DNS based service discovery on Kubernetes

[seivan]

I just tested it, and it works! But so far not satisfied with naming and general documentation so I would love some feedback and discussion here.

Also pretty sure the Kubernetes based strategies could work together in sharing code as only the fetching nodes (actual discovery) seem to differ. But could be part of a larger refactoring

config/config.exs

  config :libcluster,
  topologies: [
    k8s_example: [
      strategy: Cluster.Strategy.Kubernetes.DNS,
      config: [
        service: "web-headless",
        application_name: "myapp",
        polling_interval: 1000
        ]]]

Service.yml

apiVersion: v1
kind: Service
metadata:
  labels:
    deployment: web
  name: web-headless
  namespace: default
spec:
  clusterIP: None 
  ports:
    - port: 80
  selector:
    deployment: web

#on node1
iex --name myapp@172.17.0.6 --cookie EDABUQMNEWFJWXGCAZMV -S mix

#on node2
iex --name myapp@172.17.0.7 --cookie EDABUQMNEWFJWXGCAZMV -S mix

iex(myapp@172.17.0.7)1> Node.list
[:"myapp@172.17.0.6"]

[bitwalker]

We should document this in the README which covers the built-in strategies, otherwise this looks good to go! Sorry for the holdup!

[seivan]

@bitwalker How do you feel about the naming of the config keys? We're in an akward spot where we have to balance Erlang naming conventions together with what Kubernetes names its stuff.
e.g Node

[bitwalker]

I'm fine with them as is - you've done a good job of explaining them in the strategy documentation, and some degree of conflict in naming conventions is expected, especially between Erlang and an orchestration platform like k8s. If there appears to be confusion in the future, we can address that with better documentation if needed.

[bitwalker]

You mentioned you weren't satisfied with the quality here, I haven't taken the time to do a deep review of the code, but are there specific areas you are concerned with (beyond naming)? I will make some time to do a better review

[bitwalker]

Other than the comments I've left, this looks good to me!

[seivan]

Alright, I'll go through and fix these. Appreciate the comments!
I might need help with the "blocking" stuff you mentioned earlier, but overall it's straightforward. 👍

[seivan]

@bitwalker Sorry for the delay, other stuff came up.
I'll take a look at this this weekend.

[seivan]

@bitwalker I took care of your comments. I let the formatter deal with the spaces, extracting polling_interval/1 changed the pipe(!) and made load/1 a blocking operation.

[bitwalker]

Thanks for all your work on this!