Kubernetes Gateway API

Kuma supports Kubernetes Gateway API for configuring built-in gateway as well as traffic routing using the experimental GAMMA routing spec.

Installation

Kubernetes doesn’t include Gateway API CRDs, install them from the standard channel CRD bundle.

Gateways

Install the counter demo.

kumactl install demo | kubectl apply -f -

Add a Gateway.

The Gateway resource represents the proxy instance that handles traffic for a set of Gateway API routes.

Every Gateway refers to a GatewayClass. The GatewayClass represents the class of Gateway, in this case Kuma’s builtin edge gateway, and points to a controller that should manage these Gateways. It can also hold additional configuration.

Standard install
Custom install

For Helm and kumactl installations, a GatewayClass named kuma is automatically installed if the Gateway API CRDs are present.

If you’ve installed Kuma some other way, you can create your own GatewayClass using the controllerName: gateways.kuma.io/controller:

apiVersion: gateway.networking.k8s.io/v1
kind: GatewayClass
metadata:
  name: kuma
spec:
  controllerName: gateways.kuma.io/controller

apiVersion: gateway.networking.k8s.io/v1
kind: Gateway
metadata:
  name: kuma
  namespace: kuma-demo
spec:
  gatewayClassName: kuma
  listeners:
  - name: proxy
    port: 8080
    protocol: HTTP

When a user applies a Gateway resource, Kuma automatically creates a Deployment of built-in gateways with a corresponding Service.

kubectl get pods -n kuma-demo

NAME                          READY   STATUS    RESTARTS   AGE
redis-59c9d56fc-6gcbc         2/2     Running   0          2m8s
demo-app-5845d6447b-v7npw     2/2     Running   0          2m8s
kuma-4j6wr-58998b5576-25wl6   1/1     Running   0          30s

kubectl get svc -n kuma-demo

NAME         TYPE           CLUSTER-IP      EXTERNAL-IP   PORT(S)          AGE
redis        ClusterIP      10.43.223.223   <none>        6379/TCP         3m27s
demo-app     ClusterIP      10.43.216.203   <none>        5000/TCP         3m27s
kuma-pfh4s   LoadBalancer   10.43.122.93    172.20.0.3    8080:30627/TCP   87s

The Gateway is now accessible using the external address 172.20.0.3:8080.

Add an HTTPRoute.

HTTPRoute resources contain a set of matching criteria for HTTP requests and upstream Services to route those requests to.

apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: echo
  namespace: kuma-demo
spec:
  parentRefs:
  - group: gateway.networking.k8s.io
    kind: Gateway
    name: kuma
    namespace: kuma-demo
  rules:
  - backendRefs:
    - kind: Service
      name: demo-app
      port: 5000
      weight: 1
    matches:
    - path:
        type: PathPrefix
        value: /

After creating an HTTPRoute, accessing / forwards a request to the demo app:

curl 172.20.0.3:8080/ -i

HTTP/1.1 200 OK
x-powered-by: Express
accept-ranges: bytes
cache-control: public, max-age=0
last-modified: Tue, 20 Oct 2020 17:16:41 GMT
etag: W/"2b91-175470350a8"
content-type: text/html; charset=UTF-8
content-length: 11153
date: Fri, 18 Mar 2022 11:33:29 GMT
x-envoy-upstream-service-time: 2
server: Kuma Gateway

<html>
<head>
...

TLS termination

Gateway API supports TLS termination by using standard kubernetes.io/tls Secrets.

Here is an example

apiVersion: v1
kind: Secret
metadata:
  name: secret-tls
  namespace: kuma-demo
type: kubernetes.io/tls
data:
  tls.crt: "MIIEOzCCAyO..." # redacted
  tls.key: "MIIEowIBAAKC..." # redacted

apiVersion: gateway.networking.k8s.io/v1
kind: Gateway
metadata:
  name: kuma
  namespace: kuma-demo
spec:
  gatewayClassName: kuma
  listeners:
  - name: proxy
    port: 8080
    hostname: test.kuma.io
    protocol: HTTPS
    tls:
      certificateRefs:
      - name: secret-tls

Under the hood, Kuma CP copies the Secret to kuma-system namespace and converts it to Kuma secret . It tracks all the changes to the secret and deletes it upon deletion of the original secret.

Customization

Gateway API provides the parametersRef field on GatewayClass.spec to provide additional, implementation-specific configuration to Gateways. When using Gateway API with Kuma, you can refer to a MeshGatewayConfig resource:

apiVersion: gateway.networking.k8s.io/v1
kind: GatewayClass
metadata:
  name: kuma
spec:
  controllerName: gateways.kuma.io/controller
  parametersRef:
    kind: MeshGatewayConfig
    group: kuma.io
    name: kuma

This resource has the same structure as the MeshGatewayInstance resource except that the tags field is optional. With a MeshGatewayConfig you can then customize the generated Service and Deployment resources.

Multi-mesh

You can specify a Mesh for Gateway and HTTPRoute resources by setting the kuma.io/mesh annotation Note that HTTPRoutes must also have the annotation to reference a Gateway from a non-default Mesh.

