func forcefullyDeletePod(c clientset.Interface, pod *api.Pod) {
var zero int64
err := c.Legacy().Pods(pod.Namespace).Delete(pod.Name, &api.DeleteOptions{GracePeriodSeconds: &zero})
if err != nil {
utilruntime.HandleError(err)
}
}
func deletePods(kubeClient clientset.Interface, ns string, before unversioned.Time) (int64, error) {
items, err := kubeClient.Legacy().Pods(ns).List(api.ListOptions{})
if err != nil {
return 0, err
}
expired := unversioned.Now().After(before.Time)
var deleteOptions *api.DeleteOptions
if expired {
deleteOptions = api.NewDeleteOptions(0)
}
estimate := int64(0)
for i := range items.Items {
if items.Items[i].Spec.TerminationGracePeriodSeconds != nil {
grace := *items.Items[i].Spec.TerminationGracePeriodSeconds
if grace > estimate {
estimate = grace
}
}
err := kubeClient.Legacy().Pods(ns).Delete(items.Items[i].Name, deleteOptions)
if err != nil && !errors.IsNotFound(err) {
return 0, err
}
}
if expired {
estimate = 0
}
return estimate, nil
}
// GetNewRC returns an RC that matches the intent of the given deployment; get RCList from client interface.
// Returns nil if the new RC doesnt exist yet.
func GetNewRC(deployment extensions.Deployment, c clientset.Interface) (*api.ReplicationController, error) {
return GetNewRCFromList(deployment, c,
func(namespace string, options api.ListOptions) ([]api.ReplicationController, error) {
rcList, err := c.Legacy().ReplicationControllers(namespace).List(options)
return rcList.Items, err
})
}
func New(kubeClient clientset.Interface, resyncPeriod controller.ResyncPeriodFunc, threshold int) *GCController {
gcc := &GCController{
kubeClient: kubeClient,
threshold: threshold,
deletePod: func(namespace, name string) error {
return kubeClient.Legacy().Pods(namespace).Delete(name, api.NewDeleteOptions(0))
},
}
terminatedSelector := fields.ParseSelectorOrDie("status.phase!=" + string(api.PodPending) + ",status.phase!=" + string(api.PodRunning) + ",status.phase!=" + string(api.PodUnknown))
gcc.podStore.Store, gcc.podStoreSyncer = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
options.FieldSelector = terminatedSelector
return gcc.kubeClient.Legacy().Pods(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
options.FieldSelector = terminatedSelector
return gcc.kubeClient.Legacy().Pods(api.NamespaceAll).Watch(options)
},
},
&api.Pod{},
resyncPeriod(),
framework.ResourceEventHandlerFuncs{},
)
return gcc
}
// updateNamespaceStatusFunc will verify that the status of the namespace is correct
func updateNamespaceStatusFunc(kubeClient clientset.Interface, namespace *api.Namespace) (*api.Namespace, error) {
if namespace.DeletionTimestamp.IsZero() || namespace.Status.Phase == api.NamespaceTerminating {
return namespace, nil
}
newNamespace := api.Namespace{}
newNamespace.ObjectMeta = namespace.ObjectMeta
newNamespace.Status = namespace.Status
newNamespace.Status.Phase = api.NamespaceTerminating
return kubeClient.Legacy().Namespaces().UpdateStatus(&newNamespace)
}
func deleteResourceQuotas(kubeClient clientset.Interface, ns string) error {
resourceQuotas, err := kubeClient.Legacy().ResourceQuotas(ns).List(api.ListOptions{})
if err != nil {
return err
}
for i := range resourceQuotas.Items {
err := kubeClient.Legacy().ResourceQuotas(ns).Delete(resourceQuotas.Items[i].Name, nil)
if err != nil && !errors.IsNotFound(err) {
return err
}
}
return nil
}
func deletePersistentVolumeClaims(kubeClient clientset.Interface, ns string) error {
items, err := kubeClient.Legacy().PersistentVolumeClaims(ns).List(api.ListOptions{})
if err != nil {
return err
}
for i := range items.Items {
err := kubeClient.Legacy().PersistentVolumeClaims(ns).Delete(items.Items[i].Name, nil)
if err != nil && !errors.IsNotFound(err) {
return err
}
}
return nil
}
func getPodsForRCs(c clientset.Interface, replicationControllers []*api.ReplicationController) ([]api.Pod, error) {
allPods := []api.Pod{}
for _, rc := range replicationControllers {
selector := labels.SelectorFromSet(rc.Spec.Selector)
options := api.ListOptions{LabelSelector: selector}
podList, err := c.Legacy().Pods(rc.ObjectMeta.Namespace).List(options)
if err != nil {
return allPods, fmt.Errorf("error listing pods: %v", err)
}
allPods = append(allPods, podList.Items...)
