Thanks for the feedback Ed. I wouldn’t consider this a hack. This is essentially how you would run Flannel if you were managing your own cluster. The difference is that you could share the etcd instance the masters are using with Flannel. If you need to use an overlay like Flannel with EKS, you’d obviously want to a highly available backend and not a single node etcd cluster. I don’t know whether I mention it in this post, but the only CNI that AWS officially supports is the AWS VPC CNI. All others, including Flannel, are run at your own risk. I was motivated to write this because when EKS was first introduced, customer’s wanted to know how to use alternate CNIs, support be damned.