func createStreams(req *http.Request, w http.ResponseWriter, supportedStreamProtocols []string, idleTimeout, streamCreationTimeout time.Duration) (*context, bool) {
opts, err := newOptions(req)
if err != nil {
runtime.HandleError(err)
w.WriteHeader(http.StatusBadRequest)
fmt.Fprint(w, err.Error())
return nil, false
}
if wsstream.IsWebSocketRequest(req) {
return createWebSocketStreams(req, w, opts, idleTimeout)
}
protocol, err := httpstream.Handshake(req, w, supportedStreamProtocols)
if err != nil {
w.WriteHeader(http.StatusBadRequest)
fmt.Fprint(w, err.Error())
return nil, false
}
streamCh := make(chan streamAndReply)
upgrader := spdy.NewResponseUpgrader()
conn := upgrader.UpgradeResponse(w, req, func(stream httpstream.Stream, replySent <-chan struct{}) error {
streamCh <- streamAndReply{Stream: stream, replySent: replySent}
return nil
})
// from this point on, we can no longer call methods on response
if conn == nil {
// The upgrader is responsible for notifying the client of any errors that
// occurred during upgrading. All we can do is return here at this point
// if we weren't successful in upgrading.
return nil, false
}
conn.SetIdleTimeout(idleTimeout)
var handler protocolHandler
switch protocol {
case StreamProtocolV2Name:
handler = &v2ProtocolHandler{}
case "":
glog.V(4).Infof("Client did not request protocol negotiaion. Falling back to %q", StreamProtocolV1Name)
fallthrough
case StreamProtocolV1Name:
handler = &v1ProtocolHandler{}
}
expired := time.NewTimer(streamCreationTimeout)
ctx, err := handler.waitForStreams(streamCh, opts.expectedStreams, expired.C)
if err != nil {
runtime.HandleError(err)
return nil, false
}
ctx.conn = conn
ctx.tty = opts.tty
return ctx, true
}
func (w *etcdWatcher) decodeObject(node *etcd.Node) (runtime.Object, error) {
if obj, found := w.cache.getFromCache(node.ModifiedIndex, storage.Everything); found {
return obj, nil
}
obj, err := runtime.Decode(w.encoding, []byte(node.Value))
if err != nil {
return nil, err
}
// ensure resource version is set on the object we load from etcd
if err := w.versioner.UpdateObject(obj, node.ModifiedIndex); err != nil {
utilruntime.HandleError(fmt.Errorf("failure to version api object (%d) %#v: %v", node.ModifiedIndex, obj, err))
}
// perform any necessary transformation
if w.transform != nil {
obj, err = w.transform(obj)
if err != nil {
utilruntime.HandleError(fmt.Errorf("failure to transform api object %#v: %v", obj, err))
return nil, err
}
}
if node.ModifiedIndex != 0 {
w.cache.addToCache(node.ModifiedIndex, obj)
}
return obj, nil
}
func (w *etcdWatcher) sendDelete(res *etcd.Response) {
if res.PrevNode == nil {
utilruntime.HandleError(fmt.Errorf("unexpected nil prev node: %#v", res))
return
}
if w.include != nil && !w.include(res.PrevNode.Key) {
return
}
node := *res.PrevNode
if res.Node != nil {
// Note that this sends the *old* object with the etcd index for the time at
// which it gets deleted. This will allow users to restart the watch at the right
// index.
node.ModifiedIndex = res.Node.ModifiedIndex
}
obj, err := w.decodeObject(&node)
if err != nil {
utilruntime.HandleError(fmt.Errorf("failure to decode api object: %v\nfrom %#v %#v", err, res, res.Node))
// TODO: expose an error through watch.Interface?
// Ignore this value. If we stop the watch on a bad value, a client that uses
// the resourceVersion to resume will never be able to get past a bad value.
return
}
if !w.filter(obj) {
return
}
w.emit(watch.Event{
Type: watch.Deleted,
Object: obj,
})
}
func (w *etcdWatcher) sendAdd(res *etcd.Response) {
if res.Node == nil {
utilruntime.HandleError(fmt.Errorf("unexpected nil node: %#v", res))
return
}
if w.include != nil && !w.include(res.Node.Key) {
return
}
obj, err := w.decodeObject(res.Node)
if err != nil {
utilruntime.HandleError(fmt.Errorf("failure to decode api object: %v\n'%v' from %#v %#v", err, string(res.Node.Value), res, res.Node))
// TODO: expose an error through watch.Interface?
