Consider running epoxy-extension-server as a normal cluster deployment

[nkinkade]

Today, the epoxy-extension-server is run as a Docker container on each API machine via a systemd service named epoxy-extension-server.service. This makes troubleshooting the extension server difficult, as you have to login to each of the 3 API machines to look at logs and to debug, since you never know which one will be receiving any given request, because there is an internal loadbalancer in front of them.

Would it be possible to just run the extension-server as a normal k8s deployment, which only gets run on API machines?

Thoughts:

• Having it run as a sort of baked-in service on the API machines is nice in some ways because it makes the API servers a neat package. If we were to run it as a k8s deployment, then a core API server service then has a dependency on k8s-support builds and releases.