DNS resolution issue with the runner when deploying tf-controller into GKE

[syalioune]

Hello,

First of all, thanks for the great work on this controller.
We've tried deploying recently the tf-controller in a standard GKE clusters and hit a wall because of DNS resolution issues. The controller is not able to resolve DNS names like 10-40-129-81.podinfo.pod.cluster.local.

This issue is best described here #365 (comment) and here #462 (comment). GKE uses Cloud DNS which does not provide IP based DNS name resolution like CoreDNS.

I have a very rough quick and dirty enhancement proposal here syalioune@e107963 revolving around :

• Setup a tf-runner service in each allowed namespaces
• Add hostname=terraform_crd_name, subdomain=tf-runner fields to the runner pod so that an A record : terraform_crd_name.tf-runner.namespace.svc.cluster_domain is automatically created
• Add a SAN into the runner generated certificate

Preliminary tests shows that it works.

Before going further, I'm looking for community/maintainer feedbacks 😄

• Is this proposal in line with the project direction ?
• Would that method replace entirely pod's ip based resolution or be an alternative ?
• Should the controller be responsible for : checking the service existence in the allowed namespace ? creating the service if it does not exist ?

Cheers