// Ignore this value. If we stop the watch on a bad value, a client that uses
// the resourceVersion to resume will never be able to get past a bad value.
return
}
if !w.filter(obj) {
return
}
action := watch.Added
if res.Node.ModifiedIndex != res.Node.CreatedIndex {
action = watch.Modified
}
w.emit(watch.Event{
Type: action,
Object: obj,
})
}
// Loop infinitely, processing all service updates provided by the queue.
func (s *ServiceController) watchServices(serviceQueue *cache.DeltaFIFO) {
for {
serviceQueue.Pop(func(obj interface{}) error {
deltas, ok := obj.(cache.Deltas)
if !ok {
runtime.HandleError(fmt.Errorf("Received object from service watcher that wasn't Deltas: %+v", obj))
return nil
}
delta := deltas.Newest()
if delta == nil {
runtime.HandleError(fmt.Errorf("Received nil delta from watcher queue."))
return nil
}
err, retryDelay := s.processDelta(delta)
if retryDelay != 0 {
// Add the failed service back to the queue so we'll retry it.
runtime.HandleError(fmt.Errorf("Failed to process service delta. Retrying in %s: %v", retryDelay, err))
go func(deltas cache.Deltas, delay time.Duration) {
time.Sleep(delay)
if err := serviceQueue.AddIfNotPresent(deltas); err != nil {
runtime.HandleError(fmt.Errorf("Error requeuing service delta - will not retry: %v", err))
}
}(deltas, retryDelay)
} else if err != nil {
runtime.HandleError(fmt.Errorf("Failed to process service delta. Not retrying: %v", err))
}
return nil
})
}
}
// UpgradeResponse upgrades an HTTP response to one that supports multiplexed
// streams. newStreamHandler will be called synchronously whenever the
// other end of the upgraded connection creates a new stream.
func (u responseUpgrader) UpgradeResponse(w http.ResponseWriter, req *http.Request, newStreamHandler httpstream.NewStreamHandler) httpstream.Connection {
connectionHeader := strings.ToLower(req.Header.Get(httpstream.HeaderConnection))
upgradeHeader := strings.ToLower(req.Header.Get(httpstream.HeaderUpgrade))
if !strings.Contains(connectionHeader, strings.ToLower(httpstream.HeaderUpgrade)) || !strings.Contains(upgradeHeader, strings.ToLower(HeaderSpdy31)) {
w.WriteHeader(http.StatusBadRequest)
fmt.Fprintf(w, "unable to upgrade: missing upgrade headers in request: %#v", req.Header)
return nil
}
hijacker, ok := w.(http.Hijacker)
if !ok {
w.WriteHeader(http.StatusInternalServerError)
fmt.Fprintf(w, "unable to upgrade: unable to hijack response")
return nil
}
w.Header().Add(httpstream.HeaderConnection, httpstream.HeaderUpgrade)
w.Header().Add(httpstream.HeaderUpgrade, HeaderSpdy31)
w.WriteHeader(http.StatusSwitchingProtocols)
conn, _, err := hijacker.Hijack()
if err != nil {
runtime.HandleError(fmt.Errorf("unable to upgrade: error hijacking response: %v", err))
return nil
}
spdyConn, err := NewServerConnection(conn, newStreamHandler)
if err != nil {
runtime.HandleError(fmt.Errorf("unable to upgrade: error creating SPDY server connection: %v", err))
return nil
}
return spdyConn
}
// watchHandler watches w and keeps *resourceVersion up to date.
func (r *Reflector) watchHandler(w watch.Interface, resourceVersion *string, errc chan error, stopCh <-chan struct{}) error {
start := time.Now()
eventCount := 0
// Stopping the watcher should be idempotent and if we return from this function there's no way
// we're coming back in with the same watch interface.
defer w.Stop()
loop:
for {
select {
case <-stopCh:
return errorStopRequested
case err := <-errc:
return err
case event, ok := <-w.ResultChan():
if !ok {
break loop
}
if event.Type == watch.Error {
return apierrs.FromObject(event.Object)
}
if e, a := r.expectedType, reflect.TypeOf(event.Object); e != nil && e != a {
utilruntime.HandleError(fmt.Errorf("%s: expected type %v, but watch event object had type %v", r.name, e, a))
continue
}
meta, err := meta.Accessor(event.Object)
if err != nil {
utilruntime.HandleError(fmt.Errorf("%s: unable to understand watch event %#v", r.name, event))
continue
}
newResourceVersion := meta.GetResourceVersion()
switch event.Type {
case watch.Added:
r.store.Add(event.Object)
case watch.Modified:
r.store.Update(event.Object)
case watch.Deleted:
// TODO: Will any consumers need access to the "last known
// state", which is passed in event.Object? If so, may need
// to change this.
