func New(kubeClient client.Interface, resyncPeriod controller.ResyncPeriodFunc, threshold int) *GCController {
eventBroadcaster := record.NewBroadcaster()
eventBroadcaster.StartLogging(glog.Infof)
eventBroadcaster.StartRecordingToSink(kubeClient.Events(""))
gcc := &GCController{
kubeClient: kubeClient,
threshold: threshold,
deletePod: func(namespace, name string) error {
return kubeClient.Pods(namespace).Delete(name, api.NewDeleteOptions(0))
},
}
terminatedSelector := compileTerminatedPodSelector()
gcc.podStore.Store, gcc.podStoreSyncer = framework.NewInformer(
&cache.ListWatch{
ListFunc: func() (runtime.Object, error) {
return gcc.kubeClient.Pods(api.NamespaceAll).List(labels.Everything(), terminatedSelector)
},
WatchFunc: func(rv string) (watch.Interface, error) {
return gcc.kubeClient.Pods(api.NamespaceAll).Watch(labels.Everything(), terminatedSelector, rv)
},
},
&api.Pod{},
resyncPeriod(),
framework.ResourceEventHandlerFuncs{},
)
return gcc
}
func createPodConfigTester(mode PodConfigNotificationMode) (chan<- interface{}, <-chan kubelet.PodUpdate, *PodConfig) {
eventBroadcaster := record.NewBroadcaster()
config := NewPodConfig(mode, eventBroadcaster.NewRecorder(api.EventSource{Component: "kubelet"}))
channel := config.Channel(TestSource)
ch := config.Updates()
return channel, ch, config
}
func TestUnschedulableNodes(t *testing.T) {
etcdStorage, err := framework.NewEtcdStorage()
if err != nil {
t.Fatalf("Couldn't create etcd storage: %v", err)
}
expEtcdStorage, err := framework.NewExtensionsEtcdStorage(nil)
if err != nil {
t.Fatalf("unexpected error: %v", err)
}
storageDestinations := master.NewStorageDestinations()
storageDestinations.AddAPIGroup("", etcdStorage)
storageDestinations.AddAPIGroup("extensions", expEtcdStorage)
storageVersions := make(map[string]string)
storageVersions[""] = testapi.Default.Version()
storageVersions["extensions"] = testapi.Extensions.GroupAndVersion()
framework.DeleteAllEtcdKeys()
var m *master.Master
s := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
m.Handler.ServeHTTP(w, req)
}))
defer s.Close()
m = master.New(&master.Config{
StorageDestinations: storageDestinations,
KubeletClient: client.FakeKubeletClient{},
EnableCoreControllers: true,
EnableLogsSupport: false,
EnableUISupport: false,
EnableIndex: true,
APIPrefix: "/api",
Authorizer: apiserver.NewAlwaysAllowAuthorizer(),
AdmissionControl: admit.NewAlwaysAdmit(),
StorageVersions: storageVersions,
})
restClient := client.NewOrDie(&client.Config{Host: s.URL, Version: testapi.Default.Version()})
schedulerConfigFactory := factory.NewConfigFactory(restClient, nil)
schedulerConfig, err := schedulerConfigFactory.Create()
if err != nil {
t.Fatalf("Couldn't create scheduler config: %v", err)
}
eventBroadcaster := record.NewBroadcaster()
schedulerConfig.Recorder = eventBroadcaster.NewRecorder(api.EventSource{Component: "scheduler"})
eventBroadcaster.StartRecordingToSink(restClient.Events(""))
scheduler.New(schedulerConfig).Run()
defer close(schedulerConfig.StopEverything)
DoTestUnschedulableNodes(t, restClient, schedulerConfigFactory.NodeLister.Store)
}
func NewHorizontalController(client client.Interface, metricsClient metrics.MetricsClient) *HorizontalController {
broadcaster := record.NewBroadcaster()
broadcaster.StartRecordingToSink(client.Events(""))
recorder := broadcaster.NewRecorder(api.EventSource{Component: "horizontal-pod-autoscaler"})
return &HorizontalController{
client: client,
metricsClient: metricsClient,
eventRecorder: recorder,
}
}
func New(client client.Interface) *DeploymentController {
eventBroadcaster := record.NewBroadcaster()
eventBroadcaster.StartLogging(glog.Infof)
eventBroadcaster.StartRecordingToSink(client.Events(""))
return &DeploymentController{
client: client,
expClient: client.Extensions(),
eventRecorder: eventBroadcaster.NewRecorder(api.EventSource{Component: "deployment-controller"}),
}
}
// Run runs the specified SchedulerServer. This should never exit.
func (s *SchedulerServer) Run(_ []string) error {
if s.Kubeconfig == "" && s.Master == "" {
glog.Warningf("Neither --kubeconfig nor --master was specified. Using default API client. This might not work.")
