Add ExcludeNodeVolumeDetachingAnnotation annotation to be able to skip shouldWaitForNodeVolumes check entirely

Author: furkatgofurov7

api/v1alpha3/machine_types.go

@@ -33,6 +33,9 @@ const (
 	// ExcludeNodeDrainingAnnotation annotation explicitly skips node draining if set.
 	ExcludeNodeDrainingAnnotation = "machine.cluster.x-k8s.io/exclude-node-draining"
 
+	// ExcludeNodeVolumeDetachingAnnotation annotation explicitly skips volume detaching if set.
+	ExcludeNodeVolumeDetachingAnnotation = "machine.cluster.x-k8s.io/exclude-node-volume-detaching"
+
 	// MachineSetLabelName is the label set on machines if they're controlled by MachineSet.
 	MachineSetLabelName = "cluster.x-k8s.io/set-name"
 

api/v1alpha4/condition_consts.go

@@ -124,12 +124,17 @@ const (
 	// WaitingExternalHookReason (Severity=Info) provide evidence that we are waiting for an external hook to complete.
 	WaitingExternalHookReason = "WaitingExternalHook"
 
-	// VolumeDetachFinishedCondition reports a machine node volume detachment finished either by waiting for volumes to be
-	// detached or left alone due to volume detach timeout exceeded.
-	VolumeDetachFinishedCondition ConditionType = "VolumeDetachFinished"
+	// VolumeDetachSucceededCondition reports a machine waiting for volumes to be detached.
+	VolumeDetachSucceededCondition ConditionType = "VolumeDetachSucceeded"
 
-	// WaitingForVolumeDetachReason (Severity=Info) provide evidence that a machine node waiting for volumes to be attached.
+	// VolumeDetachTimedOutCondition reports a machine waiting for node volumes detachment timeout to be exceeded.
+	VolumeDetachTimedOutCondition ConditionType = "VolumeDetachTimedOut"
+
+	// WaitingForVolumeDetachReason (Severity=Info) provide evidence that a machine node waiting for volumes to be detached.
 	WaitingForVolumeDetachReason = "WaitingForVolumeDetach"
+
+	// WaitingForVolumeDetachTimeoutReason (Severity=Info) provide evidence that a machine node waiting for node volumes detachment timeout to be exceeded.
+	WaitingForVolumeDetachTimeoutReason = "WaitingForVolumeDetachTimeout"
 )
 
 const (

api/v1alpha4/machine_types.go

@@ -33,6 +33,9 @@ const (
 	// ExcludeNodeDrainingAnnotation annotation explicitly skips node draining if set.
 	ExcludeNodeDrainingAnnotation = "machine.cluster.x-k8s.io/exclude-node-draining"
 
+	// ExcludeNodeVolumeDetachingAnnotation annotation explicitly skips volume detaching if set.
+	ExcludeNodeVolumeDetachingAnnotation = "machine.cluster.x-k8s.io/exclude-node-volume-detaching"
+
 	// MachineSetLabelName is the label set on machines if they're controlled by MachineSet.
 	MachineSetLabelName = "cluster.x-k8s.io/set-name"
 

api/v1beta1/condition_consts.go

@@ -124,12 +124,17 @@ const (
 	// WaitingExternalHookReason (Severity=Info) provide evidence that we are waiting for an external hook to complete.
 	WaitingExternalHookReason = "WaitingExternalHook"
 
-	// VolumeDetachFinishedCondition reports a machine node volume detachment finished either by waiting for volumes to be
-	// detached or left alone due to volume detach timeout exceeded.
-	VolumeDetachFinishedCondition ConditionType = "VolumeDetachFinished"
+	// VolumeDetachSucceededCondition reports a machine waiting for volumes to be detached.
+	VolumeDetachSucceededCondition ConditionType = "VolumeDetachSucceeded"
+
+	// VolumeDetachTimedOutCondition reports a machine waiting for node volumes detachment timeout to be exceeded.
+	VolumeDetachTimedOutCondition ConditionType = "VolumeDetachTimedOut"
 
 	// WaitingForVolumeDetachReason (Severity=Info) provide evidence that a machine node waiting for volumes to be attached.
 	WaitingForVolumeDetachReason = "WaitingForVolumeDetach"
+
+	// WaitingForVolumeDetachTimeoutReason (Severity=Info) provide evidence that a machine node waiting for node volumes detachment timeout to be exceeded.
+	WaitingForVolumeDetachTimeoutReason = "WaitingForVolumeDetachTimeout"
 )
 
 const (

api/v1beta1/machine_types.go

@@ -33,6 +33,9 @@ const (
 	// ExcludeNodeDrainingAnnotation annotation explicitly skips node draining if set.
 	ExcludeNodeDrainingAnnotation = "machine.cluster.x-k8s.io/exclude-node-draining"
 
+	// ExcludeNodeVolumeDetachingAnnotation annotation explicitly skips volume detaching if set.
+	ExcludeNodeVolumeDetachingAnnotation = "machine.cluster.x-k8s.io/exclude-node-volume-detaching"
+
 	// MachineSetLabelName is the label set on machines if they're controlled by MachineSet.
 	MachineSetLabelName = "cluster.x-k8s.io/set-name"
 

internal/controllers/machine/machine_controller.go

@@ -346,24 +346,29 @@ func (r *Reconciler) reconcileDelete(ctx context.Context, cluster *clusterv1.Clu
 			conditions.MarkTrue(m, clusterv1.DrainingSucceededCondition)
 			r.recorder.Eventf(m, corev1.EventTypeNormal, "SuccessfulDrainNode", "success draining Machine's node %q", m.Status.NodeRef.Name)
 
