Introduce NodeDeletionStrategy to allow drain node when deleting cluster

Signed-off-by: Lubron Zhan <[email protected]>

Author: lubronzhan

api/v1beta2/cluster_types.go

@@ -585,6 +585,14 @@ type Topology struct {
 	// +listMapKey=name
 	// +kubebuilder:validation:MaxItems=1000
 	Variables []ClusterVariable `json:"variables,omitempty"`
+
+	// nodeDeletionStrategy specifies the strategy to delete nodes in the cluster.
+	// Avilable options:
+	// - graceful
+	// - force
+	// By default it's "force"
+	// +optional
+	NodeDeletionStrategy *NodeDrainStrategyType `json:"nodeDrainStrategy,omitempty"`
 }
 
 // ControlPlaneTopology specifies the parameters for the control plane nodes in the cluster.
@@ -901,6 +909,13 @@ type MachinePoolVariables struct {
 	Overrides []ClusterVariable `json:"overrides,omitempty"`
 }
 
+type NodeDrainStrategyType string
+
+const (
+	NodeDrainStrategyForce               NodeDrainStrategyType = "force"
+	NodeDrainStrategyGracefulWithTimeout NodeDrainStrategyType = "gracefulWithTimeout"
+)
+
 // ANCHOR_END: ClusterSpec
 
 // ANCHOR: ClusterNetwork

internal/controllers/machine/machine_controller.go

@@ -714,8 +714,8 @@ func (r *Reconciler) nodeVolumeDetachTimeoutExceeded(machine *clusterv1.Machine)
 // and if the Machine is not the last control plane node in the cluster.
 func (r *Reconciler) isDeleteNodeAllowed(ctx context.Context, cluster *clusterv1.Cluster, machine *clusterv1.Machine) error {
 	log := ctrl.LoggerFrom(ctx)
-	// Return early if the cluster is being deleted.
-	if !cluster.DeletionTimestamp.IsZero() {
+	// Return early if the cluster is being deleted and cluster's nodeDeletionStrategy is not set or set to `force`.
+	if !cluster.DeletionTimestamp.IsZero() && (cluster.Spec.Topology.NodeDeletionStrategy == nil || *cluster.Spec.Topology.NodeDeletionStrategy == clusterv1.NodeDrainStrategyForce) {
 		return errClusterIsBeingDeleted
 	}
 

internal/controllers/machine/machine_controller_test.go

@@ -2668,6 +2668,59 @@ func TestIsDeleteNodeAllowed(t *testing.T) {
 			machine:       &clusterv1.Machine{},
 			expectedError: errClusterIsBeingDeleted,
 		},
+		{
+			name: "has nodeRef and cluster is being deleted, nodeDeletionStrategy set to force",
+			cluster: &clusterv1.Cluster{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:              "test-cluster",
+					Namespace:         metav1.NamespaceDefault,
+					DeletionTimestamp: &deletionts,
+					Finalizers:        []string{clusterv1.ClusterFinalizer},
+				},
+				Spec: clusterv1.ClusterSpec{
+					Topology: &clusterv1.Topology{
+						NodeDeletionStrategy: ptr.To(clusterv1.NodeDrainStrategyForce),
+					},
+				},
+			},
+			machine:       &clusterv1.Machine{},
+			expectedError: errClusterIsBeingDeleted,
+		},
+		{
+			name: "has nodeRef and control plane is healthy, with nodeDeletionStrategy set to graceful",
+			cluster: &clusterv1.Cluster{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:      "test-cluster",
+					Namespace: metav1.NamespaceDefault,
+				},
+				Spec: clusterv1.ClusterSpec{
+					Topology: &clusterv1.Topology{
+						NodeDeletionStrategy: ptr.To(clusterv1.NodeDrainStrategyForce),
+					},
+				},
+			},
+			machine: &clusterv1.Machine{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:      "created",
+					Namespace: metav1.NamespaceDefault,
+					Labels: map[string]string{
+						clusterv1.ClusterNameLabel: "test-cluster",
+					},
+					Finalizers: []string{clusterv1.MachineFinalizer},
+				},
+				Spec: clusterv1.MachineSpec{
+					ClusterName:       "test-cluster",
+					InfrastructureRef: corev1.ObjectReference{},
+					Bootstrap:         clusterv1.Bootstrap{DataSecretName: ptr.To("data")},
+				},
+				Status: clusterv1.MachineStatus{
+					NodeRef: &corev1.ObjectReference{
+						Name: "test",
+					},
+				},
+			},
+			expectedError: nil,
+		},
 		{
 			name: "has nodeRef and control plane is healthy and externally managed",
 			cluster: &clusterv1.Cluster{