}
// This creates a client, first loading any specified kubeconfig
// file, and then overriding the Master flag, if non-empty.
kubeconfig, err := clientcmd.NewNonInteractiveDeferredLoadingClientConfig(
&clientcmd.ClientConfigLoadingRules{ExplicitPath: s.Kubeconfig},
&clientcmd.ConfigOverrides{ClusterInfo: clientcmdapi.Cluster{Server: s.Master}}).ClientConfig()
if err != nil {
return err
}
kubeconfig.QPS = 50.0
kubeconfig.Burst = 100
kubeClient, err := client.New(kubeconfig)
if err != nil {
glog.Fatalf("Invalid API configuration: %v", err)
}
go func() {
mux := http.NewServeMux()
healthz.InstallHandler(mux)
if s.EnableProfiling {
mux.HandleFunc("/debug/pprof/", pprof.Index)
mux.HandleFunc("/debug/pprof/profile", pprof.Profile)
mux.HandleFunc("/debug/pprof/symbol", pprof.Symbol)
}
mux.Handle("/metrics", prometheus.Handler())
server := &http.Server{
Addr: net.JoinHostPort(s.Address.String(), strconv.Itoa(s.Port)),
Handler: mux,
}
glog.Fatal(server.ListenAndServe())
}()
configFactory := factory.NewConfigFactory(kubeClient, util.NewTokenBucketRateLimiter(s.BindPodsQPS, s.BindPodsBurst))
config, err := s.createConfig(configFactory)
if err != nil {
glog.Fatalf("Failed to create scheduler configuration: %v", err)
}
eventBroadcaster := record.NewBroadcaster()
config.Recorder = eventBroadcaster.NewRecorder(api.EventSource{Component: "scheduler"})
eventBroadcaster.StartLogging(glog.Infof)
eventBroadcaster.StartRecordingToSink(kubeClient.Events(""))
sched := scheduler.New(config)
sched.Run()
select {}
}
func (k *KubernetesScheduler) NewPluginConfig(terminate <-chan struct{}, mux *http.ServeMux,
podsWatcher *cache.ListWatch) *PluginConfig {
// Watch and queue pods that need scheduling.
updates := make(chan queue.Entry, k.schedcfg.UpdatesBacklog)
podUpdates := &podStoreAdapter{queue.NewHistorical(updates)}
reflector := cache.NewReflector(podsWatcher, &api.Pod{}, podUpdates, 0)
// lock that guards critial sections that involve transferring pods from
// the store (cache) to the scheduling queue; its purpose is to maintain
// an ordering (vs interleaving) of operations that's easier to reason about.
kapi := &k8smScheduler{internal: k}
q := newQueuer(podUpdates)
podDeleter := &deleter{
api: kapi,
qr: q,
}
eh := &errorHandler{
api: kapi,
backoff: backoff.New(k.schedcfg.InitialPodBackoff.Duration, k.schedcfg.MaxPodBackoff.Duration),
qr: q,
}
startLatch := make(chan struct{})
eventBroadcaster := record.NewBroadcaster()
runtime.On(startLatch, func() {
eventBroadcaster.StartRecordingToSink(k.client.Events(""))
reflector.Run() // TODO(jdef) should listen for termination
podDeleter.Run(updates, terminate)
q.Run(terminate)
q.installDebugHandlers(mux)
podtask.InstallDebugHandlers(k.taskRegistry, mux)
})
return &PluginConfig{
Config: &plugin.Config{
NodeLister: nil,
Algorithm: &kubeScheduler{
api: kapi,
podUpdates: podUpdates,
},
Binder: &binder{api: kapi},
NextPod: q.yield,
Error: eh.handleSchedulingError,
Recorder: eventBroadcaster.NewRecorder(api.EventSource{Component: "scheduler"}),
},
api: kapi,
client: k.client,
qr: q,
deleter: podDeleter,
starting: startLatch,
}
}
// New returns a new service controller to keep cloud provider service resources
// (like external load balancers) in sync with the registry.
func New(cloud cloudprovider.Interface, kubeClient client.Interface, clusterName string) *ServiceController {
broadcaster := record.NewBroadcaster()
broadcaster.StartRecordingToSink(kubeClient.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),
},
}
}
// NewLoadBalancerController creates a controller for gce loadbalancers.
// - kubeClient: A kubernetes REST client.
// - clusterManager: A ClusterManager capable of creating all cloud resources
// required for L7 loadbalancing.
// - resyncPeriod: Watchers relist from the Kubernetes API server this often.
func NewLoadBalancerController(kubeClient *client.Client, clusterManager *ClusterManager, resyncPeriod time.Duration, namespace string) (*loadBalancerController, error) {
eventBroadcaster := record.NewBroadcaster()
eventBroadcaster.StartLogging(glog.Infof)
eventBroadcaster.StartRecordingToSink(kubeClient.Events(""))
lbc := loadBalancerController{
client: kubeClient,
clusterManager: clusterManager,
stopCh: make(chan struct{}),
recorder: eventBroadcaster.NewRecorder(
api.EventSource{Component: "loadbalancer-controller"}),
}
lbc.nodeQueue = NewTaskQueue(lbc.syncNodes)
lbc.ingQueue = NewTaskQueue(lbc.sync)
// Ingress watch handlers
pathHandlers := framework.ResourceEventHandlerFuncs{
AddFunc: func(obj interface{}) {
addIng := obj.(*extensions.Ingress)
lbc.recorder.Eventf(addIng, "ADD", addIng.Name)
lbc.ingQueue.enqueue(obj)
},
DeleteFunc: lbc.ingQueue.enqueue,
UpdateFunc: func(old, cur interface{}) {
if !reflect.DeepEqual(old, cur) {
glog.V(3).Infof("Ingress %v changed, syncing",
cur.(*extensions.Ingress).Name)
}
lbc.ingQueue.enqueue(cur)
},
}
lbc.ingLister.Store, lbc.ingController = framework.NewInformer(
&cache.ListWatch{
ListFunc: ingressListFunc(lbc.client, namespace),
WatchFunc: ingressWatchFunc(lbc.client, namespace),
},
&extensions.Ingress{}, resyncPeriod, pathHandlers)
// Service watch handlers
svcHandlers := framework.ResourceEventHandlerFuncs{
AddFunc: lbc.enqueueIngressForService,
UpdateFunc: func(old, cur interface{}) {
if !reflect.DeepEqual(old, cur) {
lbc.enqueueIngressForService(cur)
}
},
// Ingress deletes matter, service deletes don't.
