gateway api初体验

Live with Gateway API

本文是对分享的记录，全文使用了demosh演示工具。为了方便记录，我会转换成常用的命令和Yaml。实验介绍了linkerd 和 istio ，由于时间关系，我只尝试istio的方式。我会使用最新版本的gateway api和istio，进行实验。

准备

k8s集群

没有的话，使用仓库提供的create-cluster.sh

安装gateway api

kubectl apply -f https://github.com/kubernetes-sigs/gateway-api/releases/download/v1.2.1/experimental-install.yaml

customresourcedefinition.apiextensions.k8s.io/backendlbpolicies.gateway.networking.k8s.io created
customresourcedefinition.apiextensions.k8s.io/backendtlspolicies.gateway.networking.k8s.io created
customresourcedefinition.apiextensions.k8s.io/gatewayclasses.gateway.networking.k8s.io created
customresourcedefinition.apiextensions.k8s.io/gateways.gateway.networking.k8s.io created
customresourcedefinition.apiextensions.k8s.io/grpcroutes.gateway.networking.k8s.io created
customresourcedefinition.apiextensions.k8s.io/httproutes.gateway.networking.k8s.io created
customresourcedefinition.apiextensions.k8s.io/referencegrants.gateway.networking.k8s.io created
customresourcedefinition.apiextensions.k8s.io/tcproutes.gateway.networking.k8s.io created
customresourcedefinition.apiextensions.k8s.io/tlsroutes.gateway.networking.k8s.io created
customresourcedefinition.apiextensions.k8s.io/udproutes.gateway.networking.k8s.io created

安装isito

curl -L https://istio.io/downloadIstio | ISTIO_VERSION=1.24.2 sh - 
export PATH=$PWD/istio-1.24.2/bin:$PATH
which istioctl
istioctl version
istioctl x precheck

istioctl install --set profile=minimal -y

创建gateway

apiVersion: gateway.networking.k8s.io/v1
kind: Gateway
metadata:
  name: ingress
spec:
  gatewayClassName: istio
  listeners:
    - name: http
      protocol: HTTP
      port: 80
      allowedRoutes:
        namespaces:
          from: All

❯ k get gateway ingress
NAME      CLASS   ADDRESS      PROGRAMMED   AGE
ingress   istio   172.28.0.2   True         59s          

# 查看ip
# kubectl get gateway ingress -ojsonpath='{ .status.addresses[0].value }'

安装测试应用

原文提供的helm命令，但是拉取不下来。

helm install faces \
     -n faces \
     oci://ghcr.io/buoyantio/faces-chart \
     --version 2.0.0-rc.1 \
     --values faces/values.yaml

我找到https://github.com/BuoyantIO/faces-demo.git ，研究了一下，生成了一个output.yaml，apply了这个yaml

