"After all, the engineers only needed to refuse to fix anything, and modern industry would grind to a halt." -Michael Lewis

Enable Massive Growth

An Example Upgrade and Rollback of a Deployment image in Kubernetes

Jun 2020

In this article, I'm going to show you how to bootstrap a local kubernetes cluster with a custom image, debug it, deploy a new image, then rollback to the old image.

If you don't have a good way to get a local kubernetes cluster, you should check out: how to setup and use kind locally. Big fan.

Create Cluster, Make Base Resources

Start by creating a local kubernetes cluster using kind:

kind create cluster

This will automatically configure your kubectl cli to communicate to your local cluster (called "kind" by default).

I have created a sample repository in Docker Hub with a few different versions that we'll use for this demo. Specifically, there are two versions (both tags):

  • v2: has just an /actuator/health endpoint
  • v3: added a GET /hello endpoint

Once we have our cluster, we can first create a namespace with this yaml:

apiVersion: v1
kind: Namespace
  name: nick-sample-sb

You can apply this with:

kubectl apply -f namespace.yaml

We can then add a deployment, which will live in that created namespace, and will start with version 2:

apiVersion: apps/v1
kind: Deployment
  name: nick-sample-sb
  namespace: nick-sample-sb
  replicas: 2
      app: nick-sample-sb
        app: nick-sample-sb
        - name: nick-sample-sb
          image: nfisher23/simplesb:v2
          imagePullPolicy: Always
            - containerPort: 8080

Pretty much the same command to get this up there:

kubectl apply -f deployment.yaml

Finally, we'll expose a way for other pods in the cluster to communicate with the pods managed by the deployment (actually, the deployment manages ReplicaSets, which in turn ensure we have enough pods) with one more yaml for cluster ip:

apiVersion: v1
kind: Service
  namespace: nick-sample-sb
  name: nick-sample-sb-clusterip
    app: nick-sample-sb
  - port: 80
    targetPort: 8080
    protocol: TCP
    app: nick-sample-sb

And apply that change:

kubectl apply -f service.yaml

Note that, while I have shown these examples as being in three different yaml files, they could technically all be in one. It's usually easier to keep them more isolated, but if you're just proving things out/sandboxing, you can do so by separating them with --.


Okay, we have our namespace:

$ kubectl get namespaces | grep nick-sample-sb
nick-sample-sb       Active   29m

We also have a deployment with two pods, hopefully ready by now:

$ kubectl get deployment --namespace nick-sample-sb nick-sample-sb
nick-sample-sb   2/2     2            2           30m

We can see some more details about those pods specifically with something like:

$ kubectl get pods --namespace nick-sample-sb -o wide
NAME                              READY   STATUS    RESTARTS   AGE   IP           NODE           NOMINATED NODE   READINESS GATES
nick-sample-sb-7f65cbf9fc-bc4pw   1/1     Running   0          16m   kind-worker               
nick-sample-sb-7f65cbf9fc-kd95d   1/1     Running   0          17m   kind-worker               

Because CoreDNS is included with this version of kubernetes and it is a cluster aware DNS service, our cluster ip will automatically allow us to communicate via a DNS entry nick-sample-sb-clusterip if we are in the same namespace. Let's prove that--we can exec onto one of the pods and send a curl to the health endpoint:

$ POD=$(kubectl get pods -n nick-sample-sb | grep nick-sample | awk '{print $1}' | head -1)
$ kubectl exec --namespace nick-sample-sb $POD -- curl nick-sample-sb-clusterip/actuator/health

We can also notice that our /hello endpoint is not available:

$ kubectl exec --namespace nick-sample-sb $POD -- curl nick-sample-sb-clusterip/hello
{"timestamp":"2020-06-27T23:38:46.663+0000","status":404,"error":"Not Found","message":"No message available","path":"/hello"}

Now, we are ready to ship our code, version 3--we can do so from the cli with:

$ kubectl set image -n nick-sample-sb deployments/nick-sample-sb nick-sample-sb=nfisher23/simplesb:v3 --record
deployment.apps/nick-sample-sb image updated

This command will update to the latest version and after all the pods come up, we can see that our /hello endpoint exists:

$ POD=$(kubectl get pods -n nick-sample-sb | grep nick-sample | awk '{print $1}' | head -1)
$ kubectl exec --namespace nick-sample-sb $POD -- curl nick-sample-sb-clusterip/hello

I hope this provided you with enough information to get you started bridging the gap between applications and infra when it comes to kubernetes.

Nick Fisher is a software engineer in the Pacific Northwest. He focuses on building highly scalable and maintainable backend systems.