Bug Report

• Concourse version: 3.10.0
• Deployment type (BOSH/Docker/binary): Docker
• Infrastructure/IaaS: Kubernetes

Sometimes when my kubernetes cluster is flaky my builds fail because it can't pull the secrets from the k8s API. The kubernetes API may become unreachable for lots of reasons but usually it is only temporary so ideally when faced with that situation concourse should fail fast and retry.

Finding variable 'username': Get https://100.64.0.1:443/api/v1/namespaces/concourse-devs/secrets/some-pipeline.username: read tcp 100.96.55.56:45248->100.64.0.1:443: read: connection timed out

I think the place to add the retry would be somewhere around here:
https://github.com/concourse/atc/blob/5e111aab5cfd6041719a43f4a720a5f4d239d01c/creds/kubernetes/kubernetes.go#L71