kubectl create namespace faces
     
kubectl label namespace faces istio-injection=enabled

kubectl apply -f yaml

  • yaml

    # Source: faces-chart/templates/color.yaml
apiVersion: v1
kind: Service
metadata:
  name: "color"
  namespace: faces
  labels:
    service: "color"
spec:
  type: ClusterIP
  selector:
    service: "color"
  ports:
    - port: 80
      targetPort: http
---
# Source: faces-chart/templates/color.yaml
apiVersion: v1
kind: Service
metadata:
  name: "color2"
  namespace: faces
  labels:
    service: "color2"
spec:
  type: ClusterIP
  selector:
    service: "color2"
  ports:
    - port: 80
      targetPort: http
---
# Source: faces-chart/templates/color.yaml
apiVersion: v1
kind: Service
metadata:
  name: "color3"
  namespace: faces
  labels:
    service: "color3"
spec:
  type: ClusterIP
  selector:
    service: "color3"
  ports:
    - port: 80
      targetPort: http
---
# Source: faces-chart/templates/face.yaml
apiVersion: v1
kind: Service
metadata:
  name: face
  namespace: faces
  labels:
    service: face
spec:
  type: ClusterIP
  selector:
    service: face
  ports:
    - port: 80
      targetPort: http
---
# Source: faces-chart/templates/faces-gui.yaml
apiVersion: v1
kind: Service
metadata:
  name: faces-gui
  namespace: faces
  labels:
    service: faces-gui
spec:
  type: ClusterIP
  selector:
    service: faces-gui
  ports:
    - port: 80
      targetPort: http
---
# Source: faces-chart/templates/smiley.yaml
apiVersion: v1
kind: Service
metadata:
  name: "smiley"
  namespace: faces
  labels:
    service: "smiley"
spec:
  type: ClusterIP
  selector:
    service: "smiley"
  ports:
    - port: 80
      targetPort: http
---
# Source: faces-chart/templates/smiley.yaml
apiVersion: v1
kind: Service
metadata:
  name: "smiley2"
  namespace: faces
  labels:
    service: "smiley2"
spec:
  type: ClusterIP
  selector:
    service: "smiley2"
  ports:
    - port: 80
      targetPort: http
---
# Source: faces-chart/templates/smiley.yaml
apiVersion: v1
kind: Service
metadata:
  name: "smiley3"
  namespace: faces
  labels:
    service: "smiley3"
spec:
  type: ClusterIP
  selector:
    service: "smiley3"
  ports:
    - port: 80
      targetPort: http
---
# Source: faces-chart/templates/color.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: "color"
  namespace: faces
  labels:
    service: "color"
spec:
  replicas: 1
  selector:
    matchLabels:
      service: "color"
  template:
    metadata:
      labels:
        service: "color"
    spec:
      containers:
        - name: "color"
          image: 192.168.1.94:443/demo/faces-color:2.0.0
          imagePullPolicy: IfNotPresent
          ports:
            - name: http
              containerPort: 8000
          env:
            - name: FACES_SERVICE
              value: "color"
            - name: USER_HEADER_NAME
              value: "X-Faces-User"
            - name: DELAY_BUCKETS
              value: "0,5,10,15,20,50,200,500,1500"
            - name: ERROR_FRACTION
              value: ""
          resources:
            requests:
              cpu: 100m
              memory: 64Mi
            limits:
              cpu: 250m
              memory: 128Mi
---
# Source: faces-chart/templates/color.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: "color2"
  namespace: faces
  labels:
    service: "color2"
spec:
  replicas: 1
  selector:
    matchLabels:
      service: "color2"
  template:
    metadata:
      labels:
        service: "color2"
    spec:
      containers:
        - name: "color2"
          image: 192.168.1.94:443/demo/faces-color:2.0.0
          imagePullPolicy: IfNotPresent
          ports:
            - name: http
              containerPort: 8000
          env:
            - name: FACES_SERVICE
              value: "color"
            - name: USER_HEADER_NAME
              value: "X-Faces-User"
            - name: "COLOR"
              value: green
            - name: DELAY_BUCKETS
              value: "0,5,10,15,20,50,200,500,1500"
            - name: ERROR_FRACTION
              value: ""
          resources:
            requests:
              cpu: 100m
              memory: 64Mi
            limits:
              cpu: 250m
              memory: 128Mi
---
# Source: faces-chart/templates/color.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: "color3"
  namespace: faces
  labels:
    service: "color3"
spec:
  replicas: 1
  selector:
    matchLabels:
      service: "color3"
  template:
    metadata:
      labels:
        service: "color3"
    spec:
      containers:
        - name: "color3"
          image: 192.168.1.94:443/demo/faces-color:2.0.0
          imagePullPolicy: IfNotPresent
          ports:
            - name: http
              containerPort: 8000
          env:
            - name: FACES_SERVICE
              value: "color"
            - name: USER_HEADER_NAME
              value: "X-Faces-User"
            - name: "COLOR"
              value: darkblue
            - name: DELAY_BUCKETS
              value: "0,5,10,15,20,50,200,500,1500"
            - name: ERROR_FRACTION
              value: ""
          resources:
            requests:
              cpu: 100m
              memory: 64Mi
            limits:
              cpu: 250m
              memory: 128Mi
---
# Source: faces-chart/templates/face.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: face
  namespace: faces
  labels:
    service: face
spec:
  replicas: 1
  selector:
    matchLabels:
      service: face
  template:
    metadata:
      labels:
        service: face
    spec:
      containers:
        - name: face
          image: 192.168.1.94:443/demo/faces-workload:2.0.0
          imagePullPolicy: IfNotPresent
          ports:
            - name: http
              containerPort: 8000
          env:
            - name: FACES_SERVICE
              value: "face"
            - name: USER_HEADER_NAME
              value: "X-Faces-User"
            - name: ERROR_FRACTION
              value: ""
            - name: DELAY_BUCKETS
              value: ""
          resources:
            requests:
              cpu: 300m # The face service doesn't need much memory, but it does need more
              memory: 64Mi # CPU than the other backend services since it has to call the
            limits: # face and smiley services, then composite the results.
              cpu: 500m
              memory: 128Mi
---
# Source: faces-chart/templates/faces-gui.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: faces-gui
  namespace: faces
  labels:
    service: faces-gui
spec:
  replicas: 1
  selector:
    matchLabels:
      service: faces-gui
  template:
    metadata:
      labels:
        service: faces-gui
    spec:
      containers:
        - name: faces-gui
          image: 192.168.1.94:443/demo/faces-gui:2.0.0
          imagePullPolicy: IfNotPresent
          env:
            - name: USER_HEADER_NAME
              value: "X-Faces-User"
          ports:
            - name: http
              containerPort: 8000
          resources:
            requests:
              cpu: 50m
              memory: 64Mi
            limits:
              cpu: 100m
              memory: 128Mi
---
# Source: faces-chart/templates/smiley.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: "smiley"
  namespace: faces
  labels:
    service: "smiley"
spec:
  replicas: 1
  selector:
    matchLabels:
      service: "smiley"
  template:
    metadata:
      labels:
        service: "smiley"
    spec:
      containers:
        - name: "smiley"
          image: 192.168.1.94:443/demo/faces-workload:2.0.0
          imagePullPolicy: IfNotPresent
          ports:
            - name: http
              containerPort: 8000
          env:
            - name: FACES_SERVICE
              value: "smiley"
            - name: USER_HEADER_NAME
              value: "X-Faces-User"
            - name: DELAY_BUCKETS
              value: "0,5,10,15,20,50,200,500,1500"
            - name: ERROR_FRACTION
              value: ""
          resources:
            requests:
              cpu: 100m
              memory: 64Mi
            limits:
              cpu: 250m
              memory: 128Mi
---
# Source: faces-chart/templates/smiley.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: "smiley2"
  namespace: faces
  labels:
    service: "smiley2"
spec:
  replicas: 1
  selector:
    matchLabels:
      service: "smiley2"
  template:
    metadata:
      labels:
        service: "smiley2"
    spec:
      containers:
        - name: "smiley2"
          image: 192.168.1.94:443/demo/faces-workload:2.0.0
          imagePullPolicy: IfNotPresent
          ports:
            - name: http
              containerPort: 8000
          env:
            - name: FACES_SERVICE
              value: "smiley"
            - name: USER_HEADER_NAME
              value: "X-Faces-User"
            - name: "SMILEY"
              value: HeartEyes
            - name: DELAY_BUCKETS
              value: "0,5,10,15,20,50,200,500,1500"
            - name: ERROR_FRACTION
              value: ""
          resources:
            requests:
              cpu: 100m
              memory: 64Mi
            limits:
              cpu: 250m
              memory: 128Mi
---
# Source: faces-chart/templates/smiley.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: "smiley3"
  namespace: faces
  labels:
    service: "smiley3"
spec:
  replicas: 1
  selector:
    matchLabels:
      service: "smiley3"
  template:
    metadata:
      labels:
        service: "smiley3"
    spec:
      containers:
        - name: "smiley3"
          image: 192.168.1.94:443/demo/faces-workload:2.0.0
          imagePullPolicy: IfNotPresent
          ports:
            - name: http
              containerPort: 8000
          env:
            - name: FACES_SERVICE
              value: "smiley"
            - name: USER_HEADER_NAME
              value: "X-Faces-User"
            - name: "SMILEY"
              value: RollingEyes
            - name: DELAY_BUCKETS
              value: "0,5,10,15,20,50,200,500,1500"
            - name: ERROR_FRACTION
              value: "50"
          resources:
            requests:
              cpu: 100m
              memory: 64Mi
            limits:
              cpu: 250m
              memory: 128Mi