}
lbc.svcLister.Store, lbc.svcController = framework.NewInformer(
cache.NewListWatchFromClient(
lbc.client, "services", namespace, fields.Everything()),
&api.Service{}, resyncPeriod, svcHandlers)
nodeHandlers := framework.ResourceEventHandlerFuncs{
AddFunc: lbc.nodeQueue.enqueue,
DeleteFunc: lbc.nodeQueue.enqueue,
// Nodes are updated every 10s and we don't care, so no update handler.
}
// Node watch handlers
lbc.nodeLister.Store, lbc.nodeController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func() (runtime.Object, error) {
return lbc.client.Get().
Resource("nodes").
FieldsSelectorParam(fields.Everything()).
Do().
Get()
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return lbc.client.Get().
Prefix("watch").
Resource("nodes").
FieldsSelectorParam(fields.Everything()).
Param("resourceVersion", options.ResourceVersion).Watch()
},
},
&api.Node{}, 0, nodeHandlers)
lbc.tr = &gceTranslator{&lbc}
glog.Infof("Created new loadbalancer controller")
return &lbc, nil
}
func NewDaemonSetsController(kubeClient client.Interface, resyncPeriod controller.ResyncPeriodFunc) *DaemonSetsController {
eventBroadcaster := record.NewBroadcaster()
eventBroadcaster.StartLogging(glog.Infof)
eventBroadcaster.StartRecordingToSink(kubeClient.Events(""))
dsc := &DaemonSetsController{
kubeClient: kubeClient,
podControl: controller.RealPodControl{
KubeClient: kubeClient,
Recorder: eventBroadcaster.NewRecorder(api.EventSource{Component: "daemon-set"}),
},
expectations: controller.NewControllerExpectations(),
queue: workqueue.New(),
}
// Manage addition/update of daemon sets.
dsc.dsStore.Store, dsc.dsController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func() (runtime.Object, error) {
return dsc.kubeClient.Extensions().DaemonSets(api.NamespaceAll).List(labels.Everything())
},
WatchFunc: func(rv string) (watch.Interface, error) {
return dsc.kubeClient.Extensions().DaemonSets(api.NamespaceAll).Watch(labels.Everything(), fields.Everything(), rv)
},
},
&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() (runtime.Object, error) {
return dsc.kubeClient.Pods(api.NamespaceAll).List(labels.Everything(), fields.Everything())
},
WatchFunc: func(rv string) (watch.Interface, error) {
return dsc.kubeClient.Pods(api.NamespaceAll).Watch(labels.Everything(), fields.Everything(), rv)
},
},
&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() (runtime.Object, error) {
return dsc.kubeClient.Nodes().List(labels.Everything(), fields.Everything())
},
WatchFunc: func(rv string) (watch.Interface, error) {
return dsc.kubeClient.Nodes().Watch(labels.Everything(), fields.Everything(), rv)
},
},
&api.Node{},
resyncPeriod(),
framework.ResourceEventHandlerFuncs{
AddFunc: dsc.addNode,
UpdateFunc: dsc.updateNode,
},
)
dsc.syncHandler = dsc.syncDaemonSet
dsc.podStoreSynced = dsc.podController.HasSynced
return dsc
}
func startComponents(firstManifestURL, secondManifestURL string) (string, string) {
// Setup
servers := []string{}
glog.Infof("Creating etcd client pointing to %v", servers)
handler := delegateHandler{}
apiServer := httptest.NewServer(&handler)
etcdClient := etcd.NewClient(servers)
sleep := 4 * time.Second
ok := false
for i := 0; i < 3; i++ {
keys, err := etcdClient.Get("/", false, false)
if err != nil {
glog.Warningf("Unable to list root etcd keys: %v", err)
if i < 2 {
time.Sleep(sleep)
sleep = sleep * sleep
}
continue
}
for _, node := range keys.Node.Nodes {
if _, err := etcdClient.Delete(node.Key, true); err != nil {
glog.Fatalf("Unable delete key: %v", err)
}
}
ok = true
break
}
if !ok {
glog.Fatalf("Failed to connect to etcd")
}
cl := client.NewOrDie(&client.Config{Host: apiServer.URL, Version: testapi.Default.GroupAndVersion()})
// TODO: caesarxuchao: hacky way to specify version of Experimental client.
// We will fix this by supporting multiple group versions in Config
cl.ExtensionsClient = client.NewExtensionsOrDie(&client.Config{Host: apiServer.URL, Version: testapi.Extensions.GroupAndVersion()})
storageVersions := make(map[string]string)
etcdStorage, err := master.NewEtcdStorage(etcdClient, latest.GroupOrDie("").InterfacesFor, testapi.Default.GroupAndVersion(), etcdtest.PathPrefix())
storageVersions[""] = testapi.Default.GroupAndVersion()
if err != nil {
glog.Fatalf("Unable to get etcd storage: %v", err)
}
expEtcdStorage, err := master.NewEtcdStorage(etcdClient, latest.GroupOrDie("extensions").InterfacesFor, testapi.Extensions.GroupAndVersion(), etcdtest.PathPrefix())
storageVersions["extensions"] = testapi.Extensions.GroupAndVersion()
if err != nil {
glog.Fatalf("Unable to get etcd storage for experimental: %v", err)
}
storageDestinations := master.NewStorageDestinations()
storageDestinations.AddAPIGroup("", etcdStorage)
storageDestinations.AddAPIGroup("extensions", expEtcdStorage)
// Master
host, port, err := net.SplitHostPort(strings.TrimLeft(apiServer.URL, "http://"))
if err != nil {
glog.Fatalf("Unable to parse URL '%v': %v", apiServer.URL, err)
}
portNumber, err := strconv.Atoi(port)
if err != nil {
glog.Fatalf("Nonnumeric port? %v", err)
}
publicAddress := net.ParseIP(host)
if publicAddress == nil {
glog.Fatalf("no public address for %s", host)
}
// Create a master and install handlers into mux.
