait_for(\n assertion_fn: Callable[[], Any], timeout: float = 5.0, interval: float = 0.1\n ) -> Any:\n \"\"\"Wait for the assertion to succeed within timeout.\"\"\"\n deadline = time.time() + timeout\n last_msg = None\n while True:\n try:\n> return assertion_fn()\n\nllmisvc/test_llm_inference_service.py:1215: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\n def assert_llm_isvc_ready():\n out = get_llmisvc(\n kserve_client,\n given.metadata.name,\n given.metadata.namespace,\n given.api_version.split(\"/\")[1],\n )\n \n if \"status\" not in out:\n raise AssertionError(\"No status found in LLM inference service\")\n \n status = out[\"status\"]\n if \"conditions\" not in status:\n raise AssertionError(\"No conditions found in status\")\n \n expected_true_conditions = {\"Ready\", \"WorkloadsReady\", \"RouterReady\"}\n got_true_conditions = set()\n \n conditions = status[\"conditions\"]\n \n for condition in conditions:\n if condition.get(\"status\") == \"True\":\n got_true_conditions.add(condition.get(\"type\"))\n \n missing_conditions = expected_true_conditions - got_true_conditions\n if missing_conditions:\n> raise AssertionError(\n f\"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}\"\n )\nE AssertionError: Missing true conditions: {'Ready', 'WorkloadsReady'}, expected {'Ready', 'RouterReady', 'WorkloadsReady'}, got [{'lastTransitionTime': '2026-07-13T20:57:09Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-13T20:57:09Z', 'severity': 'Info', 'status': 'True', 'type': 'InferencePoolReady'}, {'lastTransitionTime': '2026-07-13T20:57:09Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'severity': 'Info', 'status': 'False', 'type': 'MainWorkloadReady'}, {'lastTransitionTime': '2026-07-13T20:57:01Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-13T20:57:09Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-13T20:57:33Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-13T20:57:33Z', 'severity': 'Info', 'status': 'True', 'type': 'SchedulerWorkloadReady'}, {'lastTransitionTime': '2026-07-13T20:57:09Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'WorkloadsReady'}]\n\nllmisvc/test_llm_inference_service.py:1199: AssertionError"}, "teardown": {"duration": 0.002204292999522295, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-with-refs-scheduler-managed-workload-single-cpu-model-fb-opt-125m]", "lineno": 244, "outcome": "passed", "keywords": ["test_llm_inference_service[router-with-refs-scheduler-managed-workload-single-cpu-model-fb-opt-125m]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "custom_gateway", "__wrapped__", "pytestmark", "router-with-refs-scheduler-managed-workload-single-cpu-model-fb-opt-125m", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 2.855018067995843, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 169.2792028499971, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.003105338000750635, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-managed-workload-single-cpu-model-fb-opt-125m-with-lora-hf0]", "lineno": 244, "outcome": "passed", "keywords": ["test_llm_inference_service[router-managed-workload-single-cpu-model-fb-opt-125m-with-lora-hf0]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "model_routing", "lora", "__wrapped__", "pytestmark", "router-managed-workload-single-cpu-model-fb-opt-125m-with-lora-hf0", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.15403510900068795, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 159.54795994800224, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.003320759999041911, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-managed-workload-pd-cpu-model-fb-opt-125m]", "lineno": 244, "outcome": "passed", "keywords": ["test_llm_inference_service[router-managed-workload-pd-cpu-model-fb-opt-125m]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "__wrapped__", "pytestmark", "router-managed-workload-pd-cpu-model-fb-opt-125m", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.16348403199663153, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 196.3229845629976, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0037653379986295477, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-managed-workload-single-cpu-model-fb-opt-125m-with-lora-hf1]", "lineno": 244, "outcome": "failed", "keywords": ["test_llm_inference_service[router-managed-workload-single-cpu-model-fb-opt-125m-with-lora-hf1]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "model_routing", "lora", "__wrapped__", "pytestmark", "router-managed-workload-single-cpu-model-fb-opt-125m-with-lora-hf1", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.23194146899913903, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 1011.8681546950029, "outcome": "failed", "crash": {"path": "/workspace/source/test/e2e/llmisvc/test_llm_inference_service.py", "lineno": 1115, "message": "AssertionError: Service returned 401:"}, "traceback": [{"path": "llmisvc/test_llm_inference_service.py", "lineno": 816, "message": ""}, {"path": "llmisvc/logging.py", "lineno": 40, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1119, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1215, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1115, "message": "AssertionError"}], "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python\n\ntest_case = TestCase(base_refs=['router-managed', 'workload-single-cpu', 'model-fb-opt-125m-with-lora-hf'], prompt=None, service_n... {'name': 'model-fb-opt-125m-with-lora-hf-c0d503b0'}]},\n 'status': None}, model_name='facebook/opt-125m')\n\n @pytest.mark.llminferenceservice\n @pytest.mark.asyncio(loop_scope=\"session\")\n @pytest.mark.parametrize(\n \"test_case\",\n [\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-gateway-ref\",\n \"router-with-managed-route\",\n \"model-fb-opt-125m\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"custom-route-timeout-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"router-with-refs-test\",\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n routes=[ROUTER_ROUTES[0], ROUTER_ROUTES[1]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\"router-managed\", \"workload-pd-cpu\", \"model-fb-opt-125m\"],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"custom-route-timeout-pd-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"router-with-refs-pd-test\",\n response_assertion=assert_200_with_choices,\n expected_gateway=ROUTER_GATEWAYS[1],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[1]],\n routes=[ROUTER_ROUTES[2], ROUTER_ROUTES[3]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-dp-ep-gpu\",\n \"workload-dp-ep-prefill-gpu\",\n \"model-deepseek-v2-lite\",\n ],\n prompt=\"Delve into the multifaceted implications of a fully disaggregated cloud architecture, specifically \"\n \"where the compute plane (P) and the data plane (D) are independently deployed and managed for a \"\n \"geographically distributed, high-throughput, low-latency microservices ecosystem. Beyond the \"\n \"fundamental challenges of network latency and data consistency, elaborate on the advanced \"\n \"considerations and trade-offs inherent in such a setup: 1. Network Architecture and Protocols: \"\n \"How would the network fabric and underlying protocols (e.g., RDMA, custom transport layers) need to \"\n \"evolve to support optimal performance and minimize inter-plane communication overhead, especially for \"\n \"synchronous operations? Discuss the role of network programmability (e.g., SDN, P4) in dynamically \"\n \"optimizing routing and traffic flow between P and D. 2. Advanced Data Consistency and Durability: \"\n \"Explore sophisticated data consistency models (e.g., causal consistency, strong eventual consistency) \"\n \"and their applicability in balancing performance and data integrity across a globally distributed data plane. \"\n \"Detail strategies for ensuring data durability and fault tolerance, including multi-region replication, \"\n \"intelligent partitioning, and recovery mechanisms in the event of partial or full plane failures. \"\n \"3. Dynamic Resource Orchestration and Cost Optimization: Analyze how an orchestration layer would intelligently \"\n \"manage the independent scaling of compute (P) and data (D) resources, considering fluctuating workloads, \"\n \"cost efficiency, and performance targets (e.g., using predictive analytics for resource provisioning). \"\n \"Discuss mechanisms for dynamically reallocating compute nodes to different data partitions based on \"\n \"workload patterns and data locality, potentially involving live migration strategies. \"\n \"4. Security and Compliance in a Distributed Landscape: Address the enhanced security perimeter \"\n \"challenges, including securing communication channels between P and D (encryption in transit, mutual TLS), \"\n \"fine-grained access control to data at rest and in motion, and identity management across disaggregated \"\n \"components. Discuss how such an architecture impacts compliance with regulatory frameworks (e.g., GDPR, HIPAA) \"\n \"concerning data sovereignty, privacy, and auditability. 5. Operational Complexity and Observability: \"\n \"Examine the increased complexity in monitoring, logging, and tracing across highly decoupled compute and \"\n \"data planes. What specialized tooling and practices (e.g., distributed tracing with OpenTelemetry, advanced AIOps) \"\n \"would be essential? How would incident response and troubleshooting differ in this disaggregated environment \"\n \"compared to traditional integrated systems? Consider the challenges of pinpointing root causes across \"\n \"independent failures. 6. Real-world Applicability and Future Trends: Identify specific industries \"\n \"or use cases (e.g., high-frequency trading, IoT edge processing, large language model inference) \"\n \"where the benefits of P/D disaggregation would strongly outweigh its complexities. \"\n \"Conclude by speculating on emerging technologies or paradigms (e.g., serverless compute functions \"\n \"directly interacting with object storage, in-memory disaggregation) that could further drive or \"\n \"transform P/D disaggregation in cloud computing.\",\n max_tokens=2000,\n ),\n marks=[\n pytest.mark.cluster_gpu,\n pytest.mark.cluster_nvidia,\n pytest.mark.cluster_nvidia_roce,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-no-scheduler\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"What is KServe?\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.no_scheduler,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"This test simulates DP+EP that can run on CPU, the idea is to test the LWS-based deployment, \"\n \"but without the resources requirements for DP+EP (GPUs and ROCe/IB).\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_multi_node],\n ),\n # Scheduler config tests\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-inline-config\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-inline-config-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Chat completions endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n model_name=\"Qwen/Qwen2.5-0.5B-Instruct\",\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-configmap-ref\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-configmap-ref-test\",\n before_test=[create_scheduler_configmap],\n after_test=[delete_scheduler_configmap],\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-replicas\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-ha-replicas-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-custom-template\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-custom-template-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Scheduler v0.6 \u2192 v0.7 migration tests.\n # Deploy v0.6-style configs and verify the controller migrates them\n # so the v0.7 scheduler boots successfully.\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-pd-config-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-pd-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-nonzero-threshold-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-threshold-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Precise prefix KV cache routing test\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-precise-prefix-cache-inline-config\",\n \"workload-llmd-simulator-kvcache\",\n ],\n prompt=\"KServe is a\",\n service_name=\"precise-prefix-cache-test\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Models endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=create_response_assertion(with_field=\"data\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/chat/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 LoRA adapter\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n model_name=f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/models (base + LoRA)\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=assert_models_contains(\n \"facebook/opt-125m\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n \"lora-adapter-1\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # PVC storage tests -- validate direct PVC volume mount with real vLLM serving\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-pd-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_multi_node,\n pytest.mark.pvc_storage,\n ],\n ),\n ],\n indirect=[\"test_case\"],\n ids=generate_test_id,\n )\n @log_execution\n def test_llm_inference_service(test_case: TestCase): # noqa: F811\n inject_k8s_proxy()\n \n kserve_client = KServeClient(\n config_file=os.environ.get(\"KUBECONFIG\", \"~/.kube/config\"),\n client_configuration=client.Configuration(),\n )\n \n service_name = test_case.llm_service.metadata.name\n if not test_case.llm_service.metadata.annotations:\n test_case.llm_service.metadata.annotations = {}\n \n test_case.llm_service.metadata.annotations[\n \"security.opendatahub.io/enable-auth\"\n ] = \"false\"\n prefix = test_case.log_prefix\n \n test_failed = False\n try:\n print(f\"{prefix} Creating LLMInferenceService {service_name}\")\n create_llmisvc(kserve_client, test_case.llm_service)\n print(f\"{prefix} Waiting for LLMInferenceService {service_name} to be ready\")\n wait_for_llm_isvc_ready(\n kserve_client, test_case.llm_service, test_case.wait_timeout\n )\n print(f\"{prefix} Waiting for model response from {service_name}\")\n> wait_for_model_response(\n kserve_client,\n test_case,\n test_case.wait_timeout,\n extra_headers=test_case.extra_headers,\n )\n\nllmisvc/test_llm_inference_service.py:816: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nargs = (, TestCase(base_refs=['router-managed', 'workload-sin... {'name': 'model-fb-opt-125m-with-lora-hf-c0d503b0'}]},\n 'status': None}, model_name='facebook/opt-125m'), 900)\nkwargs = {'extra_headers': {'X-Gateway-Model-Name': 'publishers/kserve-ci-e2e-test/models/facebook/opt-125m'}}\nfunc_name = 'wait_for_model_response'\ntimestamp_start = '2026-07-13T21:17:34.210783', start_time = 1783977454.211127\nduration = 905.0296046733856, timestamp_end = '2026-07-13T21:32:39.240732'\n\n @functools.wraps(func)\n def wrapper(*args, **kwargs):\n func_name = func.__name__\n \n timestamp_start = datetime.now().isoformat()\n logger.info(\n f\"[{func_name}] [{timestamp_start}] start - args={args}, kwargs={kwargs}\"\n )\n start_time = time.time()\n \n try:\n> result = func(*args, **kwargs)\n\nllmisvc/logging.py:40: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nkserve_client = \ntest_case = TestCase(base_refs=['router-managed', 'workload-single-cpu', 'model-fb-opt-125m-with-lora-hf'], prompt=None, service_n... {'name': 'model-fb-opt-125m-with-lora-hf-c0d503b0'}]},\n 'status': None}, model_name='facebook/opt-125m')\ntimeout_seconds = 900\nextra_headers = {'X-Gateway-Model-Name': 'publishers/kserve-ci-e2e-test/models/facebook/opt-125m'}\n\n @log_execution\n def wait_for_model_response(\n kserve_client: KServeClient,\n test_case: TestCase, # noqa: F811\n timeout_seconds: int = 900,\n extra_headers: Optional[Dict[str, str]] = None,\n ) -> str:\n def get_successful_response():\n try:\n if test_case.url_getter:\n service_url = test_case.url_getter(kserve_client, test_case.llm_service)\n else:\n service_url = get_llm_service_url(kserve_client, test_case.llm_service)\n except Exception as e:\n raise AssertionError(f\"\u274c Failed to get service URL: {e}\") from e\n \n model_url = service_url + test_case.endpoint\n \n headers = {\"Content-Type\": \"application/json\"}\n if extra_headers:\n headers.update(extra_headers)\n \n if test_case.payload_formatter is not None:\n test_payload = test_case.payload_formatter(test_case)\n elif test_case.prompt is not None:\n test_payload = {\n \"model\": test_case.model_name\n if not extra_headers or MODEL_ROUTING_HEADER not in extra_headers\n else extra_headers[MODEL_ROUTING_HEADER],\n \"prompt\": test_case.prompt,\n \"max_tokens\": test_case.max_tokens,\n }\n else:\n test_payload = None\n \n logger.info(f\"Calling LLM service at {model_url} with payload {test_payload}\")\n try:\n if test_payload is not None:\n response = post_with_retry(\n model_url,\n headers=headers,\n json_data=test_payload,\n timeout=test_case.response_timeout,\n )\n else:\n response = get_with_retry(\n model_url,\n headers=headers,\n timeout=test_case.response_timeout,\n )\n except Exception as e:\n logger.error(f\"\u274c Failed to call model: {e}\")\n raise AssertionError(f\"\u274c Failed to call model: {e}\") from e\n \n logger.info(f\"Model response is {response.status_code}: {response.text[:500]}\")\n \n if 200 <= response.status_code < 300:\n return response\n raise AssertionError(\n f\"Service returned {response.status_code}: {response.text}\"\n )\n \n> response = wait_for(get_successful_response, timeout=timeout_seconds, interval=5.0)\n\nllmisvc/test_llm_inference_service.py:1119: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nassertion_fn = .get_successful_response at 0x7f8e716a7560>\ntimeout = 900, interval = 5.0\n\n def wait_for(\n assertion_fn: Callable[[], Any], timeout: float = 5.0, interval: float = 0.1\n ) -> Any:\n \"\"\"Wait for the assertion to succeed within timeout.