r.store.Delete(event.Object)
default:
utilruntime.HandleError(fmt.Errorf("%s: unable to understand watch event %#v", r.name, event))
}
*resourceVersion = newResourceVersion
r.setLastSyncResourceVersion(newResourceVersion)
eventCount++
}
}
watchDuration := time.Now().Sub(start)
if watchDuration < 1*time.Second && eventCount == 0 {
glog.V(4).Infof("%s: Unexpected watch close - watch lasted less than a second and no items received", r.name)
return errors.New("very short watch")
}
glog.V(4).Infof("%s: Watch close - %v total %v items received", r.name, r.expectedType, eventCount)
return nil
}
// HandleWS implements a websocket handler.
func (s *WatchServer) HandleWS(ws *websocket.Conn) {
defer ws.Close()
done := make(chan struct{})
go wsstream.IgnoreReceives(ws, 0)
var unknown runtime.Unknown
internalEvent := &versioned.InternalEvent{}
buf := &bytes.Buffer{}
streamBuf := &bytes.Buffer{}
ch := s.watching.ResultChan()
for {
select {
case <-done:
s.watching.Stop()
return
case event, ok := <-ch:
if !ok {
// End of results.
return
}
obj := event.Object
s.fixup(obj)
if err := s.embeddedEncoder.Encode(obj, buf); err != nil {
// unexpected error
utilruntime.HandleError(fmt.Errorf("unable to encode watch object: %v", err))
return
}
// ContentType is not required here because we are defaulting to the serializer
// type
unknown.Raw = buf.Bytes()
event.Object = &unknown
// the internal event will be versioned by the encoder
*internalEvent = versioned.InternalEvent(event)
if err := s.encoder.Encode(internalEvent, streamBuf); err != nil {
// encoding error
utilruntime.HandleError(fmt.Errorf("unable to encode event: %v", err))
s.watching.Stop()
return
}
if s.useTextFraming {
if err := websocket.Message.Send(ws, streamBuf.String()); err != nil {
// Client disconnect.
s.watching.Stop()
return
}
} else {
if err := websocket.Message.Send(ws, streamBuf.Bytes()); err != nil {
// Client disconnect.
s.watching.Stop()
return
}
}
buf.Reset()
streamBuf.Reset()
}
}
}
// maybeDeleteTerminatingPod non-gracefully deletes pods that are terminating
// that should not be gracefully terminated.
func (nc *NodeController) maybeDeleteTerminatingPod(obj interface{}) {
pod, ok := obj.(*api.Pod)
if !ok {
return
}
// consider only terminating pods
if pod.DeletionTimestamp == nil {
return
}
// delete terminating pods that have not yet been scheduled
if len(pod.Spec.NodeName) == 0 {
utilruntime.HandleError(nc.forcefullyDeletePod(pod))
return
}
nodeObj, found, err := nc.nodeStore.GetByKey(pod.Spec.NodeName)
if err != nil {
// this can only happen if the Store.KeyFunc has a problem creating
// a key for the pod. If it happens once, it will happen again so
// don't bother requeuing the pod.
utilruntime.HandleError(err)
return
}
// delete terminating pods that have been scheduled on
// nonexistent nodes
if !found {
glog.Warningf("Unable to find Node: %v, deleting all assigned Pods.", pod.Spec.NodeName)
utilruntime.HandleError(nc.forcefullyDeletePod(pod))
return
}
// delete terminating pods that have been scheduled on
// nodes that do not support graceful termination
// TODO(mikedanese): this can be removed when we no longer
// guarantee backwards compatibility of master API to kubelets with
// versions less than 1.1.0
node := nodeObj.(*api.Node)
v, err := version.Parse(node.Status.NodeInfo.KubeletVersion)
if err != nil {
glog.V(0).Infof("couldn't parse verions %q of minion: %v", node.Status.NodeInfo.KubeletVersion, err)
utilruntime.HandleError(nc.forcefullyDeletePod(pod))
return
}
if gracefulDeletionVersion.GT(v) {
utilruntime.HandleError(nc.forcefullyDeletePod(pod))
return
}
}
func (w *etcdWatcher) sendModify(res *etcd.Response) {
if res.Node == nil {
glog.Errorf("unexpected nil node: %#v", res)
return
}
if w.include != nil && !w.include(res.Node.Key) {
return
}
curObj, err := w.decodeObject(res.Node)
if err != nil {
utilruntime.HandleError(fmt.Errorf("failure to decode api object: %v\n'%v' from %#v %#v", err, string(res.Node.Value), res, res.Node))
// TODO: expose an error through watch.Interface?