}
return allPods, nil
}
// NewResourceQuota creates a new resource quota admission control handler
func NewResourceQuota(client clientset.Interface) admission.Interface {
lw := &cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return client.Legacy().ResourceQuotas(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return client.Legacy().ResourceQuotas(api.NamespaceAll).Watch(options)
},
}
indexer, reflector := cache.NewNamespaceKeyedIndexerAndReflector(lw, &api.ResourceQuota{}, 0)
reflector.Run()
return createResourceQuota(client, indexer)
}
// New returns a new service controller to keep cloud provider service resources
// (like load balancers) in sync with the registry.
func New(cloud cloudprovider.Interface, kubeClient clientset.Interface, clusterName string) *ServiceController {
broadcaster := record.NewBroadcaster()
broadcaster.StartRecordingToSink(&unversioned_legacy.EventSinkImpl{kubeClient.Legacy().Events("")})
recorder := broadcaster.NewRecorder(api.EventSource{Component: "service-controller"})
return &ServiceController{
cloud: cloud,
kubeClient: kubeClient,
clusterName: clusterName,
cache: &serviceCache{serviceMap: make(map[string]*cachedService)},
eventBroadcaster: broadcaster,
eventRecorder: recorder,
nodeLister: cache.StoreToNodeLister{
Store: cache.NewStore(cache.MetaNamespaceKeyFunc),
},
}
}
// retryOnConflictError retries the specified fn if there was a conflict error
// TODO RetryOnConflict should be a generic concept in client code
func retryOnConflictError(kubeClient clientset.Interface, namespace *api.Namespace, fn updateNamespaceFunc) (result *api.Namespace, err error) {
latestNamespace := namespace
for {
result, err = fn(kubeClient, latestNamespace)
if err == nil {
return result, nil
}
if !errors.IsConflict(err) {
return nil, err
}
latestNamespace, err = kubeClient.Legacy().Namespaces().Get(latestNamespace.Name)
if err != nil {
return nil, err
}
}
return
}
// PersistentVolumeRecycler creates a new PersistentVolumeRecycler
func NewPersistentVolumeRecycler(kubeClient clientset.Interface, syncPeriod time.Duration, plugins []volume.VolumePlugin, cloud cloudprovider.Interface) (*PersistentVolumeRecycler, error) {
recyclerClient := NewRecyclerClient(kubeClient)
recycler := &PersistentVolumeRecycler{
client: recyclerClient,
kubeClient: kubeClient,
cloud: cloud,
}
if err := recycler.pluginMgr.InitPlugins(plugins, recycler); err != nil {
return nil, fmt.Errorf("Could not initialize volume plugins for PVClaimBinder: %+v", err)
}
_, volumeController := framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return kubeClient.Legacy().PersistentVolumes().List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return kubeClient.Legacy().PersistentVolumes().Watch(options)
},
},
&api.PersistentVolume{},
syncPeriod,
framework.ResourceEventHandlerFuncs{
AddFunc: func(obj interface{}) {
pv, ok := obj.(*api.PersistentVolume)
if !ok {
glog.Errorf("Error casting object to PersistentVolume: %v", obj)
return
}
recycler.reclaimVolume(pv)
},
UpdateFunc: func(oldObj, newObj interface{}) {
pv, ok := newObj.(*api.PersistentVolume)
if !ok {
glog.Errorf("Error casting object to PersistentVolume: %v", newObj)
return
}
recycler.reclaimVolume(pv)
},
},
)
recycler.volumeController = volumeController
return recycler, nil
}
// NewLimitRanger returns an object that enforces limits based on the supplied limit function
func NewLimitRanger(client clientset.Interface, limitFunc LimitFunc) admission.Interface {
lw := &cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return client.Legacy().LimitRanges(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return client.Legacy().LimitRanges(api.NamespaceAll).Watch(options)
},
}
indexer, reflector := cache.NewNamespaceKeyedIndexerAndReflector(lw, &api.LimitRange{}, 0)
reflector.Run()
return &limitRanger{
Handler: admission.NewHandler(admission.Create, admission.Update),
client: client,
limitFunc: limitFunc,
indexer: indexer,
}
}
// NewProvision creates a new namespace provision admission control handler
func NewProvision(c clientset.Interface) admission.Interface {
store := cache.NewStore(cache.MetaNamespaceKeyFunc)
reflector := cache.NewReflector(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return c.Legacy().Namespaces().List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return c.Legacy().Namespaces().Watch(options)
},
},
&api.Namespace{},
store,
0,
)
reflector.Run()
return createProvision(c, store)
}
// SyncAllPodsWithStore lists all pods and inserts them into the given store.
// Though this function is written in a generic manner, it is only used by the
// controllers for a specific purpose, to synchronously populate the store
// with the first batch of pods that would otherwise be sent by the Informer.