\"\"\"\n deadline = time.time() + timeout\n last_msg = None\n while True:\n try:\n> return assertion_fn()\n\nllmisvc/test_llm_inference_service.py:1215: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\n def get_successful_response():\n try:\n if test_case.url_getter:\n service_url = test_case.url_getter(kserve_client, test_case.llm_service)\n else:\n service_url = get_llm_service_url(kserve_client, test_case.llm_service)\n except Exception as e:\n raise AssertionError(f\"\u274c Failed to get service URL: {e}\") from e\n \n model_url = service_url + test_case.endpoint\n \n headers = {\"Content-Type\": \"application/json\"}\n if extra_headers:\n headers.update(extra_headers)\n \n if test_case.payload_formatter is not None:\n test_payload = test_case.payload_formatter(test_case)\n elif test_case.prompt is not None:\n test_payload = {\n \"model\": test_case.model_name\n if not extra_headers or MODEL_ROUTING_HEADER not in extra_headers\n else extra_headers[MODEL_ROUTING_HEADER],\n \"prompt\": test_case.prompt,\n \"max_tokens\": test_case.max_tokens,\n }\n else:\n test_payload = None\n \n logger.info(f\"Calling LLM service at {model_url} with payload {test_payload}\")\n try:\n if test_payload is not None:\n response = post_with_retry(\n model_url,\n headers=headers,\n json_data=test_payload,\n timeout=test_case.response_timeout,\n )\n else:\n response = get_with_retry(\n model_url,\n headers=headers,\n timeout=test_case.response_timeout,\n )\n except Exception as e:\n logger.error(f\"\u274c Failed to call model: {e}\")\n raise AssertionError(f\"\u274c Failed to call model: {e}\") from e\n \n logger.info(f\"Model response is {response.status_code}: {response.text[:500]}\")\n \n if 200 <= response.status_code < 300:\n return response\n> raise AssertionError(\n f\"Service returned {response.status_code}: {response.text}\"\n )\nE AssertionError: Service returned 401:\n\nllmisvc/test_llm_inference_service.py:1115: AssertionError"}, "teardown": {"duration": 0.0026280479942215607, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-custom-route-timeout-pd-scheduler-managed-workload-pd-cpu-model-fb-opt-125m]", "lineno": 244, "outcome": "passed", "keywords": ["test_llm_inference_service[router-custom-route-timeout-pd-scheduler-managed-workload-pd-cpu-model-fb-opt-125m]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "__wrapped__", "pytestmark", "router-custom-route-timeout-pd-scheduler-managed-workload-pd-cpu-model-fb-opt-125m", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.2064702599964221, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 195.2720904279995, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.003047714999411255, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-with-refs-pd-scheduler-managed-workload-pd-cpu-model-fb-opt-125m]", "lineno": 244, "outcome": "failed", "keywords": ["test_llm_inference_service[router-with-refs-pd-scheduler-managed-workload-pd-cpu-model-fb-opt-125m]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "custom_gateway", "__wrapped__", "pytestmark", "router-with-refs-pd-scheduler-managed-workload-pd-cpu-model-fb-opt-125m", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 4.363572579000902, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 904.1925835430011, "outcome": "failed", "crash": {"path": "/workspace/source/test/e2e/llmisvc/test_llm_inference_service.py", "lineno": 1199, "message": "AssertionError: Missing true conditions: {'Ready', 'WorkloadsReady'}, expected {'Ready', 'RouterReady', 'WorkloadsReady'}, got [{'lastTransitionTime': '2026-07-13T21:21:57Z', 'severity': 'Info', 'status': 'True', 'type': 'GatewaysReady'}, {'lastTransitionTime': '2026-07-13T21:21:57Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-13T21:21:57Z', 'severity': 'Info', 'status': 'True', 'type': 'InferencePoolReady'}, {'lastTransitionTime': '2026-07-13T21:23:48Z', 'severity': 'Info', 'status': 'True', 'type': 'MainWorkloadReady'}, {'lastTransitionTime': '2026-07-13T21:21:57Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'severity': 'Info', 'status': 'False', 'type': 'PrefillWorkloadReady'}, {'lastTransitionTime': '2026-07-13T21:21:57Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-13T21:21:57Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-13T21:22:22Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-13T21:22:22Z', 'severity': 'Info', 'status': 'True', 'type': 'SchedulerWorkloadReady'}, {'lastTransitionTime': '2026-07-13T21:21:57Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'WorkloadsReady'}]"}, "traceback": [{"path": "llmisvc/test_llm_inference_service.py", "lineno": 812, "message": ""}, {"path": "llmisvc/logging.py", "lineno": 40, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1204, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1215, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1199, "message": "AssertionError"}], "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python\n\ntest_case = TestCase(base_refs=['router-with-refs-pd', 'scheduler-managed', 'workload-pd-cpu', 'model-fb-opt-125m'], prompt='You a... {'name': 'model-fb-opt-125m-router-with-r-c22ea8a0'}]},\n 'status': None}, model_name='facebook/opt-125m')\n\n @pytest.mark.llminferenceservice\n @pytest.mark.asyncio(loop_scope=\"session\")\n @pytest.mark.parametrize(\n \"test_case\",\n [\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-gateway-ref\",\n \"router-with-managed-route\",\n \"model-fb-opt-125m\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"custom-route-timeout-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"router-with-refs-test\",\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n routes=[ROUTER_ROUTES[0], ROUTER_ROUTES[1]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\"router-managed\", \"workload-pd-cpu\", \"model-fb-opt-125m\"],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"custom-route-timeout-pd-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"router-with-refs-pd-test\",\n response_assertion=assert_200_with_choices,\n expected_gateway=ROUTER_GATEWAYS[1],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[1]],\n routes=[ROUTER_ROUTES[2], ROUTER_ROUTES[3]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-dp-ep-gpu\",\n \"workload-dp-ep-prefill-gpu\",\n \"model-deepseek-v2-lite\",\n ],\n prompt=\"Delve into the multifaceted implications of a fully disaggregated cloud architecture, specifically \"\n \"where the compute plane (P) and the data plane (D) are independently deployed and managed for a \"\n \"geographically distributed, high-throughput, low-latency microservices ecosystem. Beyond the \"\n \"fundamental challenges of network latency and data consistency, elaborate on the advanced \"\n \"considerations and trade-offs inherent in such a setup: 1. Network Architecture and Protocols: \"\n \"How would the network fabric and underlying protocols (e.g., RDMA, custom transport layers) need to \"\n \"evolve to support optimal performance and minimize inter-plane communication overhead, especially for \"\n \"synchronous operations? Discuss the role of network programmability (e.g., SDN, P4) in dynamically \"\n \"optimizing routing and traffic flow between P and D. 2. Advanced Data Consistency and Durability: \"\n \"Explore sophisticated data consistency models (e.g., causal consistency, strong eventual consistency) \"\n \"and their applicability in balancing performance and data integrity across a globally distributed data plane. \"\n \"Detail strategies for ensuring data durability and fault tolerance, including multi-region replication, \"\n \"intelligent partitioning, and recovery mechanisms in the event of partial or full plane failures. \"\n \"3. Dynamic Resource Orchestration and Cost Optimization: Analyze how an orchestration layer would intelligently \"\n \"manage the independent scaling of compute (P) and data (D) resources, considering fluctuating workloads, \"\n \"cost efficiency, and performance targets (e.g., using predictive analytics for resource provisioning). \"\n \"Discuss mechanisms for dynamically reallocating compute nodes to different data partitions based on \"\n \"workload patterns and data locality, potentially involving live migration strategies. \"\n \"4. Security and Compliance in a Distributed Landscape: Address the enhanced security perimeter \"\n \"challenges, including securing communication channels between P and D (encryption in transit, mutual TLS), \"\n \"fine-grained access control to data at rest and in motion, and identity management across disaggregated \"\n \"components. Discuss how such an architecture impacts compliance with regulatory frameworks (e.g., GDPR, HIPAA) \"\n \"concerning data sovereignty, privacy, and auditability. 5. Operational Complexity and Observability: \"\n \"Examine the increased complexity in monitoring, logging, and tracing across highly decoupled compute and \"\n \"data planes. What specialized tooling and practices (e.g., distributed tracing with OpenTelemetry, advanced AIOps) \"\n \"would be essential? How would incident response and troubleshooting differ in this disaggregated environment \"\n \"compared to traditional integrated systems? Consider the challenges of pinpointing root causes across \"\n \"independent failures. 6. Real-world Applicability and Future Trends: Identify specific industries \"\n \"or use cases (e.g., high-frequency trading, IoT edge processing, large language model inference) \"\n \"where the benefits of P/D disaggregation would strongly outweigh its complexities. \"\n \"Conclude by speculating on emerging technologies or paradigms (e.g., serverless compute functions \"\n \"directly interacting with object storage, in-memory disaggregation) that could further drive or \"\n \"transform P/D disaggregation in cloud computing.\",\n max_tokens=2000,\n ),\n marks=[\n pytest.mark.cluster_gpu,\n pytest.mark.cluster_nvidia,\n pytest.mark.cluster_nvidia_roce,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-no-scheduler\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"What is KServe?\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.no_scheduler,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"This test simulates DP+EP that can run on CPU, the idea is to test the LWS-based deployment, \"\n \"but without the resources requirements for DP+EP (GPUs and ROCe/IB).\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_multi_node],\n ),\n # Scheduler config tests\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-inline-config\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-inline-config-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Chat completions endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n model_name=\"Qwen/Qwen2.5-0.5B-Instruct\",\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-configmap-ref\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-configmap-ref-test\",\n before_test=[create_scheduler_configmap],\n after_test=[delete_scheduler_configmap],\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-replicas\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-ha-replicas-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-custom-template\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-custom-template-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Scheduler v0.6 \u2192 v0.7 migration tests.\n # Deploy v0.6-style configs and verify the controller migrates them\n # so the v0.7 scheduler boots successfully.\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-pd-config-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-pd-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-nonzero-threshold-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-threshold-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Precise prefix KV cache routing test\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-precise-prefix-cache-inline-config\",\n \"workload-llmd-simulator-kvcache\",\n ],\n prompt=\"KServe is a\",\n service_name=\"precise-prefix-cache-test\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Models endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=create_response_assertion(with_field=\"data\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/chat/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 LoRA adapter\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n model_name=f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/models (base + LoRA)\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=assert_models_contains(\n \"facebook/opt-125m\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n \"lora-adapter-1\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # PVC storage tests -- validate direct PVC volume mount with real vLLM serving\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-pd-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_multi_node,\n pytest.mark.pvc_storage,\n ],\n ),\n ],\n indirect=[\"test_case\"],\n ids=generate_test_id,\n )\n @log_execution\n def test_llm_inference_service(test_case: TestCase): # noqa: F811\n inject_k8s_proxy()\n \n kserve_client = KServeClient(\n config_file=os.environ.get(\"KUBECONFIG\", \"~/.kube/config\"),\n client_configuration=client.Configuration(),\n )\n \n service_name = test_case.llm_service.metadata.name\n if not test_case.llm_service.metadata.annotations:\n test_case.llm_service.metadata.annotations = {}\n \n test_case.llm_service.metadata.annotations[\n \"security.opendatahub.io/enable-auth\"\n ] = \"false\"\n prefix = test_case.log_prefix\n \n test_failed = False\n try:\n print(f\"{prefix} Creating LLMInferenceService {service_name}\")\n create_llmisvc(kserve_client, test_case.llm_service)\n print(f\"{prefix} Waiting for LLMInferenceService {service_name} to be ready\")\n> wait_for_llm_isvc_ready(\n kserve_client, test_case.llm_service, test_case.wait_timeout\n )\n\nllmisvc/test_llm_inference_service.py:812: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nargs = (, {'api_version': 'serving.kserve.io/v1alpha1',\n 'kin...h-ref-d1f07093'},\n {'name': 'model-fb-opt-125m-router-with-r-c22ea8a0'}]},\n 'status': None}, 900)\nkwargs = {}, func_name = 'wait_for_llm_isvc_ready'\ntimestamp_start = '2026-07-13T21:21:19.291784', start_time = 1783977679.2920523\nduration = 901.0002934932709, timestamp_end = '2026-07-13T21:36:20.292350'\n\n @functools.wraps(func)\n def wrapper(*args, **kwargs):\n func_name = func.__name__\n \n timestamp_start = datetime.now().isoformat()\n logger.info(\n f\"[{func_name}] [{timestamp_start}] start - args={args}, kwargs={kwargs}\"\n )\n start_time = time.time()\n \n try:\n> result = func(*args, **kwargs)\n\nllmisvc/logging.py:40: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nkserve_client = \ngiven = {'api_version': 'serving.kserve.io/v1alpha1',\n 'kind': 'LLMInferenceService',\n 'metadata': {'annotations': {'security....er-with-ref-d1f07093'},\n {'name': 'model-fb-opt-125m-router-with-r-c22ea8a0'}]},\n 'status': None}\ntimeout_seconds = 900\n\n @log_execution\n def wait_for_llm_isvc_ready(\n kserve_client: KServeClient,\n given: V1alpha1LLMInferenceService,\n timeout_seconds: int = 900,\n ) -> str:\n def assert_llm_isvc_ready():\n out = get_llmisvc(\n kserve_client,\n given.metadata.name,\n given.metadata.namespace,\n given.api_version.split(\"/\")[1],\n )\n \n if \"status\" not in out:\n raise AssertionError(\"No status found in LLM inference service\")\n \n status = out[\"status\"]\n if \"conditions\" not in status:\n raise AssertionError(\"No conditions found in status\")\n \n expected_true_conditions = {\"Ready\", \"WorkloadsReady\", \"RouterReady\"}\n got_true_conditions = set()\n \n conditions = status[\"conditions\"]\n \n for condition in conditions:\n if condition.get(\"status\") == \"True\":\n got_true_conditions.add(condition.get(\"type\"))\n \n missing_conditions = expected_true_conditions - got_true_conditions\n if missing_conditions:\n raise AssertionError(\n f\"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}\"\n )\n return True\n \n> return wait_for(assert_llm_isvc_ready, timeout=timeout_seconds, interval=1.0)\n\nllmisvc/test_llm_inference_service.py:1204: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nassertion_fn = .assert_llm_isvc_ready at 0x7fb7b5e7e2a0>\ntimeout = 900, interval = 1.0\n\n def wait_for(\n assertion_fn: Callable[[], Any], timeout: float = 5.0, interval: float = 0.1\n ) -> Any:\n \"\"\"Wait for the assertion to succeed within timeout.\"\"\"\n deadline = time.time() + timeout\n last_msg = None\n while True:\n try:\n> return assertion_fn()\n\nllmisvc/test_llm_inference_service.py:1215: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\n def assert_llm_isvc_ready():\n out = get_llmisvc(\n kserve_client,\n given.metadata.name,\n given.metadata.namespace,\n given.api_version.split(\"/\")[1],\n )\n \n if \"status\" not in out:\n raise AssertionError(\"No status found in LLM inference service\")\n \n status = out[\"status\"]\n if \"conditions\" not in status:\n raise AssertionError(\"No conditions found in status\")\n \n expected_true_conditions = {\"Ready\", \"WorkloadsReady\", \"RouterReady\"}\n got_true_conditions = set()\n \n conditions = status[\"conditions\"]\n \n for condition in conditions:\n if condition.get(\"status\") == \"True\":\n got_true_conditions.add(condition.get(\"type\"))\n \n missing_conditions = expected_true_conditions - got_true_conditions\n if missing_conditions:\n> raise AssertionError(\n f\"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}\"\n )\nE AssertionError: Missing true conditions: {'Ready', 'WorkloadsReady'}, expected {'Ready', 'RouterReady', 'WorkloadsReady'}, got [{'lastTransitionTime': '2026-07-13T21:21:57Z', 'severity': 'Info', 'status': 'True', 'type': 'GatewaysReady'}, {'lastTransitionTime': '2026-07-13T21:21:57Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-13T21:21:57Z', 'severity': 'Info', 'status': 'True', 'type': 'InferencePoolReady'}, {'lastTransitionTime': '2026-07-13T21:23:48Z', 'severity': 'Info', 'status': 'True', 'type': 'MainWorkloadReady'}, {'lastTransitionTime': '2026-07-13T21:21:57Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'severity': 'Info', 'status': 'False', 'type': 'PrefillWorkloadReady'}, {'lastTransitionTime': '2026-07-13T21:21:57Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-13T21:21:57Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-13T21:22:22Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-13T21:22:22Z', 'severity': 'Info', 'status': 'True', 'type': 'SchedulerWorkloadReady'}, {'lastTransitionTime': '2026-07-13T21:21:57Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'WorkloadsReady'}]\n\nllmisvc/test_llm_inference_service.py:1199: AssertionError"}, "teardown": {"duration": 0.0030790559976594523, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-managed-workload-single-cpu-model-pvc]", "lineno": 244, "outcome": "passed", "keywords": ["test_llm_inference_service[router-managed-workload-single-cpu-model-pvc]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "pvc_storage", "__wrapped__", "pytestmark", "router-managed-workload-single-cpu-model-pvc", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 20.379598922001605, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 161.73275701500097, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0018206399981863797, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-managed-workload-pd-cpu-model-pvc]", "lineno": 244, "outcome": "passed", "keywords": ["test_llm_inference_service[router-managed-workload-pd-cpu-model-pvc]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "pvc_storage", "__wrapped__", "pytestmark", "router-managed-workload-pd-cpu-model-pvc", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.46831684700009646, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 193.49050784800056, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0025750549975782633, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_single_node-router-no-scheduler-workload-single-cpu-model-fb-opt-125m]", "lineno": 244, "outcome": "failed", "keywords": ["test_llm_inference_service[router-no-scheduler-workload-single-cpu-model-fb-opt-125m]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "no_scheduler", "__wrapped__", "pytestmark", "router-no-scheduler-workload-single-cpu-model-fb-opt-125m", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.35430333299882477, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 902.0195227290023, "outcome": "failed", "crash": {"path": "/workspace/source/test/e2e/llmisvc/test_llm_inference_service.py", "lineno": 1199, "message": "AssertionError: Missing true conditions: {'Ready', 'WorkloadsReady'}, expected {'Ready', 'RouterReady', 'WorkloadsReady'}, got [{'lastTransitionTime': '2026-07-13T21:37:05Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-13T21:37:05Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'severity': 'Info', 'status': 'False', 'type': 'MainWorkloadReady'}, {'lastTransitionTime': '2026-07-13T21:36:43Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-13T21:37:05Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-13T21:37:05Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-13T21:37:05Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'WorkloadsReady'}]"}, "traceback": [{"path": "llmisvc/test_llm_inference_service.py", "lineno": 812, "message": ""}, {"path": "llmisvc/logging.py", "lineno": 40, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1204, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1215, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1199, "message": "AssertionError"}], "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python\n\ntest_case = TestCase(base_refs=['router-no-scheduler', 'workload-single-cpu', 'model-fb-opt-125m'], prompt='What is KServe?', serv... {'name': 'model-fb-opt-125m-llmisvc-model-7a2ca70d'}]},\n 'status': None}, model_name='facebook/opt-125m')\n\n @pytest.mark.llminferenceservice\n @pytest.mark.asyncio(loop_scope=\"session\")\n @pytest.mark.parametrize(\n \"test_case\",\n [\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-gateway-ref\",\n \"router-with-managed-route\",\n \"model-fb-opt-125m\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"custom-route-timeout-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"router-with-refs-test\",\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n routes=[ROUTER_ROUTES[0], ROUTER_ROUTES[1]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\"router-managed\", \"workload-pd-cpu\", \"model-fb-opt-125m\"],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"custom-route-timeout-pd-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"router-with-refs-pd-test\",\n response_assertion=assert_200_with_choices,\n expected_gateway=ROUTER_GATEWAYS[1],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[1]],\n routes=[ROUTER_ROUTES[2], ROUTER_ROUTES[3]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-dp-ep-gpu\",\n \"workload-dp-ep-prefill-gpu\",\n \"model-deepseek-v2-lite\",\n ],\n prompt=\"Delve into the multifaceted implications of a fully disaggregated cloud architecture, specifically \"\n \"where the compute plane (P) and the data plane (D) are independently deployed and managed for a \"\n \"geographically distributed, high-throughput, low-latency microservices ecosystem. Beyond the \"\n \"fundamental challenges of network latency and data consistency, elaborate on the advanced \"\n \"considerations and trade-offs inherent in such a setup: 1. Network Architecture and Protocols: \"\n \"How would the network fabric and underlying protocols (e.g., RDMA, custom transport layers) need to \"\n \"evolve to support optimal performance and minimize inter-plane communication overhead, especially for \"\n \"synchronous operations? Discuss the role of network programmability (e.g., SDN, P4) in dynamically \"\n \"optimizing routing and traffic flow between P and D. 2. Advanced Data Consistency and Durability: \"\n \"Explore sophisticated data consistency models (e.g., causal consistency, strong eventual consistency) \"\n \"and their applicability in balancing performance and data integrity across a globally distributed data plane. \"\n \"Detail strategies for ensuring data durability and fault tolerance, including multi-region replication, \"\n \"intelligent partitioning, and recovery mechanisms in the event of partial or full plane failures. \"\n \"3. Dynamic Resource Orchestration and Cost Optimization: Analyze how an orchestration layer would intelligently \"\n \"manage the independent scaling of compute (P) and data (D) resources, considering fluctuating workloads, \"\n \"cost efficiency, and performance targets (e.g., using predictive analytics for resource provisioning). \"\n \"Discuss mechanisms for dynamically reallocating compute nodes to different data partitions based on \"\n \"workload patterns and data locality, potentially involving live migration strategies. \"\n \"4. Security and Compliance in a Distributed Landscape: Address the enhanced security perimeter \"\n \"challenges, including securing communication channels between P and D (encryption in transit, mutual TLS), \"\n \"fine-grained access control to data at rest and in motion, and identity management across disaggregated \"\n \"components. Discuss how such an architecture impacts compliance with regulatory frameworks (e.g., GDPR, HIPAA) \"\n \"concerning data sovereignty, privacy, and auditability. 