// Ignore this value. If we stop the watch on a bad value, a client that uses
// the resourceVersion to resume will never be able to get past a bad value.
return
}
curObjPasses := w.filter(curObj)
oldObjPasses := false
var oldObj runtime.Object
if res.PrevNode != nil && res.PrevNode.Value != "" {
// Ignore problems reading the old object.
if oldObj, err = w.decodeObject(res.PrevNode); err == nil {
if err := w.versioner.UpdateObject(oldObj, res.Node.ModifiedIndex); err != nil {
utilruntime.HandleError(fmt.Errorf("failure to version api object (%d) %#v: %v", res.Node.ModifiedIndex, oldObj, err))
}
oldObjPasses = w.filter(oldObj)
}
}
// Some changes to an object may cause it to start or stop matching a filter.
// We need to report those as adds/deletes. So we have to check both the previous
// and current value of the object.
switch {
case curObjPasses && oldObjPasses:
w.emit(watch.Event{
Type: watch.Modified,
Object: curObj,
})
case curObjPasses && !oldObjPasses:
w.emit(watch.Event{
Type: watch.Added,
Object: curObj,
})
case !curObjPasses && oldObjPasses:
w.emit(watch.Event{
Type: watch.Deleted,
Object: oldObj,
})
}
// Do nothing if neither new nor old object passed the filter.
}
func getPidsForProcess(name, pidFile string) ([]int, error) {
if len(pidFile) > 0 {
if pid, err := getPidFromPidFile(pidFile); err == nil {
return []int{pid}, nil
} else {
// log the error and fall back to pidof
runtime.HandleError(err)
}
}
out, err := exec.Command("pidof", name).Output()
if err != nil {
return []int{}, fmt.Errorf("failed to find pid of %q: %v", name, err)
}
// The output of pidof is a list of pids.
pids := []int{}
for _, pidStr := range strings.Split(strings.TrimSpace(string(out)), " ") {
pid, err := strconv.Atoi(pidStr)
if err != nil {
continue
}
pids = append(pids, pid)
}
return pids, nil
}
func recordToSink(sink EventSink, event *api.Event, eventCorrelator *EventCorrelator, randGen *rand.Rand, sleepDuration time.Duration) {
// Make a copy before modification, because there could be multiple listeners.
// Events are safe to copy like this.
eventCopy := *event
event = &eventCopy
result, err := eventCorrelator.EventCorrelate(event)
if err != nil {
utilruntime.HandleError(err)
}
if result.Skip {
return
}
tries := 0
for {
if recordEvent(sink, result.Event, result.Patch, result.Event.Count > 1, eventCorrelator) {
break
}
tries++
if tries >= maxTriesPerEvent {
glog.Errorf("Unable to write event '%#v' (retry limit exceeded!)", event)
break
}
// Randomize the first sleep so that various clients won't all be
// synced up if the master goes down.
if tries == 1 {
time.Sleep(time.Duration(float64(sleepDuration) * randGen.Float64()))
} else {
time.Sleep(sleepDuration)
}
}
}
func (gcc *PodGCController) gc() {
terminatedPods, _ := gcc.podStore.List(labels.Everything())
terminatedPodCount := len(terminatedPods)
sort.Sort(byCreationTimestamp(terminatedPods))
deleteCount := terminatedPodCount - gcc.threshold
if deleteCount > terminatedPodCount {
deleteCount = terminatedPodCount
}
if deleteCount > 0 {
glog.Infof("garbage collecting %v pods", deleteCount)
}
var wait sync.WaitGroup
for i := 0; i < deleteCount; i++ {
wait.Add(1)
go func(namespace string, name string) {
defer wait.Done()
if err := gcc.deletePod(namespace, name); err != nil {
// ignore not founds
defer utilruntime.HandleError(err)
}
}(terminatedPods[i].Namespace, terminatedPods[i].Name)
}
wait.Wait()
}
// RunUntil starts a watch and handles watch events. Will restart the watch if it is closed.