// Doing this avoids more complicated forms of synchronization with the
// Informer, though it also means that the controller calling this function
// will receive "OnUpdate" events for all the pods in the store, instead of
// "OnAdd". This should be ok, since most controllers are level triggered
// and make decisions based on the contents of the store.
//
// TODO: Extend this logic to load arbitrary local state for the controllers
// instead of just pods.
func SyncAllPodsWithStore(kubeClient clientset.Interface, store cache.Store) {
var allPods *api.PodList
var err error
listOptions := api.ListOptions{LabelSelector: labels.Everything(), FieldSelector: fields.Everything()}
for {
if allPods, err = kubeClient.Legacy().Pods(api.NamespaceAll).List(listOptions); err != nil {
glog.Warningf("Retrying pod list: %v", err)
continue
}
break
}
pods := []interface{}{}
for i := range allPods.Items {
p := allPods.Items[i]
glog.V(4).Infof("Initializing store with pod %v/%v", p.Namespace, p.Name)
pods = append(pods, &p)
}
store.Replace(pods, allPods.ResourceVersion)
return
}
// NewPersistentVolumeClaimBinder creates a new PersistentVolumeClaimBinder
func NewPersistentVolumeClaimBinder(kubeClient clientset.Interface, syncPeriod time.Duration) *PersistentVolumeClaimBinder {
volumeIndex := NewPersistentVolumeOrderedIndex()
binderClient := NewBinderClient(kubeClient)
binder := &PersistentVolumeClaimBinder{
volumeIndex: volumeIndex,
client: binderClient,
}
_, volumeController := framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return kubeClient.Legacy().PersistentVolumes().List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return kubeClient.Legacy().PersistentVolumes().Watch(options)
},
},
&api.PersistentVolume{},
// TODO: Can we have much longer period here?
syncPeriod,
framework.ResourceEventHandlerFuncs{
AddFunc: binder.addVolume,
UpdateFunc: binder.updateVolume,
DeleteFunc: binder.deleteVolume,
},
)
_, claimController := framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return kubeClient.Legacy().PersistentVolumeClaims(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return kubeClient.Legacy().PersistentVolumeClaims(api.NamespaceAll).Watch(options)
},
},
&api.PersistentVolumeClaim{},
// TODO: Can we have much longer period here?
syncPeriod,
framework.ResourceEventHandlerFuncs{
AddFunc: binder.addClaim,
UpdateFunc: binder.updateClaim,
DeleteFunc: binder.deleteClaim,
},
)
binder.claimController = claimController
binder.volumeController = volumeController
return binder
}
// NewExists creates a new namespace exists admission control handler
func NewExists(c clientset.Interface) admission.Interface {
store := cache.NewStore(cache.MetaNamespaceKeyFunc)
reflector := cache.NewReflector(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return c.Legacy().Namespaces().List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return c.Legacy().Namespaces().Watch(options)
},
},
&api.Namespace{},
store,
5*time.Minute,
)
reflector.Run()
return &exists{
client: c,
store: store,
Handler: admission.NewHandler(admission.Create, admission.Update, admission.Delete),
}
}
// finalize will finalize the namespace for kubernetes
func finalizeNamespaceFunc(kubeClient clientset.Interface, namespace *api.Namespace) (*api.Namespace, error) {
namespaceFinalize := api.Namespace{}
namespaceFinalize.ObjectMeta = namespace.ObjectMeta
namespaceFinalize.Spec = namespace.Spec
finalizerSet := sets.NewString()
for i := range namespace.Spec.Finalizers {
if namespace.Spec.Finalizers[i] != api.FinalizerKubernetes {
finalizerSet.Insert(string(namespace.Spec.Finalizers[i]))
}
}
namespaceFinalize.Spec.Finalizers = make([]api.FinalizerName, 0, len(finalizerSet))
for _, value := range finalizerSet.List() {
namespaceFinalize.Spec.Finalizers = append(namespaceFinalize.Spec.Finalizers, api.FinalizerName(value))
}
namespace, err := kubeClient.Legacy().Namespaces().Finalize(&namespaceFinalize)
if err != nil {
// it was removed already, so life is good
if errors.IsNotFound(err) {
return namespace, nil
}
}
return namespace, err
}
// NewServiceAccount returns an admission.Interface implementation which limits admission of Pod CREATE requests based on the pod's ServiceAccount:
// 1. If the pod does not specify a ServiceAccount, it sets the pod's ServiceAccount to "default"
// 2. It ensures the ServiceAccount referenced by the pod exists
// 3. If LimitSecretReferences is true, it rejects the pod if the pod references Secret objects which the pod's ServiceAccount does not reference
// 4. If the pod does not contain any ImagePullSecrets, the ImagePullSecrets of the service account are added.