5. Operational Complexity and Observability: \"\n \"Examine the increased complexity in monitoring, logging, and tracing across highly decoupled compute and \"\n \"data planes. What specialized tooling and practices (e.g., distributed tracing with OpenTelemetry, advanced AIOps) \"\n \"would be essential? How would incident response and troubleshooting differ in this disaggregated environment \"\n \"compared to traditional integrated systems? Consider the challenges of pinpointing root causes across \"\n \"independent failures. 6. Real-world Applicability and Future Trends: Identify specific industries \"\n \"or use cases (e.g., high-frequency trading, IoT edge processing, large language model inference) \"\n \"where the benefits of P/D disaggregation would strongly outweigh its complexities. \"\n \"Conclude by speculating on emerging technologies or paradigms (e.g., serverless compute functions \"\n \"directly interacting with object storage, in-memory disaggregation) that could further drive or \"\n \"transform P/D disaggregation in cloud computing.\",\n max_tokens=2000,\n ),\n marks=[\n pytest.mark.cluster_gpu,\n pytest.mark.cluster_nvidia,\n pytest.mark.cluster_nvidia_roce,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-no-scheduler\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"What is KServe?\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.no_scheduler,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"This test simulates DP+EP that can run on CPU, the idea is to test the LWS-based deployment, \"\n \"but without the resources requirements for DP+EP (GPUs and ROCe/IB).\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_multi_node],\n ),\n # Scheduler config tests\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-inline-config\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-inline-config-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Chat completions endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n model_name=\"Qwen/Qwen2.5-0.5B-Instruct\",\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-configmap-ref\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-configmap-ref-test\",\n before_test=[create_scheduler_configmap],\n after_test=[delete_scheduler_configmap],\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-replicas\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-ha-replicas-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-custom-template\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-custom-template-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Scheduler v0.6 \u2192 v0.7 migration tests.\n # Deploy v0.6-style configs and verify the controller migrates them\n # so the v0.7 scheduler boots successfully.\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-pd-config-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-pd-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-nonzero-threshold-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-threshold-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Precise prefix KV cache routing test\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-precise-prefix-cache-inline-config\",\n \"workload-llmd-simulator-kvcache\",\n ],\n prompt=\"KServe is a\",\n service_name=\"precise-prefix-cache-test\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Models endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=create_response_assertion(with_field=\"data\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/chat/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 LoRA adapter\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n model_name=f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/models (base + LoRA)\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=assert_models_contains(\n \"facebook/opt-125m\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n \"lora-adapter-1\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # PVC storage tests -- validate direct PVC volume mount with real vLLM serving\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-pd-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_multi_node,\n pytest.mark.pvc_storage,\n ],\n ),\n ],\n indirect=[\"test_case\"],\n ids=generate_test_id,\n )\n @log_execution\n def test_llm_inference_service(test_case: TestCase): # noqa: F811\n inject_k8s_proxy()\n \n kserve_client = KServeClient(\n config_file=os.environ.get(\"KUBECONFIG\", \"~/.kube/config\"),\n client_configuration=client.Configuration(),\n )\n \n service_name = test_case.llm_service.metadata.name\n if not test_case.llm_service.metadata.annotations:\n test_case.llm_service.metadata.annotations = {}\n \n test_case.llm_service.metadata.annotations[\n \"security.opendatahub.io/enable-auth\"\n ] = \"false\"\n prefix = test_case.log_prefix\n \n test_failed = False\n try:\n print(f\"{prefix} Creating LLMInferenceService {service_name}\")\n create_llmisvc(kserve_client, test_case.llm_service)\n print(f\"{prefix} Waiting for LLMInferenceService {service_name} to be ready\")\n> wait_for_llm_isvc_ready(\n kserve_client, test_case.llm_service, test_case.wait_timeout\n )\n\nllmisvc/test_llm_inference_service.py:812: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nargs = (, {'api_version': 'serving.kserve.io/v1alpha1',\n 'kin...c-mod-5c1136e5'},\n {'name': 'model-fb-opt-125m-llmisvc-model-7a2ca70d'}]},\n 'status': None}, 900)\nkwargs = {}, func_name = 'wait_for_llm_isvc_ready'\ntimestamp_start = '2026-07-13T21:36:22.914688', start_time = 1783978582.9150155\nduration = 900.2256450653076, timestamp_end = '2026-07-13T21:51:23.140681'\n\n @functools.wraps(func)\n def wrapper(*args, **kwargs):\n func_name = func.__name__\n \n timestamp_start = datetime.now().isoformat()\n logger.info(\n f\"[{func_name}] [{timestamp_start}] start - args={args}, kwargs={kwargs}\"\n )\n start_time = time.time()\n \n try:\n> result = func(*args, **kwargs)\n\nllmisvc/logging.py:40: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nkserve_client = \ngiven = {'api_version': 'serving.kserve.io/v1alpha1',\n 'kind': 'LLMInferenceService',\n 'metadata': {'annotations': {'security....llmisvc-mod-5c1136e5'},\n {'name': 'model-fb-opt-125m-llmisvc-model-7a2ca70d'}]},\n 'status': None}\ntimeout_seconds = 900\n\n @log_execution\n def wait_for_llm_isvc_ready(\n kserve_client: KServeClient,\n given: V1alpha1LLMInferenceService,\n timeout_seconds: int = 900,\n ) -> str:\n def assert_llm_isvc_ready():\n out = get_llmisvc(\n kserve_client,\n given.metadata.name,\n given.metadata.namespace,\n given.api_version.split(\"/\")[1],\n )\n \n if \"status\" not in out:\n raise AssertionError(\"No status found in LLM inference service\")\n \n status = out[\"status\"]\n if \"conditions\" not in status:\n raise AssertionError(\"No conditions found in status\")\n \n expected_true_conditions = {\"Ready\", \"WorkloadsReady\", \"RouterReady\"}\n got_true_conditions = set()\n \n conditions = status[\"conditions\"]\n \n for condition in conditions:\n if condition.get(\"status\") == \"True\":\n got_true_conditions.add(condition.get(\"type\"))\n \n missing_conditions = expected_true_conditions - got_true_conditions\n if missing_conditions:\n raise AssertionError(\n f\"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}\"\n )\n return True\n \n> return wait_for(assert_llm_isvc_ready, timeout=timeout_seconds, interval=1.0)\n\nllmisvc/test_llm_inference_service.py:1204: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nassertion_fn = .assert_llm_isvc_ready at 0x7fb7b5e7c180>\ntimeout = 900, interval = 1.0\n\n def wait_for(\n assertion_fn: Callable[[], Any], timeout: float = 5.0, interval: float = 0.1\n ) -> Any:\n \"\"\"Wait for the assertion to succeed within timeout.\"\"\"\n deadline = time.time() + timeout\n last_msg = None\n while True:\n try:\n> return assertion_fn()\n\nllmisvc/test_llm_inference_service.py:1215: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\n def assert_llm_isvc_ready():\n out = get_llmisvc(\n kserve_client,\n given.metadata.name,\n given.metadata.namespace,\n given.api_version.split(\"/\")[1],\n )\n \n if \"status\" not in out:\n raise AssertionError(\"No status found in LLM inference service\")\n \n status = out[\"status\"]\n if \"conditions\" not in status:\n raise AssertionError(\"No conditions found in status\")\n \n expected_true_conditions = {\"Ready\", \"WorkloadsReady\", \"RouterReady\"}\n got_true_conditions = set()\n \n conditions = status[\"conditions\"]\n \n for condition in conditions:\n if condition.get(\"status\") == \"True\":\n got_true_conditions.add(condition.get(\"type\"))\n \n missing_conditions = expected_true_conditions - got_true_conditions\n if missing_conditions:\n> raise AssertionError(\n f\"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}\"\n )\nE AssertionError: Missing true conditions: {'Ready', 'WorkloadsReady'}, expected {'Ready', 'RouterReady', 'WorkloadsReady'}, got [{'lastTransitionTime': '2026-07-13T21:37:05Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-13T21:37:05Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'severity': 'Info', 'status': 'False', 'type': 'MainWorkloadReady'}, {'lastTransitionTime': '2026-07-13T21:36:43Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-13T21:37:05Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-13T21:37:05Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-13T21:37:05Z', 'message': 'Deployment does not have minimum availability.', 'reason': 'MinimumReplicasUnavailable', 'status': 'False', 'type': 'WorkloadsReady'}]\n\nllmisvc/test_llm_inference_service.py:1199: AssertionError"}, "teardown": {"duration": 0.0033288609993178397, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_multi_node-router-managed-workload-simulated-dp-ep-cpu-model-pvc]", "lineno": 244, "outcome": "passed", "keywords": ["test_llm_inference_service[router-managed-workload-simulated-dp-ep-cpu-model-pvc]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_multi_node", "pvc_storage", "__wrapped__", "pytestmark", "router-managed-workload-simulated-dp-ep-cpu-model-pvc", "llmd_simulator", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.24673150499438634, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 831.8194495520002, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0035062919996562414, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service.py::test_llm_inference_service[cluster_cpu-cluster_multi_node-router-managed-workload-simulated-dp-ep-cpu-model-fb-opt-125m]", "lineno": 244, "outcome": "failed", "keywords": ["test_llm_inference_service[router-managed-workload-simulated-dp-ep-cpu-model-fb-opt-125m]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_multi_node", "__wrapped__", "pytestmark", "router-managed-workload-simulated-dp-ep-cpu-model-fb-opt-125m", "llmd_simulator", "test_llm_inference_service.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.1649425890063867, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 902.5752790080005, "outcome": "failed", "crash": {"path": "/workspace/source/test/e2e/llmisvc/test_llm_inference_service.py", "lineno": 1199, "message": "AssertionError: Missing true conditions: {'Ready', 'WorkloadsReady'}, expected {'Ready', 'RouterReady', 'WorkloadsReady'}, got [{'lastTransitionTime': '2026-07-13T21:52:12Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-13T21:52:12Z', 'severity': 'Info', 'status': 'True', 'type': 'InferencePoolReady'}, {'lastTransitionTime': '2026-07-13T21:51:52Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-13T21:52:30Z', 'message': 'LWS is progressing', 'reason': 'Progressing', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-13T21:52:30Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-13T21:52:30Z', 'severity': 'Info', 'status': 'True', 'type': 'SchedulerWorkloadReady'}, {'lastTransitionTime': '2026-07-13T21:51:52Z', 'message': 'LWS is progressing', 'reason': 'Progressing', 'severity': 'Info', 'status': 'False', 'type': 'WorkerWorkloadReady'}, {'lastTransitionTime': '2026-07-13T21:51:52Z', 'message': 'LWS is progressing', 'reason': 'Progressing', 'status': 'False', 'type': 'WorkloadsReady'}]"}, "traceback": [{"path": "llmisvc/test_llm_inference_service.py", "lineno": 812, "message": ""}, {"path": "llmisvc/logging.py", "lineno": 40, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1204, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1215, "message": ""}, {"path": "llmisvc/test_llm_inference_service.py", "lineno": 1199, "message": "AssertionError"}], "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python\n\ntest_case = TestCase(base_refs=['router-managed', 'workload-simulated-dp-ep-cpu', 'model-fb-opt-125m'], prompt='This test simulate... {'name': 'model-fb-opt-125m-llmisvc-model-9f2e00e5'}]},\n 'status': None}, model_name='facebook/opt-125m')\n\n @pytest.mark.llminferenceservice\n @pytest.mark.asyncio(loop_scope=\"session\")\n @pytest.mark.parametrize(\n \"test_case\",\n [\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-gateway-ref\",\n \"router-with-managed-route\",\n \"model-fb-opt-125m\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"custom-route-timeout-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs\",\n \"scheduler-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"KServe is a\",\n service_name=\"router-with-refs-test\",\n expected_gateway=ROUTER_GATEWAYS[0],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[0]],\n routes=[ROUTER_ROUTES[0], ROUTER_ROUTES[1]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\"router-managed\", \"workload-pd-cpu\", \"model-fb-opt-125m\"],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-custom-route-timeout-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"custom-route-timeout-pd-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-with-refs-pd\",\n \"scheduler-managed\",\n \"workload-pd-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"You are an expert in Kubernetes-native machine learning serving platforms, with deep knowledge of the KServe project. \"\n \"Explain the challenges of serving large-scale models, GPU scheduling, and how KServe integrates with capabilities like multi-model serving. \"\n \"Provide a detailed comparison with open source alternatives, focusing on operational trade-offs.\",\n service_name=\"router-with-refs-pd-test\",\n response_assertion=assert_200_with_choices,\n expected_gateway=ROUTER_GATEWAYS[1],\n before_test=[\n lambda: create_router_resources(\n gateways=[ROUTER_GATEWAYS[1]],\n routes=[ROUTER_ROUTES[2], ROUTER_ROUTES[3]],\n )\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.custom_gateway,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-dp-ep-gpu\",\n \"workload-dp-ep-prefill-gpu\",\n \"model-deepseek-v2-lite\",\n ],\n prompt=\"Delve into the multifaceted implications of a fully disaggregated cloud architecture, specifically \"\n \"where the compute plane (P) and the data plane (D) are independently deployed and managed for a \"\n \"geographically distributed, high-throughput, low-latency microservices ecosystem. Beyond the \"\n \"fundamental challenges of network latency and data consistency, elaborate on the advanced \"\n \"considerations and trade-offs inherent in such a setup: 1. Network Architecture and Protocols: \"\n \"How would the network fabric and underlying protocols (e.g., RDMA, custom transport layers) need to \"\n \"evolve to support optimal performance and minimize inter-plane communication overhead, especially for \"\n \"synchronous operations? Discuss the role of network programmability (e.g., SDN, P4) in dynamically \"\n \"optimizing routing and traffic flow between P and D. 2. Advanced Data Consistency and Durability: \"\n \"Explore sophisticated data consistency models (e.g., causal consistency, strong eventual consistency) \"\n \"and their applicability in balancing performance and data integrity across a globally distributed data plane. \"\n \"Detail strategies for ensuring data durability and fault tolerance, including multi-region replication, \"\n \"intelligent partitioning, and recovery mechanisms in the event of partial or full plane failures. \"\n \"3. Dynamic Resource Orchestration and Cost Optimization: Analyze how an orchestration layer would intelligently \"\n \"manage the independent scaling of compute (P) and data (D) resources, considering fluctuating workloads, \"\n \"cost efficiency, and performance targets (e.g., using predictive analytics for resource provisioning). \"\n \"Discuss mechanisms for dynamically reallocating compute nodes to different data partitions based on \"\n \"workload patterns and data locality, potentially involving live migration strategies. \"\n \"4. Security and Compliance in a Distributed Landscape: Address the enhanced security perimeter \"\n \"challenges, including securing communication channels between P and D (encryption in transit, mutual TLS), \"\n \"fine-grained access control to data at rest and in motion, and identity management across disaggregated \"\n \"components. Discuss how such an architecture impacts compliance with regulatory frameworks (e.g., GDPR, HIPAA) \"\n \"concerning data sovereignty, privacy, and auditability. 5. Operational Complexity and Observability: \"\n \"Examine the increased complexity in monitoring, logging, and tracing across highly decoupled compute and \"\n \"data planes. What specialized tooling and practices (e.g., distributed tracing with OpenTelemetry, advanced AIOps) \"\n \"would be essential? How would incident response and troubleshooting differ in this disaggregated environment \"\n \"compared to traditional integrated systems? Consider the challenges of pinpointing root causes across \"\n \"independent failures. 6. Real-world Applicability and Future Trends: Identify specific industries \"\n \"or use cases (e.g., high-frequency trading, IoT edge processing, large language model inference) \"\n \"where the benefits of P/D disaggregation would strongly outweigh its complexities. \"\n \"Conclude by speculating on emerging technologies or paradigms (e.g., serverless compute functions \"\n \"directly interacting with object storage, in-memory disaggregation) that could further drive or \"\n \"transform P/D disaggregation in cloud computing.\",\n max_tokens=2000,\n ),\n marks=[\n pytest.mark.cluster_gpu,\n pytest.mark.cluster_nvidia,\n pytest.mark.cluster_nvidia_roce,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-no-scheduler\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"What is KServe?\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.no_scheduler,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-fb-opt-125m\",\n ],\n prompt=\"This test simulates DP+EP that can run on CPU, the idea is to test the LWS-based deployment, \"\n \"but without the resources requirements for DP+EP (GPUs and ROCe/IB).\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_multi_node],\n ),\n # Scheduler config tests\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-inline-config\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-inline-config-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Chat completions endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n model_name=\"Qwen/Qwen2.5-0.5B-Instruct\",\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=create_response_assertion(with_field=\"choices\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-configmap-ref\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-configmap-ref-test\",\n before_test=[create_scheduler_configmap],\n after_test=[delete_scheduler_configmap],\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-replicas\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-ha-replicas-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-custom-template\",\n \"workload-llmd-simulator\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-custom-template-test\",\n ),\n marks=[pytest.mark.cluster_cpu, pytest.mark.cluster_single_node],\n ),\n # Scheduler v0.6 \u2192 v0.7 migration tests.\n # Deploy v0.6-style configs and verify the controller migrates them\n # so the v0.7 scheduler boots successfully.