Merge pull request #5739 from djoshy/disable-skew-sno

MCO-2145: Disable skew enforcement on SNO clusters

Author: openshift-merge-bot[bot]

cmd/machine-config-controller/start.go

@@ -276,6 +276,7 @@ func createControllers(ctx *ctrlcommon.ControllerContext) []ctrlcommon.Controlle
 			ctx.ConfigInformerFactory.Config().V1().Schedulers(),
 			ctx.OperatorInformerFactory.Operator().V1().MachineConfigurations(),
 			ctx.InformerFactory.Machineconfiguration().V1alpha1().OSImageStreams(),
+			ctx.ConfigInformerFactory.Config().V1().Infrastructures(),
 			ctx.ClientBuilder.KubeClientOrDie("node-update-controller"),
 			ctx.ClientBuilder.MachineConfigClientOrDie("node-update-controller"),
 			ctx.FeatureGatesHandler,

pkg/controller/node/node_controller.go

@@ -120,6 +120,9 @@ type Controller struct {
 	mcopLister       mcoplistersv1.MachineConfigurationLister
 	mcopListerSynced cache.InformerSynced
 
+	infraLister       cligolistersv1.InfrastructureLister
+	infraListerSynced cache.InformerSynced
+
 	queue workqueue.TypedRateLimitingInterface[string]
 
 	fgHandler ctrlcommon.FeatureGatesHandler
@@ -144,6 +147,7 @@ func New(
 	schedulerInformer cligoinformersv1.SchedulerInformer,
 	mcopInformer mcopinformersv1.MachineConfigurationInformer,
 	osImageStreamInformer mcfginformersv1alpha1.OSImageStreamInformer,
+	infraInformer cligoinformersv1.InfrastructureInformer,
 	kubeClient clientset.Interface,
 	mcfgClient mcfgclientset.Interface,
 	fgHandler ctrlcommon.FeatureGatesHandler,
@@ -160,6 +164,7 @@ func New(
 		schedulerInformer,
 		mcopInformer,
 		osImageStreamInformer,
+		infraInformer,
 		kubeClient,
 		mcfgClient,
 		defaultUpdateDelay,
@@ -179,6 +184,7 @@ func NewWithCustomUpdateDelay(
 	schedulerInformer cligoinformersv1.SchedulerInformer,
 	mcopInformer mcopinformersv1.MachineConfigurationInformer,
 	osImageStreamInformer mcfginformersv1alpha1.OSImageStreamInformer,
+	infraInformer cligoinformersv1.InfrastructureInformer,
 	kubeClient clientset.Interface,
 	mcfgClient mcfgclientset.Interface,
 	updateDelay time.Duration,
@@ -196,6 +202,7 @@ func NewWithCustomUpdateDelay(
 		schedulerInformer,
 		mcopInformer,
 		osImageStreamInformer,
+		infraInformer,
 		kubeClient,
 		mcfgClient,
 		updateDelay,
@@ -216,6 +223,7 @@ func newController(
 	schedulerInformer cligoinformersv1.SchedulerInformer,
 	mcopInformer mcopinformersv1.MachineConfigurationInformer,
 	osImageStreamInformer mcfginformersv1alpha1.OSImageStreamInformer,
+	infraInformer cligoinformersv1.InfrastructureInformer,
 	kubeClient clientset.Interface,
 	mcfgClient mcfgclientset.Interface,
 	updateDelay time.Duration,
@@ -297,6 +305,9 @@ func newController(
 	ctrl.mcopLister = mcopInformer.Lister()
 	ctrl.mcopListerSynced = mcopInformer.Informer().HasSynced
 
+	ctrl.infraLister = infraInformer.Lister()
+	ctrl.infraListerSynced = infraInformer.Informer().HasSynced
+
 	// Only initialize OSImageStream lister if feature gate is enabled
 	if ctrl.osStreamsFgEnabled {
 		ctrl.osImageStreamLister = osImageStreamInformer.Lister()
@@ -314,6 +325,7 @@ func (ctrl *Controller) Run(workers int, stopCh <-chan struct{}) {
 	syncers := []cache.InformerSynced{
 		ctrl.ccListerSynced, ctrl.mcListerSynced, ctrl.mcpListerSynced, ctrl.moscListerSynced,
 		ctrl.mosbListerSynced, ctrl.nodeListerSynced, ctrl.schedulerListerSynced, ctrl.mcopListerSynced,
+		ctrl.infraListerSynced,
 	}
 	// Only wait for the OSImageStream informer to sync if the feature is enabled
 	if ctrl.osStreamsFgEnabled {
@@ -1951,8 +1963,8 @@ func (ctrl *Controller) deleteMachineConfiguration(_ any) {
 
