Refactor aws-network-topology-operator to put finalizer on the AWSCluster instead of Cluster

[alex-dabija]

Task

Refactor aws-network-topology-operator to put finalizer on the AWSCluster instead of Cluster.

Towards epic.

Background

Currently, the aws-network-topology-operator adds a finalizer on the Cluster resources. This leads to a race condition because in certain situations the AWSCluster resource could be deleted before the operator is ready to remove the finalizer from the Cluster resource and it still expects the AWSCluster resource to be around.

This probably doesn't happen that often because in order for the AWSCluster resource to be deleted the transit gateway attachment needs to be removed.

Update 2023-08: When implementing this, ensure #2714 (comment) is mentioned as use case, and gets fixed.

Tasks

Preview Give feedback

1. https://github.com/giantswarm/giantswarm/issues/28489

[bdehri]

It looks like finalizer is put on both Cluster and AWSCluster. So, what we want to do is put the finalizer only on AWSCluster ?

func (g *Cluster) AddFinalizer(ctx context.Context, capiCluster *capi.Cluster, finalizer string) error {
  originalCluster := capiCluster.DeepCopy()
  controllerutil.AddFinalizer(capiCluster, finalizer)
  if err := g.Client.Patch(ctx, capiCluster, client.MergeFrom(originalCluster)); err != nil {
    return err
  }
  capaCluster, err := g.GetAWSCluster(ctx, types.NamespacedName{Name: capiCluster.Name, Namespace: capiCluster.Namespace})
  if err != nil {
    return err
  }
  originalCAPACluster := capaCluster.DeepCopy()
  controllerutil.AddFinalizer(capaCluster, finalizer)
  return g.Client.Patch(ctx, capaCluster, client.MergeFrom(originalCAPACluster))
}

[fiunchinho]

I think the controller should only reconcile AWSClusters and put the finalizer only to AWSClusters.

[mnitchev]

Done in https://github.com/giantswarm/aws-resolver-rules-operator/releases/tag/v0.13.0