-			// After node draining, make sure either volumes are detached or left alone after volume detach timeout
-			// exceeded before deleting the Node.
-			if conditions.Get(m, clusterv1.VolumeDetachFinishedCondition) == nil {
-				conditions.MarkFalse(m, clusterv1.VolumeDetachFinishedCondition, clusterv1.WaitingForVolumeDetachReason, clusterv1.ConditionSeverityInfo, "Waiting for node volumes to be detached")
-			}
-			if r.nodeDrainTimeoutExceeded(m) || r.volumeDetachTimeoutExceeded(m) {
-				conditions.MarkTrue(m, clusterv1.VolumeDetachFinishedCondition)
-				r.recorder.Eventf(m, corev1.EventTypeNormal, "NodeVolumeUndetached", "node volumes detach timeout exceeded for Machine's node %q, moving on with node deletion", m.Status.NodeRef.Name)
-			} else if ok, err := r.shouldWaitForNodeVolumes(ctx, cluster, m.Status.NodeRef.Name); ok || err != nil {
-				if err != nil {
-					r.recorder.Eventf(m, corev1.EventTypeWarning, "FailedWaitForVolumeDetach", "error wait for volume detach, node %q: %v", m.Status.NodeRef.Name, err)
-					return ctrl.Result{}, err
+			if _, exists := m.ObjectMeta.Annotations[clusterv1.ExcludeNodeVolumeDetachingAnnotation]; !exists {
+				// After node draining, if ExcludeNodeVolumeDetachingAnnotation does not exist, make sure either volumes are
+				// detached or left alone after volume detach timeout exceeded before deleting the Node.
+				if conditions.Get(m, clusterv1.VolumeDetachTimedOutCondition) == nil {
+					conditions.MarkFalse(m, clusterv1.VolumeDetachTimedOutCondition, clusterv1.WaitingForVolumeDetachTimeoutReason, clusterv1.ConditionSeverityInfo, "Waiting for node volumes detachment timeout to be exceeded")
+				}
+				if !r.volumeDetachTimeoutExceeded(m) {
+					if conditions.Get(m, clusterv1.VolumeDetachSucceededCondition) == nil {
+						conditions.MarkFalse(m, clusterv1.VolumeDetachSucceededCondition, clusterv1.WaitingForVolumeDetachReason, clusterv1.ConditionSeverityInfo, "Waiting for node volumes to be detached")
+					}
+					if ok, err := r.shouldWaitForNodeVolumes(ctx, cluster, m.Status.NodeRef.Name); ok || err != nil {
+						if err != nil {
+							r.recorder.Eventf(m, corev1.EventTypeWarning, "FailedWaitForVolumeDetach", "error wait for volume detach, node %q: %v", m.Status.NodeRef.Name, err)
+							return ctrl.Result{}, err
+						}
+						log.Info("Waiting for node volumes to be detached", "node", klog.KRef("", m.Status.NodeRef.Name))
+						return ctrl.Result{}, nil
+					}
+					conditions.MarkTrue(m, clusterv1.VolumeDetachSucceededCondition)
+					r.recorder.Eventf(m, corev1.EventTypeNormal, "NodeVolumesDetached", "success waiting for node volumes detach Machine's node %q", m.Status.NodeRef.Name)
 				}
-				log.Info("Waiting for node volumes to be detached", "node", klog.KRef("", m.Status.NodeRef.Name))
-				return ctrl.Result{}, nil
-			} else {
-				conditions.MarkTrue(m, clusterv1.VolumeDetachFinishedCondition)
-				r.recorder.Eventf(m, corev1.EventTypeNormal, "NodeVolumesDetached", "success waiting for node volumes detach Machine's node %q", m.Status.NodeRef.Name)
+				conditions.MarkTrue(m, clusterv1.VolumeDetachTimedOutCondition)
+				r.recorder.Eventf(m, corev1.EventTypeNormal, "NodeVolumesDetachTimedOut", "node volumes detach timeout exceeded for Machine's node %q, moving on with node deletion", m.Status.NodeRef.Name)
 			}
 		}
 	}
@@ -461,13 +466,13 @@ func (r *Reconciler) volumeDetachTimeoutExceeded(machine *clusterv1.Machine) boo
 		return false
 	}
 
-	// if the volume detaching finished condition does not exist
-	if conditions.Get(machine, clusterv1.VolumeDetachFinishedCondition) == nil {
+	// if the volume detaching succeeded condition does not exist
+	if conditions.Get(machine, clusterv1.VolumeDetachSucceededCondition) == nil {
 		return false
 	}
 
 	now := time.Now()
-	firstTimeDetach := conditions.GetLastTransitionTime(machine, clusterv1.VolumeDetachFinishedCondition)
+	firstTimeDetach := conditions.GetLastTransitionTime(machine, clusterv1.VolumeDetachSucceededCondition)
 	diff := now.Sub(firstTimeDetach.Time)
 	return diff.Seconds() >= machine.Spec.VolumeDetachTimeout.Seconds()
 }