// RunUntil starts a goroutine and returns immediately. It will exit when stopCh is closed.
func (r *Reflector) RunUntil(stopCh <-chan struct{}) {
glog.V(3).Infof("Starting reflector %v (%s) from %s", r.expectedType, r.resyncPeriod, r.name)
go wait.Until(func() {
if err := r.ListAndWatch(stopCh); err != nil {
utilruntime.HandleError(err)
}
}, r.period, stopCh)
}
// Run starts a watch and handles watch events. Will restart the watch if it is closed.
// Run starts a goroutine and returns immediately.
func (r *Reflector) Run() {
glog.V(3).Infof("Starting reflector %v (%s) from %s", r.expectedType, r.resyncPeriod, r.name)
go wait.Until(func() {
if err := r.ListAndWatch(wait.NeverStop); err != nil {
utilruntime.HandleError(err)
}
}, r.period, wait.NeverStop)
}
// cleanupOrphanedPods deletes pods that are bound to nodes that don't
// exist.
func (nc *NodeController) cleanupOrphanedPods() {
pods, err := nc.podStore.List(labels.Everything())
if err != nil {
utilruntime.HandleError(err)
return
}
for _, pod := range pods {
if pod.Spec.NodeName == "" {
continue
}
if _, exists, _ := nc.nodeStore.Store.GetByKey(pod.Spec.NodeName); exists {
continue
}
if err := nc.forcefullyDeletePod(pod); err != nil {
utilruntime.HandleError(err)
}
}
}
// ObjectReplenenishmentDeleteFunc will replenish on every delete
func ObjectReplenishmentDeleteFunc(options *ReplenishmentControllerOptions) func(obj interface{}) {
return func(obj interface{}) {
metaObject, err := meta.Accessor(obj)
if err != nil {
tombstone, ok := obj.(cache.DeletedFinalStateUnknown)
if !ok {
glog.Errorf("replenishment controller could not get object from tombstone %+v, could take up to %v before quota is replenished", obj, options.ResyncPeriod())
utilruntime.HandleError(err)
return
}
metaObject, err = meta.Accessor(tombstone.Obj)
if err != nil {
glog.Errorf("replenishment controller tombstone contained object that is not a meta %+v, could take up to %v before quota is replenished", tombstone.Obj, options.ResyncPeriod())
utilruntime.HandleError(err)
return
}
}
options.ReplenishmentFunc(options.GroupKind, metaObject.GetNamespace(), nil)
}
}
func (w *etcdWatcher) sendResult(res *etcd.Response) {
switch res.Action {
case EtcdCreate, EtcdGet:
w.sendAdd(res)
case EtcdSet, EtcdCAS:
w.sendModify(res)
case EtcdDelete, EtcdExpire, EtcdCAD:
w.sendDelete(res)
default:
utilruntime.HandleError(fmt.Errorf("unknown action: %v", res.Action))
}
}
func (*v1ProtocolHandler) waitForStreams(streams <-chan streamAndReply, expectedStreams int, expired <-chan time.Time) (*context, error) {
ctx := &context{}
receivedStreams := 0
replyChan := make(chan struct{})
stop := make(chan struct{})
defer close(stop)
WaitForStreams:
for {
select {
case stream := <-streams:
streamType := stream.Headers().Get(api.StreamType)
switch streamType {
case api.StreamTypeError:
ctx.errorStream = stream
// This defer statement shouldn't be here, but due to previous refactoring, it ended up in
// here. This is what 1.0.x kubelets do, so we're retaining that behavior. This is fixed in
// the v2ProtocolHandler.
defer stream.Reset()
go waitStreamReply(stream.replySent, replyChan, stop)
case api.StreamTypeStdin:
ctx.stdinStream = stream
go waitStreamReply(stream.replySent, replyChan, stop)
case api.StreamTypeStdout:
ctx.stdoutStream = stream
go waitStreamReply(stream.replySent, replyChan, stop)
case api.StreamTypeStderr:
ctx.stderrStream = stream
go waitStreamReply(stream.replySent, replyChan, stop)
default:
runtime.HandleError(fmt.Errorf("Unexpected stream type: %q", streamType))
}
case <-replyChan:
receivedStreams++
if receivedStreams == expectedStreams {
break WaitForStreams
}
case <-expired:
// TODO find a way to return the error to the user. Maybe use a separate
// stream to report errors?
return nil, errors.New("timed out waiting for client to create streams")
}
}
if ctx.stdinStream != nil {
ctx.stdinStream.Close()
}
return ctx, nil
}
// ServeAttach handles requests to attach to a container. After creating/receiving the required
// streams, it delegates the actual attaching to attacher.