// 5. If MountServiceAccountToken is true, it adds a VolumeMount with the pod's ServiceAccount's api token secret to containers
func NewServiceAccount(cl clientset.Interface) *serviceAccount {
serviceAccountsIndexer, serviceAccountsReflector := cache.NewNamespaceKeyedIndexerAndReflector(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return cl.Legacy().ServiceAccounts(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return cl.Legacy().ServiceAccounts(api.NamespaceAll).Watch(options)
},
},
&api.ServiceAccount{},
0,
)
tokenSelector := fields.SelectorFromSet(map[string]string{client.SecretType: string(api.SecretTypeServiceAccountToken)})
secretsIndexer, secretsReflector := cache.NewNamespaceKeyedIndexerAndReflector(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
options.FieldSelector = tokenSelector
return cl.Legacy().Secrets(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
options.FieldSelector = tokenSelector
return cl.Legacy().Secrets(api.NamespaceAll).Watch(options)
},
},
&api.Secret{},
0,
)
return &serviceAccount{
Handler: admission.NewHandler(admission.Create),
// TODO: enable this once we've swept secret usage to account for adding secret references to service accounts
LimitSecretReferences: false,
// Auto mount service account API token secrets
MountServiceAccountToken: true,
// Reject pod creation until a service account token is available
RequireAPIToken: true,
client: cl,
serviceAccounts: serviceAccountsIndexer,
serviceAccountsReflector: serviceAccountsReflector,
secrets: secretsIndexer,
secretsReflector: secretsReflector,
}
}
// NewDeploymentController creates a new DeploymentController.
func NewDeploymentController(client clientset.Interface, resyncPeriod controller.ResyncPeriodFunc) *DeploymentController {
eventBroadcaster := record.NewBroadcaster()
eventBroadcaster.StartLogging(glog.Infof)
// TODO: remove the wrapper when every clients have moved to use the clientset.
eventBroadcaster.StartRecordingToSink(&unversioned_legacy.EventSinkImpl{client.Legacy().Events("")})
dc := &DeploymentController{
client: client,
eventRecorder: eventBroadcaster.NewRecorder(api.EventSource{Component: "deployment-controller"}),
queue: workqueue.New(),
podExpectations: controller.NewControllerExpectations(),
rcExpectations: controller.NewControllerExpectations(),
}
dc.dStore.Store, dc.dController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return dc.client.Extensions().Deployments(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return dc.client.Extensions().Deployments(api.NamespaceAll).Watch(options)
},
},
&extensions.Deployment{},
FullDeploymentResyncPeriod,
framework.ResourceEventHandlerFuncs{
AddFunc: func(obj interface{}) {
d := obj.(*extensions.Deployment)
glog.V(4).Infof("Adding deployment %s", d.Name)
dc.enqueueDeployment(obj)
},
UpdateFunc: func(old, cur interface{}) {
oldD := old.(*extensions.Deployment)
glog.V(4).Infof("Updating deployment %s", oldD.Name)
// Resync on deployment object relist.
dc.enqueueDeployment(cur)
},
// This will enter the sync loop and no-op, because the deployment has been deleted from the store.
DeleteFunc: func(obj interface{}) {
d := obj.(*extensions.Deployment)
glog.V(4).Infof("Deleting deployment %s", d.Name)
dc.enqueueDeployment(obj)
},
},
)
dc.rcStore.Store, dc.rcController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return dc.client.Legacy().ReplicationControllers(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return dc.client.Legacy().ReplicationControllers(api.NamespaceAll).Watch(options)
},
},
&api.ReplicationController{},
resyncPeriod(),
framework.ResourceEventHandlerFuncs{
AddFunc: dc.addRC,
UpdateFunc: dc.updateRC,
DeleteFunc: dc.deleteRC,
},
)
dc.podStore.Store, dc.podController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return dc.client.Legacy().Pods(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return dc.client.Legacy().Pods(api.NamespaceAll).Watch(options)
},
},
&api.Pod{},
resyncPeriod(),
framework.ResourceEventHandlerFuncs{
// When pod updates (becomes ready), we need to enqueue deployment
UpdateFunc: dc.updatePod,
// When pod is deleted, we need to update deployment's expectations
DeleteFunc: dc.deletePod,
},
)
dc.syncHandler = dc.syncDeployment
dc.rcStoreSynced = dc.rcController.HasSynced
dc.podStoreSynced = dc.podController.HasSynced
return dc
}
// NewNamespaceController creates a new NamespaceController
func NewNamespaceController(kubeClient clientset.Interface, versions *unversioned.APIVersions, resyncPeriod time.Duration) *NamespaceController {
var controller *framework.Controller
_, controller = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return kubeClient.Legacy().Namespaces().List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return kubeClient.Legacy().Namespaces().Watch(options)
},
},
&api.Namespace{},
// TODO: Can we have much longer period here?