\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-pd-config-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-pd-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-v06-nonzero-threshold-migration\",\n \"workload-llmd-simulator-pd\",\n ],\n prompt=\"KServe is a\",\n service_name=\"scheduler-v06-threshold-migration-test\",\n response_assertion=assert_200_with_choices,\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Precise prefix KV cache routing test\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"scheduler-with-precise-prefix-cache-inline-config\",\n \"workload-llmd-simulator-kvcache\",\n ],\n prompt=\"KServe is a\",\n service_name=\"precise-prefix-cache-test\",\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Models endpoint coverage\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=create_response_assertion(with_field=\"data\"),\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/chat/completions\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\"facebook/opt-125m\"),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n peers=[\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-llmd-simulator\",\n \"model-qwen2.5-0.5b\",\n ],\n endpoint=\"/v1/chat/completions\",\n prompt=\"What is KServe?\",\n payload_formatter=chat_completions_payload,\n response_assertion=assert_model_field_matches(\n \"Qwen/Qwen2.5-0.5B-Instruct\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/Qwen/Qwen2.5-0.5B-Instruct\",\n },\n ),\n ],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.llmd_simulator,\n pytest.mark.model_routing,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 LoRA adapter\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/completions\",\n prompt=\"KServe is a\",\n model_name=f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n payload_formatter=completions_payload,\n response_assertion=assert_model_field_matches(\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\"\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # Model-based routing via X-Gateway-Model-Name header \u2014 /v1/models (base + LoRA)\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-fb-opt-125m-with-lora-hf\",\n ],\n endpoint=\"/v1/models\",\n response_assertion=assert_models_contains(\n \"facebook/opt-125m\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n \"lora-adapter-1\",\n f\"publishers/{KSERVE_TEST_NAMESPACE}/models/lora-adapter-1\",\n ),\n url_getter=get_model_routing_url,\n extra_headers={\n MODEL_ROUTING_HEADER: f\"publishers/{KSERVE_TEST_NAMESPACE}/models/facebook/opt-125m\",\n },\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.model_routing,\n pytest.mark.lora,\n ],\n ),\n # PVC storage tests -- validate direct PVC volume mount with real vLLM serving\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-single-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-pd-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n response_assertion=assert_200_with_choices,\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_single_node,\n pytest.mark.pvc_storage,\n ],\n ),\n pytest.param(\n TestCase(\n base_refs=[\n \"router-managed\",\n \"workload-simulated-dp-ep-cpu\",\n \"model-pvc\",\n ],\n prompt=\"KServe is a\",\n before_test=[ensure_pvc_with_model],\n ),\n marks=[\n pytest.mark.cluster_cpu,\n pytest.mark.cluster_multi_node,\n pytest.mark.pvc_storage,\n ],\n ),\n ],\n indirect=[\"test_case\"],\n ids=generate_test_id,\n )\n @log_execution\n def test_llm_inference_service(test_case: TestCase): # noqa: F811\n inject_k8s_proxy()\n \n kserve_client = KServeClient(\n config_file=os.environ.get(\"KUBECONFIG\", \"~/.kube/config\"),\n client_configuration=client.Configuration(),\n )\n \n service_name = test_case.llm_service.metadata.name\n if not test_case.llm_service.metadata.annotations:\n test_case.llm_service.metadata.annotations = {}\n \n test_case.llm_service.metadata.annotations[\n \"security.opendatahub.io/enable-auth\"\n ] = \"false\"\n prefix = test_case.log_prefix\n \n test_failed = False\n try:\n print(f\"{prefix} Creating LLMInferenceService {service_name}\")\n create_llmisvc(kserve_client, test_case.llm_service)\n print(f\"{prefix} Waiting for LLMInferenceService {service_name} to be ready\")\n> wait_for_llm_isvc_ready(\n kserve_client, test_case.llm_service, test_case.wait_timeout\n )\n\nllmisvc/test_llm_inference_service.py:812: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nargs = (, {'api_version': 'serving.kserve.io/v1alpha1',\n 'kin...pu-ll-b2c82424'},\n {'name': 'model-fb-opt-125m-llmisvc-model-9f2e00e5'}]},\n 'status': None}, 900)\nkwargs = {}, func_name = 'wait_for_llm_isvc_ready'\ntimestamp_start = '2026-07-13T21:51:25.092848', start_time = 1783979485.0931387\nduration = 900.2995824813843, timestamp_end = '2026-07-13T22:06:25.392723'\n\n @functools.wraps(func)\n def wrapper(*args, **kwargs):\n func_name = func.__name__\n \n timestamp_start = datetime.now().isoformat()\n logger.info(\n f\"[{func_name}] [{timestamp_start}] start - args={args}, kwargs={kwargs}\"\n )\n start_time = time.time()\n \n try:\n> result = func(*args, **kwargs)\n\nllmisvc/logging.py:40: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nkserve_client = \ngiven = {'api_version': 'serving.kserve.io/v1alpha1',\n 'kind': 'LLMInferenceService',\n 'metadata': {'annotations': {'security....p-ep-cpu-ll-b2c82424'},\n {'name': 'model-fb-opt-125m-llmisvc-model-9f2e00e5'}]},\n 'status': None}\ntimeout_seconds = 900\n\n @log_execution\n def wait_for_llm_isvc_ready(\n kserve_client: KServeClient,\n given: V1alpha1LLMInferenceService,\n timeout_seconds: int = 900,\n ) -> str:\n def assert_llm_isvc_ready():\n out = get_llmisvc(\n kserve_client,\n given.metadata.name,\n given.metadata.namespace,\n given.api_version.split(\"/\")[1],\n )\n \n if \"status\" not in out:\n raise AssertionError(\"No status found in LLM inference service\")\n \n status = out[\"status\"]\n if \"conditions\" not in status:\n raise AssertionError(\"No conditions found in status\")\n \n expected_true_conditions = {\"Ready\", \"WorkloadsReady\", \"RouterReady\"}\n got_true_conditions = set()\n \n conditions = status[\"conditions\"]\n \n for condition in conditions:\n if condition.get(\"status\") == \"True\":\n got_true_conditions.add(condition.get(\"type\"))\n \n missing_conditions = expected_true_conditions - got_true_conditions\n if missing_conditions:\n raise AssertionError(\n f\"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}\"\n )\n return True\n \n> return wait_for(assert_llm_isvc_ready, timeout=timeout_seconds, interval=1.0)\n\nllmisvc/test_llm_inference_service.py:1204: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nassertion_fn = .assert_llm_isvc_ready at 0x7fb7b5e7f920>\ntimeout = 900, interval = 1.0\n\n def wait_for(\n assertion_fn: Callable[[], Any], timeout: float = 5.0, interval: float = 0.1\n ) -> Any:\n \"\"\"Wait for the assertion to succeed within timeout.\"\"\"\n deadline = time.time() + timeout\n last_msg = None\n while True:\n try:\n> return assertion_fn()\n\nllmisvc/test_llm_inference_service.py:1215: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\n def assert_llm_isvc_ready():\n out = get_llmisvc(\n kserve_client,\n given.metadata.name,\n given.metadata.namespace,\n given.api_version.split(\"/\")[1],\n )\n \n if \"status\" not in out:\n raise AssertionError(\"No status found in LLM inference service\")\n \n status = out[\"status\"]\n if \"conditions\" not in status:\n raise AssertionError(\"No conditions found in status\")\n \n expected_true_conditions = {\"Ready\", \"WorkloadsReady\", \"RouterReady\"}\n got_true_conditions = set()\n \n conditions = status[\"conditions\"]\n \n for condition in conditions:\n if condition.get(\"status\") == \"True\":\n got_true_conditions.add(condition.get(\"type\"))\n \n missing_conditions = expected_true_conditions - got_true_conditions\n if missing_conditions:\n> raise AssertionError(\n f\"Missing true conditions: {missing_conditions}, expected {expected_true_conditions}, got {conditions}\"\n )\nE AssertionError: Missing true conditions: {'Ready', 'WorkloadsReady'}, expected {'Ready', 'RouterReady', 'WorkloadsReady'}, got [{'lastTransitionTime': '2026-07-13T21:52:12Z', 'severity': 'Info', 'status': 'True', 'type': 'HTTPRoutesReady'}, {'lastTransitionTime': '2026-07-13T21:52:12Z', 'severity': 'Info', 'status': 'True', 'type': 'InferencePoolReady'}, {'lastTransitionTime': '2026-07-13T21:51:52Z', 'severity': 'Info', 'status': 'True', 'type': 'PresetsCombined'}, {'lastTransitionTime': '2026-07-13T21:52:30Z', 'message': 'LWS is progressing', 'reason': 'Progressing', 'status': 'False', 'type': 'Ready'}, {'lastTransitionTime': '2026-07-13T21:52:30Z', 'status': 'True', 'type': 'RouterReady'}, {'lastTransitionTime': '2026-07-13T21:52:30Z', 'severity': 'Info', 'status': 'True', 'type': 'SchedulerWorkloadReady'}, {'lastTransitionTime': '2026-07-13T21:51:52Z', 'message': 'LWS is progressing', 'reason': 'Progressing', 'severity': 'Info', 'status': 'False', 'type': 'WorkerWorkloadReady'}, {'lastTransitionTime': '2026-07-13T21:51:52Z', 'message': 'LWS is progressing', 'reason': 'Progressing', 'status': 'False', 'type': 'WorkloadsReady'}]\n\nllmisvc/test_llm_inference_service.py:1199: AssertionError"}, "teardown": {"duration": 0.0028353520028758794, "outcome": "passed", "longrepr": "[gw0] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service_conversion.py::TestLLMInferenceServiceConversion::test_v1alpha1_to_v1alpha2_conversion", "lineno": 212, "outcome": "passed", "keywords": ["test_v1alpha1_to_v1alpha2_conversion", "cluster_single_node", "cluster_cpu", "pytestmark", "TestLLMInferenceServiceConversion", "conversion", "llminferenceservice", "test_llm_inference_service_conversion.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.04783940900233574, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 0.06908022199786501, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.04773964399646502, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service_conversion.py::TestLLMInferenceServiceConversion::test_v1alpha2_to_v1alpha1_conversion", "lineno": 303, "outcome": "passed", "keywords": ["test_v1alpha2_to_v1alpha1_conversion", "cluster_single_node", "cluster_cpu", "pytestmark", "TestLLMInferenceServiceConversion", "conversion", "llminferenceservice", "test_llm_inference_service_conversion.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.043457904997922014, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 0.07023426199884852, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.017290114003117196, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service_conversion.py::TestLLMInferenceServiceConversion::test_criticality_preservation_via_annotations", "lineno": 394, "outcome": "passed", "keywords": ["test_criticality_preservation_via_annotations", "cluster_single_node", "cluster_cpu", "pytestmark", "TestLLMInferenceServiceConversion", "conversion", "llminferenceservice", "test_llm_inference_service_conversion.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.048811617998580914, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 0.07531941900379024, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.05320364300132496, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service_conversion.py::TestLLMInferenceServiceConversion::test_lora_criticality_preservation", "lineno": 531, "outcome": "passed", "keywords": ["test_lora_criticality_preservation", "cluster_single_node", "cluster_cpu", "pytestmark", "TestLLMInferenceServiceConversion", "conversion", "llminferenceservice", "test_llm_inference_service_conversion.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.0426624269966851, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 0.1575445159978699, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.030830381998384837, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service_conversion.py::TestLLMInferenceServiceConversion::test_round_trip_conversion_preserves_fields", "lineno": 680, "outcome": "passed", "keywords": ["test_round_trip_conversion_preserves_fields", "cluster_single_node", "cluster_cpu", "pytestmark", "TestLLMInferenceServiceConversion", "conversion", "llminferenceservice", "test_llm_inference_service_conversion.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.04013296400080435, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 0.3174123309945571, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.11340860999916913, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_inference_service_stop.py::test_llm_stop_feature[cluster_cpu-cluster_single_node-router-managed-workload-single-cpu-model-fb-opt-125m]", "lineno": 39, "outcome": "passed", "keywords": ["test_llm_stop_feature[router-managed-workload-single-cpu-model-fb-opt-125m]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "__wrapped__", "pytestmark", "router-managed-workload-single-cpu-model-fb-opt-125m", "test_llm_inference_service_stop.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.1725303109997185, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 307.7370326820019, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0020025609992444515, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_lora_adapters.py::test_llm_with_lora_adapters[cluster_cpu-single-lora-adapter-hf]", "lineno": 203, "outcome": "passed", "keywords": ["test_llm_with_lora_adapters[single-lora-adapter-hf]", "parametrize", "llminferenceservice", "cluster_cpu", "lora", "__wrapped__", "pytestmark", "single-lora-adapter-hf", "test_llm_lora_adapters.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.03126971600431716, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 163.85403823399974, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0019582360037020408, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_lora_adapters.py::test_llm_with_lora_adapters[cluster_cpu-multiple-lora-adapters]", "lineno": 203, "outcome": "passed", "keywords": ["test_llm_with_lora_adapters[multiple-lora-adapters]", "parametrize", "llminferenceservice", "cluster_cpu", "lora", "__wrapped__", "pytestmark", "multiple-lora-adapters", "test_llm_lora_adapters.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.04717933099891525, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 194.29080298799818, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0019396579955355264, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_llm_tls.py::test_llm_tls_resources[cluster_cpu-cluster_single_node-router-managed-workload-single-cpu-model-fb-opt-125m]", "lineno": 92, "outcome": "passed", "keywords": ["test_llm_tls_resources[router-managed-workload-single-cpu-model-fb-opt-125m]", "parametrize", "asyncio", "llminferenceservice", "cluster_cpu", "cluster_single_node", "__wrapped__", "pytestmark", "router-managed-workload-single-cpu-model-fb-opt-125m", "test_llm_tls.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.30059704400628107, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 165.9871261130029, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.0020575040034600534, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_prestop_hook.py::test_prestop_hook[cluster_cpu-cluster_single_node-router-managed-workload-single-cpu-model-fb-opt-125m]", "lineno": 40, "outcome": "passed", "keywords": ["test_prestop_hook[router-managed-workload-single-cpu-model-fb-opt-125m]", "parametrize", "llminferenceservice", "cluster_cpu", "cluster_single_node", "__wrapped__", "pytestmark", "router-managed-workload-single-cpu-model-fb-opt-125m", "test_prestop_hook.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.19642177199420985, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 280.920274326003, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "teardown": {"duration": 0.00590490699687507, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_rolling_upgrade.py::test_rolling_upgrade_coordination[cluster_cpu-cluster_single_node-router-managed-workload-llmd-simulator-model-fb-opt-125m]", "lineno": 40, "outcome": "error", "keywords": ["test_rolling_upgrade_coordination[router-managed-workload-llmd-simulator-model-fb-opt-125m]", "parametrize", "llminferenceservice", "cluster_cpu", "cluster_single_node", "__wrapped__", "pytestmark", "router-managed-workload-llmd-simulator-model-fb-opt-125m", "llmd_simulator", "test_rolling_upgrade.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.05458470400481019, "outcome": "failed", "crash": {"path": "/workspace/source/python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/rest.py", "lineno": 238, "message": "kubernetes.client.exceptions.ApiException: (500)\nReason: Internal Server Error\nHTTP response headers: HTTPHeaderDict({'Audit-Id': '190f2227-5483-4ba6-8b9f-ee7976c421ca', 'Cache-Control': 'no-cache, private', 'Content-Type': 'application/json', 'Strict-Transport-Security': 'max-age=31536000; includeSubDomains; preload', 'X-Kubernetes-Pf-Flowschema-Uid': 'c30fff65-8b8a-47f0-8303-5f205881ee46', 'X-Kubernetes-Pf-Prioritylevel-Uid': '1d0f3173-41ee-4f4d-bdbf-9bfbca43ee13', 'Date': 'Mon, 13 Jul 2026 22:11:24 GMT', 'Content-Length': '833'})\nHTTP response body: {\"kind\":\"Status\",\"apiVersion\":\"v1\",\"metadata\":{},\"status\":\"Failure\",\"message\":\"Internal error occurred: failed calling webhook \\\"llminferenceserviceconfig.kserve-webhook-server.v1alpha1.validator\\\": failed to call webhook: Post \\\"https://llmisvc-webhook-server-service.kserve.svc:443/validate-serving-kserve-io-v1alpha1-llminferenceserviceconfig?timeout=10s\\\": no endpoints available for service \\\"llmisvc-webhook-server-service\\\"\",\"reason\":\"InternalError\",\"details\":{\"causes\":[{\"message\":\"failed calling webhook \\\"llminferenceserviceconfig.kserve-webhook-server.v1alpha1.validator\\\": failed to call webhook: Post \\\"https://llmisvc-webhook-server-service.kserve.svc:443/validate-serving-kserve-io-v1alpha1-llminferenceserviceconfig?timeout=10s\\\": no endpoints available for service \\\"llmisvc-webhook-server-service\\\"\"}]},\"code\":500}"}, "traceback": [{"path": "common/gateway_proxy_istio.py", "lineno": 183, "message": ""}, {"path": "llmisvc/fixtures.py", "lineno": 1476, "message": ""}, {"path": "llmisvc/fixtures.py", "lineno": 1436, "message": ""}, {"path": "llmisvc/fixtures.py", "lineno": 1615, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api/custom_objects_api.py", "lineno": 231, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api/custom_objects_api.py", "lineno": 354, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py", "lineno": 348, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py", "lineno": 180, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py", "lineno": 391, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/rest.py", "lineno": 279, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/rest.py", "lineno": 238, "message": "ApiException"}], "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python\n\nkserve_client = \nllm_config = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'router-managed...grade-64e041bd', 'namespace': 'kserve-ci-e2e-test'}, 'spec': {'router': {'gateway': {}, 'route': {}, 'scheduler': {}}}}\nnamespace = 'kserve-ci-e2e-test'\n\n def _create_or_update_llmisvc_config(kserve_client, llm_config, namespace=None):\n \"\"\"Create or update an LLMInferenceServiceConfig resource.\"\"\"\n version = llm_config[\"apiVersion\"].split(\"/\")[1]\n \n if namespace is None:\n namespace = llm_config.get(\"metadata\", {}).get(\"namespace\", \"default\")\n \n name = llm_config.get(\"metadata\", {}).get(\"name\")\n if not name:\n raise ValueError(\"LLMInferenceServiceConfig must have a name in metadata\")\n \n logger.info(f\"Checking LLMInferenceServiceConfig {name} in namespace {namespace}\")\n \n try:\n> existing_config = kserve_client.api_instance.get_namespaced_custom_object(\n constants.KSERVE_GROUP,\n version,\n namespace,\n KSERVE_PLURAL_LLMINFERENCESERVICECONFIG,\n name,\n )\n\nllmisvc/fixtures.py:1589: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \ngroup = 'serving.kserve.io', version = 'v1alpha1'\nnamespace = 'kserve-ci-e2e-test', plural = 'llminferenceserviceconfigs'\nname = 'router-managed-rolling-upgrade-64e041bd'\nkwargs = {'_return_http_data_only': True}\n\n def get_namespaced_custom_object(self, group, version, namespace, plural, name, **kwargs): # noqa: E501\n \"\"\"get_namespaced_custom_object # noqa: E501\n \n Returns a namespace scoped custom object # noqa: E501\n This method makes a synchronous HTTP request by default. To make an\n asynchronous HTTP request, please pass async_req=True\n >>> thread = api.get_namespaced_custom_object(group, version, namespace, plural, name, async_req=True)\n >>> result = thread.get()\n \n :param async_req bool: execute request asynchronously\n :param str group: the custom resource's group (required)\n :param str version: the custom resource's version (required)\n :param str namespace: The custom resource's namespace (required)\n :param str plural: the custom resource's plural name. For TPRs this would be lowercase plural kind. (required)\n :param str name: the custom object's name (required)\n :param _preload_content: if False, the urllib3.HTTPResponse object will\n be returned without reading/decoding response\n data. Default is True.