I know it is unrelated, but the first thing that came to mind reading about the Endpoint Kubernetes object (as described in the Services) is that it is like assigning multiple IP addresses to an old-school DNS name which results in the DNS servers rotating around through them in “round robin” fashion (if they even still do that, it was like two decades ago that I worked with DNS like that). Here’s the excerpt that triggered that thought:
An Endpoint is an object that gets IP addresses of individual Pods assigned to it. The Endpoint object is then in turn referenced by a kubernetes Service, so that the Service has a record of the internal IPs of Pods in order to be able to communicate with them.
I haven’t figured out if the way connections get distributed is a matter of Service policy or not. I imagine it is.