Added optional delays to starting controller managers. #22669

cmd/kube-controller-manager/app/controllermanager.go

@@ -75,6 +75,11 @@ import (
 	"github.com/spf13/pflag"
 )
 
+const (
+	// Jitter used when starting controller managers
+	ControllerStartJitter = 1.0
+)
+
 // NewControllerManagerCommand creates a *cobra.Command object with default parameters
 func NewControllerManagerCommand() *cobra.Command {
 	s := options.NewCMServer()
@@ -186,6 +191,7 @@ func Run(s *options.CMServer) error {
 func StartControllers(s *options.CMServer, kubeClient *client.Client, kubeconfig *restclient.Config, stop <-chan struct{}) error {
 	go endpointcontroller.NewEndpointController(clientset.NewForConfigOrDie(restclient.AddUserAgent(kubeconfig, "endpoint-controller")), ResyncPeriod(s)).
 		Run(s.ConcurrentEndpointSyncs, wait.NeverStop)
+	time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 
 	go replicationcontroller.NewReplicationManager(
 		clientset.NewForConfigOrDie(restclient.AddUserAgent(kubeconfig, "replication-controller")),
@@ -193,10 +199,12 @@ func StartControllers(s *options.CMServer, kubeClient *client.Client, kubeconfig
 		replicationcontroller.BurstReplicas,
 		s.LookupCacheSizeForRC,
 	).Run(s.ConcurrentRCSyncs, wait.NeverStop)
+	time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 
 	if s.TerminatedPodGCThreshold > 0 {
 		go gc.New(clientset.NewForConfigOrDie(restclient.AddUserAgent(kubeconfig, "garbage-collector")), ResyncPeriod(s), s.TerminatedPodGCThreshold).
 			Run(wait.NeverStop)
+		time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 	}
 
 	cloud, err := cloudprovider.InitCloudProvider(s.CloudProvider, s.CloudConfigFile)
@@ -211,11 +219,13 @@ func StartControllers(s *options.CMServer, kubeClient *client.Client, kubeconfig
 		flowcontrol.NewTokenBucketRateLimiter(s.DeletingPodsQps, s.DeletingPodsBurst),
 		s.NodeMonitorGracePeriod.Duration, s.NodeStartupGracePeriod.Duration, s.NodeMonitorPeriod.Duration, clusterCIDR, s.AllocateNodeCIDRs)
 	nodeController.Run(s.NodeSyncPeriod.Duration)
+	time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 
 	serviceController := servicecontroller.New(cloud, clientset.NewForConfigOrDie(restclient.AddUserAgent(kubeconfig, "service-controller")), s.ClusterName)
 	if err := serviceController.Run(s.ServiceSyncPeriod.Duration, s.NodeSyncPeriod.Duration); err != nil {
 		glog.Errorf("Failed to start service controller: %v", err)
 	}
+	time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 
 	if s.AllocateNodeCIDRs {
 		if cloud == nil {
@@ -225,6 +235,7 @@ func StartControllers(s *options.CMServer, kubeClient *client.Client, kubeconfig
 		} else {
 			routeController := routecontroller.New(routes, clientset.NewForConfigOrDie(restclient.AddUserAgent(kubeconfig, "route-controller")), s.ClusterName, clusterCIDR)
 			routeController.Run(s.NodeSyncPeriod.Duration)
+			time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 		}
 	} else {
 		glog.Infof("allocate-node-cidrs set to %v, node controller not creating routes", s.AllocateNodeCIDRs)
@@ -249,6 +260,7 @@ func StartControllers(s *options.CMServer, kubeClient *client.Client, kubeconfig
 		GroupKindsToReplenish:     groupKindsToReplenish,
 	}
 	go resourcequotacontroller.NewResourceQuotaController(resourceQuotaControllerOptions).Run(s.ConcurrentResourceQuotaSyncs, wait.NeverStop)
+	time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 
 	// If apiserver is not running we should wait for some time and fail only then. This is particularly
 	// important when we start apiserver and controller manager at the same time.
@@ -279,6 +291,7 @@ func StartControllers(s *options.CMServer, kubeClient *client.Client, kubeconfig
 	}
 	namespaceController := namespacecontroller.NewNamespaceController(namespaceKubeClient, namespaceClientPool, groupVersionResources, s.NamespaceSyncPeriod.Duration, api.FinalizerKubernetes)
 	go namespaceController.Run(s.ConcurrentNamespaceSyncs, wait.NeverStop)
+	time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 
 	groupVersion := "extensions/v1beta1"
 	resources, found := resourceMap[groupVersion]
@@ -297,30 +310,35 @@ func StartControllers(s *options.CMServer, kubeClient *client.Client, kubeconfig
 			)
 			go podautoscaler.NewHorizontalController(hpaClient.Core(), hpaClient.Extensions(), hpaClient, metricsClient, s.HorizontalPodAutoscalerSyncPeriod.Duration).
 				Run(wait.NeverStop)
+			time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 		}
 