场景

base

gui-route.yaml

---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: faces-gui-route
  namespace: faces
spec:
  parentRefs:
    - name: ingress
      namespace: default
  rules:
  - matches:
    - path:
        type: PathPrefix
        value: /gui/
    backendRefs:
    - name: faces-gui
      port: 80
    filters:
    - type: URLRewrite
      urlRewrite:
        path:
          type: ReplaceFullPath
          replaceFullPath: /

face-route.yaml

---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: face-route
  namespace: faces
spec:
  parentRefs:
    - name: ingress
      namespace: default
      kind: Gateway
  rules:
  - matches:
    - path:
        type: PathPrefix
        value: /face
    filters:
      - type: URLRewrite
        urlRewrite:
          path:
            type: ReplacePrefixMatch
            replacePrefixMatch: /
    backendRefs:
    - name: face
      port: 80

http://localhost/gui 打开浏览器可以看到页面

image.png

unconditional

smiley-route.yaml

apiVersion: gateway.networking.k8s.io/v1beta1
kind: HTTPRoute
metadata:
  name: smiley-route
  namespace: faces
spec:
  parentRefs:
    - kind: Service
      group: ""
      name: smiley
      port: 80
  rules:
    - backendRefs:
        - name: smiley2
          group: ""
          port: 80

smiley2服务是微笑的表情

image.png

color-route.yaml

apiVersion: gateway.networking.k8s.io/v1alpha2
kind: GRPCRoute
metadata:
  name: color-route
  namespace: faces
spec:
  parentRefs:
    - name: color
      kind: Service
      group: ""
      port: 80
  rules:
    - backendRefs:
        - name: color2
          port: 80

color returns blue, but color2 returns green

image.png

canary

先删除两条路由

kubectl delete -n faces httproute smiley-route
kubectl delete -n faces grpcroute color-route

smiley-canary-10.yaml 灰度10%

apiVersion: gateway.networking.k8s.io/v1beta1
kind: HTTPRoute
metadata:
  name: smiley-canary
  namespace: faces
spec:
  parentRefs:
    - name: smiley
      kind: Service
      group: ""
      port: 80
  rules:
    - backendRefs:
        - name: smiley
          port: 80
          weight: 90
        - name: smiley2
          port: 80
          weight: 10