 // syncBootImageSkewEnforcementMetric updates the mcc_boot_image_skew_enforcement_none metric
 // based on the current BootImageSkewEnforcementStatus mode in MachineConfiguration.
-// The metric is set to 1 when mode is "None", indicating that scaling operations may
-// not be successful.
+// The metric is set to 1 when mode is "None" on non-SNO clusters, indicating that scaling
+// operations may not be successful. On SNO clusters, None is the default and the alert is suppressed.
 func (ctrl *Controller) syncBootImageSkewEnforcementMetric(obj any) {
 
 	mcop, ok := obj.(*opv1.MachineConfiguration)
@@ -1962,6 +1974,17 @@ func (ctrl *Controller) syncBootImageSkewEnforcementMetric(obj any) {
 	}
 
 	if mcop.Status.BootImageSkewEnforcementStatus.Mode == opv1.BootImageSkewEnforcementModeStatusNone {
+		infra, err := ctrl.infraLister.Get("cluster")
+		if err != nil {
+			klog.Warningf("Failed to get infrastructure for skew enforcement metric: %v", err)
+			ctrlcommon.MCCBootImageSkewEnforcementNone.Set(0)
+			return
+		}
+		// On SNO clusters, None is the default; suppress the alert.
+		if infra.Status.ControlPlaneTopology == configv1.SingleReplicaTopologyMode {
+			ctrlcommon.MCCBootImageSkewEnforcementNone.Set(0)
+			return
+		}
 		ctrlcommon.MCCBootImageSkewEnforcementNone.Set(1)
 	} else {
 		ctrlcommon.MCCBootImageSkewEnforcementNone.Set(0)

pkg/controller/node/node_controller_test.go

@@ -112,14 +112,15 @@ func (f *fixture) newControllerWithStopChan(stopCh <-chan struct{}) *Controller
 	oi := operatorinformer.NewSharedInformerFactory(operatorClient, noResyncPeriodFunc())
 	c := NewWithCustomUpdateDelay(i.Machineconfiguration().V1().ControllerConfigs(), i.Machineconfiguration().V1().MachineConfigs(), i.Machineconfiguration().V1().MachineConfigPools(), k8sI.Core().V1().Nodes(),
 		k8sI.Core().V1().Pods(), i.Machineconfiguration().V1().MachineOSConfigs(), i.Machineconfiguration().V1().MachineOSBuilds(), i.Machineconfiguration().V1().MachineConfigNodes(), ci.Config().V1().Schedulers(), oi.Operator().V1().MachineConfigurations(),
-		i.Machineconfiguration().V1alpha1().OSImageStreams(), f.kubeclient, f.client, time.Millisecond, f.fgHandler)
+		i.Machineconfiguration().V1alpha1().OSImageStreams(), ci.Config().V1().Infrastructures(), f.kubeclient, f.client, time.Millisecond, f.fgHandler)
 
 	c.ccListerSynced = alwaysReady
 	c.mcpListerSynced = alwaysReady
 	c.nodeListerSynced = alwaysReady
 	c.schedulerListerSynced = alwaysReady
 	c.mcopListerSynced = alwaysReady
 	c.osImageStreamListerSynced = alwaysReady
+	c.infraListerSynced = alwaysReady
 	c.eventRecorder = &record.FakeRecorder{}
 
 	i.Start(stopCh)

pkg/operator/sync.go

@@ -2367,7 +2367,7 @@ func (optr *Operator) syncMachineConfiguration(_ *renderConfig, _ *configv1.Clus
 	}
 
 	// Update skew enforcement status if needed
-	optr.syncBootImageSkewEnforcementStatus(mcop, newMachineConfigurationStatus, supportsBootImageUpdates)
+	optr.syncBootImageSkewEnforcementStatus(mcop, newMachineConfigurationStatus, infra, supportsBootImageUpdates)
 
 	newMachineConfigurationStatus.ObservedGeneration = mcop.GetGeneration()
 	// Check if any changes are required in the Status before making the API call.
@@ -2533,7 +2533,7 @@ func (optr *Operator) syncPreBuiltImageMachineConfigs() error {
 
