I am a huge fan of ECS. It just works. You have to write a tiny bit of tooling, but it's the kind of glue code you'd probably write on top of kube anyway. Failu

I am a huge fan of ECS. It just works. You have to write a tiny bit of tooling, but it's the kind of glue code you'd probably write on top of kube anyway. Failures are straightforward and visible. There's no debugging layer after layer of network redirects. It's much harder for an errant config to "helpfully" punch a hole through your firewall. You get the same uptime and scalability benefits as k8s or EKS. Everything is still a container.