func ServeAttach(w http.ResponseWriter, req *http.Request, attacher Attacher, podName string, uid types.UID, container string, idleTimeout, streamCreationTimeout time.Duration, supportedProtocols []string) {
ctx, ok := createStreams(req, w, supportedProtocols, idleTimeout, streamCreationTimeout)
if !ok {
// error is handled by createStreams
return
}
defer ctx.conn.Close()
err := attacher.AttachContainer(podName, uid, container, ctx.stdinStream, ctx.stdoutStream, ctx.stderrStream, ctx.tty)
if err != nil {
msg := fmt.Sprintf("error attaching to container: %v", err)
runtime.HandleError(errors.New(msg))
fmt.Fprint(ctx.errorStream, msg)
}
}
// errorJSONFatal renders an error to the response, and if codec fails will render plaintext.
// Returns the HTTP status code of the error.
func errorJSONFatal(err error, codec runtime.Encoder, w http.ResponseWriter) int {
utilruntime.HandleError(fmt.Errorf("apiserver was unable to write a JSON response: %v", err))
status := errToAPIStatus(err)
code := int(status.Code)
output, err := runtime.Encode(codec, status)
if err != nil {
w.WriteHeader(code)
fmt.Fprintf(w, "%s: %s", status.Reason, status.Message)
return code
}
w.Header().Set("Content-Type", "application/json")
w.WriteHeader(code)
w.Write(output)
return code
}
func (*v2ProtocolHandler) waitForStreams(streams <-chan streamAndReply, expectedStreams int, expired <-chan time.Time) (*context, error) {
ctx := &context{}
receivedStreams := 0
replyChan := make(chan struct{})
stop := make(chan struct{})
defer close(stop)
WaitForStreams:
for {
select {
case stream := <-streams:
streamType := stream.Headers().Get(api.StreamType)
switch streamType {
case api.StreamTypeError:
ctx.errorStream = stream
go waitStreamReply(stream.replySent, replyChan, stop)
case api.StreamTypeStdin:
ctx.stdinStream = stream
go waitStreamReply(stream.replySent, replyChan, stop)
case api.StreamTypeStdout:
ctx.stdoutStream = stream
go waitStreamReply(stream.replySent, replyChan, stop)
case api.StreamTypeStderr:
ctx.stderrStream = stream
go waitStreamReply(stream.replySent, replyChan, stop)
default:
runtime.HandleError(fmt.Errorf("Unexpected stream type: %q", streamType))
}
case <-replyChan:
receivedStreams++
if receivedStreams == expectedStreams {
break WaitForStreams
}
case <-expired:
// TODO find a way to return the error to the user. Maybe use a separate
// stream to report errors?
return nil, errors.New("timed out waiting for client to create streams")
}
}
return ctx, nil
}
// etcdGetInitialWatchState turns an etcd Get request into a watch equivalent
func etcdGetInitialWatchState(ctx context.Context, client etcd.KeysAPI, key string, recursive bool, quorum bool, incoming chan<- *etcd.Response) (resourceVersion uint64, err error) {
opts := etcd.GetOptions{
Recursive: recursive,
Sort: false,
Quorum: quorum,
}
resp, err := client.Get(ctx, key, &opts)
if err != nil {
if !etcdutil.IsEtcdNotFound(err) {
utilruntime.HandleError(fmt.Errorf("watch was unable to retrieve the current index for the provided key (%q): %v", key, err))
return resourceVersion, toStorageErr(err, key, 0)
}
if etcdError, ok := err.(etcd.Error); ok {
resourceVersion = etcdError.Index
}
return resourceVersion, nil
}
resourceVersion = resp.Index
convertRecursiveResponse(resp.Node, resp, incoming)
return
}
// serveWatch handles serving requests to the server
// TODO: the functionality in this method and in WatchServer.Serve is not cleanly decoupled.
