Canary Rollouts using SMI Traffic Split

Managing Canary rollouts using SMI Taffic Split

This guide demonstrates how to perform Canary rollouts using the SMI Traffic Split configuration.

Prerequisites

  • Kubernetes cluster running Kubernetes v1.19.0 or greater.
  • Have FSM installed.
  • Have kubectl available to interact with the API server.
  • Have fsm CLI available for managing the service mesh.

Demonstration

Explanation

In this demo, we use two applications, curl and httpbin implemented with Pipy, to act as client and server respectively. The service has two versions, v1 and v2, which are simulated by deploying httpbin-v1 and httpbin-v2.

Observant viewers may notice the frequent use of Pipy to implement httpbin functionalities in demonstrations. This is because web services implemented with Pipy can easily customize response content, facilitating the observation of test results.

Prerequisites

  • Kubernetes cluster
  • kubectl CLI

Installing the Service Mesh

Download the FSM CLI.

system=$(uname -s | tr '[:upper:]' '[:lower:]')
arch=$(uname -m | sed -E 's/x86_/amd/' | sed -E 's/aarch/arm/')
release=v1.2.3
curl -L https://github.com/flomesh-io/fsm/releases/download/${release}/fsm-${release}-${system}-${arch}.tar.gz | tar -vxzf -
cp ./${system}-amd64/fsm /usr/local/bin/fsm

Install the service mesh and wait for all components to run successfully.

fsm install --timeout 120s

Deploying the Sample Application

The curl and httpbin applications run in their respective namespaces, which are managed by the service mesh through the fsm namespace add xxx command.

kubectl create ns httpbin
kubectl create ns curl
fsm namespace add httpbin curl

Deploy the v1 version of httpbin which returns Hi, I am v1! for all HTTP requests. Other applications access httpbin through the Service httpbin.

kubectl apply -n httpbin -f - <<EOF
apiVersion: v1
kind: Service
metadata:
  name: httpbin
spec:
  ports:
    - name: pipy
      port: 8080
      targetPort: 8080
      protocol: TCP
  selector:
    app: pipy
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: httpbin-v1
  labels:
    app: pipy
    version: v1
spec:
  replicas: 1
  selector:
    matchLabels:
      app: pipy
      version: v1
  template:
    metadata:
      labels:
        app: pipy
        version: v1
    spec:
      containers:
        - name: pipy
          image: flomesh/pipy:latest
          ports:
            - name: pipy
              containerPort: 8080
          command:
            - pipy
            - -e
            - |
              pipy()
              .listen(8080)
              .serveHTTP(new Message('Hi, I am v1!\n'))
EOF

Deploy the curl application.

kubectl apply -n curl -f - <<EOF
apiVersion: v1
kind: Service
metadata:
  name: curl
  labels:
    app: curl
    service: curl
spec:
  ports:
    - name: http
      port: 80
  selector:
    app: curl
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: curl
spec:
  replicas: 1
  selector:
    matchLabels:
      app: curl
  template:
    metadata:
      labels:
        app: curl
    spec:
      containers:
      - image: curlimages/curl
        imagePullPolicy: IfNotPresent
        name: curl
        command: ["sleep", "365d"]
EOF

Wait for all applications to run successfully.

kubectl wait --for=condition=ready pod --all -A

Test application access.

curl_client="$(kubectl get pod -n curl -l app=curl -o jsonpath='{.items[0].metadata.name}')"
# send four request
kubectl exec "$curl_client" -n curl -c curl -- curl -s httpbin.httpbin:8080 httpbin.httpbin:8080 httpbin.httpbin:8080 httpbin.httpbin:8080

Expected results are displayed.

Hi, I am v1!
Hi, I am v1!
Hi, I am v1!
Hi, I am v1!

Next, deploy the v2 version of httpbin.

Deploying Version v2

The v2 version of httpbin returns Hi, I am v2! for all HTTP requests. Before deployment, we need to set the default traffic split strategy, otherwise, the new version instances would be accessible through the Service httpbin.

Create Service httpbin-v1 with an additional version tag in its selector compared to Service httpbin. Currently, both have the same endpoints.

kubectl apply -n httpbin -f - <<EOF
apiVersion: v1
kind: Service
metadata:
  name: httpbin-v1
spec:
  ports:
    - name: pipy
      port: 8080
      targetPort: 8080
      protocol: TCP
  selector:
    app: pipy
    version: v1
EOF

Apply the TrafficSplit strategy to route all traffic to httpbin-v1.

kubectl apply -n httpbin -f - <<EOF
apiVersion: split.smi-spec.io/v1alpha4
kind: TrafficSplit
metadata:
  name: httpbin-split
spec:
  service: httpbin
  backends:
  - service: httpbin-v1
    weight: 100
EOF