m := master.New(&master.Config{
StorageDestinations: storageDestinations,
KubeletClient: fakeKubeletClient{},
EnableCoreControllers: true,
EnableLogsSupport: false,
EnableProfiling: true,
APIPrefix: "/api",
APIGroupPrefix: "/apis",
Authorizer: apiserver.NewAlwaysAllowAuthorizer(),
AdmissionControl: admit.NewAlwaysAdmit(),
ReadWritePort: portNumber,
PublicAddress: publicAddress,
CacheTimeout: 2 * time.Second,
StorageVersions: storageVersions,
})
handler.delegate = m.Handler
// Scheduler
schedulerConfigFactory := factory.NewConfigFactory(cl, nil)
schedulerConfig, err := schedulerConfigFactory.Create()
if err != nil {
glog.Fatalf("Couldn't create scheduler config: %v", err)
}
eventBroadcaster := record.NewBroadcaster()
schedulerConfig.Recorder = eventBroadcaster.NewRecorder(api.EventSource{Component: "scheduler"})
eventBroadcaster.StartLogging(glog.Infof)
eventBroadcaster.StartRecordingToSink(cl.Events(""))
scheduler.New(schedulerConfig).Run()
// ensure the service endpoints are sync'd several times within the window that the integration tests wait
go endpointcontroller.NewEndpointController(cl, controller.NoResyncPeriodFunc).
Run(3, util.NeverStop)
// TODO: Write an integration test for the replication controllers watch.
go replicationControllerPkg.NewReplicationManager(cl, controller.NoResyncPeriodFunc, replicationControllerPkg.BurstReplicas).
Run(3, util.NeverStop)
nodeController := nodecontroller.NewNodeController(nil, cl, 5*time.Minute, util.NewFakeRateLimiter(), util.NewFakeRateLimiter(),
40*time.Second, 60*time.Second, 5*time.Second, nil, false)
nodeController.Run(5 * time.Second)
cadvisorInterface := new(cadvisor.Fake)
// Kubelet (localhost)
testRootDir := makeTempDirOrDie("kubelet_integ_1.", "")
configFilePath := makeTempDirOrDie("config", testRootDir)
glog.Infof("Using %s as root dir for kubelet #1", testRootDir)
fakeDocker1.VersionInfo = docker.Env{"ApiVersion=1.20"}
kcfg := kubeletapp.SimpleKubelet(
cl,
&fakeDocker1,
"localhost",
testRootDir,
firstManifestURL,
"127.0.0.1",
10250, /* KubeletPort */
0, /* ReadOnlyPort */
api.NamespaceDefault,
empty_dir.ProbeVolumePlugins(),
nil,
cadvisorInterface,
configFilePath,
nil,
kubecontainer.FakeOS{},
1*time.Second, /* FileCheckFrequency */
1*time.Second, /* HTTPCheckFrequency */
10*time.Second, /* MinimumGCAge */
3*time.Second, /* NodeStatusUpdateFrequency */
10*time.Second /* SyncFrequency */)
kubeletapp.RunKubelet(kcfg, nil)
// Kubelet (machine)
// Create a second kubelet so that the guestbook example's two redis slaves both
// have a place they can schedule.
testRootDir = makeTempDirOrDie("kubelet_integ_2.", "")
glog.Infof("Using %s as root dir for kubelet #2", testRootDir)
fakeDocker2.VersionInfo = docker.Env{"ApiVersion=1.20"}
kcfg = kubeletapp.SimpleKubelet(
cl,
&fakeDocker2,
"127.0.0.1",
testRootDir,
secondManifestURL,
"127.0.0.1",
10251, /* KubeletPort */
0, /* ReadOnlyPort */
api.NamespaceDefault,
empty_dir.ProbeVolumePlugins(),
nil,
cadvisorInterface,
"",
nil,
kubecontainer.FakeOS{},
1*time.Second, /* FileCheckFrequency */
1*time.Second, /* HTTPCheckFrequency */
10*time.Second, /* MinimumGCAge */
3*time.Second, /* NodeStatusUpdateFrequency */
10*time.Second /* SyncFrequency */)
kubeletapp.RunKubelet(kcfg, nil)
return apiServer.URL, configFilePath
}
// NewNodeController returns a new node controller to sync instances from cloudprovider.