\n :param _request_timeout: timeout setting for this request. If one\n number provided, it will be total request\n timeout. It can also be a pair (tuple) of\n (connection, read) timeouts.\n :return: object\n If the method is called asynchronously,\n returns the request thread.\n \"\"\"\n kwargs['_return_http_data_only'] = True\n> return self.get_namespaced_custom_object_with_http_info(group, version, namespace, plural, name, **kwargs) # noqa: E501\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api/custom_objects_api.py:1632: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \ngroup = 'serving.kserve.io', version = 'v1alpha1'\nnamespace = 'kserve-ci-e2e-test', plural = 'llminferenceserviceconfigs'\nname = 'router-managed-rolling-upgrade-64e041bd'\nkwargs = {'_return_http_data_only': True}\nlocal_var_params = {'_return_http_data_only': True, 'all_params': ['group', 'version', 'namespace', 'plural', 'name', 'async_req', ...], 'auth_settings': ['BearerToken'], 'body_params': None, ...}\nall_params = ['group', 'version', 'namespace', 'plural', 'name', 'async_req', ...]\nkey = '_return_http_data_only', val = True, collection_formats = {}\npath_params = {'group': 'serving.kserve.io', 'name': 'router-managed-rolling-upgrade-64e041bd', 'namespace': 'kserve-ci-e2e-test', 'plural': 'llminferenceserviceconfigs', ...}\nquery_params = []\n\n def get_namespaced_custom_object_with_http_info(self, group, version, namespace, plural, name, **kwargs): # noqa: E501\n \"\"\"get_namespaced_custom_object # noqa: E501\n \n Returns a namespace scoped custom object # noqa: E501\n This method makes a synchronous HTTP request by default. To make an\n asynchronous HTTP request, please pass async_req=True\n >>> thread = api.get_namespaced_custom_object_with_http_info(group, version, namespace, plural, name, async_req=True)\n >>> result = thread.get()\n \n :param async_req bool: execute request asynchronously\n :param str group: the custom resource's group (required)\n :param str version: the custom resource's version (required)\n :param str namespace: The custom resource's namespace (required)\n :param str plural: the custom resource's plural name. For TPRs this would be lowercase plural kind. (required)\n :param str name: the custom object's name (required)\n :param _return_http_data_only: response data without head status code\n and headers\n :param _preload_content: if False, the urllib3.HTTPResponse object will\n be returned without reading/decoding response\n data. Default is True.\n :param _request_timeout: timeout setting for this request. If one\n number provided, it will be total request\n timeout. It can also be a pair (tuple) of\n (connection, read) timeouts.\n :return: tuple(object, status_code(int), headers(HTTPHeaderDict))\n If the method is called asynchronously,\n returns the request thread.\n \"\"\"\n \n local_var_params = locals()\n \n all_params = [\n 'group',\n 'version',\n 'namespace',\n 'plural',\n 'name'\n ]\n all_params.extend(\n [\n 'async_req',\n '_return_http_data_only',\n '_preload_content',\n '_request_timeout'\n ]\n )\n \n for key, val in six.iteritems(local_var_params['kwargs']):\n if key not in all_params:\n raise ApiTypeError(\n \"Got an unexpected keyword argument '%s'\"\n \" to method get_namespaced_custom_object\" % key\n )\n local_var_params[key] = val\n del local_var_params['kwargs']\n # verify the required parameter 'group' is set\n if self.api_client.client_side_validation and ('group' not in local_var_params or # noqa: E501\n local_var_params['group'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `group` when calling `get_namespaced_custom_object`\") # noqa: E501\n # verify the required parameter 'version' is set\n if self.api_client.client_side_validation and ('version' not in local_var_params or # noqa: E501\n local_var_params['version'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `version` when calling `get_namespaced_custom_object`\") # noqa: E501\n # verify the required parameter 'namespace' is set\n if self.api_client.client_side_validation and ('namespace' not in local_var_params or # noqa: E501\n local_var_params['namespace'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `namespace` when calling `get_namespaced_custom_object`\") # noqa: E501\n # verify the required parameter 'plural' is set\n if self.api_client.client_side_validation and ('plural' not in local_var_params or # noqa: E501\n local_var_params['plural'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `plural` when calling `get_namespaced_custom_object`\") # noqa: E501\n # verify the required parameter 'name' is set\n if self.api_client.client_side_validation and ('name' not in local_var_params or # noqa: E501\n local_var_params['name'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `name` when calling `get_namespaced_custom_object`\") # noqa: E501\n \n collection_formats = {}\n \n path_params = {}\n if 'group' in local_var_params:\n path_params['group'] = local_var_params['group'] # noqa: E501\n if 'version' in local_var_params:\n path_params['version'] = local_var_params['version'] # noqa: E501\n if 'namespace' in local_var_params:\n path_params['namespace'] = local_var_params['namespace'] # noqa: E501\n if 'plural' in local_var_params:\n path_params['plural'] = local_var_params['plural'] # noqa: E501\n if 'name' in local_var_params:\n path_params['name'] = local_var_params['name'] # noqa: E501\n \n query_params = []\n \n header_params = {}\n \n form_params = []\n local_var_files = {}\n \n body_params = None\n # HTTP header `Accept`\n header_params['Accept'] = self.api_client.select_header_accept(\n ['application/json']) # noqa: E501\n \n # Authentication setting\n auth_settings = ['BearerToken'] # noqa: E501\n \n> return self.api_client.call_api(\n '/apis/{group}/{version}/namespaces/{namespace}/{plural}/{name}', 'GET',\n path_params,\n query_params,\n header_params,\n body=body_params,\n post_params=form_params,\n files=local_var_files,\n response_type='object', # noqa: E501\n auth_settings=auth_settings,\n async_req=local_var_params.get('async_req'),\n _return_http_data_only=local_var_params.get('_return_http_data_only'), # noqa: E501\n _preload_content=local_var_params.get('_preload_content', True),\n _request_timeout=local_var_params.get('_request_timeout'),\n collection_formats=collection_formats)\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api/custom_objects_api.py:1739: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nresource_path = '/apis/{group}/{version}/namespaces/{namespace}/{plural}/{name}'\nmethod = 'GET'\npath_params = {'group': 'serving.kserve.io', 'name': 'router-managed-rolling-upgrade-64e041bd', 'namespace': 'kserve-ci-e2e-test', 'plural': 'llminferenceserviceconfigs', ...}\nquery_params = []\nheader_params = {'Accept': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\nbody = None, post_params = [], files = {}, response_type = 'object'\nauth_settings = ['BearerToken'], async_req = None, _return_http_data_only = True\ncollection_formats = {}, _preload_content = True, _request_timeout = None\n_host = None\n\n def call_api(self, resource_path, method,\n path_params=None, query_params=None, header_params=None,\n body=None, post_params=None, files=None,\n response_type=None, auth_settings=None, async_req=None,\n _return_http_data_only=None, collection_formats=None,\n _preload_content=True, _request_timeout=None, _host=None):\n \"\"\"Makes the HTTP request (synchronous) and returns deserialized data.\n \n To make an async_req request, set the async_req parameter.\n \n :param resource_path: Path to method endpoint.\n :param method: Method to call.\n :param path_params: Path parameters in the url.\n :param query_params: Query parameters in the url.\n :param header_params: Header parameters to be\n placed in the request header.\n :param body: Request body.\n :param post_params dict: Request post form parameters,\n for `application/x-www-form-urlencoded`, `multipart/form-data`.\n :param auth_settings list: Auth Settings names for the request.\n :param response: Response data type.\n :param files dict: key -> filename, value -> filepath,\n for `multipart/form-data`.\n :param async_req bool: execute request asynchronously\n :param _return_http_data_only: response data without head status code\n and headers\n :param collection_formats: dict of collection formats for path, query,\n header, and post parameters.\n :param _preload_content: if False, the urllib3.HTTPResponse object will\n be returned without reading/decoding response\n data. Default is True.\n :param _request_timeout: timeout setting for this request. If one\n number provided, it will be total request\n timeout. It can also be a pair (tuple) of\n (connection, read) timeouts.\n :return:\n If async_req parameter is True,\n the request will be called asynchronously.\n The method will return the request thread.\n If parameter async_req is False or missing,\n then the method will return the response directly.\n \"\"\"\n if not async_req:\n> return self.__call_api(resource_path, method,\n path_params, query_params, header_params,\n body, post_params, files,\n response_type, auth_settings,\n _return_http_data_only, collection_formats,\n _preload_content, _request_timeout, _host)\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py:348: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nresource_path = '/apis/serving.kserve.io/v1alpha1/namespaces/kserve-ci-e2e-test/llminferenceserviceconfigs/router-managed-rolling-upgrade-64e041bd'\nmethod = 'GET'\npath_params = [('group', 'serving.kserve.io'), ('version', 'v1alpha1'), ('namespace', 'kserve-ci-e2e-test'), ('plural', 'llminferenceserviceconfigs'), ('name', 'router-managed-rolling-upgrade-64e041bd')]\nquery_params = []\nheader_params = {'Accept': 'application/json', 'Content-Type': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\nbody = None, post_params = [], files = {}, response_type = 'object'\nauth_settings = ['BearerToken'], _return_http_data_only = True\ncollection_formats = {}, _preload_content = True, _request_timeout = None\n_host = None\n\n def __call_api(\n self, resource_path, method, path_params=None,\n query_params=None, header_params=None, body=None, post_params=None,\n files=None, response_type=None, auth_settings=None,\n _return_http_data_only=None, collection_formats=None,\n _preload_content=True, _request_timeout=None, _host=None):\n \n config = self.configuration\n \n # header parameters\n header_params = header_params or {}\n header_params.update(self.default_headers)\n if self.cookie:\n header_params['Cookie'] = self.cookie\n if header_params:\n header_params = self.sanitize_for_serialization(header_params)\n header_params = dict(self.parameters_to_tuples(header_params,\n collection_formats))\n \n # path parameters\n if path_params:\n path_params = self.sanitize_for_serialization(path_params)\n path_params = self.parameters_to_tuples(path_params,\n collection_formats)\n for k, v in path_params:\n # specified safe chars, encode everything\n resource_path = resource_path.replace(\n '{%s}' % k,\n quote(str(v), safe=config.safe_chars_for_path_param)\n )\n \n # query parameters\n if query_params:\n query_params = self.sanitize_for_serialization(query_params)\n query_params = self.parameters_to_tuples(query_params,\n collection_formats)\n \n # post parameters\n if post_params or files:\n post_params = post_params if post_params else []\n post_params = self.sanitize_for_serialization(post_params)\n post_params = self.parameters_to_tuples(post_params,\n collection_formats)\n post_params.extend(self.files_parameters(files))\n \n # auth setting\n self.update_params_for_auth(header_params, query_params, auth_settings)\n \n # body\n if body:\n body = self.sanitize_for_serialization(body)\n \n # request url\n if _host is None:\n url = self.configuration.host + resource_path\n else:\n # use server/host defined in path or operation instead\n url = _host + resource_path\n \n # perform request and return response\n> response_data = self.request(\n method, url, query_params=query_params, headers=header_params,\n post_params=post_params, body=body,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout)\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py:180: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nmethod = 'GET'\nurl = 'https://abc8a0cf66a9a44cba4c61b8a1456aeb-75321b9aa6cba46b.elb.us-east-1.amazonaws.com:6443/apis/serving.kserve.io/v1alpha1/namespaces/kserve-ci-e2e-test/llminferenceserviceconfigs/router-managed-rolling-upgrade-64e041bd'\nquery_params = []\nheaders = {'Accept': 'application/json', 'Content-Type': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\npost_params = [], body = None, _preload_content = True, _request_timeout = None\n\n def request(self, method, url, query_params=None, headers=None,\n post_params=None, body=None, _preload_content=True,\n _request_timeout=None):\n \"\"\"Makes the HTTP request using RESTClient.\"\"\"\n if method == \"GET\":\n> return self.rest_client.GET(url,\n query_params=query_params,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout,\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py:373: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nurl = 'https://abc8a0cf66a9a44cba4c61b8a1456aeb-75321b9aa6cba46b.elb.us-east-1.amazonaws.com:6443/apis/serving.kserve.io/v1alpha1/namespaces/kserve-ci-e2e-test/llminferenceserviceconfigs/router-managed-rolling-upgrade-64e041bd'\nheaders = {'Accept': 'application/json', 'Content-Type': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\nquery_params = [], _preload_content = True, _request_timeout = None\n\n def GET(self, url, headers=None, query_params=None, _preload_content=True,\n _request_timeout=None):\n> return self.request(\"GET\", url,\n headers=headers,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout,\n query_params=query_params)\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/rest.py:244: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nmethod = 'GET'\nurl = 'https://abc8a0cf66a9a44cba4c61b8a1456aeb-75321b9aa6cba46b.elb.us-east-1.amazonaws.com:6443/apis/serving.kserve.io/v1alpha1/namespaces/kserve-ci-e2e-test/llminferenceserviceconfigs/router-managed-rolling-upgrade-64e041bd'\nquery_params = []\nheaders = {'Accept': 'application/json', 'Content-Type': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\nbody = None, post_params = {}, _preload_content = True, _request_timeout = None\n\n def request(self, method, url, query_params=None, headers=None,\n body=None, post_params=None, _preload_content=True,\n _request_timeout=None):\n \"\"\"Perform requests.\n \n :param method: http request method\n :param url: http request url\n :param query_params: query parameters in the url\n :param headers: http request headers\n :param body: request json body, for `application/json`\n :param post_params: request post parameters,\n `application/x-www-form-urlencoded`\n and `multipart/form-data`\n :param _preload_content: if False, the urllib3.HTTPResponse object will\n be returned without reading/decoding response\n data. Default is True.\n :param _request_timeout: timeout setting for this request. If one\n number provided, it will be total request\n timeout. It can also be a pair (tuple) of\n (connection, read) timeouts.\n \"\"\"\n method = method.upper()\n assert method in ['GET', 'HEAD', 'DELETE', 'POST', 'PUT',\n 'PATCH', 'OPTIONS']\n \n if post_params and body:\n raise ApiValueError(\n \"body parameter cannot be used with post_params parameter.\"\n )\n \n post_params = post_params or {}\n headers = headers or {}\n \n timeout = None\n if _request_timeout:\n if isinstance(_request_timeout, (int, ) if six.PY3 else (int, long)): # noqa: E501,F821\n timeout = urllib3.Timeout(total=_request_timeout)\n elif (isinstance(_request_timeout, tuple) and\n len(_request_timeout) == 2):\n timeout = urllib3.Timeout(\n connect=_request_timeout[0], read=_request_timeout[1])\n \n if 'Content-Type' not in headers:\n headers['Content-Type'] = 'application/json'\n \n try:\n # For `POST`, `PUT`, `PATCH`, `OPTIONS`, `DELETE`\n if method in ['POST', 'PUT', 'PATCH', 'OPTIONS', 'DELETE']:\n if query_params:\n url += '?' + urlencode(query_params)\n if (re.search('json', headers['Content-Type'], re.IGNORECASE) or\n headers['Content-Type'] == 'application/apply-patch+yaml'):\n if headers['Content-Type'] == 'application/json-patch+json':\n if not isinstance(body, list):\n headers['Content-Type'] = \\\n 'application/strategic-merge-patch+json'\n request_body = None\n if body is not None:\n request_body = json.dumps(body)\n r = self.pool_manager.request(\n method, url,\n body=request_body,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n elif headers['Content-Type'] == 'application/x-www-form-urlencoded': # noqa: E501\n r = self.pool_manager.request(\n method, url,\n fields=post_params,\n encode_multipart=False,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n elif headers['Content-Type'] == 'multipart/form-data':\n # must del headers['Content-Type'], or the correct\n # Content-Type which generated by urllib3 will be\n # overwritten.\n del headers['Content-Type']\n r = self.pool_manager.request(\n method, url,\n fields=post_params,\n encode_multipart=True,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n # Pass a `string` parameter directly in the body to support\n # other content types than Json when `body` argument is\n # provided in serialized form\n elif isinstance(body, str) or isinstance(body, bytes):\n request_body = body\n r = self.pool_manager.request(\n method, url,\n body=request_body,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n else:\n # Cannot generate the request from given parameters\n msg = \"\"\"Cannot prepare a request message for provided\n arguments. Please check that your arguments match\n declared content type.\"\"\"\n raise ApiException(status=0, reason=msg)\n # For `GET`, `HEAD`\n else:\n r = self.pool_manager.request(method, url,\n fields=query_params,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n except urllib3.exceptions.SSLError as e:\n msg = \"{0}\\n{1}\".format(type(e).__name__, str(e))\n raise ApiException(status=0, reason=msg)\n \n if _preload_content:\n r = RESTResponse(r)\n \n # In the python 3, the response.data is bytes.\n # we need to decode it to string.\n if six.PY3:\n r.data = r.data.decode('utf8')\n \n # log response body\n logger.debug(\"response body: %s\", r.data)\n \n if not 200 <= r.status <= 299:\n> raise ApiException(http_resp=r)\nE kubernetes.client.exceptions.ApiException: (404)\nE Reason: Not Found\nE HTTP response headers: HTTPHeaderDict({'Audit-Id': '6c80fc16-04c3-408a-8276-c22a0c8d46ac', 'Cache-Control': 'no-cache, private', 'Content-Type': 'application/json', 'Strict-Transport-Security': 'max-age=31536000; includeSubDomains; preload', 'X-Kubernetes-Pf-Flowschema-Uid': 'c30fff65-8b8a-47f0-8303-5f205881ee46', 'X-Kubernetes-Pf-Prioritylevel-Uid': '1d0f3173-41ee-4f4d-bdbf-9bfbca43ee13', 'Date': 'Mon, 13 Jul 2026 22:11:24 GMT', 'Content-Length': '336'})\nE HTTP response body: {\"kind\":\"Status\",\"apiVersion\":\"v1\",\"metadata\":{},\"status\":\"Failure\",\"message\":\"llminferenceserviceconfigs.serving.kserve.io \\\"router-managed-rolling-upgrade-64e041bd\\\" not found\",\"reason\":\"NotFound\",\"details\":{\"name\":\"router-managed-rolling-upgrade-64e041bd\",\"group\":\"serving.kserve.io\",\"kind\":\"llminferenceserviceconfigs\"},\"code\":404}\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/rest.py:238: ApiException\n\nDuring handling of the above exception, another exception occurred:\n\nrequest = >\n\n @pytest.fixture(autouse=True)\n def ensure_gateway_proxy_memory(request):\n \"\"\"After test setup creates gateways, patch them for proxy memory.\"\"\"\n if not GATEWAY_PROXY_MEMORY:\n return\n \n # Let test_case (llmisvc) create gateways first\n \n if \"test_case\" in request.fixturenames:\n> request.getfixturevalue(\"test_case\")\n\ncommon/gateway_proxy_istio.py:183: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nrequest = >\n\n @pytest.fixture(scope=\"function\")\n def test_case(request):\n tc = request.param\n \n inject_k8s_proxy()\n \n kserve_client = KServeClient(\n config_file=os.environ.get(\"KUBECONFIG\", \"~/.kube/config\"),\n client_configuration=client.Configuration(),\n )\n \n # Execute before test hooks\n try:\n for func in tc.before_test:\n func()\n except Exception as before_test_error:\n raise RuntimeError(\n f\"Failed to execute before test hook: {before_test_error}\"\n ) from before_test_error\n \n try:\n> _setup_test_case_service(kserve_client, tc, request.node.name)\n\nllmisvc/fixtures.py:1476: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nkserve_client = \ntc = TestCase(base_refs=['router-managed', 'workload-llmd-simulator', 'model-fb-opt-125m'], prompt='KServe is a', service_n...None, expected_gateway=None, before_test=[], after_test=[], peers=[], llm_service=None, model_name='facebook/opt-125m')\ntest_node_name = 'test_rolling_upgrade_coordination[router-managed-workload-llmd-simulator-model-fb-opt-125m]'\npeer_index = None\n\n def _setup_test_case_service(kserve_client, tc, test_node_name, peer_index=None):\n \"\"\"Create LLMInferenceServiceConfigs and build the LLMInferenceService for a TestCase.