Set failure only on instance error when no nodeRef

Setting a failure message is permanent and requires user action to get
rid of. We cannot know if the error is permanent though. The idea with
this commit is that if the machine has a nodeRef, we know that the
instance was working at some point, so the error could well be
temporary. Therefore we do not set failure in this case.
On the other hand, if the machine does not have a nodeRef, we set
failure since it is more likely that there is some configuration error
that will not go away.

Remove dead code and improve logging slightly in similar cases.

Author: lentzi90

controllers/openstackmachine_controller.go

@@ -334,14 +334,6 @@ func (r *OpenStackMachineReconciler) reconcileNormal(ctx context.Context, scope
 		return ctrl.Result{}, err
 	}
 
-	// Set an error message if we couldn't find the instance.
-	if instanceStatus == nil {
-		err = errors.New("OpenStack instance not found")
-		openStackMachine.SetFailure(capierrors.UpdateMachineError, err)
-		conditions.MarkFalse(openStackMachine, infrav1.InstanceReadyCondition, infrav1.InstanceNotFoundReason, clusterv1.ConditionSeverityError, "")
-		return ctrl.Result{}, nil
-	}
-
 	// TODO(sbueringer) From CAPA: TODO(ncdc): move this validation logic into a validating webhook (for us: create validation logic in webhook)
 
 	openStackMachine.Spec.ProviderID = pointer.String(fmt.Sprintf("openstack:///%s", instanceStatus.ID()))
@@ -364,10 +356,13 @@ func (r *OpenStackMachineReconciler) reconcileNormal(ctx context.Context, scope
 		conditions.MarkTrue(openStackMachine, infrav1.InstanceReadyCondition)
 		openStackMachine.Status.Ready = true
 	case infrav1.InstanceStateError:
-		// Error is unexpected, thus we report error and never retry
+		// If the machine has a NodeRef the error could be something temporary.
+		// If not, it is more likely a configuration error so we set failure and never retry.
 		scope.Logger().Info("Machine instance state is ERROR", "id", instanceStatus.ID())
-		err = fmt.Errorf("instance state %q is unexpected", instanceStatus.State())
-		openStackMachine.SetFailure(capierrors.UpdateMachineError, err)
+		if machine.Status.NodeRef == nil {
+			err = fmt.Errorf("instance state %q is unexpected", instanceStatus.State())
+			openStackMachine.SetFailure(capierrors.UpdateMachineError, err)
+		}
 		conditions.MarkFalse(openStackMachine, infrav1.InstanceReadyCondition, infrav1.InstanceStateErrorReason, clusterv1.ConditionSeverityError, "")
 		return ctrl.Result{}, nil
 	case infrav1.InstanceStateDeleted:
@@ -429,6 +424,8 @@ func (r *OpenStackMachineReconciler) reconcileNormal(ctx context.Context, scope
 func (r *OpenStackMachineReconciler) getOrCreate(logger logr.Logger, cluster *clusterv1.Cluster, openStackCluster *infrav1.OpenStackCluster, machine *clusterv1.Machine, openStackMachine *infrav1.OpenStackMachine, computeService *compute.Service, userData string) (*compute.InstanceStatus, error) {
 	instanceStatus, err := computeService.GetInstanceStatusByName(openStackMachine, openStackMachine.Name)
 	if err != nil {
+		logger.Info("Unable to get OpenStack instance", "Machine", openStackMachine.Name)
+		conditions.MarkFalse(openStackMachine, infrav1.InstanceReadyCondition, infrav1.InstanceNotFoundReason, clusterv1.ConditionSeverityError, err.Error())
 		return nil, err
 	}