Leandro Proença

Posted on Mar 12, 2023

Kubernetes 101, part IV, deployments

#kubernetes #docker

In the third part, we went a bit further and learned how Kubernetes employs self-healing capabilities by using the controller pattern through the utilisation of workload resources such as ReplicaSets.

However, it's well-known that most applications require to be updated frequently.

ReplicaSets are useful only for applications that do NOT need to be updated/deployed constantly. Almost always we're going to need another object that Kubernetes provides as a workload resource: the deployment object.

🚀 Deployment

The Kubernetes Deployment object brings a Deployment Controller that acts like the ReplicaSet Controller, but additionaly it allows a variety of update strategies so the application won't suffer any downtime.

Creating the Deployment

We'll follow the same process as for the ReplicaSet, but referencing kind: Deployment instead:



kind: Deployment
apiVersion: apps/v1
metadata:
  name: nginx
  labels:
    app: nginx
spec:
  ### The number of replicas of the same identical Pod
  replicas: 2
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
    spec:
      containers:
      - name: nginx
        image: nginx

At first glance, it works exactly like ReplicaSet. Let's check the deployment object in kubectl:



$ kubectl get deployments
NAME    READY   UP-TO-DATE   AVAILABLE   AGE
nginx   2/2     2            2           11s

The output explains that we have 2 of 2 replicas (Pods) ready.

Checking the Pods:



$ kubectl get pods
NAME                    READY   STATUS    RESTARTS   AGE
nginx-76d6c9b8c-sk28k   1/1     Running   0          7s
nginx-76d6c9b8c-f25nc   1/1     Running   0          7s

Note that, like ReplicaSet, each Pod gets a random unique ID in the cluster (sk28k and f25nc respectively).

Until now, it's exactly a ReplicaSet, nothing different from it. Time to see the differences.

Updating a Deployment

A Deployment allows updates with no downtime. As for our example with NGINX, let's demonstrate a simple update like changing the NGINX version in the Pod.

First, check the current NGINX version by entering to one of the Pods in the Deployment and executing nginx -v:



$ kubectl exec -it deploy/nginx -- nginx -v
nginx version: nginx/1.23.3

Next, we change the Pod image through the Deployment by issuing the command kubectl set image:



$ kubectl set image deployment/nginx nginx=nginx:1.22.1
deployment.apps/nginx image updated

Checking the Pods, we can see that the new Pods is being created while the current Pods are still running.

Hence, the application is up and running. No downtime!



$ kubectl get pods
NAME                     READY   STATUS              RESTARTS   AGE
nginx-76d6c9b8c-sk28k    1/1     Running             0          14m
nginx-76d6c9b8c-f25nc    1/1     Running             0          14m
nginx-756fbff4c9-t2pjb   0/1     ContainerCreating   0          8s

Then, after all Pods are successfully updated:



$ kubectl get pods
NAME                     READY   STATUS    RESTARTS   AGE
nginx-756fbff4c9-t2pjb   1/1     Running   0          19s
nginx-756fbff4c9-44g6w   1/1     Running   0          8s

Wanna confirm that?



kubectl exec -it deploy/nginx -- nginx -v
nginx version: nginx/1.22.1

Such a big YAY! 🚀

Bonus: rollout status

kubectl provides a command called rollout status that can be used during deployments, so the command keeps blocked until the deployment is finished.

It's very useful in CI/CD systems:



### Update the image
$ kubectl set image deployment/nginx nginx=nginx:1.23.1
deployment.apps/nginx image updated

### Check the rollout deployment (blocks until it is finished)
$ kubectl rollout status deployment/nginx
Waiting for deployment "nginx" rollout to finish: 1 out of 2 new replicas have been updated...
Waiting for deployment "nginx" rollout to finish: 1 out of 2 new replicas have been updated...
Waiting for deployment "nginx" rollout to finish: 1 old replicas are pending termination...
Waiting for deployment "nginx" rollout to finish: 1 old replicas are pending termination...
deployment "nginx" successfully rolled out

### Checking the NGINX version
$ kubectl exec -it deploy/nginx -- nginx -v
nginx version: nginx/1.23.1

Kubernetes is wonderful, isn't it?

🖼️ Pictures or didn't happen

Okay, sometimes is worth seeing the explanation in pictures for a better comprehension.

👉 The Deployment Controller starts creating a new Pod:

👉 Then, one of the old replicas starts terminating, and a new one starts being created:

👉 At the final stage, we have two new running replicas along with two old terminated replicas:

👉 And finally, the application was successfully rolled out:

↩️ Rollbacks

Not rare, when faced in situation where updates are delivered with bugs, we may want to rollback to an earlier version of our application.

The kubectl rollout brings a command history where Kubernetes persist all versions given a specific deployment:



$ kubectl rollout history deployment/nginx
deployment.apps/nginx
REVISION  CHANGE-CAUSE
1         <none>
2         <none>
3         <none>
4         <none>

It's natural to think that the current version is the last one (4), but we can check that by using kubectl describe:



$ kubectl describe deployment/nginx | grep revision
Annotations:            deployment.kubernetes.io/revision: 4

How about rolling back to the previous version (3)?



## Rollback to the previous version
$ kubectl rollout undo deployment/nginx 
deployment.apps/nginx rolled back

### Or simply going to a specific version
$ kubectl rollout undo deployment/nginx --to-revision=3

### Follow the deployment
$ kubectl rollout status deployment/nginx
Waiting for deployment "nginx" rollout to finish: 1 old replicas are pending termination...
Waiting for deployment "nginx" rollout to finish: 1 old replicas are pending termination...
deployment "nginx" successfully rolled out

### Check the NGINX version
$ kubectl exec -it deploy/nginx -- nginx -v
nginx version: nginx/1.23.3

😱 Unbelievable! 😱 Rollbacks in Kubernetes is so easy that we don't have excuses to not doing rollbacks when needed.

Why a system rollback is so important

I've seen a lot of projects that struggle at rollbacks. Sometimes because the rollback process is very hard. And sometimes because people don't have the proficiency in managing rollbacks.

Many people prefer doing Git reverts over application rollback.

That's a very wrong move IMHO.

Git reverts can be hard due to conflicts. The revert will take its time through all the CI/CD pipeline again. Until the revert is accepted to go to production, the application is bleeding with sometimes unacceptable bugs.

Invest time and resources having a good rollback system. If you're using Kubernetes, there's no excuse to not performing rollbacks, just experiment yourself how straightforward it is.

Scaling replicas in a Deployment

Additionally, we can scale up (or down) the number of replicas given a Deployment:



$ kubectl scale deployment/nginx --replicas=5
deployment.apps/nginx scaled

Then check the Pods:



$ kubectl get pods
NAME                     READY   STATUS    RESTARTS   AGE
nginx-799c5dd65b-5ncg2   1/1     Running   0          4m12s
nginx-799c5dd65b-qm2vb   1/1     Running   0          4m8s
nginx-799c5dd65b-pfngc   1/1     Running   0          8s
nginx-799c5dd65b-w49q4   1/1     Running   0          8s
nginx-799c5dd65b-ph6wv   1/1     Running   0          8s