Kubernetes

High Availability

High availability is about implementing your service in such a way that it will continue to work in the face of the failure of various components. It also helps you to perform upgrades while minimizing service interruption.

One of the Kubernetes features that can be used in a highly available service is the Deployment. A Deployment is a template for creating a Pod, and Kubernetes handles creating and destroying Pods according to the details in the template, such as the number of pods to create (so-called “replicas”).

Creating a Deployment

In the following example, we are going to create a Deployment.

First, delete the pod we previously created.

$ kubectl delete pod webserver
pod "webserver" deleted

Try to refresh the page at EXTERNAL IP of the loadbalancer service. It should fail because the pod has already been deleted.
```
$ kubectl get service
NAME           TYPE           CLUSTER-IP       EXTERNAL-IP     PORT(S)        AGE
nginxservice   LoadBalancer   10.254.210.248   103.6.252.244   80:31086/TCP   67m
```
You can also examine the endpoints that the service has discovered. These are the IPs of the Pods where the service is forwarding incoming requests. There should be no endpoints since we just deleted the webserver Pod.
```
$ kubectl get endpoints
NAME           ENDPOINTS         AGE
nginxservice   <none>            69m
```

Create a deployment. Save the following yaml as deployment.yaml, and create it using kubectl apply -f deployment.yaml.

apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    run: webserver
  name: webserver
spec:
  replicas: 2
  selector:
    matchLabels:
      run: webserver
  template:
    metadata:
      labels:
        run: webserver
    spec:
      containers:
      - image: nginx
        name: nginx

Note that we’re specifying replicas: 2 here, to say we want two pods to be running in this deployment.

Check the status of the new deployment, it may take a little while for both pods to become ready (the 2/2 under READY).

$ kubectl get deployment
NAME        READY   UP-TO-DATE   AVAILABLE   AGE
webserver   2/2     2            2           66s

Show the pods themselves.

kubectl get pods
NAME                         READY   STATUS    RESTARTS   AGE
webserver-56f66697cc-b8gvq   1/1     Running   0          106s
webserver-56f66697cc-h4stn   1/1     Running   0          106s

Browse to the page at EXTERNAL IP of the loadbalancer service. It should now be working. As we created pods in the deployment with the same label as the previous single pod (run=webserver), the loadbalancer knows how to find them.
```
$ kubectl get service
NAME           TYPE           CLUSTER-IP       EXTERNAL-IP     PORT(S)        AGE
nginxservice   LoadBalancer   10.254.210.248   103.6.252.244   80:31086/TCP   67m
```
If you check the endpoints again, you can see the IPs of the Pods that were discovered by the service.
```
$ kubectl get endpoints
NAME           ENDPOINTS                       AGE
nginxservice   10.100.1.31:80,10.100.1.32:80   67m
```

Delete a pod

Select a pod to be deleted and show the pods again.

$ kubectl delete po webserver-56f66697cc-h4stn
pod "webserver-56f66697cc-h4stn" deleted

$ kubectl get pods
NAME                         READY   STATUS              RESTARTS   AGE
webserver-56f66697cc-b8gvq   1/1     Running             0          2m35s
webserver-56f66697cc-n8vdf   0/1     ContainerCreating   0          3s

You can see that Kubernetes has created a new pod to ensure that there are two replicas.

The nginx web page should continue working even with one pod down.

Scaling up pods

You can also scale up the number of pods by editing the deployment config file.

Edit deployment.yaml and change replicas from 2 to 3
Submit the new config
```
kubectl apply -f deployment.yaml
```

Show the pods. You should see a new container being created. E.g.

$ kubectl get po
NAME                         READY   STATUS              RESTARTS   AGE
webserver-56f66697cc-b8gvq   1/1     Running             0          5m52s
webserver-56f66697cc-n8vdf   1/1     Running             0          3m20s
webserver-56f66697cc-rqzj7   0/1     ContainerCreating   0          4s

You can also scale the deployment using the kubectl scale command. Here we scale back down to two replicas, and one of the three existing pods will be deleted.

$ kubectl scale deployment webserver --replicas=2
deployment.extensions/webserver scaled

$ kubectl get po
NAME                         READY   STATUS    RESTARTS   AGE
webserver-56f66697cc-b8gvq   1/1     Running   0          10m
webserver-56f66697cc-n8vdf   1/1     Running   0          7m33s

More information

For more information, refer to:

Deployment