401 missing_api_key | Missing/malformed bearer header | Authorization: Bearer ... |
401 admin_authentication_required / invalid_admin_credentials | Missing/expired console session or incorrect login | Sign in again on the private operator origin; verify the deployment-owned username/password secret |
429 admin_login_rate_limited | Global or per-username rolling login bucket is full | Wait for Retry-After; investigate repeated failures before raising the limit |
503 admin_login_throttle_unavailable | Redis-backed login enforcement is unavailable | Restore Redis and readiness before allowing password login |
429 oidc_login_rate_limited | The global OIDC initiation bucket is full | Wait for Retry-After; investigate initiation traffic before raising the limit |
503 oidc_login_throttle_unavailable | Redis-backed OIDC initiation enforcement is unavailable | Restore Redis and readiness before allowing SSO initiation |
401 oidc_login_failed | State expired/replayed, token validation failed, or subject is not mapped | Start a new SSO login; verify exact issuer/subject mapping and IdP client registration |
503 oidc_provider_unavailable | Discovery, token, or key endpoint is unavailable/unsafe | Verify IdP availability, HTTPS endpoints, issuer equality, and client secret |
503 admin_audit_unavailable | The pre-action audit event could not be persisted | Restore Postgres/audit readiness; the mutation did not run |
X-Gateway-Audit-Status: failed | The outcome event failed after processing | Correlate X-Request-ID, inspect target state, and do not retry blindly |
401 invalid_api_key | Wrong or rotated token | Key secret and /admin/keys status |
| Previous token stopped working | Rotation overlap expired or another rotation replaced the prior generation | Install the newest one-time token; inspect previous_token_valid_until metadata |
403 endpoint_not_allowed | Key lacks exact route scope | allowed_endpoints including /v1 form |
| 403 model denial | Alias not granted | /v1/gateway/info and key allowed_models |
400 unsupported_param | Endpoint/model rejects a field | Remove field or choose compatible alias |
| 400 output-limit error | TPM/budget needs maximum output | Add max_completion_tokens or max_output_tokens |
| 429 rate limit | RPM/TPM exhausted | Retry-After, key limits, traffic fan-out |
| 429 budget exceeded | Available budget cannot cover reservation | /admin/budgets, output cap, period reset |
| 503 limit backend unavailable | Redis cannot be reached | /ready, Redis network/TLS/credentials |
shared_cooldown_unavailable log event | Redis cooldown read/write failed; direct-HTTP routing is using process-local cooldown state | Restore Redis and readiness; verify healthy providers/fallbacks continue serving |
model_catalog_ready: false with serving_ready: true | Replica is lagging or cannot activate the latest graph, but retains a safe runtime | model_catalog_activation_failed, /admin/model-catalog propagation status, database health, repair or reset pending revision |
data_retention_ready: false | Lease, purge, or retention-state persistence failed | /admin/data-governance, data_retention_failed, Postgres health |
| 502/504 provider error | Upstream authentication/network/timeout | Provider status, secret, API base, timeout |
/health works but traffic fails | Liveness is not readiness | Private /ready component flags |
| Usage missing | Log queue exhausted during DB outage | gateway_request_logs_dropped_total |
| Cost missing | Provider usage/cost unavailable or route unpriced | costed_requests, model pricing snapshot |
| Config reload changes one instance | Reload is process-local | Deploy immutable config to every process |