func NewNodeController(
cloud cloudprovider.Interface,
kubeClient client.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(kubeClient.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() (runtime.Object, error) {
return nc.kubeClient.Pods(api.NamespaceAll).List(labels.Everything(), fields.Everything())
},
WatchFunc: func(rv string) (watch.Interface, error) {
return nc.kubeClient.Pods(api.NamespaceAll).Watch(labels.Everything(), fields.Everything(), rv)
},
},
&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() (runtime.Object, error) {
return nc.kubeClient.Nodes().List(labels.Everything(), fields.Everything())
},
WatchFunc: func(rv string) (watch.Interface, error) {
return nc.kubeClient.Nodes().Watch(labels.Everything(), fields.Everything(), rv)
},
},
&api.Node{},
controller.NoResyncPeriodFunc(),
framework.ResourceEventHandlerFuncs{},
)
return nc
}
// RunKubelet is responsible for setting up and running a kubelet. It is used in three different applications:
// 1 Integration tests
// 2 Kubelet binary
// 3 Standalone 'kubernetes' binary
// Eventually, #2 will be replaced with instances of #3
func RunKubelet(kcfg *KubeletConfig, builder KubeletBuilder) error {
kcfg.Hostname = nodeutil.GetHostname(kcfg.HostnameOverride)
if len(kcfg.NodeName) == 0 {
// Query the cloud provider for our node name, default to Hostname
nodeName := kcfg.Hostname
if kcfg.Cloud != nil {
var err error
instances, ok := kcfg.Cloud.Instances()
if !ok {
return fmt.Errorf("failed to get instances from cloud provider")
}
nodeName, err = instances.CurrentNodeName(kcfg.Hostname)
if err != nil {
return fmt.Errorf("error fetching current instance name from cloud provider: %v", err)
}
glog.V(2).Infof("cloud provider determined current node name to be %s", nodeName)
}
kcfg.NodeName = nodeName
}
eventBroadcaster := record.NewBroadcaster()
kcfg.Recorder = eventBroadcaster.NewRecorder(api.EventSource{Component: "kubelet", Host: kcfg.NodeName})
eventBroadcaster.StartLogging(glog.V(3).Infof)
if kcfg.KubeClient != nil {
glog.V(4).Infof("Sending events to api server.")
if kcfg.EventRecordQPS == 0.0 {
eventBroadcaster.StartRecordingToSink(kcfg.KubeClient.Events(""))
} else {
eventClient := *kcfg.KubeClient
eventClient.Throttle = util.NewTokenBucketRateLimiter(kcfg.EventRecordQPS, kcfg.EventBurst)
eventBroadcaster.StartRecordingToSink(eventClient.Events(""))
}
} else {
glog.Warning("No api server defined - no events will be sent to API server.")
}
privilegedSources := capabilities.PrivilegedSources{
HostNetworkSources: kcfg.HostNetworkSources,
HostPIDSources: kcfg.HostPIDSources,
HostIPCSources: kcfg.HostIPCSources,
}
capabilities.Setup(kcfg.AllowPrivileged, privilegedSources, 0)
credentialprovider.SetPreferredDockercfgPath(kcfg.RootDirectory)
if builder == nil {
builder = createAndInitKubelet
}
if kcfg.OSInterface == nil {
kcfg.OSInterface = kubecontainer.RealOS{}
}
k, podCfg, err := builder(kcfg)
if err != nil {
return fmt.Errorf("failed to create kubelet: %v", err)
}
util.ApplyRLimitForSelf(kcfg.MaxOpenFiles)
// process pods and exit.
if kcfg.Runonce {
if _, err := k.RunOnce(podCfg.Updates()); err != nil {
return fmt.Errorf("runonce failed: %v", err)
}
glog.Infof("Started kubelet as runonce")
} else {
startKubelet(k, podCfg, kcfg)
glog.Infof("Started kubelet")
}
return nil
}
func NewJobController(kubeClient client.Interface, resyncPeriod controller.ResyncPeriodFunc) *JobController {
eventBroadcaster := record.NewBroadcaster()
eventBroadcaster.StartLogging(glog.Infof)
eventBroadcaster.StartRecordingToSink(kubeClient.Events(""))
jm := &JobController{
kubeClient: kubeClient,
podControl: controller.RealPodControl{
KubeClient: kubeClient,
Recorder: eventBroadcaster.NewRecorder(api.EventSource{Component: "job"}),
},
expectations: controller.NewControllerExpectations(),
queue: workqueue.New(),
}
jm.jobStore.Store, jm.jobController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func() (runtime.Object, error) {
return jm.kubeClient.Extensions().Jobs(api.NamespaceAll).List(labels.Everything(), fields.Everything())
},
WatchFunc: func(rv string) (watch.Interface, error) {
return jm.kubeClient.Extensions().Jobs(api.NamespaceAll).Watch(labels.Everything(), fields.Everything(), rv)
},
},
&extensions.Job{},
// TODO: Can we have much longer period here?
replicationcontroller.FullControllerResyncPeriod,
framework.ResourceEventHandlerFuncs{
AddFunc: jm.enqueueController,
UpdateFunc: func(old, cur interface{}) {
job := cur.(*extensions.Job)
for _, c := range job.Status.Conditions {
if c.Type == extensions.JobComplete && c.Status == api.ConditionTrue {
return
}
}
jm.enqueueController(cur)
},
DeleteFunc: jm.enqueueController,
},
)
jm.podStore.Store, jm.podController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func() (runtime.Object, error) {
return jm.kubeClient.Pods(api.NamespaceAll).List(labels.Everything(), fields.Everything())
},
WatchFunc: func(rv string) (watch.Interface, error) {
return jm.kubeClient.Pods(api.NamespaceAll).Watch(labels.Everything(), fields.Everything(), rv)
},
},
&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
}
// NewReplicationManager creates a new ReplicationManager.
func NewReplicationManager(kubeClient client.Interface, resyncPeriod controller.ResyncPeriodFunc, burstReplicas int) *ReplicationManager {
eventBroadcaster := record.NewBroadcaster()
eventBroadcaster.StartLogging(glog.Infof)
eventBroadcaster.StartRecordingToSink(kubeClient.Events(""))
rm := &ReplicationManager{
kubeClient: kubeClient,
podControl: controller.RealPodControl{
KubeClient: kubeClient,
Recorder: eventBroadcaster.NewRecorder(api.EventSource{Component: "replication-controller"}),
},
burstReplicas: burstReplicas,
expectations: controller.NewControllerExpectations(),
queue: workqueue.New(),
}
rm.rcStore.Store, rm.rcController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func() (runtime.Object, error) {
return rm.kubeClient.ReplicationControllers(api.NamespaceAll).List(labels.Everything())
},
WatchFunc: func(rv string) (watch.Interface, error) {
return rm.kubeClient.ReplicationControllers(api.NamespaceAll).Watch(labels.Everything(), fields.Everything(), rv)
},
},
&api.ReplicationController{},
// TODO: Can we have much longer period here?
