In Kubernetes, applications are very easily updated. For this, Kubernetes uses rolling updates, which means that traffic to a container is first drained before the container is replaced. During an upgrade of the application, Kubernetes will deploy an additional pod and run it through some specified probes.

A probe is a diagnostic that is periodically performed on a pod to check its status. During the upgrading or creation of a pod, Kubernetes brings up the additional pod and makes sure that it passes the liveness and readiness probes.

If the newly created pod succeeds with both probes, the traffic to a single old pod is terminated and traffic to the new pod is opened. For this termination, Kubernetes uses a termination grace period. During this period, the 2 connection to the load balancer is stopped and active connections are processed successfully, and new traffic is routed to a running pod. The default grace period is 30 seconds, during which the pod will be in a termination state and all old traffic to this pod is redirected to the other pods.

This process continues until all pods are replaced with the new version. All of this is default behavior within Azure Kubernetes. A deployment is simply triggered by adjusting the deployment file and applying it with the same command as used previously:

Kubectl apply -f [file]

By default, httpGet probes are added to pods that are being exposed, but they can also be customized by adding the readiness probe or liveness probe to the deployment:

readinessProbe:
          httpGet:
             path: /
             port: 80
             initialDelaySeconds: 5
             periodSeconds: 5
             successThreshold: 1

This readiness probe performs an httpGet request on the pod and has the following options:

• path: The path it should call for the httpGet request.
• port: The port number it should use for the call. This is also configured in our deployment file.
• initialDelaySeconds: The seconds it waits before running the probe once the container is started.

• periodSeconds: The number of seconds the probe waits before it times out.
• successThreshold: The amount of success required for the probe minimum value is 1.

As mentioned, a deployment has a default rolling upgrade scenario configured. The configuration of the rolling deployment can be retrieved by using the following command:

kubectl describe deployment kubernetes-deployment