\n \n Returns a list of created config names for cleanup tracking.\n \"\"\"\n missing_refs = [\n ref for ref in tc.base_refs if ref not in LLMINFERENCESERVICE_CONFIGS\n ]\n if missing_refs:\n raise ValueError(\n f\"Missing base_refs in LLMINFERENCESERVICE_CONFIGS: {missing_refs}\"\n )\n if not tc.service_name:\n suffix = f\"-peer-{peer_index}\" if peer_index is not None else \"\"\n tc.service_name = generate_service_name(test_node_name + suffix, tc.base_refs)\n if tc.model_name == \"default/model\":\n tc.model_name = _get_model_name_from_configs(tc.base_refs)\n \n created_configs = []\n unique_base_refs = []\n for base_ref in tc.base_refs:\n unique_config_name = generate_k8s_safe_suffix(base_ref, [tc.service_name])\n unique_base_refs.append(unique_config_name)\n \n unique_config_body = {\n \"apiVersion\": \"serving.kserve.io/v1alpha1\",\n \"kind\": \"LLMInferenceServiceConfig\",\n \"metadata\": {\n \"name\": unique_config_name,\n \"namespace\": KSERVE_TEST_NAMESPACE,\n },\n \"spec\": LLMINFERENCESERVICE_CONFIGS[base_ref],\n }\n \n> _create_or_update_llmisvc_config(\n kserve_client, unique_config_body, KSERVE_TEST_NAMESPACE\n )\n\nllmisvc/fixtures.py:1436: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nkserve_client = \nllm_config = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'router-managed...grade-64e041bd', 'namespace': 'kserve-ci-e2e-test'}, 'spec': {'router': {'gateway': {}, 'route': {}, 'scheduler': {}}}}\nnamespace = 'kserve-ci-e2e-test'\n\n def _create_or_update_llmisvc_config(kserve_client, llm_config, namespace=None):\n \"\"\"Create or update an LLMInferenceServiceConfig resource.\"\"\"\n version = llm_config[\"apiVersion\"].split(\"/\")[1]\n \n if namespace is None:\n namespace = llm_config.get(\"metadata\", {}).get(\"namespace\", \"default\")\n \n name = llm_config.get(\"metadata\", {}).get(\"name\")\n if not name:\n raise ValueError(\"LLMInferenceServiceConfig must have a name in metadata\")\n \n logger.info(f\"Checking LLMInferenceServiceConfig {name} in namespace {namespace}\")\n \n try:\n existing_config = kserve_client.api_instance.get_namespaced_custom_object(\n constants.KSERVE_GROUP,\n version,\n namespace,\n KSERVE_PLURAL_LLMINFERENCESERVICECONFIG,\n name,\n )\n \n llm_config[\"metadata\"] = existing_config[\"metadata\"]\n \n outputs = kserve_client.api_instance.replace_namespaced_custom_object(\n constants.KSERVE_GROUP,\n version,\n namespace,\n KSERVE_PLURAL_LLMINFERENCESERVICECONFIG,\n name,\n llm_config,\n )\n logger.info(f\"\u2713 Successfully updated LLMInferenceServiceConfig {name}\")\n return outputs\n \n except client.rest.ApiException as e:\n if e.status == 404: # Not found - create it\n logger.info(\n f\"Resource not found, creating LLMInferenceServiceConfig {name}\"\n )\n> outputs = kserve_client.api_instance.create_namespaced_custom_object(\n constants.KSERVE_GROUP,\n version,\n namespace,\n KSERVE_PLURAL_LLMINFERENCESERVICECONFIG,\n llm_config,\n )\n\nllmisvc/fixtures.py:1615: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \ngroup = 'serving.kserve.io', version = 'v1alpha1'\nnamespace = 'kserve-ci-e2e-test', plural = 'llminferenceserviceconfigs'\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'router-managed...grade-64e041bd', 'namespace': 'kserve-ci-e2e-test'}, 'spec': {'router': {'gateway': {}, 'route': {}, 'scheduler': {}}}}\nkwargs = {'_return_http_data_only': True}\n\n def create_namespaced_custom_object(self, group, version, namespace, plural, body, **kwargs): # noqa: E501\n \"\"\"create_namespaced_custom_object # noqa: E501\n \n Creates a namespace scoped Custom object # noqa: E501\n This method makes a synchronous HTTP request by default. To make an\n asynchronous HTTP request, please pass async_req=True\n >>> thread = api.create_namespaced_custom_object(group, version, namespace, plural, body, async_req=True)\n >>> result = thread.get()\n \n :param async_req bool: execute request asynchronously\n :param str group: The custom resource's group name (required)\n :param str version: The custom resource's version (required)\n :param str namespace: The custom resource's namespace (required)\n :param str plural: The custom resource's plural name. For TPRs this would be lowercase plural kind. (required)\n :param object body: The JSON schema of the Resource to create. (required)\n :param str pretty: If 'true', then the output is pretty printed.\n :param str dry_run: When present, indicates that modifications should not be persisted. An invalid or unrecognized dryRun directive will result in an error response and no further processing of the request. Valid values are: - All: all dry run stages will be processed\n :param str field_manager: fieldManager is a name associated with the actor or entity that is making these changes. The value must be less than or 128 characters long, and only contain printable characters, as defined by https://golang.org/pkg/unicode/#IsPrint.\n :param str field_validation: fieldValidation instructs the server on how to handle objects in the request (POST/PUT/PATCH) containing unknown or duplicate fields. Valid values are: - Ignore: This will ignore any unknown fields that are silently dropped from the object, and will ignore all but the last duplicate field that the decoder encounters. This is the default behavior prior to v1.23. - Warn: This will send a warning via the standard warning response header for each unknown field that is dropped from the object, and for each duplicate field that is encountered. The request will still succeed if there are no other errors, and will only persist the last of any duplicate fields. This is the default in v1.23+ - Strict: This will fail the request with a BadRequest error if any unknown fields would be dropped from the object, or if any duplicate fields are present. The error returned from the server will contain all unknown and duplicate fields encountered. (optional)\n :param _preload_content: if False, the urllib3.HTTPResponse object will\n be returned without reading/decoding response\n data. Default is True.\n :param _request_timeout: timeout setting for this request. If one\n number provided, it will be total request\n timeout. It can also be a pair (tuple) of\n (connection, read) timeouts.\n :return: object\n If the method is called asynchronously,\n returns the request thread.\n \"\"\"\n kwargs['_return_http_data_only'] = True\n> return self.create_namespaced_custom_object_with_http_info(group, version, namespace, plural, body, **kwargs) # noqa: E501\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api/custom_objects_api.py:231: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \ngroup = 'serving.kserve.io', version = 'v1alpha1'\nnamespace = 'kserve-ci-e2e-test', plural = 'llminferenceserviceconfigs'\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'router-managed...grade-64e041bd', 'namespace': 'kserve-ci-e2e-test'}, 'spec': {'router': {'gateway': {}, 'route': {}, 'scheduler': {}}}}\nkwargs = {'_return_http_data_only': True}\nlocal_var_params = {'_return_http_data_only': True, 'all_params': ['group', 'version', 'namespace', 'plural', 'body', 'pretty', ...], 'au...64e041bd', 'namespace': 'kserve-ci-e2e-test'}, 'spec': {'router': {'gateway': {}, 'route': {}, 'scheduler': {}}}}, ...}\nall_params = ['group', 'version', 'namespace', 'plural', 'body', 'pretty', ...]\nkey = '_return_http_data_only', val = True, collection_formats = {}\npath_params = {'group': 'serving.kserve.io', 'namespace': 'kserve-ci-e2e-test', 'plural': 'llminferenceserviceconfigs', 'version': 'v1alpha1'}\nquery_params = []\n\n def create_namespaced_custom_object_with_http_info(self, group, version, namespace, plural, body, **kwargs): # noqa: E501\n \"\"\"create_namespaced_custom_object # noqa: E501\n \n Creates a namespace scoped Custom object # noqa: E501\n This method makes a synchronous HTTP request by default. To make an\n asynchronous HTTP request, please pass async_req=True\n >>> thread = api.create_namespaced_custom_object_with_http_info(group, version, namespace, plural, body, async_req=True)\n >>> result = thread.get()\n \n :param async_req bool: execute request asynchronously\n :param str group: The custom resource's group name (required)\n :param str version: The custom resource's version (required)\n :param str namespace: The custom resource's namespace (required)\n :param str plural: The custom resource's plural name. For TPRs this would be lowercase plural kind. (required)\n :param object body: The JSON schema of the Resource to create. (required)\n :param str pretty: If 'true', then the output is pretty printed.\n :param str dry_run: When present, indicates that modifications should not be persisted. An invalid or unrecognized dryRun directive will result in an error response and no further processing of the request. Valid values are: - All: all dry run stages will be processed\n :param str field_manager: fieldManager is a name associated with the actor or entity that is making these changes. The value must be less than or 128 characters long, and only contain printable characters, as defined by https://golang.org/pkg/unicode/#IsPrint.\n :param str field_validation: fieldValidation instructs the server on how to handle objects in the request (POST/PUT/PATCH) containing unknown or duplicate fields. Valid values are: - Ignore: This will ignore any unknown fields that are silently dropped from the object, and will ignore all but the last duplicate field that the decoder encounters. This is the default behavior prior to v1.23. - Warn: This will send a warning via the standard warning response header for each unknown field that is dropped from the object, and for each duplicate field that is encountered. The request will still succeed if there are no other errors, and will only persist the last of any duplicate fields. This is the default in v1.23+ - Strict: This will fail the request with a BadRequest error if any unknown fields would be dropped from the object, or if any duplicate fields are present. The error returned from the server will contain all unknown and duplicate fields encountered. (optional)\n :param _return_http_data_only: response data without head status code\n and headers\n :param _preload_content: if False, the urllib3.HTTPResponse object will\n be returned without reading/decoding response\n data. Default is True.\n :param _request_timeout: timeout setting for this request. If one\n number provided, it will be total request\n timeout. It can also be a pair (tuple) of\n (connection, read) timeouts.\n :return: tuple(object, status_code(int), headers(HTTPHeaderDict))\n If the method is called asynchronously,\n returns the request thread.\n \"\"\"\n \n local_var_params = locals()\n \n all_params = [\n 'group',\n 'version',\n 'namespace',\n 'plural',\n 'body',\n 'pretty',\n 'dry_run',\n 'field_manager',\n 'field_validation'\n ]\n all_params.extend(\n [\n 'async_req',\n '_return_http_data_only',\n '_preload_content',\n '_request_timeout'\n ]\n )\n \n for key, val in six.iteritems(local_var_params['kwargs']):\n if key not in all_params:\n raise ApiTypeError(\n \"Got an unexpected keyword argument '%s'\"\n \" to method create_namespaced_custom_object\" % key\n )\n local_var_params[key] = val\n del local_var_params['kwargs']\n # verify the required parameter 'group' is set\n if self.api_client.client_side_validation and ('group' not in local_var_params or # noqa: E501\n local_var_params['group'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `group` when calling `create_namespaced_custom_object`\") # noqa: E501\n # verify the required parameter 'version' is set\n if self.api_client.client_side_validation and ('version' not in local_var_params or # noqa: E501\n local_var_params['version'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `version` when calling `create_namespaced_custom_object`\") # noqa: E501\n # verify the required parameter 'namespace' is set\n if self.api_client.client_side_validation and ('namespace' not in local_var_params or # noqa: E501\n local_var_params['namespace'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `namespace` when calling `create_namespaced_custom_object`\") # noqa: E501\n # verify the required parameter 'plural' is set\n if self.api_client.client_side_validation and ('plural' not in local_var_params or # noqa: E501\n local_var_params['plural'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `plural` when calling `create_namespaced_custom_object`\") # noqa: E501\n # verify the required parameter 'body' is set\n if self.api_client.client_side_validation and ('body' not in local_var_params or # noqa: E501\n local_var_params['body'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `body` when calling `create_namespaced_custom_object`\") # noqa: E501\n \n collection_formats = {}\n \n path_params = {}\n if 'group' in local_var_params:\n path_params['group'] = local_var_params['group'] # noqa: E501\n if 'version' in local_var_params:\n path_params['version'] = local_var_params['version'] # noqa: E501\n if 'namespace' in local_var_params:\n path_params['namespace'] = local_var_params['namespace'] # noqa: E501\n if 'plural' in local_var_params:\n path_params['plural'] = local_var_params['plural'] # noqa: E501\n \n query_params = []\n if 'pretty' in local_var_params and local_var_params['pretty'] is not None: # noqa: E501\n query_params.append(('pretty', local_var_params['pretty'])) # noqa: E501\n if 'dry_run' in local_var_params and local_var_params['dry_run'] is not None: # noqa: E501\n query_params.append(('dryRun', local_var_params['dry_run'])) # noqa: E501\n if 'field_manager' in local_var_params and local_var_params['field_manager'] is not None: # noqa: E501\n query_params.append(('fieldManager', local_var_params['field_manager'])) # noqa: E501\n if 'field_validation' in local_var_params and local_var_params['field_validation'] is not None: # noqa: E501\n query_params.append(('fieldValidation', local_var_params['field_validation'])) # noqa: E501\n \n header_params = {}\n \n form_params = []\n local_var_files = {}\n \n body_params = None\n if 'body' in local_var_params:\n body_params = local_var_params['body']\n # HTTP header `Accept`\n header_params['Accept'] = self.api_client.select_header_accept(\n ['application/json']) # noqa: E501\n \n # Authentication setting\n auth_settings = ['BearerToken'] # noqa: E501\n \n> return self.api_client.call_api(\n '/apis/{group}/{version}/namespaces/{namespace}/{plural}', 'POST',\n path_params,\n query_params,\n header_params,\n body=body_params,\n post_params=form_params,\n files=local_var_files,\n response_type='object', # noqa: E501\n auth_settings=auth_settings,\n async_req=local_var_params.get('async_req'),\n _return_http_data_only=local_var_params.get('_return_http_data_only'), # noqa: E501\n _preload_content=local_var_params.get('_preload_content', True),\n _request_timeout=local_var_params.get('_request_timeout'),\n collection_formats=collection_formats)\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api/custom_objects_api.py:354: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nresource_path = '/apis/{group}/{version}/namespaces/{namespace}/{plural}'\nmethod = 'POST'\npath_params = {'group': 'serving.kserve.io', 'namespace': 'kserve-ci-e2e-test', 'plural': 'llminferenceserviceconfigs', 'version': 'v1alpha1'}\nquery_params = []\nheader_params = {'Accept': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'router-managed...grade-64e041bd', 'namespace': 'kserve-ci-e2e-test'}, 'spec': {'router': {'gateway': {}, 'route': {}, 'scheduler': {}}}}\npost_params = [], files = {}, response_type = 'object'\nauth_settings = ['BearerToken'], async_req = None, _return_http_data_only = True\ncollection_formats = {}, _preload_content = True, _request_timeout = None\n_host = None\n\n def call_api(self, resource_path, method,\n path_params=None, query_params=None, header_params=None,\n body=None, post_params=None, files=None,\n response_type=None, auth_settings=None, async_req=None,\n _return_http_data_only=None, collection_formats=None,\n _preload_content=True, _request_timeout=None, _host=None):\n \"\"\"Makes the HTTP request (synchronous) and returns deserialized data.\n \n To make an async_req request, set the async_req parameter.\n \n :param resource_path: Path to method endpoint.\n :param method: Method to call.\n :param path_params: Path parameters in the url.\n :param query_params: Query parameters in the url.\n :param header_params: Header parameters to be\n placed in the request header.\n :param body: Request body.\n :param post_params dict: Request post form parameters,\n for `application/x-www-form-urlencoded`, `multipart/form-data`.\n :param auth_settings list: Auth Settings names for the request.\n :param response: Response data type.\n :param files dict: key -> filename, value -> filepath,\n for `multipart/form-data`.\n :param async_req bool: execute request asynchronously\n :param _return_http_data_only: response data without head status code\n and headers\n :param collection_formats: dict of collection formats for path, query,\n header, and post parameters.\n :param _preload_content: if False, the urllib3.HTTPResponse object will\n be returned without reading/decoding response\n data. Default is True.\n :param _request_timeout: timeout setting for this request. If one\n number provided, it will be total request\n timeout. It can also be a pair (tuple) of\n (connection, read) timeouts.\n :return:\n If async_req parameter is True,\n the request will be called asynchronously.\n The method will return the request thread.\n If parameter async_req is False or missing,\n then the method will return the response directly.\n \"\"\"\n if not async_req:\n> return self.__call_api(resource_path, method,\n path_params, query_params, header_params,\n body, post_params, files,\n response_type, auth_settings,\n _return_http_data_only, collection_formats,\n _preload_content, _request_timeout, _host)\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py:348: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nresource_path = '/apis/serving.kserve.io/v1alpha1/namespaces/kserve-ci-e2e-test/llminferenceserviceconfigs'\nmethod = 'POST'\npath_params = [('group', 'serving.kserve.io'), ('version', 'v1alpha1'), ('namespace', 'kserve-ci-e2e-test'), ('plural', 'llminferenceserviceconfigs')]\nquery_params = []\nheader_params = {'Accept': 'application/json', 'Content-Type': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'router-managed...grade-64e041bd', 'namespace': 'kserve-ci-e2e-test'}, 'spec': {'router': {'gateway': {}, 'route': {}, 'scheduler': {}}}}\npost_params = [], files = {}, response_type = 'object'\nauth_settings = ['BearerToken'], _return_http_data_only = True\ncollection_formats = {}, _preload_content = True, _request_timeout = None\n_host = None\n\n def __call_api(\n self, resource_path, method, path_params=None,\n query_params=None, header_params=None, body=None, post_params=None,\n files=None, response_type=None, auth_settings=None,\n _return_http_data_only=None, collection_formats=None,\n _preload_content=True, _request_timeout=None, _host=None):\n \n config = self.configuration\n \n # header parameters\n header_params = header_params or {}\n header_params.update(self.default_headers)\n if self.cookie:\n header_params['Cookie'] = self.cookie\n if header_params:\n header_params = self.sanitize_for_serialization(header_params)\n header_params = dict(self.parameters_to_tuples(header_params,\n collection_formats))\n \n # path parameters\n if path_params:\n path_params = self.sanitize_for_serialization(path_params)\n path_params = self.parameters_to_tuples(path_params,\n collection_formats)\n for k, v in path_params:\n # specified safe chars, encode everything\n resource_path = resource_path.replace(\n '{%s}' % k,\n quote(str(v), safe=config.safe_chars_for_path_param)\n )\n \n # query parameters\n if query_params:\n query_params = self.sanitize_for_serialization(query_params)\n query_params = self.parameters_to_tuples(query_params,\n collection_formats)\n \n # post parameters\n if post_params or files:\n post_params = post_params if post_params else []\n post_params = self.sanitize_for_serialization(post_params)\n post_params = self.parameters_to_tuples(post_params,\n collection_formats)\n post_params.extend(self.files_parameters(files))\n \n # auth setting\n self.update_params_for_auth(header_params, query_params, auth_settings)\n \n # body\n if body:\n body = self.sanitize_for_serialization(body)\n \n # request url\n if _host is None:\n url = self.configuration.host + resource_path\n else:\n # use server/host defined in path or operation instead\n url = _host + resource_path\n \n # perform request and return response\n> response_data = self.request(\n method, url, query_params=query_params, headers=header_params,\n post_params=post_params, body=body,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout)\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py:180: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nmethod = 'POST'\nurl = 'https://abc8a0cf66a9a44cba4c61b8a1456aeb-75321b9aa6cba46b.elb.us-east-1.amazonaws.com:6443/apis/serving.kserve.io/v1alpha1/namespaces/kserve-ci-e2e-test/llminferenceserviceconfigs'\nquery_params = []\nheaders = {'Accept': 'application/json', 'Content-Type': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\npost_params = []\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'router-managed...grade-64e041bd', 'namespace': 'kserve-ci-e2e-test'}, 'spec': {'router': {'gateway': {}, 'route': {}, 'scheduler': {}}}}\n_preload_content = True, _request_timeout = None\n\n def request(self, method, url, query_params=None, headers=None,\n post_params=None, body=None, _preload_content=True,\n _request_timeout=None):\n \"\"\"Makes the HTTP request using RESTClient.