Add support for node annotation to disable node fencing.

pkg/controller/machinehealthcheck/machinehealthcheck_controller.go

@@ -28,10 +28,11 @@ import (
 )
 
 const (
-	machineAnnotationKey       = "machine.openshift.io/machine"
-	machineRebootAnnotationKey = "healthchecking.openshift.io/machine-remediation-reboot"
-	ownerControllerKind        = "MachineSet"
-	remediationStrategyReboot  = healthcheckingv1alpha1.RemediationStrategyType("reboot")
+	machineAnnotationKey           = "machine.openshift.io/machine"
+	machineRebootAnnotationKey     = "healthchecking.openshift.io/machine-remediation-reboot"
+	ownerControllerKind            = "MachineSet"
+	remediationStrategyReboot      = healthcheckingv1alpha1.RemediationStrategyType("reboot")
+	disableRemediationAnotationKey = "healthchecking.openshift.io/disabled"
 )
 
 // Add creates a new MachineHealthCheck Controller and adds it to the Manager. The Manager will set fields on the Controller
@@ -132,6 +133,12 @@ func (r *ReconcileMachineHealthCheck) Reconcile(request reconcile.Request) (reco
 	}
 
 	for _, hc := range allMachineHealthChecks.Items {
+		if hasMachineRemediationDisabled(machine) {
+			glog.Infof("Machine %s has a matching %s annotation set to <true>, remediation is skipped.",
+				machineKey, disableRemediationAnotationKey)
+			continue
+		}
+
 		if hasMatchingLabels(&hc, machine) {
 			glog.V(4).Infof("Machine %s has a matching machineHealthCheck: %s", machineKey, hc.Name)
 			return remediate(r, hc.Spec.RemediationStrategy, machine)
@@ -330,6 +337,11 @@ func hasMachineSetOwner(machine mapiv1.Machine) bool {
 	return false
 }
 
+func hasMachineRemediationDisabled(machine mapiv1.Machine) bool {
+	skipRemediation := machine.GetAnnotations()[disableRemediationAnotationKey] == "true"
+	return skipRemediation
+}
+
 func hasMatchingLabels(machineHealthCheck *healthcheckingv1alpha1.MachineHealthCheck, machine *mapiv1.Machine) bool {
 	selector, err := metav1.LabelSelectorAsSelector(&machineHealthCheck.Spec.Selector)
 	if err != nil {

pkg/controller/machinehealthcheck/machinehealthcheck_controller_test.go

@@ -332,6 +332,32 @@ func TestHasMachineSetOwner(t *testing.T) {
 
 }
 
+func TestHasMachineRemediationDisabled(t *testing.T) {
+	machineWithRemediationDisabled := maotesting.NewMachine("machineWithRemediationDisabled", "node")
+	machineWithRemediationEnabled := maotesting.NewMachine("machineWithRemediationEnabled", "node")
+	machineWithRemediationDisabled.Annotations[disableRemediationAnotationKey] = "true"
+
+	testsCases := []struct {
+		machine  *mapiv1alpha1.Machine
+		expected bool
+	}{
+		{
+			machine:  machineWithRemediationDisabled,
+			expected: true,
+		},
+		{
+			machine:  machineWithRemediationEnabled,
+			expected: false,
+		},
+	}
+
+	for _, tc := range testsCases {
+		if got := hasMachineRemediationDisabled(*tc.machine); got != tc.expected {
+			t.Errorf("Test case: Machine %s. Expected: %t, got: %t", tc.machine.Name, tc.expected, got)
+		}
+	}
+}
+
 func TestUnhealthyForTooLong(t *testing.T) {
 	nodeUnhealthyForTooLong := maotesting.NewNode("nodeUnhealthyForTooLong", false)
 	nodeRecentlyUnhealthy := maotesting.NewNode("nodeRecentlyUnhealthy", false)