There are some other pieces missing. The cluster-autoscaler implementation is quite basic. It doesn't have a way to managing multiple clusters inside the same p

There are some other pieces missing. The cluster-autoscaler implementation is quite basic. It doesn't have a way to managing multiple clusters inside the same project and no way to add custom tags. That makes it harder to track resources. The permissions/security is quite simple. Read/write tokens with full permissions on the project is very wide. Adding S3 like service without more security would be interesting. These are the most important ones, but I remember there were a few more nice to have feature (like server groups for scaling up an down and the ability to retrieve the userdata, image and ssh keys for a running server.