AutoDevops - Broken cluster state when adding a private cluster that is inaccessible

Summary

If you add a cluster where the Master API is not accessible, it'll save and look like it worked but deployments will never work because these 3 ENV vars will never get added even after rectifying the connection:

  • KUBE_TOKEN
  • KUBE_NAMESPACE
  • KUBECONFIG

Steps to reproduce

  • Create a private cluster in GKE and whitelist your own personal IP address
  • Add that cluster to Gitlab
  • Try to install Helm (get error)
  • Remove IP whitelist (so Gitlab can now connect) and install Helm
  • Try to deploy any AutoDevops project and you'll get an error with kubectl that it can't connect to: localhost:8080 (because kubeconfig is empty)

What is the current bug behavior?

On deployment you get an error that kubectl cannot connect to a cluster on localhost:8080 (missing kubeconfig). These three ENV vars are missing:

  • KUBE_TOKEN
  • KUBE_NAMESPACE
  • KUBECONFIG

What is the expected correct behavior?

Upon fixing the connection to your cluster, Gitlab should set/use those 3 env vars on deployments

Relevant logs and/or screenshots

$ ensure_namespace
The connection to the server localhost:8080 was refused - did you specify the right host or port?
ERROR: Job failed: exit code 1

Possible fixes

I think upon adding a cluster, it should check that it's able to connect to it first.