FullControllerResyncPeriod,
framework.ResourceEventHandlerFuncs{
AddFunc: rm.enqueueController,
UpdateFunc: func(old, cur interface{}) {
// You might imagine that we only really need to enqueue the
// controller when Spec changes, but it is safer to sync any
// time this function is triggered. That way a full informer
// resync can requeue any controllers 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
// controllers 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 rcs that haven't met expectations yet won't
// sync, and all the listing is done using local stores.
oldRC := old.(*api.ReplicationController)
curRC := cur.(*api.ReplicationController)
if oldRC.Status.Replicas != curRC.Status.Replicas {
glog.V(4).Infof("Observed updated replica count for rc: %v, %d->%d", curRC.Name, oldRC.Status.Replicas, curRC.Status.Replicas)
}
rm.enqueueController(cur)
},
// This will enter the sync loop and no-op, because the controller has been deleted from the store.
// Note that deleting a controller immediately after scaling it to 0 will not work. The recommended
// way of achieving this is by performing a `stop` operation on the controller.
DeleteFunc: rm.enqueueController,
},
)
rm.podStore.Store, rm.podController = framework.NewInformer(
&cache.ListWatch{
ListFunc: func() (runtime.Object, error) {
return rm.kubeClient.Pods(api.NamespaceAll).List(labels.Everything(), fields.Everything())
},
WatchFunc: func(rv string) (watch.Interface, error) {
return rm.kubeClient.Pods(api.NamespaceAll).Watch(labels.Everything(), fields.Everything(), rv)
},
},
&api.Pod{},
resyncPeriod(),
framework.ResourceEventHandlerFuncs{
AddFunc: rm.addPod,
// This invokes the rc for every pod change, eg: host assignment. Though this might seem like overkill
// the most frequent pod update is status, and the associated rc will only list from local storage, so
// it should be ok.
UpdateFunc: rm.updatePod,
DeleteFunc: rm.deletePod,
},
)
rm.syncHandler = rm.syncReplicationController
rm.podStoreSynced = rm.podController.HasSynced
return rm
}
func TestScheduler(t *testing.T) {
eventBroadcaster := record.NewBroadcaster()
defer eventBroadcaster.StartLogging(t.Logf).Stop()
errS := errors.New("scheduler")
errB := errors.New("binder")
table := []struct {
injectBindError error
sendPod *api.Pod
algo algorithm.ScheduleAlgorithm
expectErrorPod *api.Pod
expectAssumedPod *api.Pod
expectError error
expectBind *api.Binding
eventReason string
}{
{
sendPod: podWithID("foo", ""),
algo: mockScheduler{"machine1", nil},
expectBind: &api.Binding{ObjectMeta: api.ObjectMeta{Name: "foo"}, Target: api.ObjectReference{Kind: "Node", Name: "machine1"}},
expectAssumedPod: podWithID("foo", "machine1"),
eventReason: "Scheduled",
}, {
sendPod: podWithID("foo", ""),
algo: mockScheduler{"machine1", errS},
expectError: errS,
expectErrorPod: podWithID("foo", ""),
eventReason: "FailedScheduling",
}, {
sendPod: podWithID("foo", ""),
algo: mockScheduler{"machine1", nil},
expectBind: &api.Binding{ObjectMeta: api.ObjectMeta{Name: "foo"}, Target: api.ObjectReference{Kind: "Node", Name: "machine1"}},
injectBindError: errB,
expectError: errB,
expectErrorPod: podWithID("foo", ""),
eventReason: "FailedScheduling",
},
}
for i, item := range table {
var gotError error
var gotPod *api.Pod
var gotAssumedPod *api.Pod
var gotBinding *api.Binding
c := &Config{
Modeler: &FakeModeler{
AssumePodFunc: func(pod *api.Pod) {
gotAssumedPod = pod
},
},
NodeLister: algorithm.FakeNodeLister(
api.NodeList{Items: []api.Node{{ObjectMeta: api.ObjectMeta{Name: "machine1"}}}},
),
Algorithm: item.algo,
Binder: fakeBinder{func(b *api.Binding) error {
gotBinding = b
return item.injectBindError
}},
Error: func(p *api.Pod, err error) {
gotPod = p
gotError = err
},
NextPod: func() *api.Pod {
return item.sendPod
},
Recorder: eventBroadcaster.NewRecorder(api.EventSource{Component: "scheduler"}),
}
s := New(c)
called := make(chan struct{})
events := eventBroadcaster.StartEventWatcher(func(e *api.Event) {
if e, a := item.eventReason, e.Reason; e != a {
t.Errorf("%v: expected %v, got %v", i, e, a)
}
close(called)
})
s.scheduleOne()
if e, a := item.expectAssumedPod, gotAssumedPod; !reflect.DeepEqual(e, a) {
t.Errorf("%v: assumed pod: wanted %v, got %v", i, e, a)
}
if e, a := item.expectErrorPod, gotPod; !reflect.DeepEqual(e, a) {
t.Errorf("%v: error pod: wanted %v, got %v", i, e, a)
}
if e, a := item.expectError, gotError; !reflect.DeepEqual(e, a) {
t.Errorf("%v: error: wanted %v, got %v", i, e, a)
}
if e, a := item.expectBind, gotBinding; !reflect.DeepEqual(e, a) {
t.Errorf("%v: error: %s", i, util.ObjectDiff(e, a))
}
<-called
events.Stop()
}
}
func TestSchedulerForgetAssumedPodAfterDelete(t *testing.T) {
eventBroadcaster := record.NewBroadcaster()
defer eventBroadcaster.StartLogging(t.Logf).Stop()
// Setup modeler so we control the contents of all 3 stores: assumed,
// scheduled and queued
scheduledPodStore := cache.NewStore(cache.MetaNamespaceKeyFunc)
scheduledPodLister := &cache.StoreToPodLister{Store: scheduledPodStore}
queuedPodStore := cache.NewFIFO(cache.MetaNamespaceKeyFunc)
queuedPodLister := &cache.StoreToPodLister{Store: queuedPodStore}
modeler := NewSimpleModeler(queuedPodLister, scheduledPodLister)
// Create a fake clock used to timestamp entries and calculate ttl. Nothing
// will expire till we flip to something older than the ttl, at which point
// all entries inserted with fakeTime will expire.