\"\"\"\n if method == \"GET\":\n return self.rest_client.GET(url,\n query_params=query_params,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout,\n headers=headers)\n elif method == \"HEAD\":\n return self.rest_client.HEAD(url,\n query_params=query_params,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout,\n headers=headers)\n elif method == \"OPTIONS\":\n return self.rest_client.OPTIONS(url,\n query_params=query_params,\n headers=headers,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout)\n elif method == \"POST\":\n> return self.rest_client.POST(url,\n query_params=query_params,\n headers=headers,\n post_params=post_params,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout,\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py:391: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nurl = 'https://abc8a0cf66a9a44cba4c61b8a1456aeb-75321b9aa6cba46b.elb.us-east-1.amazonaws.com:6443/apis/serving.kserve.io/v1alpha1/namespaces/kserve-ci-e2e-test/llminferenceserviceconfigs'\nheaders = {'Accept': 'application/json', 'Content-Type': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\nquery_params = [], post_params = []\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'router-managed...grade-64e041bd', 'namespace': 'kserve-ci-e2e-test'}, 'spec': {'router': {'gateway': {}, 'route': {}, 'scheduler': {}}}}\n_preload_content = True, _request_timeout = None\n\n def POST(self, url, headers=None, query_params=None, post_params=None,\n body=None, _preload_content=True, _request_timeout=None):\n> return self.request(\"POST\", url,\n headers=headers,\n query_params=query_params,\n post_params=post_params,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout,\n body=body)\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/rest.py:279: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nmethod = 'POST'\nurl = 'https://abc8a0cf66a9a44cba4c61b8a1456aeb-75321b9aa6cba46b.elb.us-east-1.amazonaws.com:6443/apis/serving.kserve.io/v1alpha1/namespaces/kserve-ci-e2e-test/llminferenceserviceconfigs'\nquery_params = []\nheaders = {'Accept': 'application/json', 'Content-Type': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'router-managed...grade-64e041bd', 'namespace': 'kserve-ci-e2e-test'}, 'spec': {'router': {'gateway': {}, 'route': {}, 'scheduler': {}}}}\npost_params = {}, _preload_content = True, _request_timeout = None\n\n def request(self, method, url, query_params=None, headers=None,\n body=None, post_params=None, _preload_content=True,\n _request_timeout=None):\n \"\"\"Perform requests.\n \n :param method: http request method\n :param url: http request url\n :param query_params: query parameters in the url\n :param headers: http request headers\n :param body: request json body, for `application/json`\n :param post_params: request post parameters,\n `application/x-www-form-urlencoded`\n and `multipart/form-data`\n :param _preload_content: if False, the urllib3.HTTPResponse object will\n be returned without reading/decoding response\n data. Default is True.\n :param _request_timeout: timeout setting for this request. If one\n number provided, it will be total request\n timeout. It can also be a pair (tuple) of\n (connection, read) timeouts.\n \"\"\"\n method = method.upper()\n assert method in ['GET', 'HEAD', 'DELETE', 'POST', 'PUT',\n 'PATCH', 'OPTIONS']\n \n if post_params and body:\n raise ApiValueError(\n \"body parameter cannot be used with post_params parameter.\"\n )\n \n post_params = post_params or {}\n headers = headers or {}\n \n timeout = None\n if _request_timeout:\n if isinstance(_request_timeout, (int, ) if six.PY3 else (int, long)): # noqa: E501,F821\n timeout = urllib3.Timeout(total=_request_timeout)\n elif (isinstance(_request_timeout, tuple) and\n len(_request_timeout) == 2):\n timeout = urllib3.Timeout(\n connect=_request_timeout[0], read=_request_timeout[1])\n \n if 'Content-Type' not in headers:\n headers['Content-Type'] = 'application/json'\n \n try:\n # For `POST`, `PUT`, `PATCH`, `OPTIONS`, `DELETE`\n if method in ['POST', 'PUT', 'PATCH', 'OPTIONS', 'DELETE']:\n if query_params:\n url += '?' + urlencode(query_params)\n if (re.search('json', headers['Content-Type'], re.IGNORECASE) or\n headers['Content-Type'] == 'application/apply-patch+yaml'):\n if headers['Content-Type'] == 'application/json-patch+json':\n if not isinstance(body, list):\n headers['Content-Type'] = \\\n 'application/strategic-merge-patch+json'\n request_body = None\n if body is not None:\n request_body = json.dumps(body)\n r = self.pool_manager.request(\n method, url,\n body=request_body,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n elif headers['Content-Type'] == 'application/x-www-form-urlencoded': # noqa: E501\n r = self.pool_manager.request(\n method, url,\n fields=post_params,\n encode_multipart=False,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n elif headers['Content-Type'] == 'multipart/form-data':\n # must del headers['Content-Type'], or the correct\n # Content-Type which generated by urllib3 will be\n # overwritten.\n del headers['Content-Type']\n r = self.pool_manager.request(\n method, url,\n fields=post_params,\n encode_multipart=True,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n # Pass a `string` parameter directly in the body to support\n # other content types than Json when `body` argument is\n # provided in serialized form\n elif isinstance(body, str) or isinstance(body, bytes):\n request_body = body\n r = self.pool_manager.request(\n method, url,\n body=request_body,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n else:\n # Cannot generate the request from given parameters\n msg = \"\"\"Cannot prepare a request message for provided\n arguments. Please check that your arguments match\n declared content type.\"\"\"\n raise ApiException(status=0, reason=msg)\n # For `GET`, `HEAD`\n else:\n r = self.pool_manager.request(method, url,\n fields=query_params,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n except urllib3.exceptions.SSLError as e:\n msg = \"{0}\\n{1}\".format(type(e).__name__, str(e))\n raise ApiException(status=0, reason=msg)\n \n if _preload_content:\n r = RESTResponse(r)\n \n # In the python 3, the response.data is bytes.\n # we need to decode it to string.\n if six.PY3:\n r.data = r.data.decode('utf8')\n \n # log response body\n logger.debug(\"response body: %s\", r.data)\n \n if not 200 <= r.status <= 299:\n> raise ApiException(http_resp=r)\nE kubernetes.client.exceptions.ApiException: (500)\nE Reason: Internal Server Error\nE HTTP response headers: HTTPHeaderDict({'Audit-Id': '190f2227-5483-4ba6-8b9f-ee7976c421ca', 'Cache-Control': 'no-cache, private', 'Content-Type': 'application/json', 'Strict-Transport-Security': 'max-age=31536000; includeSubDomains; preload', 'X-Kubernetes-Pf-Flowschema-Uid': 'c30fff65-8b8a-47f0-8303-5f205881ee46', 'X-Kubernetes-Pf-Prioritylevel-Uid': '1d0f3173-41ee-4f4d-bdbf-9bfbca43ee13', 'Date': 'Mon, 13 Jul 2026 22:11:24 GMT', 'Content-Length': '833'})\nE HTTP response body: {\"kind\":\"Status\",\"apiVersion\":\"v1\",\"metadata\":{},\"status\":\"Failure\",\"message\":\"Internal error occurred: failed calling webhook \\\"llminferenceserviceconfig.kserve-webhook-server.v1alpha1.validator\\\": failed to call webhook: Post \\\"https://llmisvc-webhook-server-service.kserve.svc:443/validate-serving-kserve-io-v1alpha1-llminferenceserviceconfig?timeout=10s\\\": no endpoints available for service \\\"llmisvc-webhook-server-service\\\"\",\"reason\":\"InternalError\",\"details\":{\"causes\":[{\"message\":\"failed calling webhook \\\"llminferenceserviceconfig.kserve-webhook-server.v1alpha1.validator\\\": failed to call webhook: Post \\\"https://llmisvc-webhook-server-service.kserve.svc:443/validate-serving-kserve-io-v1alpha1-llminferenceserviceconfig?timeout=10s\\\": no endpoints available for service \\\"llmisvc-webhook-server-service\\\"\"}]},\"code\":500}\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/rest.py:238: ApiException"}, "teardown": {"duration": 0.0002267439995193854, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}, {"nodeid": "llmisvc/test_storage_version_migration.py::TestStorageVersionMigration::test_storage_version_migration_after_simulated_upgrade", "lineno": 113, "outcome": "failed", "keywords": ["test_storage_version_migration_after_simulated_upgrade", "cluster_single_node", "cluster_cpu", "pytestmark", "TestStorageVersionMigration", "conversion", "llminferenceservice", "test_storage_version_migration.py", "llmisvc/__init__.py", "e2e"], "setup": {"duration": 0.058629597995604854, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}, "call": {"duration": 0.03731475200038403, "outcome": "failed", "crash": {"path": "/workspace/source/python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/rest.py", "lineno": 238, "message": "kubernetes.client.exceptions.ApiException: (500)\nReason: Internal Server Error\nHTTP response headers: HTTPHeaderDict({'Audit-Id': '9ba752cc-fbc7-4911-b3e2-604b9a48b2e5', 'Cache-Control': 'no-cache, private', 'Content-Type': 'application/json', 'Strict-Transport-Security': 'max-age=31536000; includeSubDomains; preload', 'X-Kubernetes-Pf-Flowschema-Uid': 'c30fff65-8b8a-47f0-8303-5f205881ee46', 'X-Kubernetes-Pf-Prioritylevel-Uid': '1d0f3173-41ee-4f4d-bdbf-9bfbca43ee13', 'Date': 'Mon, 13 Jul 2026 22:11:24 GMT', 'Content-Length': '833'})\nHTTP response body: {\"kind\":\"Status\",\"apiVersion\":\"v1\",\"metadata\":{},\"status\":\"Failure\",\"message\":\"Internal error occurred: failed calling webhook \\\"llminferenceserviceconfig.kserve-webhook-server.v1alpha1.validator\\\": failed to call webhook: Post \\\"https://llmisvc-webhook-server-service.kserve.svc:443/validate-serving-kserve-io-v1alpha1-llminferenceserviceconfig?timeout=10s\\\": no endpoints available for service \\\"llmisvc-webhook-server-service\\\"\",\"reason\":\"InternalError\",\"details\":{\"causes\":[{\"message\":\"failed calling webhook \\\"llminferenceserviceconfig.kserve-webhook-server.v1alpha1.validator\\\": failed to call webhook: Post \\\"https://llmisvc-webhook-server-service.kserve.svc:443/validate-serving-kserve-io-v1alpha1-llminferenceserviceconfig?timeout=10s\\\": no endpoints available for service \\\"llmisvc-webhook-server-service\\\"\"}]},\"code\":500}"}, "traceback": [{"path": "llmisvc/test_storage_version_migration.py", "lineno": 151, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api/custom_objects_api.py", "lineno": 231, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api/custom_objects_api.py", "lineno": 354, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py", "lineno": 348, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py", "lineno": 180, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py", "lineno": 391, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/rest.py", "lineno": 279, "message": ""}, {"path": "../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/rest.py", "lineno": 238, "message": "ApiException"}], "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python\n\nself = \n\n @pytest.mark.cluster_cpu\n @pytest.mark.cluster_single_node\n def test_storage_version_migration_after_simulated_upgrade(self):\n \"\"\"Test that storage version migration runs successfully after controller restart.\n \n Simulates an upgrade by:\n 1. Creating a resource via v1alpha1 API\n 2. Patching CRD storedVersions to include the stale v1alpha1 version\n 3. Restarting the controller (which triggers migration on startup)\n 4. Verifying storedVersions is cleaned up to only contain v1alpha2\n \"\"\"\n # 1. Create a config resource via v1alpha1 so we have something to migrate\n config_name = \"migration-test-config\"\n config = {\n \"apiVersion\": f\"{constants.KSERVE_GROUP}/{constants.KSERVE_V1ALPHA1_VERSION}\",\n \"kind\": \"LLMInferenceServiceConfig\",\n \"metadata\": {\n \"name\": config_name,\n \"namespace\": self.namespace,\n },\n \"spec\": {\n \"model\": {\"uri\": OPT_125M_MODEL_URI, \"name\": \"facebook/opt-125m\"},\n \"router\": {\"route\": {}},\n \"template\": {\n \"containers\": [\n {\n \"name\": \"main\",\n \"image\": \"public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0\",\n \"resources\": {\n \"limits\": {\"cpu\": \"1\", \"memory\": \"2Gi\"},\n \"requests\": {\"cpu\": \"100m\", \"memory\": \"512Mi\"},\n },\n }\n ]\n },\n },\n }\n> self.kserve_client.api_instance.create_namespaced_custom_object(\n constants.KSERVE_GROUP,\n constants.KSERVE_V1ALPHA1_VERSION,\n self.namespace,\n KSERVE_PLURAL_LLMINFERENCESERVICECONFIG,\n config,\n )\n\nllmisvc/test_storage_version_migration.py:151: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \ngroup = 'serving.kserve.io', version = 'v1alpha1'\nnamespace = 'kserve-ci-e2e-test', plural = 'llminferenceserviceconfigs'\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'migration-test...ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0', 'name': 'main', 'resources': {'limits': {...}, 'requests': {...}}}]}}}\nkwargs = {'_return_http_data_only': True}\n\n def create_namespaced_custom_object(self, group, version, namespace, plural, body, **kwargs): # noqa: E501\n \"\"\"create_namespaced_custom_object # noqa: E501\n \n Creates a namespace scoped Custom object # noqa: E501\n This method makes a synchronous HTTP request by default. To make an\n asynchronous HTTP request, please pass async_req=True\n >>> thread = api.create_namespaced_custom_object(group, version, namespace, plural, body, async_req=True)\n >>> result = thread.get()\n \n :param async_req bool: execute request asynchronously\n :param str group: The custom resource's group name (required)\n :param str version: The custom resource's version (required)\n :param str namespace: The custom resource's namespace (required)\n :param str plural: The custom resource's plural name. For TPRs this would be lowercase plural kind. (required)\n :param object body: The JSON schema of the Resource to create. (required)\n :param str pretty: If 'true', then the output is pretty printed.\n :param str dry_run: When present, indicates that modifications should not be persisted. An invalid or unrecognized dryRun directive will result in an error response and no further processing of the request. Valid values are: - All: all dry run stages will be processed\n :param str field_manager: fieldManager is a name associated with the actor or entity that is making these changes. The value must be less than or 128 characters long, and only contain printable characters, as defined by https://golang.org/pkg/unicode/#IsPrint.\n :param str field_validation: fieldValidation instructs the server on how to handle objects in the request (POST/PUT/PATCH) containing unknown or duplicate fields. Valid values are: - Ignore: This will ignore any unknown fields that are silently dropped from the object, and will ignore all but the last duplicate field that the decoder encounters. This is the default behavior prior to v1.23. - Warn: This will send a warning via the standard warning response header for each unknown field that is dropped from the object, and for each duplicate field that is encountered. The request will still succeed if there are no other errors, and will only persist the last of any duplicate fields. This is the default in v1.23+ - Strict: This will fail the request with a BadRequest error if any unknown fields would be dropped from the object, or if any duplicate fields are present. The error returned from the server will contain all unknown and duplicate fields encountered. (optional)\n :param _preload_content: if False, the urllib3.HTTPResponse object will\n be returned without reading/decoding response\n data. Default is True.\n :param _request_timeout: timeout setting for this request. If one\n number provided, it will be total request\n timeout. It can also be a pair (tuple) of\n (connection, read) timeouts.\n :return: object\n If the method is called asynchronously,\n returns the request thread.\n \"\"\"\n kwargs['_return_http_data_only'] = True\n> return self.create_namespaced_custom_object_with_http_info(group, version, namespace, plural, body, **kwargs) # noqa: E501\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api/custom_objects_api.py:231: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \ngroup = 'serving.kserve.io', version = 'v1alpha1'\nnamespace = 'kserve-ci-e2e-test', plural = 'llminferenceserviceconfigs'\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'migration-test...ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0', 'name': 'main', 'resources': {'limits': {...}, 'requests': {...}}}]}}}\nkwargs = {'_return_http_data_only': True}\nlocal_var_params = {'_return_http_data_only': True, 'all_params': ['group', 'version', 'namespace', 'plural', 'body', 'pretty', ...], 'au...ers': [{'image': 'public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0', 'name': 'main', 'resources': {...}}]}}}, ...}\nall_params = ['group', 'version', 'namespace', 'plural', 'body', 'pretty', ...]\nkey = '_return_http_data_only', val = True, collection_formats = {}\npath_params = {'group': 'serving.kserve.io', 'namespace': 'kserve-ci-e2e-test', 'plural': 'llminferenceserviceconfigs', 'version': 'v1alpha1'}\nquery_params = []\n\n def create_namespaced_custom_object_with_http_info(self, group, version, namespace, plural, body, **kwargs): # noqa: E501\n \"\"\"create_namespaced_custom_object # noqa: E501\n \n Creates a namespace scoped Custom object # noqa: E501\n This method makes a synchronous HTTP request by default. To make an\n asynchronous HTTP request, please pass async_req=True\n >>> thread = api.create_namespaced_custom_object_with_http_info(group, version, namespace, plural, body, async_req=True)\n >>> result = thread.get()\n \n :param async_req bool: execute request asynchronously\n :param str group: The custom resource's group name (required)\n :param str version: The custom resource's version (required)\n :param str namespace: The custom resource's namespace (required)\n :param str plural: The custom resource's plural name. For TPRs this would be lowercase plural kind. (required)\n :param object body: The JSON schema of the Resource to create. (required)\n :param str pretty: If 'true', then the output is pretty printed.\n :param str dry_run: When present, indicates that modifications should not be persisted. An invalid or unrecognized dryRun directive will result in an error response and no further processing of the request. Valid values are: - All: all dry run stages will be processed\n :param str field_manager: fieldManager is a name associated with the actor or entity that is making these changes. The value must be less than or 128 characters long, and only contain printable characters, as defined by https://golang.org/pkg/unicode/#IsPrint.\n :param str field_validation: fieldValidation instructs the server on how to handle objects in the request (POST/PUT/PATCH) containing unknown or duplicate fields. Valid values are: - Ignore: This will ignore any unknown fields that are silently dropped from the object, and will ignore all but the last duplicate field that the decoder encounters. This is the default behavior prior to v1.23. - Warn: This will send a warning via the standard warning response header for each unknown field that is dropped from the object, and for each duplicate field that is encountered. The request will still succeed if there are no other errors, and will only persist the last of any duplicate fields. This is the default in v1.23+ - Strict: This will fail the request with a BadRequest error if any unknown fields would be dropped from the object, or if any duplicate fields are present. The error returned from the server will contain all unknown and duplicate fields encountered. (optional)\n :param _return_http_data_only: response data without head status code\n and headers\n :param _preload_content: if False, the urllib3.HTTPResponse object will\n be returned without reading/decoding response\n data. Default is True.\n :param _request_timeout: timeout setting for this request. If one\n number provided, it will be total request\n timeout. It can also be a pair (tuple) of\n (connection, read) timeouts.\n :return: tuple(object, status_code(int), headers(HTTPHeaderDict))\n If the method is called asynchronously,\n returns the request thread.