 		if containsResource(resources, "daemonsets") {
 			glog.Infof("Starting daemon set controller")
 			go daemon.NewDaemonSetsController(clientset.NewForConfigOrDie(restclient.AddUserAgent(kubeconfig, "daemon-set-controller")), ResyncPeriod(s), s.LookupCacheSizeForDaemonSet).
 				Run(s.ConcurrentDaemonSetSyncs, wait.NeverStop)
+			time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 		}
 
 		if containsResource(resources, "jobs") {
 			glog.Infof("Starting job controller")
 			go job.NewJobController(clientset.NewForConfigOrDie(restclient.AddUserAgent(kubeconfig, "job-controller")), ResyncPeriod(s)).
 				Run(s.ConcurrentJobSyncs, wait.NeverStop)
+			time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 		}
 
 		if containsResource(resources, "deployments") {
 			glog.Infof("Starting deployment controller")
 			go deployment.NewDeploymentController(clientset.NewForConfigOrDie(restclient.AddUserAgent(kubeconfig, "deployment-controller")), ResyncPeriod(s)).
 				Run(s.ConcurrentDeploymentSyncs, wait.NeverStop)
+			time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 		}
 
 		if containsResource(resources, "replicasets") {
 			glog.Infof("Starting ReplicaSet controller")
 			go replicaset.NewReplicaSetController(clientset.NewForConfigOrDie(restclient.AddUserAgent(kubeconfig, "replicaset-controller")), ResyncPeriod(s), replicaset.BurstReplicas, s.LookupCacheSizeForRS).
 				Run(s.ConcurrentRSSyncs, wait.NeverStop)
+			time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 		}
 	}
 
@@ -332,6 +350,7 @@ func StartControllers(s *options.CMServer, kubeClient *client.Client, kubeconfig
 
 	pvclaimBinder := persistentvolumecontroller.NewPersistentVolumeClaimBinder(clientset.NewForConfigOrDie(restclient.AddUserAgent(kubeconfig, "persistent-volume-binder")), s.PVClaimBinderSyncPeriod.Duration)
 	pvclaimBinder.Run()
+	time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 
 	pvRecycler, err := persistentvolumecontroller.NewPersistentVolumeRecycler(
 		clientset.NewForConfigOrDie(restclient.AddUserAgent(kubeconfig, "persistent-volume-recycler")),
@@ -344,6 +363,7 @@ func StartControllers(s *options.CMServer, kubeClient *client.Client, kubeconfig
 		glog.Fatalf("Failed to start persistent volume recycler: %+v", err)
 	}
 	pvRecycler.Run()
+	time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 
 	if provisioner != nil {
 		pvController, err := persistentvolumecontroller.NewPersistentVolumeProvisionerController(persistentvolumecontroller.NewControllerClient(clientset.NewForConfigOrDie(restclient.AddUserAgent(kubeconfig, "persistent-volume-provisioner"))), s.PVClaimBinderSyncPeriod.Duration, s.ClusterName, volumePlugins, provisioner, cloud)
@@ -351,6 +371,7 @@ func StartControllers(s *options.CMServer, kubeClient *client.Client, kubeconfig
 			glog.Fatalf("Failed to start persistent volume provisioner controller: %+v", err)
 		}
 		pvController.Run()
+		time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 	}
 
 	var rootCA []byte
@@ -379,6 +400,7 @@ func StartControllers(s *options.CMServer, kubeClient *client.Client, kubeconfig
 					RootCA:         rootCA,
 				},
 			).Run()
+			time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 		}
 	}
 
