Support kubernetes based peer identification

[trashhalo]

I just wanted to point out that for users running in kubernetes which I assume is a majority, you can use a headless service to identify peers. This would allow you to not have redis and reduce the deployed surface area of the application.

I know this because libcluster the distributed cluster library for elixir uses this. The code can be seen here. https://github.com/bitwalker/libcluster/blob/main/lib/strategy/kubernetes_dns.ex

1. Create a kubernetes service that targets the pods
2. Use dns to inspect service this gives you the ips of your peers

Feel free to close if not helpful! just wanted to let you know

[trashhalo]

Documentation for DNS records for headless services. https://github.com/kubernetes/dns/blob/master/docs/specification.md#24---records-for-a-headless-service

[codeboten]

Thanks for the suggestion @trashhalo, it's a great idea, we may implement this in the future

[kentquirk]

Thank you for the suggestion. As of recent versions, Refinery uses Redis for a lot more than peer discovery, so this change would not allow us to avoid it.

It would enable us to reduce a bit of pubsub traffic, in the cases where systems are running on k8s (which is the majority but by no means all installations).