Introduce feature gates for the scheduler component

Signed-off-by: iawia002 <[email protected]>
Signed-off-by: Xinzhao Xu <[email protected]>

artifacts/deploy/karmada-scheduler.yaml

@@ -28,7 +28,7 @@ spec:
             - --kubeconfig=/etc/kubeconfig
             - --bind-address=0.0.0.0
             - --secure-port=10351
-            - --failover=true
+            - --feature-gates=Failover=true
             - --enable-scheduler-estimator=true
             - --v=4
           volumeMounts:

charts/templates/karmada_scheduler.yaml

@@ -55,7 +55,7 @@ spec:
             - --kubeconfig=/etc/kubeconfig
             - --bind-address=0.0.0.0
             - --secure-port=10351
-            - --failover=true
+            - --feature-gates=Failover=true
           volumeMounts:
           {{- include "karmada.kubeconfig.volumeMount" . | nindent 12 }}
           resources:

cmd/scheduler/app/options/options.go

@@ -8,6 +8,7 @@ import (
 	"k8s.io/client-go/tools/leaderelection/resourcelock"
 	componentbaseconfig "k8s.io/component-base/config"
 
+	"github.com/karmada-io/karmada/pkg/features"
 	"github.com/karmada-io/karmada/pkg/util"
 )
 
@@ -33,9 +34,6 @@ type Options struct {
 	// SecurePort is the port that the server serves at.
 	SecurePort int
 
-	// Failover indicates if scheduler should reschedule on cluster failure.
-	Failover bool
-
 	// KubeAPIQPS is the QPS to use while talking with karmada-apiserver.
 	KubeAPIQPS float32
 	// KubeAPIBurst is the burst to allow while talking with karmada-apiserver.
@@ -76,10 +74,10 @@ func (o *Options) AddFlags(fs *pflag.FlagSet) {
 	fs.StringVar(&o.Master, "master", o.Master, "The address of the Kubernetes API server. Overrides any value in KubeConfig. Only required if out-of-cluster.")
 	fs.StringVar(&o.BindAddress, "bind-address", defaultBindAddress, "The IP address on which to listen for the --secure-port port.")
 	fs.IntVar(&o.SecurePort, "secure-port", defaultPort, "The secure port on which to serve HTTPS.")
-	fs.BoolVar(&o.Failover, "failover", false, "Reschedule on cluster failure.")
 	fs.Float32Var(&o.KubeAPIQPS, "kube-api-qps", 40.0, "QPS to use while talking with karmada-apiserver. Doesn't cover events and node heartbeat apis which rate limiting is controlled by a different set of flags.")
 	fs.IntVar(&o.KubeAPIBurst, "kube-api-burst", 60, "Burst to use while talking with karmada-apiserver. Doesn't cover events and node heartbeat apis which rate limiting is controlled by a different set of flags.")
 	fs.BoolVar(&o.EnableSchedulerEstimator, "enable-scheduler-estimator", false, "Enable calling cluster scheduler estimator for adjusting replicas.")
 	fs.DurationVar(&o.SchedulerEstimatorTimeout.Duration, "scheduler-estimator-timeout", 3*time.Second, "Specifies the timeout period of calling the scheduler estimator service.")
 	fs.IntVar(&o.SchedulerEstimatorPort, "scheduler-estimator-port", defaultEstimatorPort, "The secure port on which to connect the accurate scheduler estimator.")
+	features.FeatureGate.AddFlag(fs)
 }

cmd/scheduler/app/scheduler.go

@@ -66,7 +66,6 @@ func run(opts *options.Options, stopChan <-chan struct{}) error {
 		cancel()
 	}()
 
-	scheduler.Failover = opts.Failover
 	sched := scheduler.NewScheduler(dynamicClientSet, karmadaClient, kubeClientSet, opts)
 	if !opts.LeaderElection.LeaderElect {
 		sched.Run(ctx)

pkg/features/features.go

@@ -0,0 +1,24 @@
+package features
+
+import (
+	"k8s.io/apimachinery/pkg/util/runtime"
+	"k8s.io/component-base/featuregate"
+)
+
+const (
+	// Failover indicates if scheduler should reschedule on cluster failure.
+	Failover featuregate.Feature = "Failover"
+)
+
+var (
+	// FeatureGate is a shared global scheduler FeatureGate.
+	FeatureGate featuregate.MutableFeatureGate = featuregate.NewFeatureGate()
+
+	defaultFeatureGates = map[featuregate.Feature]featuregate.FeatureSpec{
+		Failover: {Default: false, PreRelease: featuregate.Alpha},
+	}
+)
+
+func init() {
+	runtime.Must(FeatureGate.Add(defaultFeatureGates))
+}

pkg/scheduler/scheduler.go

@@ -24,6 +24,7 @@ import (
 	policyv1alpha1 "github.com/karmada-io/karmada/pkg/apis/policy/v1alpha1"
 	workv1alpha2 "github.com/karmada-io/karmada/pkg/apis/work/v1alpha2"
 	estimatorclient "github.com/karmada-io/karmada/pkg/estimator/client"
+	"github.com/karmada-io/karmada/pkg/features"
 	karmadaclientset "github.com/karmada-io/karmada/pkg/generated/clientset/versioned"
 	informerfactory "github.com/karmada-io/karmada/pkg/generated/informers/externalversions"
 	clusterlister "github.com/karmada-io/karmada/pkg/generated/listers/cluster/v1alpha1"
@@ -68,10 +69,6 @@ const (
 	scheduleSuccessMessage = "the binding has been scheduled"
 )
 
-// Failover indicates if the scheduler should performs re-scheduler in case of cluster failure.
-// TODO(RainbowMango): Remove the temporary solution by introducing feature flag
-var Failover bool
-
 // Scheduler is the scheduler schema, which is used to schedule a specific resource to specific clusters
 type Scheduler struct {
 	DynamicClient          dynamic.Interface
@@ -412,7 +409,7 @@ func (s *Scheduler) scheduleNext() bool {
 		klog.Infof("Reschedule binding(%s) as replicas scaled down or scaled up", keys)
 		metrics.BindingSchedule(string(ScaleSchedule), metrics.SinceInSeconds(start), err)
 	case FailoverSchedule:
-		if Failover {
+		if features.FeatureGate.Enabled(features.Failover) {
 			err = s.rescheduleOne(keys)
 			klog.Infof("Reschedule binding(%s) as cluster failure", keys)
 			metrics.BindingSchedule(string(FailoverSchedule), metrics.SinceInSeconds(start), err)
@@ -554,9 +551,9 @@ func (s *Scheduler) updateCluster(_, newObj interface{}) {
 
 	// Check if cluster becomes failure
 	if meta.IsStatusConditionPresentAndEqual(newCluster.Status.Conditions, clusterv1alpha1.ClusterConditionReady, metav1.ConditionFalse) {
-		klog.Infof("Found cluster(%s) failure and failover flag is %v", newCluster.Name, Failover)
+		klog.Infof("Found cluster(%s) failure and failover flag is %v", newCluster.Name, features.FeatureGate.Enabled(features.Failover))
 
-		if Failover { // Trigger reschedule on cluster failure only when flag is true.
+		if features.FeatureGate.Enabled(features.Failover) { // Trigger reschedule on cluster failure only when flag is true.
 			s.enqueueAffectedBinding(newCluster.Name)
 			s.enqueueAffectedClusterBinding(newCluster.Name)
 			return