Merge pull request #8564 from sbueringer/pr-improve-kcp-error-log

🌱 KCP: improve error log on scale up when Machine doesn't have a nodeRef

Author: k8s-ci-robot

controlplane/kubeadm/internal/controllers/controller_test.go

@@ -2321,6 +2321,10 @@ func createMachineNodePair(name string, cluster *clusterv1.Cluster, kcp *control
 }
 
 func setMachineHealthy(m *clusterv1.Machine) {
+	m.Status.NodeRef = &corev1.ObjectReference{
+		Kind: "Node",
+		Name: "node-1",
+	}
 	conditions.MarkTrue(m, controlplanev1.MachineAPIServerPodHealthyCondition)
 	conditions.MarkTrue(m, controlplanev1.MachineControllerManagerPodHealthyCondition)
 	conditions.MarkTrue(m, controlplanev1.MachineSchedulerPodHealthyCondition)

controlplane/kubeadm/internal/controllers/scale.go

@@ -200,9 +200,17 @@ loopmachines:
 			}
 		}
 
-		for _, condition := range allMachineHealthConditions {
-			if err := preflightCheckCondition("machine", machine, condition); err != nil {
-				machineErrors = append(machineErrors, err)
+		if machine.Status.NodeRef == nil {
+			// The conditions will only ever be set on a Machine if we're able to correlate a Machine to a Node.
+			// Correlating Machines to Nodes requires the nodeRef to be set.
+			// Instead of confusing users with errors about that the conditions are not set, let's point them
+			// towards the unset nodeRef (which is the root cause of the conditions not being there).
+			machineErrors = append(machineErrors, errors.Errorf("Machine %s does not have a corresponding Node yet (Machine.status.nodeRef not set)", machine.Name))
+		} else {
+			for _, condition := range allMachineHealthConditions {
+				if err := preflightCheckCondition("Machine", machine, condition); err != nil {
+					machineErrors = append(machineErrors, err)
+				}
 			}
 		}
 	}

controlplane/kubeadm/internal/controllers/scale_test.go

@@ -492,12 +492,28 @@ func TestPreflightChecks(t *testing.T) {
 			},
 			expectResult: ctrl.Result{RequeueAfter: deleteRequeueAfter},
 		},
+		{
+			name: "control plane without a nodeRef should requeue",
+			kcp:  &controlplanev1.KubeadmControlPlane{},
+			machines: []*clusterv1.Machine{
+				{
+					Status: clusterv1.MachineStatus{
+						NodeRef: nil,
+					},
+				},
+			},
+			expectResult: ctrl.Result{RequeueAfter: preflightFailedRequeueAfter},
+		},
 		{
 			name: "control plane with an unhealthy machine condition should requeue",
 			kcp:  &controlplanev1.KubeadmControlPlane{},
 			machines: []*clusterv1.Machine{
 				{
 					Status: clusterv1.MachineStatus{
+						NodeRef: &corev1.ObjectReference{
+							Kind: "Node",
+							Name: "node-1",
+						},
 						Conditions: clusterv1.Conditions{
 							*conditions.FalseCondition(controlplanev1.MachineAPIServerPodHealthyCondition, "fooReason", clusterv1.ConditionSeverityError, ""),
 							*conditions.TrueCondition(controlplanev1.MachineControllerManagerPodHealthyCondition),
@@ -523,6 +539,10 @@ func TestPreflightChecks(t *testing.T) {
 			machines: []*clusterv1.Machine{
 				{
 					Status: clusterv1.MachineStatus{
+						NodeRef: &corev1.ObjectReference{
+							Kind: "Node",
+							Name: "node-1",
+						},
 						Conditions: clusterv1.Conditions{
 							*conditions.TrueCondition(controlplanev1.MachineAPIServerPodHealthyCondition),
 							*conditions.TrueCondition(controlplanev1.MachineControllerManagerPodHealthyCondition),