Gateway API (Envoy Gateway)

Kubernetes Gateway API ingress controller powered by Envoy Gateway.

Agentic Friendly

Component Category

Networking / ingress and routing

Component Description

Gateway API is the Kubernetes-native traffic-routing standard used by the platform. Envoy Gateway implements the specification with Envoy proxy as the data plane. The component manages a shared Gateway resource and per-component HTTPRoute resources for traffic routing.

Why It Is Used

In BullSequana AI, Gateway API provides the primary ingress path for all platform endpoints. Its routing model supports advanced traffic policies, header-based routing, and native TLS termination through the Kubernetes Gateway API specification.

Learn More

Deployment notes

Gateway API deploys Envoy Gateway v1.8.1 into the gateway-api namespace at sync wave 4 in the common tier. It creates a GatewayClass, the shared Gateway, and an EnvoyProxy configuration that runs two Envoy replicas behind a LoadBalancer Service. Individual components define their own HTTPRoute resources to register their endpoints with the gateway.

The shared Gateway has an HTTPS listener for *.<base-domain>, which terminates TLS with the wildcard certificate, and an HTTP listener. The HTTP listener admits only the platform's catch-all redirect route, so every plain-HTTP request receives 301 to HTTPS. Application routes bind to the HTTPS listener.

GATEWAY_API_GATEWAY_LOAD_BALANCER_SOURCE_RANGES takes a comma-separated list of CIDRs that may reach the load balancer. Empty, the default, leaves it open. A ClientTrafficPolicy drops request headers whose name contains an underscore, instead of rejecting the request, so such headers can neither break nor impersonate platform headers.

The internal model gateway, kserve-internal, uses the same GatewayClass in the same namespace, with a ClusterIP Service. See Inference.

Interacts With

  • cert-manager, which provisions TLS certificates referenced by the Gateway resource.
  • External DNS, which creates DNS records for endpoints exposed through the gateway.
  • KServe internal gateway, which shares the GatewayClass and carries model traffic from the BSQAI API.
  • Gateway API Inference Extension, whose InferencePool resources the Envoy Gateway controller reads for model-aware routing.
  • All platform components, which define HTTPRoute resources to route traffic through the shared gateway.

On this page