可以看到笑脸比较少。

image.png

模拟灰度发布

smiley-canary-10.yaml → smiley-canary-50.yaml → smiley-canary-90.yaml → smiley-canary-100.yaml

流量比例从90-10，50-50，10-90，0-100，最后缩容到0，kubectl scale -n faces deploy/smiley --replicas=0

apiVersion: gateway.networking.k8s.io/v1beta1
kind: HTTPRoute
metadata:
  name: smiley-canary
  namespace: faces
spec:
  parentRefs:
    - name: smiley
      kind: Service
      group: ""
      port: 80
  rules:
    - backendRefs:
        - name: smiley
          port: 80
          weight: 9~~0 50 10~~ 0
        - name: smiley2
          port: 80
          weight: 1~~0 50 90~~ 100

灰度完成之后的效果，但是这样流量转发，出了问题不好排查。

image.png

回滚

kubectl scale -n faces deploy/smiley --replicas=1
kubectl rollout status -n faces deploy
kubectl delete -n faces httproute smiley-canary

灰度gRPC也是一样的操作

apiVersion: gateway.networking.k8s.io/v1alpha2
kind: GRPCRoute
metadata:
  name: color-canary
  namespace: faces
spec:
  parentRefs:
    - name: color
      kind: Service
      group: ""
      port: 80
  rules:
    - backendRefs:
        - name: color
          port: 80
          weight: 50
        - name: color2
          port: 80
          weight: 50

image.png

Dynamic Routing

可以基于path, headers, or query parameters灰度，这里是用edge cell or a center cell的方式。

smiley-edge.yaml

apiVersion: gateway.networking.k8s.io/v1alpha2
kind: GRPCRoute
metadata:
  name: color-edge
  namespace: faces
spec:
  parentRefs:
    - name: color
      kind: Service
      group: ""
      port: 80
  rules:
    - matches:
      - method:
          service: ColorService
          method: Edge
      backendRefs:
        - name: color2
          port: 80
    - backendRefs:
        - name: color
          port: 80

image.png

Dynamic Canarying

smiley-edge-canary-50.yaml 根据path灰度，全部流量的20%，在分成50%和50%

apiVersion: gateway.networking.k8s.io/v1beta1
kind: HTTPRoute
metadata:
  name: smiley-edge
  namespace: faces
spec:
  parentRefs:
    - name: smiley
      kind: Service
      group: ""
      port: 80
  rules:
    - matches:
      - path:
          value: "/edge"
      backendRefs:
        - name: smiley3
          port: 80
          weight: 50
        - name: smiley2
          port: 80
          weight: 50
    - backendRefs:
        - name: smiley
          port: 80

image.png

Rollback

smiley3 有时候有问题。

apiVersion: gateway.networking.k8s.io/v1beta1
kind: HTTPRoute
metadata:
  name: smiley-edge
  namespace: faces
spec:
  parentRefs:
    - name: smiley
      kind: Service
      group: ""
      port: 80
  rules:
    - matches:
      - path:
          value: "/edge"
      backendRefs:
        - name: smiley2
          port: 80
    - backendRefs:
        - name: smiley
          port: 80

image.png

abtest

smiley-ab.yaml，header里带有x-faces-user: heart，会返回爱心

apiVersion: gateway.networking.k8s.io/v1beta1
kind: HTTPRoute
metadata:
  name: smiley-a-b
  namespace: faces
spec:
  parentRefs:
    - name: smiley
      kind: Service
      group: ""
      port: 80
  rules:
    - matches:
        - headers:
            - name: "x-faces-user"
              value: "heart"
      backendRefs:
        - name: smiley2
          port: 80
    - backendRefs:
        - name: smiley
          port: 80

image.png

清除

kubectl delete -n faces httproute smiley-a-b
kubectl delete -n faces grpcroute color-a-b

timeout

face-route.yaml 增加了200ms的超时。

---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: face-route
  namespace: faces
spec:
  parentRefs:
    - name: ingress
      namespace: default
      kind: Gateway
  rules:
  - matches:
    - path:
        type: PathPrefix
        value: /face
    filters:
      - type: URLRewrite
        urlRewrite:
          path:
            type: ReplacePrefixMatch
            replacePrefixMatch: /
    backendRefs:
    - name: face
      port: 80
    timeouts:
      request: 200ms

可以看到有超时的结果了

image.png

© 2025 · Built with Gatsby