Implement metadata propagation on the workload cluster nodes

Signed-off-by: Danil Grigorev <[email protected]>

bootstrap/api/v1alpha1/rke2config_types.go

@@ -68,6 +68,14 @@ type RKE2AgentConfig struct {
 	//+optional
 	NodeLabels []string `json:"nodeLabels,omitempty"`
 
+	// NodeAnnotations  Aplying annotations on created nodes post bootstrap phase
+	//
+	// Unfortunately it is not possible to apply annotations via kubelet
+	// using current bootstrap configurations.
+	// Issue: https://github.com/kubernetes/kubernetes/issues/108046
+	//+optional
+	NodeAnnotations map[string]string `json:"nodeAnnotations,omitempty"`
+
 	// NodeTaints Registering kubelet with set of taints.
 	//+optional
 	NodeTaints []string `json:"nodeTaints,omitempty"`

controlplane/api/v1alpha1/condition_consts.go

@@ -54,9 +54,16 @@ const (
 	// is up to date. Whe this condition is false, the RKE2ControlPlane is executing a rolling upgrade.
 	MachinesSpecUpToDateCondition clusterv1.ConditionType = "MachinesSpecUpToDate"
 
+	// NodeMetadataUpToDate documents that the metadata of the nodes controlled by the RKE2 machines
+	// is up to date. Whe this condition is false, the node metadata is not propagated.
+	NodeMetadataUpToDate clusterv1.ConditionType = "NodeMetadataUpToDate"
+
 	// MachineAgentHealthyCondition reports a machine's rke2 agent's operational status.
 	MachineAgentHealthyCondition clusterv1.ConditionType = "AgentHealthy"
 
+	// NodePatchFailedReason (Severity=Error) documents reasong why Node object could not be patched
+	NodePatchFailedReason = "NodePatchFailed"
+
 	// PodInspectionFailedReason documents a failure in inspecting the pod status.
 	PodInspectionFailedReason = "PodInspectionFailed"
 

controlplane/internal/controllers/rke2controlplane_controller.go

@@ -649,7 +649,7 @@ func (r *RKE2ControlPlaneReconciler) reconcileKubeconfig(
 
 // reconcileControlPlaneConditions is responsible of reconciling conditions reporting the status of static pods and
 // the status of the etcd cluster.
-func (r *RKE2ControlPlaneReconciler) reconcileControlPlaneConditions(ctx context.Context, controlPlane *rke2.ControlPlane) (ctrl.Result, error) {
+func (r *RKE2ControlPlaneReconciler) reconcileControlPlaneConditions(ctx context.Context, controlPlane *rke2.ControlPlane) (res ctrl.Result, retErr error) {
 	logger := log.FromContext(ctx)
 
 	readyCPMachines := controlPlane.Machines.Filter(collections.IsReady())
@@ -679,17 +679,32 @@ func (r *RKE2ControlPlaneReconciler) reconcileControlPlaneConditions(ctx context
 
 	workloadCluster, err := r.managementCluster.GetWorkloadCluster(ctx, util.ObjectKey(controlPlane.Cluster))
 	if err != nil {
-		logger.Info("Unable to get Workload cluster")
+		logger.Error(err, "Unable to get Workload cluster")
 
 		return ctrl.Result{}, errors.Wrap(err, "cannot get remote client to workload cluster")
 	}
 
+	defer func() {
+		// Always attempt to Patch the Machine conditions after each reconcile.
+		if err := controlPlane.PatchMachines(ctx); err != nil {
+			retErr = kerrors.NewAggregate([]error{retErr, err})
+		}
+	}()
+
+	if err := workloadCluster.InitWorkload(ctx, controlPlane); err != nil {
+		logger.Error(err, "Unable to initialize workload cluster")
+
+		return ctrl.Result{}, err
+	}
+
 	// Update conditions status
-	workloadCluster.UpdateAgentConditions(ctx, controlPlane)
-	workloadCluster.UpdateEtcdConditions(ctx, controlPlane)
+	workloadCluster.UpdateAgentConditions(controlPlane)
+	workloadCluster.UpdateEtcdConditions(controlPlane)
+
+	// Patch nodes metadata
+	if err := workloadCluster.UpdateNodeMetadata(ctx, controlPlane); err != nil {
+		logger.Error(err, "Unable to update node metadata")
 
-	// Patch machines with the updated conditions.
-	if err := controlPlane.PatchMachines(ctx); err != nil {
 		return ctrl.Result{}, err
 	}
 
@@ -721,11 +736,11 @@ func (r *RKE2ControlPlaneReconciler) upgradeControlPlane(
 		return ctrl.Result{}, err
 	}
 
-	status, err := workloadCluster.ClusterStatus(ctx)
-	if err != nil {
+	if err := workloadCluster.InitWorkload(ctx, controlPlane); err != nil {
 		return ctrl.Result{}, err
 	}
 
+	status := workloadCluster.ClusterStatus()
 	if status.Nodes <= *rcp.Spec.Replicas {
 		// scaleUp ensures that we don't continue scaling up while waiting for Machines to have NodeRefs
 		return r.scaleUpControlPlane(ctx, cluster, rcp, controlPlane)

pkg/rke2/control_plane.go

@@ -352,6 +352,7 @@ func (c *ControlPlane) PatchMachines(ctx context.Context) error {
 			if err := helper.Patch(ctx, machine, patch.WithOwnedConditions{Conditions: []clusterv1.ConditionType{
 				controlplanev1.MachineAgentHealthyCondition,
 				controlplanev1.MachineEtcdMemberHealthyCondition,
+				controlplanev1.NodeMetadataUpToDate,
 			}}); err != nil {
 				errList = append(errList, errors.Wrapf(err, "failed to patch machine %s", machine.Name))
 			}

pkg/rke2/management_cluster.go

@@ -113,7 +113,5 @@ func (m *Management) GetWorkloadCluster(ctx context.Context, clusterKey ctrlclie
 		return nil, &RemoteClusterConnectionError{Name: clusterKey.String(), Err: err}
 	}
 
-	return &Workload{
-		Client: c,
-	}, nil
+	return NewWorkload(c), nil
 }