Then deploy the new version.

kubectl apply -n httpbin -f - <<EOF
apiVersion: v1
kind: Service
metadata:
  name: httpbin-v2
spec:
  ports:
    - name: pipy
      port: 8080
      targetPort: 8080
      protocol: TCP
  selector:
    app: pipy
    version: v2
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: httpbin-v2
  labels:
    app: pipy
    version: v2
spec:
  replicas: 1
  selector:
    matchLabels:
      app: pipy
      version: v2
  template:
    metadata:
      labels:
        app: pipy
        version: v2
    spec:
      containers:
        - name: pipy
          image: flomesh/pipy:latest
          ports:
            - name: pipy
              containerPort: 8080
          command:
            - pipy
            - -e
            - |
              pipy()
              .listen(8080)
              .serveHTTP(new Message('Hi, I am v2!\n'))
EOF

Wait for the new version to run successfully.

kubectl wait --for=condition=ready pod -n httpbin -l version=v2

If you send requests again, v1 httpbin still handles them.

kubectl exec "$curl_client" -n curl -c curl -- curl -s httpbin.httpbin:8080 httpbin.httpbin:8080 httpbin.httpbin:8080 httpbin.httpbin:8080
Hi, I am v1!
Hi, I am v1!
Hi, I am v1!
Hi, I am v1!

Canary Release

Modify the traffic split strategy to direct 25% of the traffic to version v2.

kubectl apply -n httpbin -f - <<EOF
apiVersion: split.smi-spec.io/v1alpha4
kind: TrafficSplit
metadata:
  name: httpbin-split
spec:
  service: httpbin
  backends:
  - service: httpbin-v1
    weight: 75
  - service: httpbin-v2
    weight: 25
EOF

Sending requests again shows that 1/4 of the traffic is handled by version v2.

kubectl exec "$curl_client" -n curl -c curl -- curl -s httpbin.httpbin:8080 httpbin.httpbin:8080 httpbin.httpbin:8080 httpbin.httpbin:8080
Hi, I am v1!
Hi, I am v2!
Hi, I am v1!
Hi, I am v1!

Advanced Canary Release

Suppose in version v2, /test endpoint functionality has been updated. For risk control, we want only a portion of the traffic to access the /test endpoint of the v2 version, while other endpoints like /demo should only access the v1 version.

We need to introduce another resource to define the traffic visiting the /test endpoint: a route definition. Here we define two routes:

  • httpbin-test: Traffic using the GET method to access /test.
  • httpbin-all: Traffic using the GET method to access /, note this is a prefix match.
kubectl apply -n httpbin -f - <<EOF
apiVersion: specs.smi-spec.io/v1alpha4
kind: HTTPRouteGroup
metadata:
  name: httpbin-test
spec:
  matches:
  - name: test
    pathRegex: "/test"
    methods:
    - GET
---
apiVersion: specs.smi-spec.io/v1alpha4
kind: HTTPRouteGroup
metadata:
  name: httpbin-all
spec:
  matches:
  - name: test
    pathRegex: ".*"
    methods:
    - GET
EOF

Then

update the traffic split strategy, associating the routes to the split. Also, create a new policy.

kubectl apply -n httpbin -f - <<EOF
apiVersion: split.smi-spec.io/v1alpha4
kind: TrafficSplit
metadata:
  name: httpbin-split
spec:
  service: httpbin
  matches:
  - name: httpbin-test
    kind: HTTPRouteGroup
  backends:
  - service: httpbin-v1
    weight: 75
  - service: httpbin-v2
    weight: 25
---
apiVersion: split.smi-spec.io/v1alpha4
kind: TrafficSplit
metadata:
  name: httpbin-all
spec:
  service: httpbin
  matches:
  - name: httpbin-all
    kind: HTTPRouteGroup
  backends:
  - service: httpbin-v1
    weight: 100
EOF

Now, when accessing the /test endpoint, only 25% of the traffic goes to the new version.

kubectl exec "$curl_client" -n curl -c curl -- curl -s httpbin.httpbin:8080/test httpbin.httpbin:8080/test httpbin.httpbin:8080/test httpbin.httpbin:8080/test
Hi, I am v1!
Hi, I am v2!
Hi, I am v1!
Hi, I am v1!

And requests to the /demo endpoint all go to the old v1 version.

kubectl exec "$curl_client" -n curl -c curl -- curl -s httpbin.httpbin:8080/demo httpbin.httpbin:8080/demo httpbin.httpbin:8080/demo httpbin.httpbin:8080/demo
Hi, I am v1!
Hi, I am v1!
Hi, I am v1!
Hi, I am v1!

This meets the expectations.

Feedback

Was this page helpful?


Last modified April 11, 2024: update versions of fsm and pipy (cea5b3e)