self = <test_api_keys.TestAPIKeyModelInference object at 0x7fb43f07c6d0> model_completions_url = 'https://maas.apps.b6eed26b-8fd3-48dc-8fce-4c8c0972e457.prod.konfluxeaas.com/llm/facebook-opt-125m-simulated/v1/completions' api_key_headers = {'Authorization': 'Bearer sk-oai-1Y4HICU0ZBVldktdK_oeHSnAE90NBT81KZxigRaxq7ZQYuLSSmRPF8hAqqoZS', 'Content-Type': 'application/json'} inference_model_name = 'facebook/opt-125m' def test_api_key_model_access_success( self, model_completions_url: str, api_key_headers: dict, inference_model_name: str, ): """Test 11: Valid API key can access model endpoint - verify 200 response. Subscription is bound on the key at mint (see conftest ``api_key`` fixture). """ r = requests.post( model_completions_url, headers=api_key_headers, json={ "model": inference_model_name, "prompt": "Hello world", "max_tokens": 10, }, timeout=60, verify=TLS_VERIFY, ) > assert r.status_code == 200, f"Expected 200, got {r.status_code}: {r.text}" E AssertionError: Expected 200, got 500: Internal Server Error. E E assert 500 == 200 E + where 500 = <Response [500]>.status_code test/e2e/tests/test_api_keys.py:472: AssertionErrorself = <test_api_keys.TestAPIKeyModelInference object at 0x7fb43f07cc10> model_completions_url = 'https://maas.apps.b6eed26b-8fd3-48dc-8fce-4c8c0972e457.prod.konfluxeaas.com/llm/facebook-opt-125m-simulated/v1/completions' inference_model_name = 'facebook/opt-125m' def test_invalid_api_key_rejected( self, model_completions_url: str, inference_model_name: str, ): """Test 12: Invalid API key should be rejected with 403.""" invalid_headers = { "Authorization": "Bearer sk-oai-invalid-key-12345", "Content-Type": "application/json", } r = requests.post( model_completions_url, headers=invalid_headers, json={ "model": inference_model_name, "prompt": "Test", "max_tokens": 5, }, timeout=30, verify=TLS_VERIFY, ) > assert r.status_code == 403, f"Expected 403 for invalid key, got {r.status_code}: {r.text}" E AssertionError: Expected 403 for invalid key, got 500: Internal Server Error. E E assert 500 == 403 E + where 500 = <Response [500]>.status_code test/e2e/tests/test_api_keys.py:505: AssertionErrorself = <test_api_keys.TestAPIKeyModelInference object at 0x7fb43f074070> api_keys_base_url = 'https://maas.apps.b6eed26b-8fd3-48dc-8fce-4c8c0972e457.prod.konfluxeaas.com/maas-api/v1/api-keys' model_completions_url = 'https://maas.apps.b6eed26b-8fd3-48dc-8fce-4c8c0972e457.prod.konfluxeaas.com/llm/facebook-opt-125m-simulated/v1/completions' headers = {'Authorization': 'Bearer eyJhbGciOiJSUzI1NiIsImtpZCI6IjkxendNWFRwWUh4SDRIR2dySVlHUlpsbmFPTVQ2ZHhuWnFuMDc5ckpSc0UifQ.e...XTD8EFEJJWBOngmvrpdkfoY_mSKjNZ8DmdqzE4Xq3dOZkDCXhMeyi2hxOpsDTNav9IdBzBiqh05dbsm0A', 'Content-Type': 'application/json'} inference_model_name = 'facebook/opt-125m' def test_revoked_api_key_rejected( self, api_keys_base_url: str, model_completions_url: str, headers: dict, inference_model_name: str, ): """Test 14: Revoked API key should be rejected with 403.""" # Create a new key designated = SIMULATOR_SUBSCRIPTION r_create = requests.post( api_keys_base_url, headers=headers, json={"name": "test-revoke-inference", "subscription": designated}, timeout=30, verify=TLS_VERIFY, ) assert r_create.status_code in (200, 201), f"Failed to create key: {r_create.text}" data = r_create.json() key = data["key"] key_id = data["id"] # Verify key works first key_headers = {"Authorization": f"Bearer {key}", "Content-Type": "application/json"} r_test = requests.post( model_completions_url, headers=key_headers, json={"model": inference_model_name, "prompt": "Test", "max_tokens": 5}, timeout=60, verify=TLS_VERIFY, ) # Key should work (200) or might fail for other reasons - we just need to test revocation initial_status = r_test.status_code print(f"[inference] Key before revoke: HTTP {initial_status}") # Revoke the key r_revoke = requests.delete( f"{api_keys_base_url}/{key_id}", headers=headers, timeout=30, verify=TLS_VERIFY, ) assert r_revoke.status_code == 200, f"Failed to revoke: {r_revoke.text}" assert r_revoke.json().get("status") == "revoked" # Wait for revocation to propagate time.sleep(2) # Try to use revoked key r_revoked = requests.post( model_completions_url, headers=key_headers, json={"model": inference_model_name, "prompt": "Test", "max_tokens": 5}, timeout=30, verify=TLS_VERIFY, ) > assert r_revoked.status_code == 403, f"Expected 403 for revoked key, got {r_revoked.status_code}: {r_revoked.text}" E AssertionError: Expected 403 for revoked key, got 500: Internal Server Error. E E assert 500 == 403 E + where 500 = <Response [500]>.status_code test/e2e/tests/test_api_keys.py:588: AssertionError/workspace/source/test/e2e/tests/test_api_keys.py:624: Chat completions returned 500self = <test_api_keys.TestAPIKeyRevocationE2E object at 0x7fb43f0743d0> api_keys_base_url = 'https://maas.apps.b6eed26b-8fd3-48dc-8fce-4c8c0972e457.prod.konfluxeaas.com/maas-api/v1/api-keys' model_completions_url = 'https://maas.apps.b6eed26b-8fd3-48dc-8fce-4c8c0972e457.prod.konfluxeaas.com/llm/facebook-opt-125m-simulated/v1/completions' headers = {'Authorization': 'Bearer eyJhbGciOiJSUzI1NiIsImtpZCI6IjkxendNWFRwWUh4SDRIR2dySVlHUlpsbmFPTVQ2ZHhuWnFuMDc5ckpSc0UifQ.e...XTD8EFEJJWBOngmvrpdkfoY_mSKjNZ8DmdqzE4Xq3dOZkDCXhMeyi2hxOpsDTNav9IdBzBiqh05dbsm0A', 'Content-Type': 'application/json'} inference_model_name = 'facebook/opt-125m' def test_revoke_then_create_new_key_works( self, api_keys_base_url: str, model_completions_url: str, headers: dict, inference_model_name: str, ): """After revoking a key, a newly created key should still work for inference.""" designated = SIMULATOR_SUBSCRIPTION # Create key A r_a = requests.post( api_keys_base_url, headers=headers, json={"name": "test-revoke-remint-a", "subscription": designated}, timeout=30, verify=TLS_VERIFY, ) assert r_a.status_code in (200, 201), f"Failed to create key A: {r_a.text}" key_a = r_a.json()["key"] key_a_id = r_a.json()["id"] # Revoke key A r_revoke = requests.delete(f"{api_keys_base_url}/{key_a_id}", headers=headers, timeout=30, verify=TLS_VERIFY) assert r_revoke.status_code == 200 # Create key B (new key after revocation) r_b = requests.post( api_keys_base_url, headers=headers, json={"name": "test-revoke-remint-b", "subscription": designated}, timeout=30, verify=TLS_VERIFY, ) assert r_b.status_code in (200, 201), f"Failed to create key B: {r_b.text}" key_b = r_b.json()["key"] # Poll until revoked key A is rejected (revocation may take time to propagate) max_wait = 30 poll_interval = 0.5 deadline = time.monotonic() + max_wait while True: remaining = deadline - time.monotonic() if remaining <= 0: break r_a_inf = requests.post( model_completions_url, headers={"Authorization": f"Bearer {key_a}", "Content-Type": "application/json"}, json={"model": inference_model_name, "prompt": "Test", "max_tokens": 5}, timeout=min(remaining, 10), verify=TLS_VERIFY, ) if r_a_inf.status_code == 403: break time.sleep(poll_interval) > assert r_a_inf.status_code == 403, ( f"Revoked key A should be rejected within {max_wait}s, got {r_a_inf.status_code}" ) E AssertionError: Revoked key A should be rejected within 30s, got 500 E assert 500 == 403 E + where 500 = <Response [500]>.status_code test/e2e/tests/test_api_keys.py:712: AssertionErrorself = <test_api_keys.TestAPIKeyRevocationE2E object at 0x7fb43f049520> api_keys_base_url = 'https://maas.apps.b6eed26b-8fd3-48dc-8fce-4c8c0972e457.prod.konfluxeaas.com/maas-api/v1/api-keys' model_completions_url = 'https://maas.apps.b6eed26b-8fd3-48dc-8fce-4c8c0972e457.prod.konfluxeaas.com/llm/facebook-opt-125m-simulated/v1/completions' headers = {'Authorization': 'Bearer eyJhbGciOiJSUzI1NiIsImtpZCI6IjkxendNWFRwWUh4SDRIR2dySVlHUlpsbmFPTVQ2ZHhuWnFuMDc5ckpSc0UifQ.e...XTD8EFEJJWBOngmvrpdkfoY_mSKjNZ8DmdqzE4Xq3dOZkDCXhMeyi2hxOpsDTNav9IdBzBiqh05dbsm0A', 'Content-Type': 'application/json'} inference_model_name = 'facebook/opt-125m' def test_revoke_keys_rejected_at_gateway( self, api_keys_base_url: str, model_completions_url: str, headers: dict, inference_model_name: str, ): """After individually revoking keys, they should be rejected at the gateway.""" designated = SIMULATOR_SUBSCRIPTION # Create 3 keys, capturing plaintext and IDs keys = [] for i in range(3): r = requests.post( api_keys_base_url, headers=headers, json={"name": f"test-revoke-gw-{i}", "subscription": designated}, timeout=30, verify=TLS_VERIFY, ) assert r.status_code in (200, 201), f"Failed to create key {i}: {r.text}" keys.append({"id": r.json()["id"], "key": r.json()["key"]}) # Smoke-test: verify at least one key works before revocation r_smoke = requests.post( model_completions_url, headers={"Authorization": f"Bearer {keys[0]['key']}", "Content-Type": "application/json"}, json={"model": inference_model_name, "prompt": "Test", "max_tokens": 5}, timeout=60, verify=TLS_VERIFY, ) > assert r_smoke.status_code == 200, ( f"Key should work before revoke, got {r_smoke.status_code}: {r_smoke.text}" ) E AssertionError: Key should work before revoke, got 500: Internal Server Error. E E assert 500 == 200 E + where 500 = <Response [500]>.status_code test/e2e/tests/test_api_keys.py:782: AssertionError