resyncPeriod,
framework.ResourceEventHandlerFuncs{
AddFunc: func(obj interface{}) {
namespace := obj.(*api.Namespace)
if err := syncNamespace(kubeClient, versions, namespace); err != nil {
if estimate, ok := err.(*contentRemainingError); ok {
go func() {
// Estimate is the aggregate total of TerminationGracePeriodSeconds, which defaults to 30s
// for pods. However, most processes will terminate faster - within a few seconds, probably
// with a peak within 5-10s. So this division is a heuristic that avoids waiting the full
// duration when in many cases things complete more quickly. The extra second added is to
// ensure we never wait 0 seconds.
t := estimate.Estimate/2 + 1
glog.V(4).Infof("Content remaining in namespace %s, waiting %d seconds", namespace.Name, t)
time.Sleep(time.Duration(t) * time.Second)
if err := controller.Requeue(namespace); err != nil {
utilruntime.HandleError(err)
}
}()
return
}
utilruntime.HandleError(err)
}
},
UpdateFunc: func(oldObj, newObj interface{}) {
namespace := newObj.(*api.Namespace)
if err := syncNamespace(kubeClient, versions, namespace); err != nil {
if estimate, ok := err.(*contentRemainingError); ok {
go func() {
t := estimate.Estimate/2 + 1
glog.V(4).Infof("Content remaining in namespace %s, waiting %d seconds", namespace.Name, t)
time.Sleep(time.Duration(t) * time.Second)
if err := controller.Requeue(namespace); err != nil {
utilruntime.HandleError(err)
}
}()
return
}
utilruntime.HandleError(err)
}
},
},
)
return &NamespaceController{
controller: controller,
}
}
// NewReplicaSetController creates a new ReplicaSetController.
func NewReplicaSetController(kubeClient clientset.Interface, resyncPeriod controller.ResyncPeriodFunc, burstReplicas int) *ReplicaSetController {
eventBroadcaster := record.NewBroadcaster()
eventBroadcaster.StartLogging(glog.Infof)
eventBroadcaster.StartRecordingToSink(&unversioned_legacy.EventSinkImpl{kubeClient.Legacy().Events("")})
rsc := &ReplicaSetController{
kubeClient: kubeClient,
podControl: controller.RealPodControl{
KubeClient: kubeClient,
Recorder: eventBroadcaster.NewRecorder(api.EventSource{Component: "replicaset-controller"}),
},
burstReplicas: burstReplicas,
expectations: controller.NewControllerExpectations(),
queue: workqueue.New(),
}
rsc.rsStore.Store, rsc.rsController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return rsc.kubeClient.Extensions().ReplicaSets(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return rsc.kubeClient.Extensions().ReplicaSets(api.NamespaceAll).Watch(options)
},
},
&extensions.ReplicaSet{},
// TODO: Can we have much longer period here?
FullControllerResyncPeriod,
framework.ResourceEventHandlerFuncs{
AddFunc: rsc.enqueueReplicaSet,
UpdateFunc: func(old, cur interface{}) {
// You might imagine that we only really need to enqueue the
// replica set when Spec changes, but it is safer to sync any
// time this function is triggered. That way a full informer
// resync can requeue any replica set that don't yet have pods
// but whose last attempts at creating a pod have failed (since
// we don't block on creation of pods) instead of those
// replica sets stalling indefinitely. Enqueueing every time
// does result in some spurious syncs (like when Status.Replica
// is updated and the watch notification from it retriggers
// this function), but in general extra resyncs shouldn't be
// that bad as ReplicaSets that haven't met expectations yet won't
// sync, and all the listing is done using local stores.
oldRS := old.(*extensions.ReplicaSet)
curRS := cur.(*extensions.ReplicaSet)
if oldRS.Status.Replicas != curRS.Status.Replicas {
glog.V(4).Infof("Observed updated replica count for ReplicaSet: %v, %d->%d", curRS.Name, oldRS.Status.Replicas, curRS.Status.Replicas)
}
rsc.enqueueReplicaSet(cur)
},
// This will enter the sync loop and no-op, because the replica set has been deleted from the store.
// Note that deleting a replica set immediately after scaling it to 0 will not work. The recommended
// way of achieving this is by performing a `stop` operation on the replica set.
DeleteFunc: rsc.enqueueReplicaSet,
},
)
rsc.podStore.Store, rsc.podController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return rsc.kubeClient.Legacy().Pods(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return rsc.kubeClient.Legacy().Pods(api.NamespaceAll).Watch(options)
},
},
&api.Pod{},
resyncPeriod(),
framework.ResourceEventHandlerFuncs{
AddFunc: rsc.addPod,
// This invokes the ReplicaSet for every pod change, eg: host assignment. Though this might seem like
// overkill the most frequent pod update is status, and the associated ReplicaSet will only list from
// local storage, so it should be ok.
