[BUG] gke-deploy / PodDisruptionBudget

[feitnomore]

Failing to use PodDisruptionBudget

When using PodDisruptionBudget with spec: maxUnavailable, although resource is being deployed, the builder can't identify it as ready/healthy.

Affected builder image

gcr.io/cloud-builders/gke-deploy

Expected Behavior

Trigger should pass however builder fails do identify the PodDisruptionBudget is working fine.

Actual Behavior

According to resource/ready.go under podDisruptionBudgetIsReady, one of the things that is being checked is status.desiredHealthy == spec.minAvailable, however if we deploy something using maxUnavailable instead of minAvailabe, that is supported on Kubernetes 1.17+, builder will always return false here:

minAvailable, ok, err := unstructured.NestedInt64(obj.Object, "spec", "minAvailable")
if err != nil {
	return false, fmt.Errorf("failed to get spec.minAvailable field: %v", err)
}
if !ok {
	return false, nil
}

Steps to Reproduce the Problem

Use maxUnavailable instead of minAvailable.

apiVersion: policy/v1beta1
kind: PodDisruptionBudget
metadata:
  name: nodejs04
  namespace: default
spec:
  maxUnavailable: 1
  selector:
    matchLabels:
      app: nodejs04

Additional Info

Suggestion here is to remove status.desiredHealthy == spec.minAvailable and minAvailable retrieval, and leave only checking is status.currentHealthy >= status.desiredHealthy as I think it should be sufficient.

If you are ok with the suggestion, I can try to submit a PR myself.

Thanks

[feitnomore]

PR merged into master.