ttl := 30 * time.Second
fakeTime := time.Date(2009, time.November, 10, 23, 0, 0, 0, time.UTC)
fakeClock := &util.FakeClock{Time: fakeTime}
ttlPolicy := &cache.TTLPolicy{Ttl: ttl, Clock: fakeClock}
assumedPodsStore := cache.NewFakeExpirationStore(
cache.MetaNamespaceKeyFunc, nil, ttlPolicy, fakeClock)
modeler.assumedPods = &cache.StoreToPodLister{Store: assumedPodsStore}
// Port is the easiest way to cause a fit predicate failure
podPort := 8080
firstPod := podWithPort("foo", "", podPort)
// Create the scheduler config
algo := NewGenericScheduler(
map[string]algorithm.FitPredicate{"PodFitsHostPorts": predicates.PodFitsHostPorts},
[]algorithm.PriorityConfig{},
modeler.PodLister(),
rand.New(rand.NewSource(time.Now().UnixNano())))
var gotBinding *api.Binding
c := &Config{
Modeler: modeler,
NodeLister: algorithm.FakeNodeLister(
api.NodeList{Items: []api.Node{{ObjectMeta: api.ObjectMeta{Name: "machine1"}}}},
),
Algorithm: algo,
Binder: fakeBinder{func(b *api.Binding) error {
scheduledPodStore.Add(podWithPort(b.Name, b.Target.Name, podPort))
gotBinding = b
return nil
}},
NextPod: func() *api.Pod {
return queuedPodStore.Pop().(*api.Pod)
},
Error: func(p *api.Pod, err error) {
t.Errorf("Unexpected error when scheduling pod %+v: %v", p, err)
},
Recorder: eventBroadcaster.NewRecorder(api.EventSource{Component: "scheduler"}),
}
// First scheduling pass should schedule the pod
s := New(c)
called := make(chan struct{})
events := eventBroadcaster.StartEventWatcher(func(e *api.Event) {
if e, a := "Scheduled", e.Reason; e != a {
t.Errorf("expected %v, got %v", e, a)
}
close(called)
})
queuedPodStore.Add(firstPod)
// queuedPodStore: [foo:8080]
// scheduledPodStore: []
// assumedPods: []
s.scheduleOne()
// queuedPodStore: []
// scheduledPodStore: [foo:8080]
// assumedPods: [foo:8080]
pod, exists, _ := scheduledPodStore.GetByKey("foo")
if !exists {
t.Errorf("Expected scheduled pod store to contain pod")
}
pod, exists, _ = queuedPodStore.GetByKey("foo")
if exists {
t.Errorf("Did not expect a queued pod, found %+v", pod)
}
pod, exists, _ = assumedPodsStore.GetByKey("foo")
if !exists {
t.Errorf("Assumed pod store should contain stale pod")
}
expectBind := &api.Binding{
ObjectMeta: api.ObjectMeta{Name: "foo"},
Target: api.ObjectReference{Kind: "Node", Name: "machine1"},
}
if ex, ac := expectBind, gotBinding; !reflect.DeepEqual(ex, ac) {
t.Errorf("Expected exact match on binding: %s", util.ObjectDiff(ex, ac))
}
<-called
events.Stop()
scheduledPodStore.Delete(pod)
_, exists, _ = assumedPodsStore.Get(pod)
if !exists {
t.Errorf("Expected pod %#v in assumed pod store", pod)
}
secondPod := podWithPort("bar", "", podPort)
queuedPodStore.Add(secondPod)
// queuedPodStore: [bar:8080]
// scheduledPodStore: []
// assumedPods: [foo:8080]
// Second scheduling pass will fail to schedule if the store hasn't expired
// the deleted pod. This would normally happen with a timeout.