UpdateFunc: rsc.updatePod,
DeleteFunc: rsc.deletePod,
},
)
rsc.syncHandler = rsc.syncReplicaSet
rsc.podStoreSynced = rsc.podController.HasSynced
return rsc
}
// NewNodeController returns a new node controller to sync instances from cloudprovider.
func NewNodeController(
cloud cloudprovider.Interface,
kubeClient clientset.Interface,
podEvictionTimeout time.Duration,
deletionEvictionLimiter util.RateLimiter,
terminationEvictionLimiter util.RateLimiter,
nodeMonitorGracePeriod time.Duration,
nodeStartupGracePeriod time.Duration,
nodeMonitorPeriod time.Duration,
clusterCIDR *net.IPNet,
allocateNodeCIDRs bool) *NodeController {
eventBroadcaster := record.NewBroadcaster()
recorder := eventBroadcaster.NewRecorder(api.EventSource{Component: "controllermanager"})
eventBroadcaster.StartLogging(glog.Infof)
if kubeClient != nil {
glog.Infof("Sending events to api server.")
eventBroadcaster.StartRecordingToSink(&unversioned_legacy.EventSinkImpl{kubeClient.Legacy().Events("")})
} else {
glog.Infof("No api server defined - no events will be sent to API server.")
}
if allocateNodeCIDRs && clusterCIDR == nil {
glog.Fatal("NodeController: Must specify clusterCIDR if allocateNodeCIDRs == true.")
}
evictorLock := sync.Mutex{}
nc := &NodeController{
cloud: cloud,
knownNodeSet: make(sets.String),
kubeClient: kubeClient,
recorder: recorder,
podEvictionTimeout: podEvictionTimeout,
maximumGracePeriod: 5 * time.Minute,
evictorLock: &evictorLock,
podEvictor: NewRateLimitedTimedQueue(deletionEvictionLimiter),
terminationEvictor: NewRateLimitedTimedQueue(terminationEvictionLimiter),
nodeStatusMap: make(map[string]nodeStatusData),
nodeMonitorGracePeriod: nodeMonitorGracePeriod,
nodeMonitorPeriod: nodeMonitorPeriod,
nodeStartupGracePeriod: nodeStartupGracePeriod,
lookupIP: net.LookupIP,
now: unversioned.Now,
clusterCIDR: clusterCIDR,
allocateNodeCIDRs: allocateNodeCIDRs,
forcefullyDeletePod: func(p *api.Pod) { forcefullyDeletePod(kubeClient, p) },
}
nc.podStore.Store, nc.podController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return nc.kubeClient.Legacy().Pods(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return nc.kubeClient.Legacy().Pods(api.NamespaceAll).Watch(options)
},
},
&api.Pod{},
controller.NoResyncPeriodFunc(),
framework.ResourceEventHandlerFuncs{
AddFunc: nc.maybeDeleteTerminatingPod,
UpdateFunc: func(_, obj interface{}) { nc.maybeDeleteTerminatingPod(obj) },
},
)
nc.nodeStore.Store, nc.nodeController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return nc.kubeClient.Legacy().Nodes().List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return nc.kubeClient.Legacy().Nodes().Watch(options)
},
},
&api.Node{},
controller.NoResyncPeriodFunc(),
framework.ResourceEventHandlerFuncs{},
)
nc.daemonSetStore.Store, nc.daemonSetController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return nc.kubeClient.Extensions().DaemonSets(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return nc.kubeClient.Extensions().DaemonSets(api.NamespaceAll).Watch(options)
},
},
&api.Node{},
controller.NoResyncPeriodFunc(),
framework.ResourceEventHandlerFuncs{},
)
return nc
}
// syncNamespace orchestrates deletion of a Namespace and its associated content.
func syncNamespace(kubeClient clientset.Interface, versions *unversioned.APIVersions, namespace *api.Namespace) error {
if namespace.DeletionTimestamp == nil {
return nil
}
// multiple controllers may edit a namespace during termination
// first get the latest state of the namespace before proceeding
// if the namespace was deleted already, don't do anything
namespace, err := kubeClient.Legacy().Namespaces().Get(namespace.Name)
if err != nil {
if errors.IsNotFound(err) {
return nil
}
return err
}
glog.V(4).Infof("Syncing namespace %s", namespace.Name)
// ensure that the status is up to date on the namespace
// if we get a not found error, we assume the namespace is truly gone
namespace, err = retryOnConflictError(kubeClient, namespace, updateNamespaceStatusFunc)
if err != nil {
if errors.IsNotFound(err) {
return nil
}
return err
}
// if the namespace is already finalized, delete it
if finalized(namespace) {
err = kubeClient.Legacy().Namespaces().Delete(namespace.Name, nil)
if err != nil && !errors.IsNotFound(err) {
return err
}
return nil
}
// there may still be content for us to remove
estimate, err := deleteAllContent(kubeClient, versions, namespace.Name, *namespace.DeletionTimestamp)
if err != nil {
return err
}
if estimate > 0 {
return &contentRemainingError{estimate}
}
// we have removed content, so mark it finalized by us
result, err := retryOnConflictError(kubeClient, namespace, finalizeNamespaceFunc)
if err != nil {
return err
}
// now check if all finalizers have reported that we delete now
if finalized(result) {
err = kubeClient.Legacy().Namespaces().Delete(namespace.Name, nil)
if err != nil && !errors.IsNotFound(err) {
return err
}
}
return nil
}
func NewDaemonSetsController(kubeClient clientset.Interface, resyncPeriod controller.ResyncPeriodFunc) *DaemonSetsController {
eventBroadcaster := record.NewBroadcaster()
eventBroadcaster.StartLogging(glog.Infof)
// TODO: remove the wrapper when every clients have moved to use the clientset.