@@ -386,6 +408,7 @@ func StartControllers(s *options.CMServer, kubeClient *client.Client, kubeconfig
 		clientset.NewForConfigOrDie(restclient.AddUserAgent(kubeconfig, "service-account-controller")),
 		serviceaccountcontroller.DefaultServiceAccountsControllerOptions(),
 	).Run()
+	time.Sleep(wait.Jitter(s.ControllerStartInterval.Duration, ControllerStartJitter))
 
 	select {}
 }

cmd/kube-controller-manager/app/options/options.go

@@ -84,6 +84,7 @@ func NewCMServer() *CMServer {
 			KubeAPIQPS:              20.0,
 			KubeAPIBurst:            30,
 			LeaderElection:          leaderelection.DefaultLeaderElectionConfiguration(),
+			ControllerStartInterval: unversioned.Duration{0 * time.Second},
 		},
 	}
 	return &s
@@ -146,5 +147,6 @@ func (s *CMServer) AddFlags(fs *pflag.FlagSet) {
 	fs.StringVar(&s.RootCAFile, "root-ca-file", s.RootCAFile, "If set, this root certificate authority will be included in service account's token secret. This must be a valid PEM-encoded CA bundle.")
 	fs.Float32Var(&s.KubeAPIQPS, "kube-api-qps", s.KubeAPIQPS, "QPS to use while talking with kubernetes apiserver")
 	fs.IntVar(&s.KubeAPIBurst, "kube-api-burst", s.KubeAPIBurst, "Burst to use while talking with kubernetes apiserver")
+	fs.DurationVar(&s.ControllerStartInterval.Duration, "controller-start-interval", s.ControllerStartInterval.Duration, "Interval between starting controller managers.")
 	leaderelection.BindFlags(&s.LeaderElection, fs)
 }

docs/admin/kube-controller-manager.md

@@ -67,6 +67,7 @@ kube-controller-manager
       --concurrent-replicaset-syncs=5: The number of replica sets that are allowed to sync concurrently. Larger number = more responsive replica management, but more CPU (and network) load
       --concurrent-resource-quota-syncs=5: The number of resource quotas that are allowed to sync concurrently. Larger number = more responsive quota management, but more CPU (and network) load
       --concurrent_rc_syncs=5: The number of replication controllers that are allowed to sync concurrently. Larger number = more responsive replica management, but more CPU (and network) load
+      --controller-start-interval=0: Interval between starting controller managers.
       --daemonset-lookup-cache-size=1024: The the size of lookup cache for daemonsets. Larger number = more responsive daemonsets, but more MEM load.
       --deleting-pods-burst=10: Number of nodes on which pods are bursty deleted in case of node failure. For more details look into RateLimiter.
       --deleting-pods-qps=0.1: Number of nodes per second on which pods are deleted in case of node failure.
@@ -108,7 +109,7 @@ kube-controller-manager
       --terminated-pod-gc-threshold=12500: Number of terminated pods that can exist before the terminated pod garbage collector starts deleting terminated pods. If <= 0, the terminated pod garbage collector is disabled.
 ```
 
-###### Auto generated by spf13/cobra on 29-Feb-2016
+###### Auto generated by spf13/cobra on 5-Apr-2016
 
 
 <!-- BEGIN MUNGE: GENERATED_ANALYTICS -->

hack/verify-flags/known-flags.txt

@@ -65,6 +65,7 @@ conntrack-tcp-timeout-established
 contain-pod-resources
 container-port
 container-runtime
+controller-start-interval
 cors-allowed-origins
 cpu-cfs-quota
 cpu-percent

pkg/apis/componentconfig/deep_copy_generated.go

@@ -127,6 +127,9 @@ func DeepCopy_componentconfig_KubeControllerManagerConfiguration(in KubeControll
 	if err := DeepCopy_componentconfig_VolumeConfiguration(in.VolumeConfiguration, &out.VolumeConfiguration, c); err != nil {
 		return err
 	}
+	if err := unversioned.DeepCopy_unversioned_Duration(in.ControllerStartInterval, &out.ControllerStartInterval, c); err != nil {
+		return err
+	}
 	return nil
 }
 

pkg/apis/componentconfig/types.go

@@ -510,6 +510,8 @@ type KubeControllerManagerConfiguration struct {
 	LeaderElection LeaderElectionConfiguration `json:"leaderElection"`
 	// vloumeConfiguration holds configuration for volume related features.
 	VolumeConfiguration VolumeConfiguration `json:"volumeConfiguration"`
+	// How long to wait between starting controller managers
+	ControllerStartInterval unversioned.Duration `json:"controllerStartInterval"`
 }
 
 // VolumeConfiguration contains *all* enumerated flags meant to configure all volume