func serveWatch(watcher watch.Interface, scope RequestScope, req *restful.Request, res *restful.Response, timeout time.Duration) {
// negotiate for the stream serializer
serializer, err := negotiateOutputStreamSerializer(req.Request, scope.Serializer)
if err != nil {
scope.err(err, res.ResponseWriter, req.Request)
return
}
if serializer.Framer == nil {
scope.err(fmt.Errorf("no framer defined for %q available for embedded encoding", serializer.MediaType), res.ResponseWriter, req.Request)
return
}
encoder := scope.Serializer.EncoderForVersion(serializer.Serializer, scope.Kind.GroupVersion())
useTextFraming := serializer.EncodesAsText
// find the embedded serializer matching the media type
embeddedEncoder := scope.Serializer.EncoderForVersion(serializer.Embedded.Serializer, scope.Kind.GroupVersion())
server := &WatchServer{
watching: watcher,
scope: scope,
useTextFraming: useTextFraming,
mediaType: serializer.MediaType,
framer: serializer.Framer,
encoder: encoder,
embeddedEncoder: embeddedEncoder,
fixup: func(obj runtime.Object) {
if err := setSelfLink(obj, req, scope.Namer); err != nil {
utilruntime.HandleError(fmt.Errorf("failed to set link for object %v: %v", reflect.TypeOf(obj), err))
}
},
t: &realTimeoutFactory{timeout},
}
server.ServeHTTP(res.ResponseWriter, req.Request)
}
// worker runs a worker thread that just dequeues items, processes them, and marks them done.
// It enforces that the syncHandler is never invoked concurrently with the same key.
func (rq *ResourceQuotaController) worker() {
workFunc := func() bool {
key, quit := rq.queue.Get()
if quit {
return true
}
defer rq.queue.Done(key)
err := rq.syncHandler(key.(string))
if err == nil {
rq.queue.Forget(key)
return false
}
utilruntime.HandleError(err)
rq.queue.AddRateLimited(key)
return false
}
for {
if quit := workFunc(); quit {
glog.Infof("resource quota controller worker shutting down")
return
}
}
}
// errToAPIStatus converts an error to an unversioned.Status object.
func errToAPIStatus(err error) *unversioned.Status {
switch t := err.(type) {
case statusError:
status := t.Status()
if len(status.Status) == 0 {
status.Status = unversioned.StatusFailure
}
if status.Code == 0 {
switch status.Status {
case unversioned.StatusSuccess:
status.Code = http.StatusOK
case unversioned.StatusFailure:
status.Code = http.StatusInternalServerError
}
}
//TODO: check for invalid responses
return &status
default:
status := http.StatusInternalServerError
switch {
//TODO: replace me with NewConflictErr
case storage.IsTestFailed(err):
status = http.StatusConflict
}
// Log errors that were not converted to an error status
// by REST storage - these typically indicate programmer
// error by not using pkg/api/errors, or unexpected failure
// cases.
runtime.HandleError(fmt.Errorf("apiserver received an error that is not an unversioned.Status: %v", err))
return &unversioned.Status{
Status: unversioned.StatusFailure,
Code: int32(status),
Reason: unversioned.StatusReasonUnknown,
Message: err.Error(),
}
}
}
// write renders a returned runtime.Object to the response as a stream or an encoded object. If the object
// returned by the response implements rest.ResourceStreamer that interface will be used to render the
// response. The Accept header and current API version will be passed in, and the output will be copied
// directly to the response body. If content type is returned it is used, otherwise the content type will
// be "application/octet-stream". All other objects are sent to standard JSON serialization.
func write(statusCode int, gv unversioned.GroupVersion, s runtime.NegotiatedSerializer, object runtime.Object, w http.ResponseWriter, req *http.Request) {
if stream, ok := object.(rest.ResourceStreamer); ok {
out, flush, contentType, err := stream.InputStream(gv.String(), req.Header.Get("Accept"))
if err != nil {
errorNegotiated(err, s, gv, w, req)
return
}
if out == nil {
// No output provided - return StatusNoContent
w.WriteHeader(http.StatusNoContent)
return
}
defer out.Close()
if wsstream.IsWebSocketRequest(req) {
r := wsstream.NewReader(out, true)
if err := r.Copy(w, req); err != nil {
utilruntime.HandleError(fmt.Errorf("error encountered while streaming results via websocket: %v", err))
}
return
}
if len(contentType) == 0 {
contentType = "application/octet-stream"
}
w.Header().Set("Content-Type", contentType)
w.WriteHeader(statusCode)
writer := w.(io.Writer)
if flush {
writer = flushwriter.Wrap(w)
}
io.Copy(writer, out)
return
}
writeNegotiated(s, gv, w, req, statusCode, object)
}
// manageJob is the core method responsible for managing the number of running
// pods according to what is specified in the job.Spec.