eventBroadcaster.StartRecordingToSink(&unversioned_legacy.EventSinkImpl{kubeClient.Legacy().Events("")})
dsc := &DaemonSetsController{
kubeClient: kubeClient,
podControl: controller.RealPodControl{
KubeClient: kubeClient,
Recorder: eventBroadcaster.NewRecorder(api.EventSource{Component: "daemon-set"}),
},
burstReplicas: BurstReplicas,
expectations: controller.NewControllerExpectations(),
queue: workqueue.New(),
}
// Manage addition/update of daemon sets.
dsc.dsStore.Store, dsc.dsController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return dsc.kubeClient.Extensions().DaemonSets(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return dsc.kubeClient.Extensions().DaemonSets(api.NamespaceAll).Watch(options)
},
},
&extensions.DaemonSet{},
// TODO: Can we have much longer period here?
FullDaemonSetResyncPeriod,
framework.ResourceEventHandlerFuncs{
AddFunc: func(obj interface{}) {
ds := obj.(*extensions.DaemonSet)
glog.V(4).Infof("Adding daemon set %s", ds.Name)
dsc.enqueueDaemonSet(obj)
},
UpdateFunc: func(old, cur interface{}) {
oldDS := old.(*extensions.DaemonSet)
glog.V(4).Infof("Updating daemon set %s", oldDS.Name)
dsc.enqueueDaemonSet(cur)
},
DeleteFunc: func(obj interface{}) {
ds := obj.(*extensions.DaemonSet)
glog.V(4).Infof("Deleting daemon set %s", ds.Name)
dsc.enqueueDaemonSet(obj)
},
},
)
// Watch for creation/deletion of pods. The reason we watch is that we don't want a daemon set to create/delete
// more pods until all the effects (expectations) of a daemon set's create/delete have been observed.
dsc.podStore.Store, dsc.podController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return dsc.kubeClient.Legacy().Pods(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return dsc.kubeClient.Legacy().Pods(api.NamespaceAll).Watch(options)
},
},
&api.Pod{},
resyncPeriod(),
framework.ResourceEventHandlerFuncs{
AddFunc: dsc.addPod,
UpdateFunc: dsc.updatePod,
DeleteFunc: dsc.deletePod,
},
)
// Watch for new nodes or updates to nodes - daemon pods are launched on new nodes, and possibly when labels on nodes change,
dsc.nodeStore.Store, dsc.nodeController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return dsc.kubeClient.Legacy().Nodes().List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return dsc.kubeClient.Legacy().Nodes().Watch(options)
},
},
&api.Node{},
resyncPeriod(),
framework.ResourceEventHandlerFuncs{
AddFunc: dsc.addNode,
UpdateFunc: dsc.updateNode,
},
)
dsc.syncHandler = dsc.syncDaemonSet
dsc.podStoreSynced = dsc.podController.HasSynced
return dsc
}
// IncrementUsage updates the supplied ResourceQuotaStatus object based on the incoming operation
// Return true if the usage must be recorded prior to admitting the new resource
// Return an error if the operation should not pass admission control
func IncrementUsage(a admission.Attributes, status *api.ResourceQuotaStatus, client clientset.Interface) (bool, error) {
// on update, the only resource that can modify the value of a quota is pods
// so if your not a pod, we exit quickly
if a.GetOperation() == admission.Update && a.GetResource() != api.Resource("pods") {
return false, nil
}
var errs []error
dirty := true
set := map[api.ResourceName]bool{}
for k := range status.Hard {
set[k] = true
}
obj := a.GetObject()
// handle max counts for each kind of resource (pods, services, replicationControllers, etc.)
if a.GetOperation() == admission.Create {
resourceName := resourceToResourceName[a.GetResource()]
hard, hardFound := status.Hard[resourceName]
if hardFound {
used, usedFound := status.Used[resourceName]
if !usedFound {
return false, fmt.Errorf("quota usage stats are not yet known, unable to admit resource until an accurate count is completed.")