 // syncBootImageSkewEnforcementStatus determines the appropriate BootImageSkewEnforcementStatus based on
 // the MachineConfiguration spec, platform defaults, and cluster version information.
-func (optr *Operator) syncBootImageSkewEnforcementStatus(mcop *opv1.MachineConfiguration, newMachineConfigurationStatus *opv1.MachineConfigurationStatus, supportsBootImageUpdates bool) {
+func (optr *Operator) syncBootImageSkewEnforcementStatus(mcop *opv1.MachineConfiguration, newMachineConfigurationStatus *opv1.MachineConfigurationStatus, infra *configv1.Infrastructure, supportsBootImageUpdates bool) {
 	// React to any changes in boot image skew enforcement configuration
 	if !optr.fgHandler.Enabled(features.FeatureGateBootImageSkewEnforcement) {
 		return
@@ -2552,6 +2552,9 @@ func (optr *Operator) syncBootImageSkewEnforcementStatus(mcop *opv1.MachineConfi
 		} else { // only other possible opinion is "None"
 			newMachineConfigurationStatus.BootImageSkewEnforcementStatus = apihelpers.GetSkewEnforcementStatusNone()
 		}
+	} else if infra.Status.ControlPlaneTopology == configv1.SingleReplicaTopologyMode {
+		// On SNO clusters, default to None since there are no MachineSets to scale
+		newMachineConfigurationStatus.BootImageSkewEnforcementStatus = apihelpers.GetSkewEnforcementStatusNone()
 	} else if supportsBootImageUpdates {
 		// If an "All" option is specified and BootImageSkewEnforcementStatus is empty or not set to Automatic => set Mode to Automatic.
 		if apihelpers.HasMAPIMachineSetManagerWithMode(newMachineConfigurationStatus.ManagedBootImagesStatus.MachineManagers, opv1.MachineSets, opv1.All) {

pkg/operator/sync_test.go

@@ -137,6 +137,12 @@ func withPlatformType(platformType configv1.PlatformType) infraOption {
 	}
 }
 
+func withControlPlaneTopology(topology configv1.TopologyMode) infraOption {
+	return func(infra *configv1.Infrastructure) {
+		infra.Status.ControlPlaneTopology = topology
+	}
+}
+
 type kubeCloudConfigOption func(*corev1.ConfigMap)
 
 func buildKubeCloudConfig(opts ...kubeCloudConfigOption) *corev1.ConfigMap {
@@ -728,6 +734,32 @@ func TestSyncMachineConfiguration(t *testing.T) {
 			},
 			expectedSkewEnforcementStatus: apihelpers.GetSkewEnforcementStatusAutomaticWithOCPVersion("4.18.0"),
 		},
+		{
+			name:               "SNO cluster, no skew enforcement spec, skew enforcement defaults to None",
+			infra:              buildInfra(withPlatformType(configv1.AWSPlatformType), withControlPlaneTopology(configv1.SingleReplicaTopologyMode)),
+			mcop:               buildMachineConfigurationWithNoBootImageConfiguration(),
+			clusterVersion:     buildClusterVersion("4.18.0"),
+			annotationExpected: true,
+			expectedManagedBootImagesStatus: opv1.ManagedBootImages{
+				MachineManagers: []opv1.MachineManager{
+					{Resource: opv1.MachineSets, APIGroup: opv1.MachineAPI, Selection: opv1.MachineManagerSelector{Mode: opv1.All}},
+				},
+			},
+			expectedSkewEnforcementStatus: apihelpers.GetSkewEnforcementStatusNone(),
+		},
+		{
+			name:           "SNO cluster, spec defines manual mode, status should reflect spec",
+			infra:          buildInfra(withPlatformType(configv1.AWSPlatformType), withControlPlaneTopology(configv1.SingleReplicaTopologyMode)),
+			mcop:           buildMachineConfigurationWithSkewEnforcementManual("4.17.0"),
+			clusterVersion: buildClusterVersion("4.18.0"),
+			annotationExpected: true,
+			expectedManagedBootImagesStatus: opv1.ManagedBootImages{
+				MachineManagers: []opv1.MachineManager{
+					{Resource: opv1.MachineSets, APIGroup: opv1.MachineAPI, Selection: opv1.MachineManagerSelector{Mode: opv1.All}},
+				},
+			},
+			expectedSkewEnforcementStatus: apihelpers.GetSkewEnforcementStatusManualWithOCPVersion("4.17.0"),
+		},
 	}
 
 	for _, tc := range cases {

test/e2e-bootstrap/bootstrap_test.go

@@ -626,6 +626,7 @@ func createControllers(ctx *ctrlcommon.ControllerContext) []ctrlcommon.Controlle
 			ctx.ConfigInformerFactory.Config().V1().Schedulers(),
 			ctx.OperatorInformerFactory.Operator().V1().MachineConfigurations(),
 			ctx.InformerFactory.Machineconfiguration().V1alpha1().OSImageStreams(),
+			ctx.ConfigInformerFactory.Config().V1().Infrastructures(),
 			ctx.ClientBuilder.KubeClientOrDie("node-update-controller"),
 			ctx.ClientBuilder.MachineConfigClientOrDie("node-update-controller"),
 			ctx.FeatureGatesHandler,