Cross-mesh

Cross-mesh gateways are supported with Gateway API. You’ll just need to create a corresponding GatewayClass pointing to a MeshGatewayConfig that sets crossMesh: true:

apiVersion: gateway.networking.k8s.io/v1
kind: GatewayClass
metadata:
  name: kuma-cross-mesh
spec:
  controllerName: gateways.kuma.io/controller
  parametersRef:
    group: kuma.io
    kind: MeshGatewayConfig
    name: default-cross-mesh
---
apiVersion: kuma.io/v1alpha1
kind: MeshGatewayConfig
metadata:
  name: default-cross-mesh
spec:
  crossMesh: true

and then reference it in your Gateway:

apiVersion: gateway.networking.k8s.io/v1
kind: Gateway
metadata:
  name: kuma
  namespace: default
spec:
  gatewayClassName: kuma-cross-mesh
  listeners:
  - name: proxy
    port: 8080
    protocol: HTTP

Multi-zone Deployments

The Gateway API supports multi-zone deployments, but with some limitations:

Gateway API resources like Gateway, ReferenceGrant, and HTTPRoute must be created in non-global zones.
Only services deployed within the same Kubernetes cluster, such as the HTTPRoute, can be referenced via backendRef.

Important Note: This limitation exist because, Kuma currently only allows referencing as backendRefs Kubernetes Services.

This is a temporary limitation. We’re actively working on extending backendRef to support Kuma’s MeshServices. Once this feature is complete, you’ll be able to reference services across different clusters within your mesh.

To better visualize this limitation here’s an example scenario that describes how you could configure multi-zone deployments with the Gateway API. In this example, you have the following resources:
- Two zones (zone-1 and zone-2) in separate Kubernetes clusters
- Gateway with listener on port 8080 deployed in zone-1
- Two services:
  - A service named backend deployed in each zone
  - A service named db deployed only in zone-2
```
 
 flowchart TD
     subgraph c2["k8s-cluster-2"]
         subgraph z2["zone-2"]
             subgraph c1z2s1["Service"]
                 b2(backend)
             end
             subgraph c1z2s2["Service"]
                 db(db)
             end
         end
     end
     subgraph c1["k8s-cluster-1"]
         subgraph z1["zone-1"]
             subgraph Gateway
                 listener(:8080)
             end
             subgraph Service
                 b1(backend)
             end
         end
     end
   
```
If you deploy multi-zone with Gateway API, the following will occur:
- If you create an HTTPRoute with a backendRef targeting the backend service in k8s-cluster-1, it will only route traffic to the backend service in k8s-cluster-1.
```
 
 flowchart TD
     subgraph c2["k8s-cluster-2"]
         subgraph z2["zone-2"]
             subgraph c1z2s1["Service"]
                 backend2(backend)
             end
             subgraph c1z2s2["Service"]
                 db(db)
             end
         end
     end
     subgraph c1["k8s-cluster-1"]
         subgraph z1["zone-1"]
             subgraph Service
                 backend1(backend)
             end
             subgraph Gateway
                 listener(:8080)
             end
             subgraph HTTPRoute
                 route1(/)
             end
             route1--"❌"-->backend2
             linkStyle 0 stroke:red,color:red,stroke-dasharray: 5 5;
             route1-->backend1
             listener-->route1
         end
     end
   
```
- Similarly, if you create an HTTPRoute with a backendRef pointing to the db service in k8s-cluster-1, it will result in a HTTPRoute with a ResolvedRefs status condition of BackendNotFound because service db is not present in k8s-cluster-1.
```
 
 flowchart TD
     subgraph c2["k8s-cluster-2"]
         subgraph z2["zone-2"]
             subgraph c1z2s1["Service"]
                 backend2(backend)
             end
             subgraph c1z2s2["Service"]
                 db(db)
             end
         end
     end
     subgraph c1["k8s-cluster-1"]
         subgraph z1["zone-1"]
             subgraph Gateway
                 listener(:8080)
             end
             subgraph HTTPRoute
                 route1(/)
             end
             subgraph Service
                 backend1(backend)
             end
             route1--"❌"-->db
             linkStyle 0 stroke:red,color:red,stroke-dasharray: 5 5;
             listener-->route1
         end
     end
   
```

Service to service routing

Kuma also supports routing between services with HTTPRoute in conformance with the GAMMA specifications.

GAMMA is a Gateway API subproject focused on mesh implementations of Gateway API and extending the Gateway API resources to mesh use cases.

The key feature of HTTPRoute for mesh routing is specifying a Kubernetes Service as the parentRef, as opposed to a Gateway. All requests to this Service are then filtered and routed as specified in the HTTPRoute.

apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: canary-demo-app
  namespace: kuma-demo
spec:
  parentRefs:
  - name: demo-app
    port: 5000
    kind: Service
  rules:
  - backendRefs:
    - name: demo-app-v1
      port: 5000
    - name: demo-app-v2
      port: 5000

The namespace of the HTTPRoute is key. If the route’s namespace and the parentRef’s namespace are identical, Kuma applies the route to requests from all workloads. If the route’s namespace differs from its parentRef’s namespace, the HTTPRoute applies only to requests from workloads in the route’s namespace.

Remember to tag your Service ports with appProtocol: http to use them in an HTTPRoute!

Because of how Kuma maps resources at the moment, the combination of the HTTPRoutes name and namespace and the parent Service name and namespace must be no more than 249 characters.

How it works

Kuma includes controllers that reconcile Gateway API CRDs and convert them into the corresponding Kuma CRDs. This is why in the GUI, Kuma MeshGateways/MeshHTTPRoutes/MeshTCPRoutes are visible and not Kubernetes Gateway API resources.

Kubernetes Gateway API resources serve as the source of truth for Kuma gateways and routes. Any edits to the corresponding Kuma resources are overwritten.