}
if used.Value() >= hard.Value() {
errs = append(errs, fmt.Errorf("limited to %s %s", hard.String(), resourceName))
dirty = false
} else {
status.Used[resourceName] = *resource.NewQuantity(used.Value()+int64(1), resource.DecimalSI)
}
}
}
if a.GetResource() == api.Resource("pods") {
for _, resourceName := range []api.ResourceName{api.ResourceMemory, api.ResourceCPU} {
// ignore tracking the resource if it's not in the quota document
if !set[resourceName] {
continue
}
hard, hardFound := status.Hard[resourceName]
if !hardFound {
continue
}
// if we do not yet know how much of the current resource is used, we cannot accept any request
used, usedFound := status.Used[resourceName]
if !usedFound {
return false, fmt.Errorf("unable to admit pod until quota usage stats are calculated.")
}
// the amount of resource being requested, or an error if it does not make a request that is tracked
pod := obj.(*api.Pod)
delta, err := resourcequotacontroller.PodRequests(pod, resourceName)
if err != nil {
return false, fmt.Errorf("%s is limited by quota, must make explicit request.", resourceName)
}
// if this operation is an update, we need to find the delta usage from the previous state
if a.GetOperation() == admission.Update {
oldPod, err := client.Legacy().Pods(a.GetNamespace()).Get(pod.Name)
if err != nil {
return false, err
}
// if the previous version of the resource made a resource request, we need to subtract the old request
// from the current to get the actual resource request delta. if the previous version of the pod
// made no request on the resource, then we get an err value. we ignore the err value, and delta
// will just be equal to the total resource request on the pod since there is nothing to subtract.
oldRequest, err := resourcequotacontroller.PodRequests(oldPod, resourceName)
if err == nil {
err = delta.Sub(*oldRequest)
if err != nil {
return false, err
}
}
}
newUsage := used.Copy()
newUsage.Add(*delta)
// make the most precise comparison possible
newUsageValue := newUsage.Value()
hardUsageValue := hard.Value()
if newUsageValue <= resource.MaxMilliValue && hardUsageValue <= resource.MaxMilliValue {
newUsageValue = newUsage.MilliValue()
hardUsageValue = hard.MilliValue()
}
if newUsageValue > hardUsageValue {
errs = append(errs, fmt.Errorf("%s quota is %s, current usage is %s, requesting %s.", resourceName, hard.String(), used.String(), delta.String()))
dirty = false
} else {
status.Used[resourceName] = *newUsage
}
}
}
return dirty, utilerrors.NewAggregate(errs)
}
func NewJobController(kubeClient clientset.Interface, resyncPeriod controller.ResyncPeriodFunc) *JobController {
eventBroadcaster := record.NewBroadcaster()
eventBroadcaster.StartLogging(glog.Infof)
// TODO: remove the wrapper when every clients have moved to use the clientset.
eventBroadcaster.StartRecordingToSink(&unversioned_legacy.EventSinkImpl{kubeClient.Legacy().Events("")})
jm := &JobController{
kubeClient: kubeClient,
podControl: controller.RealPodControl{
KubeClient: kubeClient,
Recorder: eventBroadcaster.NewRecorder(api.EventSource{Component: "job-controller"}),
},
expectations: controller.NewControllerExpectations(),
queue: workqueue.New(),
recorder: eventBroadcaster.NewRecorder(api.EventSource{Component: "job-controller"}),
}
jm.jobStore.Store, jm.jobController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return jm.kubeClient.Extensions().Jobs(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return jm.kubeClient.Extensions().Jobs(api.NamespaceAll).Watch(options)
},
},
&extensions.Job{},
// TODO: Can we have much longer period here?
replicationcontroller.FullControllerResyncPeriod,
framework.ResourceEventHandlerFuncs{
AddFunc: jm.enqueueController,
UpdateFunc: func(old, cur interface{}) {
if job := cur.(*extensions.Job); !isJobFinished(job) {
jm.enqueueController(job)
}
},
DeleteFunc: jm.enqueueController,
},
)
jm.podStore.Store, jm.podController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return jm.kubeClient.Legacy().Pods(api.NamespaceAll).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return jm.kubeClient.Legacy().Pods(api.NamespaceAll).Watch(options)
},
},
&api.Pod{},
resyncPeriod(),
framework.ResourceEventHandlerFuncs{
AddFunc: jm.addPod,
UpdateFunc: jm.updatePod,
DeleteFunc: jm.deletePod,
},
)
jm.updateHandler = jm.updateJobStatus
jm.syncHandler = jm.syncJob
jm.podStoreSynced = jm.podController.HasSynced
return jm
}
func deleteEvents(kubeClient clientset.Interface, ns string) error {
return kubeClient.Legacy().Events(ns).DeleteCollection(nil, api.ListOptions{})
}