//expirationPolicy.NeverExpire = util.NewStringSet()
fakeClock.Time = fakeClock.Time.Add(ttl + 1)
called = make(chan struct{})
events = eventBroadcaster.StartEventWatcher(func(e *api.Event) {
if e, a := "Scheduled", e.Reason; e != a {
t.Errorf("expected %v, got %v", e, a)
}
close(called)
})
s.scheduleOne()
expectBind = &api.Binding{
ObjectMeta: api.ObjectMeta{Name: "bar"},
Target: api.ObjectReference{Kind: "Node", Name: "machine1"},
}
if ex, ac := expectBind, gotBinding; !reflect.DeepEqual(ex, ac) {
t.Errorf("Expected exact match on binding: %s", util.ObjectDiff(ex, ac))
}
<-called
events.Stop()
}
// NewProxyServerDefault creates a new ProxyServer object with default parameters.
func NewProxyServerDefault(config *ProxyServerConfig) (*ProxyServer, error) {
protocol := utiliptables.ProtocolIpv4
if config.BindAddress.To4() == nil {
protocol = utiliptables.ProtocolIpv6
}
// We ommit creation of pretty much everything if we run in cleanup mode
if config.CleanupAndExit {
execer := exec.New()
dbus := utildbus.New()
IptInterface := utiliptables.New(execer, dbus, protocol)
return &ProxyServer{
Config: config,
IptInterface: IptInterface,
}, nil
}
// TODO(vmarmol): Use container config for this.
var oomAdjuster *oom.OomAdjuster
if config.OOMScoreAdj != 0 {
oomAdjuster := oom.NewOomAdjuster()
if err := oomAdjuster.ApplyOomScoreAdj(0, config.OOMScoreAdj); err != nil {
glog.V(2).Info(err)
}
}
if config.ResourceContainer != "" {
// Run in its own container.
if err := util.RunInResourceContainer(config.ResourceContainer); err != nil {
glog.Warningf("Failed to start in resource-only container %q: %v", config.ResourceContainer, err)
} else {
glog.V(2).Infof("Running in resource-only container %q", config.ResourceContainer)
}
}
// Create a Kube Client
// define api config source
if config.Kubeconfig == "" && config.Master == "" {
glog.Warningf("Neither --kubeconfig nor --master was specified. Using default API client. This might not work.")
}
// This creates a client, first loading any specified kubeconfig
// file, and then overriding the Master flag, if non-empty.
kubeconfig, err := clientcmd.NewNonInteractiveDeferredLoadingClientConfig(
&clientcmd.ClientConfigLoadingRules{ExplicitPath: config.Kubeconfig},
&clientcmd.ConfigOverrides{ClusterInfo: clientcmdapi.Cluster{Server: config.Master}}).ClientConfig()
if err != nil {
return nil, err
}
client, err := kubeclient.New(kubeconfig)
if err != nil {
glog.Fatalf("Invalid API configuration: %v", err)
}
// Create event recorder
hostname := nodeutil.GetHostname(config.HostnameOverride)
eventBroadcaster := record.NewBroadcaster()
recorder := eventBroadcaster.NewRecorder(api.EventSource{Component: "kube-proxy", Host: hostname})
eventBroadcaster.StartRecordingToSink(client.Events(""))
// Create a iptables utils.
execer := exec.New()
dbus := utildbus.New()
iptInterface := utiliptables.New(execer, dbus, protocol)
var proxier proxy.ProxyProvider
var endpointsHandler proxyconfig.EndpointsConfigHandler
useIptablesProxy := false
if mayTryIptablesProxy(config.ProxyMode, client.Nodes(), hostname) {
var err error
// guaranteed false on error, error only necessary for debugging
useIptablesProxy, err = iptables.ShouldUseIptablesProxier()
if err != nil {
glog.Errorf("Can't determine whether to use iptables proxy, using userspace proxier: %v", err)
}
}
if useIptablesProxy {
glog.V(2).Info("Using iptables Proxier.")
execer := exec.New()
proxierIptables, err := iptables.NewProxier(iptInterface, execer, config.SyncPeriod, config.MasqueradeAll)
if err != nil {
glog.Fatalf("Unable to create proxier: %v", err)
}
proxier = proxierIptables
endpointsHandler = proxierIptables
// No turning back. Remove artifacts that might still exist from the userspace Proxier.
glog.V(2).Info("Tearing down userspace rules. Errors here are acceptable.")
userspace.CleanupLeftovers(iptInterface)
} else {
glog.V(2).Info("Using userspace Proxier.")
// This is a proxy.LoadBalancer which NewProxier needs but has methods we don't need for
// our config.EndpointsConfigHandler.
loadBalancer := userspace.NewLoadBalancerRR()
// set EndpointsConfigHandler to our loadBalancer
endpointsHandler = loadBalancer
proxierUserspace, err := userspace.NewProxier(loadBalancer, config.BindAddress, iptInterface, config.PortRange, config.SyncPeriod, config.UDPIdleTimeout)
if err != nil {
glog.Fatalf("Unable to create proxier: %v", err)
}
proxier = proxierUserspace
// Remove artifacts from the pure-iptables Proxier.
glog.V(2).Info("Tearing down pure-iptables proxy rules. Errors here are acceptable.")
iptables.CleanupLeftovers(iptInterface)
}
iptInterface.AddReloadFunc(proxier.Sync)
// Create configs (i.e. Watches for Services and Endpoints)
// Note: RegisterHandler() calls need to happen before creation of Sources because sources
// only notify on changes, and the initial update (on process start) may be lost if no handlers
// are registered yet.
serviceConfig := proxyconfig.NewServiceConfig()
serviceConfig.RegisterHandler(proxier)
endpointsConfig := proxyconfig.NewEndpointsConfig()
endpointsConfig.RegisterHandler(endpointsHandler)
proxyconfig.NewSourceAPI(
client,
30*time.Second,
serviceConfig.Channel("api"),
endpointsConfig.Channel("api"),
)
config.nodeRef = &api.ObjectReference{
Kind: "Node",
Name: hostname,
UID: types.UID(hostname),
Namespace: "",
}
return NewProxyServer(config, client, endpointsConfig, endpointsHandler, iptInterface, oomAdjuster, proxier, recorder, serviceConfig)
}