This repository holds a bash script that allows you to perform canary deployments on a Kubernetes cluster. It is part of the respective Codefresh blog post

The script expects you to have an existing deployment and service on your K8s cluster. It does the following:

1. Reads the existing deployment from the cluster to a yml file
2. Changes the name of the deployment and the docker image to a new version
3. Deploys 1 replica for the new version (the canary)
4. Waits for some time (it is configurable) and checks the number of restarts
5. If everything is ok it adds more canaries and scales down the production instances
6. The cycle continues until all replicas used by the service are canaries (the production replicas are zero)

If something goes wrong (the pods have restarts) the scripts deletes all canaries and scales back the production version to the original number of replicas

Of course during the wait period when both deployments are active, you are free to run your own additional checks or integration tests to see if the new deployment is ok.

The canary percentage is configurable. The script will automatically calculate the phase

Example:

• Production instance has 5 replicas
• User enters canary waves to 35%
• Script calculates 35% is about 2 pods

As a convention the script expects

1. The name of your deployment to be $APP_NAME-$VERSION
2. Your service has a metadata label that shows which deployment is currently "in production"

Notice that the new color deployment created by the script will follow the same conventions. This way each subsequent pipeline you run will work in the same manner.

You can see examples of the labels with the sample application:

• service
• deployment

The script needs one environment variable called KUBE_CONTEXT that selects the K8s cluster that will be used (if you have more than one)

The rest of the parameters are provided as command line arguments

Here is an example:

./k8s-canary-rollout.sh myService myApp 20 my-namespace 73df943 30 

The script also comes with a Dockerfile that allows you to use it as a Docker image in any Docker based workflow such as Codefresh.

For the KUBE_CONTEXT environment variable just use the name of your cluster as found in the Codefresh Kubernetes dashboard. For the rest of the arguments you need to define them as parameters in your codefresh.yml file.

 canaryDeploy:
    title: "Deploying new version ${{CF_SHORT_REVISION}}"
    image: codefresh/k8s-canary:master
    environment:
      - WORKING_VOLUME=.
      - SERVICE_NAME=my-demo-app
      - DEPLOYMENT_NAME=my-demo-app
      - TRAFFIC_INCREMENT=20
      - NEW_VERSION=${{CF_SHORT_REVISION}}
      - SLEEP_SECONDS=40
      - NAMESPACE=canary
      - KUBE_CONTEXT=myDemoAKSCluster

The CF_SHORT_REVISION variable is offered by Codefresh and contains the git hash of the version that was just pushed. See all variables in the official documentation

The canary step is now deployed in dockerhub as well

https://hub.docker.com/r/codefresh/k8s-canary/

Further improvements

• Make the script create an initial deployment/service if nothing is deployed in the kubernetes cluster
• Add more complex and configurable healthchecks