\n \"\"\"\n \n local_var_params = locals()\n \n all_params = [\n 'group',\n 'version',\n 'namespace',\n 'plural',\n 'body',\n 'pretty',\n 'dry_run',\n 'field_manager',\n 'field_validation'\n ]\n all_params.extend(\n [\n 'async_req',\n '_return_http_data_only',\n '_preload_content',\n '_request_timeout'\n ]\n )\n \n for key, val in six.iteritems(local_var_params['kwargs']):\n if key not in all_params:\n raise ApiTypeError(\n \"Got an unexpected keyword argument '%s'\"\n \" to method create_namespaced_custom_object\" % key\n )\n local_var_params[key] = val\n del local_var_params['kwargs']\n # verify the required parameter 'group' is set\n if self.api_client.client_side_validation and ('group' not in local_var_params or # noqa: E501\n local_var_params['group'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `group` when calling `create_namespaced_custom_object`\") # noqa: E501\n # verify the required parameter 'version' is set\n if self.api_client.client_side_validation and ('version' not in local_var_params or # noqa: E501\n local_var_params['version'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `version` when calling `create_namespaced_custom_object`\") # noqa: E501\n # verify the required parameter 'namespace' is set\n if self.api_client.client_side_validation and ('namespace' not in local_var_params or # noqa: E501\n local_var_params['namespace'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `namespace` when calling `create_namespaced_custom_object`\") # noqa: E501\n # verify the required parameter 'plural' is set\n if self.api_client.client_side_validation and ('plural' not in local_var_params or # noqa: E501\n local_var_params['plural'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `plural` when calling `create_namespaced_custom_object`\") # noqa: E501\n # verify the required parameter 'body' is set\n if self.api_client.client_side_validation and ('body' not in local_var_params or # noqa: E501\n local_var_params['body'] is None): # noqa: E501\n raise ApiValueError(\"Missing the required parameter `body` when calling `create_namespaced_custom_object`\") # noqa: E501\n \n collection_formats = {}\n \n path_params = {}\n if 'group' in local_var_params:\n path_params['group'] = local_var_params['group'] # noqa: E501\n if 'version' in local_var_params:\n path_params['version'] = local_var_params['version'] # noqa: E501\n if 'namespace' in local_var_params:\n path_params['namespace'] = local_var_params['namespace'] # noqa: E501\n if 'plural' in local_var_params:\n path_params['plural'] = local_var_params['plural'] # noqa: E501\n \n query_params = []\n if 'pretty' in local_var_params and local_var_params['pretty'] is not None: # noqa: E501\n query_params.append(('pretty', local_var_params['pretty'])) # noqa: E501\n if 'dry_run' in local_var_params and local_var_params['dry_run'] is not None: # noqa: E501\n query_params.append(('dryRun', local_var_params['dry_run'])) # noqa: E501\n if 'field_manager' in local_var_params and local_var_params['field_manager'] is not None: # noqa: E501\n query_params.append(('fieldManager', local_var_params['field_manager'])) # noqa: E501\n if 'field_validation' in local_var_params and local_var_params['field_validation'] is not None: # noqa: E501\n query_params.append(('fieldValidation', local_var_params['field_validation'])) # noqa: E501\n \n header_params = {}\n \n form_params = []\n local_var_files = {}\n \n body_params = None\n if 'body' in local_var_params:\n body_params = local_var_params['body']\n # HTTP header `Accept`\n header_params['Accept'] = self.api_client.select_header_accept(\n ['application/json']) # noqa: E501\n \n # Authentication setting\n auth_settings = ['BearerToken'] # noqa: E501\n \n> return self.api_client.call_api(\n '/apis/{group}/{version}/namespaces/{namespace}/{plural}', 'POST',\n path_params,\n query_params,\n header_params,\n body=body_params,\n post_params=form_params,\n files=local_var_files,\n response_type='object', # noqa: E501\n auth_settings=auth_settings,\n async_req=local_var_params.get('async_req'),\n _return_http_data_only=local_var_params.get('_return_http_data_only'), # noqa: E501\n _preload_content=local_var_params.get('_preload_content', True),\n _request_timeout=local_var_params.get('_request_timeout'),\n collection_formats=collection_formats)\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api/custom_objects_api.py:354: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nresource_path = '/apis/{group}/{version}/namespaces/{namespace}/{plural}'\nmethod = 'POST'\npath_params = {'group': 'serving.kserve.io', 'namespace': 'kserve-ci-e2e-test', 'plural': 'llminferenceserviceconfigs', 'version': 'v1alpha1'}\nquery_params = []\nheader_params = {'Accept': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'migration-test...ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0', 'name': 'main', 'resources': {'limits': {...}, 'requests': {...}}}]}}}\npost_params = [], files = {}, response_type = 'object'\nauth_settings = ['BearerToken'], async_req = None, _return_http_data_only = True\ncollection_formats = {}, _preload_content = True, _request_timeout = None\n_host = None\n\n def call_api(self, resource_path, method,\n path_params=None, query_params=None, header_params=None,\n body=None, post_params=None, files=None,\n response_type=None, auth_settings=None, async_req=None,\n _return_http_data_only=None, collection_formats=None,\n _preload_content=True, _request_timeout=None, _host=None):\n \"\"\"Makes the HTTP request (synchronous) and returns deserialized data.\n \n To make an async_req request, set the async_req parameter.\n \n :param resource_path: Path to method endpoint.\n :param method: Method to call.\n :param path_params: Path parameters in the url.\n :param query_params: Query parameters in the url.\n :param header_params: Header parameters to be\n placed in the request header.\n :param body: Request body.\n :param post_params dict: Request post form parameters,\n for `application/x-www-form-urlencoded`, `multipart/form-data`.\n :param auth_settings list: Auth Settings names for the request.\n :param response: Response data type.\n :param files dict: key -> filename, value -> filepath,\n for `multipart/form-data`.\n :param async_req bool: execute request asynchronously\n :param _return_http_data_only: response data without head status code\n and headers\n :param collection_formats: dict of collection formats for path, query,\n header, and post parameters.\n :param _preload_content: if False, the urllib3.HTTPResponse object will\n be returned without reading/decoding response\n data. Default is True.\n :param _request_timeout: timeout setting for this request. If one\n number provided, it will be total request\n timeout. It can also be a pair (tuple) of\n (connection, read) timeouts.\n :return:\n If async_req parameter is True,\n the request will be called asynchronously.\n The method will return the request thread.\n If parameter async_req is False or missing,\n then the method will return the response directly.\n \"\"\"\n if not async_req:\n> return self.__call_api(resource_path, method,\n path_params, query_params, header_params,\n body, post_params, files,\n response_type, auth_settings,\n _return_http_data_only, collection_formats,\n _preload_content, _request_timeout, _host)\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py:348: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nresource_path = '/apis/serving.kserve.io/v1alpha1/namespaces/kserve-ci-e2e-test/llminferenceserviceconfigs'\nmethod = 'POST'\npath_params = [('group', 'serving.kserve.io'), ('version', 'v1alpha1'), ('namespace', 'kserve-ci-e2e-test'), ('plural', 'llminferenceserviceconfigs')]\nquery_params = []\nheader_params = {'Accept': 'application/json', 'Content-Type': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'migration-test...ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0', 'name': 'main', 'resources': {'limits': {...}, 'requests': {...}}}]}}}\npost_params = [], files = {}, response_type = 'object'\nauth_settings = ['BearerToken'], _return_http_data_only = True\ncollection_formats = {}, _preload_content = True, _request_timeout = None\n_host = None\n\n def __call_api(\n self, resource_path, method, path_params=None,\n query_params=None, header_params=None, body=None, post_params=None,\n files=None, response_type=None, auth_settings=None,\n _return_http_data_only=None, collection_formats=None,\n _preload_content=True, _request_timeout=None, _host=None):\n \n config = self.configuration\n \n # header parameters\n header_params = header_params or {}\n header_params.update(self.default_headers)\n if self.cookie:\n header_params['Cookie'] = self.cookie\n if header_params:\n header_params = self.sanitize_for_serialization(header_params)\n header_params = dict(self.parameters_to_tuples(header_params,\n collection_formats))\n \n # path parameters\n if path_params:\n path_params = self.sanitize_for_serialization(path_params)\n path_params = self.parameters_to_tuples(path_params,\n collection_formats)\n for k, v in path_params:\n # specified safe chars, encode everything\n resource_path = resource_path.replace(\n '{%s}' % k,\n quote(str(v), safe=config.safe_chars_for_path_param)\n )\n \n # query parameters\n if query_params:\n query_params = self.sanitize_for_serialization(query_params)\n query_params = self.parameters_to_tuples(query_params,\n collection_formats)\n \n # post parameters\n if post_params or files:\n post_params = post_params if post_params else []\n post_params = self.sanitize_for_serialization(post_params)\n post_params = self.parameters_to_tuples(post_params,\n collection_formats)\n post_params.extend(self.files_parameters(files))\n \n # auth setting\n self.update_params_for_auth(header_params, query_params, auth_settings)\n \n # body\n if body:\n body = self.sanitize_for_serialization(body)\n \n # request url\n if _host is None:\n url = self.configuration.host + resource_path\n else:\n # use server/host defined in path or operation instead\n url = _host + resource_path\n \n # perform request and return response\n> response_data = self.request(\n method, url, query_params=query_params, headers=header_params,\n post_params=post_params, body=body,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout)\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py:180: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nmethod = 'POST'\nurl = 'https://abc8a0cf66a9a44cba4c61b8a1456aeb-75321b9aa6cba46b.elb.us-east-1.amazonaws.com:6443/apis/serving.kserve.io/v1alpha1/namespaces/kserve-ci-e2e-test/llminferenceserviceconfigs'\nquery_params = []\nheaders = {'Accept': 'application/json', 'Content-Type': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\npost_params = []\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'migration-test...ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0', 'name': 'main', 'resources': {'limits': {...}, 'requests': {...}}}]}}}\n_preload_content = True, _request_timeout = None\n\n def request(self, method, url, query_params=None, headers=None,\n post_params=None, body=None, _preload_content=True,\n _request_timeout=None):\n \"\"\"Makes the HTTP request using RESTClient.\"\"\"\n if method == \"GET\":\n return self.rest_client.GET(url,\n query_params=query_params,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout,\n headers=headers)\n elif method == \"HEAD\":\n return self.rest_client.HEAD(url,\n query_params=query_params,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout,\n headers=headers)\n elif method == \"OPTIONS\":\n return self.rest_client.OPTIONS(url,\n query_params=query_params,\n headers=headers,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout)\n elif method == \"POST\":\n> return self.rest_client.POST(url,\n query_params=query_params,\n headers=headers,\n post_params=post_params,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout,\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/api_client.py:391: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nurl = 'https://abc8a0cf66a9a44cba4c61b8a1456aeb-75321b9aa6cba46b.elb.us-east-1.amazonaws.com:6443/apis/serving.kserve.io/v1alpha1/namespaces/kserve-ci-e2e-test/llminferenceserviceconfigs'\nheaders = {'Accept': 'application/json', 'Content-Type': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\nquery_params = [], post_params = []\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'migration-test...ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0', 'name': 'main', 'resources': {'limits': {...}, 'requests': {...}}}]}}}\n_preload_content = True, _request_timeout = None\n\n def POST(self, url, headers=None, query_params=None, post_params=None,\n body=None, _preload_content=True, _request_timeout=None):\n> return self.request(\"POST\", url,\n headers=headers,\n query_params=query_params,\n post_params=post_params,\n _preload_content=_preload_content,\n _request_timeout=_request_timeout,\n body=body)\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/rest.py:279: \n_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \n\nself = \nmethod = 'POST'\nurl = 'https://abc8a0cf66a9a44cba4c61b8a1456aeb-75321b9aa6cba46b.elb.us-east-1.amazonaws.com:6443/apis/serving.kserve.io/v1alpha1/namespaces/kserve-ci-e2e-test/llminferenceserviceconfigs'\nquery_params = []\nheaders = {'Accept': 'application/json', 'Content-Type': 'application/json', 'User-Agent': 'OpenAPI-Generator/32.0.1/python'}\nbody = {'apiVersion': 'serving.kserve.io/v1alpha1', 'kind': 'LLMInferenceServiceConfig', 'metadata': {'name': 'migration-test...ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.19.0', 'name': 'main', 'resources': {'limits': {...}, 'requests': {...}}}]}}}\npost_params = {}, _preload_content = True, _request_timeout = None\n\n def request(self, method, url, query_params=None, headers=None,\n body=None, post_params=None, _preload_content=True,\n _request_timeout=None):\n \"\"\"Perform requests.\n \n :param method: http request method\n :param url: http request url\n :param query_params: query parameters in the url\n :param headers: http request headers\n :param body: request json body, for `application/json`\n :param post_params: request post parameters,\n `application/x-www-form-urlencoded`\n and `multipart/form-data`\n :param _preload_content: if False, the urllib3.HTTPResponse object will\n be returned without reading/decoding response\n data. Default is True.\n :param _request_timeout: timeout setting for this request. If one\n number provided, it will be total request\n timeout. It can also be a pair (tuple) of\n (connection, read) timeouts.\n \"\"\"\n method = method.upper()\n assert method in ['GET', 'HEAD', 'DELETE', 'POST', 'PUT',\n 'PATCH', 'OPTIONS']\n \n if post_params and body:\n raise ApiValueError(\n \"body parameter cannot be used with post_params parameter.\"\n )\n \n post_params = post_params or {}\n headers = headers or {}\n \n timeout = None\n if _request_timeout:\n if isinstance(_request_timeout, (int, ) if six.PY3 else (int, long)): # noqa: E501,F821\n timeout = urllib3.Timeout(total=_request_timeout)\n elif (isinstance(_request_timeout, tuple) and\n len(_request_timeout) == 2):\n timeout = urllib3.Timeout(\n connect=_request_timeout[0], read=_request_timeout[1])\n \n if 'Content-Type' not in headers:\n headers['Content-Type'] = 'application/json'\n \n try:\n # For `POST`, `PUT`, `PATCH`, `OPTIONS`, `DELETE`\n if method in ['POST', 'PUT', 'PATCH', 'OPTIONS', 'DELETE']:\n if query_params:\n url += '?' + urlencode(query_params)\n if (re.search('json', headers['Content-Type'], re.IGNORECASE) or\n headers['Content-Type'] == 'application/apply-patch+yaml'):\n if headers['Content-Type'] == 'application/json-patch+json':\n if not isinstance(body, list):\n headers['Content-Type'] = \\\n 'application/strategic-merge-patch+json'\n request_body = None\n if body is not None:\n request_body = json.dumps(body)\n r = self.pool_manager.request(\n method, url,\n body=request_body,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n elif headers['Content-Type'] == 'application/x-www-form-urlencoded': # noqa: E501\n r = self.pool_manager.request(\n method, url,\n fields=post_params,\n encode_multipart=False,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n elif headers['Content-Type'] == 'multipart/form-data':\n # must del headers['Content-Type'], or the correct\n # Content-Type which generated by urllib3 will be\n # overwritten.\n del headers['Content-Type']\n r = self.pool_manager.request(\n method, url,\n fields=post_params,\n encode_multipart=True,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n # Pass a `string` parameter directly in the body to support\n # other content types than Json when `body` argument is\n # provided in serialized form\n elif isinstance(body, str) or isinstance(body, bytes):\n request_body = body\n r = self.pool_manager.request(\n method, url,\n body=request_body,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n else:\n # Cannot generate the request from given parameters\n msg = \"\"\"Cannot prepare a request message for provided\n arguments. Please check that your arguments match\n declared content type.\"\"\"\n raise ApiException(status=0, reason=msg)\n # For `GET`, `HEAD`\n else:\n r = self.pool_manager.request(method, url,\n fields=query_params,\n preload_content=_preload_content,\n timeout=timeout,\n headers=headers)\n except urllib3.exceptions.SSLError as e:\n msg = \"{0}\\n{1}\".format(type(e).__name__, str(e))\n raise ApiException(status=0, reason=msg)\n \n if _preload_content:\n r = RESTResponse(r)\n \n # In the python 3, the response.data is bytes.\n # we need to decode it to string.\n if six.PY3:\n r.data = r.data.decode('utf8')\n \n # log response body\n logger.debug(\"response body: %s\", r.data)\n \n if not 200 <= r.status <= 299:\n> raise ApiException(http_resp=r)\nE kubernetes.client.exceptions.ApiException: (500)\nE Reason: Internal Server Error\nE HTTP response headers: HTTPHeaderDict({'Audit-Id': '9ba752cc-fbc7-4911-b3e2-604b9a48b2e5', 'Cache-Control': 'no-cache, private', 'Content-Type': 'application/json', 'Strict-Transport-Security': 'max-age=31536000; includeSubDomains; preload', 'X-Kubernetes-Pf-Flowschema-Uid': 'c30fff65-8b8a-47f0-8303-5f205881ee46', 'X-Kubernetes-Pf-Prioritylevel-Uid': '1d0f3173-41ee-4f4d-bdbf-9bfbca43ee13', 'Date': 'Mon, 13 Jul 2026 22:11:24 GMT', 'Content-Length': '833'})\nE HTTP response body: {\"kind\":\"Status\",\"apiVersion\":\"v1\",\"metadata\":{},\"status\":\"Failure\",\"message\":\"Internal error occurred: failed calling webhook \\\"llminferenceserviceconfig.kserve-webhook-server.v1alpha1.validator\\\": failed to call webhook: Post \\\"https://llmisvc-webhook-server-service.kserve.svc:443/validate-serving-kserve-io-v1alpha1-llminferenceserviceconfig?timeout=10s\\\": no endpoints available for service \\\"llmisvc-webhook-server-service\\\"\",\"reason\":\"InternalError\",\"details\":{\"causes\":[{\"message\":\"failed calling webhook \\\"llminferenceserviceconfig.kserve-webhook-server.v1alpha1.validator\\\": failed to call webhook: Post \\\"https://llmisvc-webhook-server-service.kserve.svc:443/validate-serving-kserve-io-v1alpha1-llminferenceserviceconfig?timeout=10s\\\": no endpoints available for service \\\"llmisvc-webhook-server-service\\\"\"}]},\"code\":500}\n\n../../python/kserve/.venv/lib64/python3.11/site-packages/kubernetes/client/rest.py:238: ApiException"}, "teardown": {"duration": 2.0584711139963474, "outcome": "passed", "longrepr": "[gw1] linux -- Python 3.11.13 /workspace/source/python/kserve/.venv/bin/python"}}], "warnings": [{"message": "The event_loop fixture provided by pytest-asyncio has been redefined in\n/workspace/source/test/e2e/conftest.py:43\nReplacing the event_loop fixture with a custom implementation is deprecated\nand will lead to errors in the future.\nIf you want to request an asyncio event loop with a scope other than function\nscope, use the \"scope\" argument to the asyncio mark when marking the tests.\nIf you want to return different types of event loops, use the event_loop_policy\nfixture.\n", "category": "DeprecationWarning", "when": "runtest", "filename": "/workspace/source/python/kserve/.venv/lib64/python3.11/site-packages/pytest_asyncio/plugin.py", "lineno": 761}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The event_loop fixture provided by pytest-asyncio has been redefined in\n/workspace/source/test/e2e/conftest.py:43\nReplacing the event_loop fixture with a custom implementation is deprecated\nand will lead to errors in the future.\nIf you want to request an asyncio event loop with a scope other than function\nscope, use the \"scope\" argument to the asyncio mark when marking the tests.\nIf you want to return different types of event loops, use the event_loop_policy\nfixture.\n", "category": "DeprecationWarning", "when": "runtest", "filename": "/workspace/source/python/kserve/.venv/lib64/python3.11/site-packages/pytest_asyncio/plugin.py", "lineno": 761}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service_stop.py", "lineno": 40}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_inference_service.py", "lineno": 245}, {"message": "The test is marked with '@pytest.mark.asyncio' but it is not an async function. Please remove the asyncio mark. If the test is not marked explicitly, check for global marks applied via 'pytestmark'.", "category": "PytestWarning", "when": "runtest", "filename": "llmisvc/test_llm_tls.py", "lineno": 93}]}