Finally we have gotten to the last step of the upgrade process which is upgrading our Nodes.
There are two ways to provision and manage your worker nodes - self-managed node groups and managed node groups. In this workshop eksctl was configured to use the managed node groups. This was helpful here as managed node groups make this easier for us by automating both the AWS and the Kubernetes side of the process.
The way that managed node groups does this is:
If we instead had used a self-managed node group then we need to do the Kubernetes taint and draining steps ourselves to ensure Kubernetes knows that Node is going away and can manage that process gracefully in order for such an upgrade to be non-disruptive.
The first step only applies to if we are using the cluster autoscaler. We don’t want conflicting Node scaling actions during our upgrade so we should scale that to zero to suspend it during this process using the command below. Unless you have done that chapter in the workshop and left it deployed you can skip this step.
kubectl scale deployments/cluster-autoscaler --replicas=0 -n kube-system
We can then trigger the MNG upgrade process by running the following eksctl command:
eksctl upgrade nodegroup --name=nodegroup --cluster=eksworkshop-eksctl --kubernetes-version=1.18
In another Terminal tab you can follow the progress with:
kubectl get nodes --watch
You’ll notice the new nodes come up (three one in each AZ), one of the older nodes go STATUS SchedulingDisabled, then eventually that node go away and another new node come up to replace it and so on as described in the process above until all the old Nodes have gone away. Then it’ll scale back down from 6 Nodes to the original 3.