[EKS]: How to upgrade worker nodes created by eksctl?

[mrichman]

What's the procedure for updating existing worker nodes created by eksctl? The AWS docs want us to update the existing stack with this template: https://amazon-eks.s3-us-west-2.amazonaws.com/cloudformation/2018-12-10/amazon-eks-nodegroup.yaml.

This is obviously not the same template that eksctl uses.

Should this be done using eksctl somehow, or should we be updating the CloudFormation stack that eksctl creates?

Assume I've already upgraded my cluster (i.e. from 1.10 to 1.11) and I've replaced kube-dns with CoreDNS.

[whereisaaron]

@mrichman the current approach with eksctl is to create a new updated node group, e.g. with a newer AMI for a new version of k8s or different options, then drain and delete the old node group.

There is discussion of adding a --replaces option to create nodegroup so this can be a one-step process.
eksctl-io/eksctl#443

So the eksctl process is more like the 'Migrating to a New Worker Node Group' AWS documentation. But easier... 😄
https://docs.aws.amazon.com/eks/latest/userguide/migrate-stack.html

[astleychen]

Can someone advise how to migrate an on-demand node group to spot-instanced node group over eksctl? From doc here, it seems not able to modify the CFn directly to make the change.

[tabern]

By design - standard eksctl node groups are immutable. See https://eksctl.io/usage/managing-nodegroups/#nodegroup-immutability

However, if you create EKS Managed Node Groups using eksctl these can be upgraded.

Create an eksctl MNG eksctl create nodegroup --managed or eksctl create cluster --managed
Upgrade an eksctl MNG eksctl upgrade nodegroup --name=<mng name> --cluster=<cluster>