func (jm *JobController) manageJob(activePods []*api.Pod, succeeded int32, job *batch.Job) int32 {
var activeLock sync.Mutex
active := int32(len(activePods))
parallelism := *job.Spec.Parallelism
jobKey, err := controller.KeyFunc(job)
if err != nil {
glog.Errorf("Couldn't get key for job %#v: %v", job, err)
return 0
}
if active > parallelism {
diff := active - parallelism
jm.expectations.ExpectDeletions(jobKey, int(diff))
glog.V(4).Infof("Too many pods running job %q, need %d, deleting %d", jobKey, parallelism, diff)
// Sort the pods in the order such that not-ready < ready, unscheduled
// < scheduled, and pending < running. This ensures that we delete pods
// in the earlier stages whenever possible.
sort.Sort(controller.ActivePods(activePods))
active -= diff
wait := sync.WaitGroup{}
wait.Add(int(diff))
for i := int32(0); i < diff; i++ {
go func(ix int32) {
defer wait.Done()
if err := jm.podControl.DeletePod(job.Namespace, activePods[ix].Name, job); err != nil {
defer utilruntime.HandleError(err)
// Decrement the expected number of deletes because the informer won't observe this deletion
jm.expectations.DeletionObserved(jobKey)
activeLock.Lock()
active++
activeLock.Unlock()
}
}(i)
}
wait.Wait()
} else if active < parallelism {
wantActive := int32(0)
if job.Spec.Completions == nil {
// Job does not specify a number of completions. Therefore, number active
// should be equal to parallelism, unless the job has seen at least
// once success, in which leave whatever is running, running.
if succeeded > 0 {
wantActive = active
} else {
wantActive = parallelism
}
} else {
// Job specifies a specific number of completions. Therefore, number
// active should not ever exceed number of remaining completions.
wantActive = *job.Spec.Completions - succeeded
if wantActive > parallelism {
wantActive = parallelism
}
}
diff := wantActive - active
if diff < 0 {
glog.Errorf("More active than wanted: job %q, want %d, have %d", jobKey, wantActive, active)
diff = 0
}
jm.expectations.ExpectCreations(jobKey, int(diff))
glog.V(4).Infof("Too few pods running job %q, need %d, creating %d", jobKey, wantActive, diff)
active += diff
wait := sync.WaitGroup{}
wait.Add(int(diff))
for i := int32(0); i < diff; i++ {
go func() {
defer wait.Done()
if err := jm.podControl.CreatePods(job.Namespace, &job.Spec.Template, job); err != nil {
defer utilruntime.HandleError(err)
// Decrement the expected number of creates because the informer won't observe this pod
jm.expectations.CreationObserved(jobKey)
activeLock.Lock()
active--
activeLock.Unlock()
}
}()
}
wait.Wait()
}
return active
}
// syncJob will sync the job with the given key if it has had its expectations fulfilled, meaning
// it did not expect to see any more of its pods created or deleted. This function is not meant to be invoked
// concurrently with the same key.
func (jm *JobController) syncJob(key string) error {
startTime := time.Now()
defer func() {
glog.V(4).Infof("Finished syncing job %q (%v)", key, time.Now().Sub(startTime))
}()
if !jm.podStoreSynced() {
// Sleep so we give the pod reflector goroutine a chance to run.
time.Sleep(replicationcontroller.PodStoreSyncedPollPeriod)
glog.V(4).Infof("Waiting for pods controller to sync, requeuing job %v", key)
jm.queue.Add(key)
return nil
}
obj, exists, err := jm.jobStore.Store.GetByKey(key)
if !exists {
glog.V(4).Infof("Job has been deleted: %v", key)
jm.expectations.DeleteExpectations(key)
return nil
}
if err != nil {
glog.Errorf("Unable to retrieve job %v from store: %v", key, err)
jm.queue.Add(key)
return err
}
job := *obj.(*batch.Job)
// Check the expectations of the job before counting active pods, otherwise a new pod can sneak in
// and update the expectations after we've retrieved active pods from the store. If a new pod enters
// the store after we've checked the expectation, the job sync is just deferred till the next relist.
jobKey, err := controller.KeyFunc(&job)
if err != nil {
glog.Errorf("Couldn't get key for job %#v: %v", job, err)
return err
}
jobNeedsSync := jm.expectations.SatisfiedExpectations(jobKey)
selector, _ := unversioned.LabelSelectorAsSelector(job.Spec.Selector)
podList, err := jm.podStore.Pods(job.Namespace).List(selector)
if err != nil {
glog.Errorf("Error getting pods for job %q: %v", key, err)
jm.queue.Add(key)
return err
}
activePods := controller.FilterActivePods(podList.Items)
active := int32(len(activePods))
succeeded, failed := getStatus(podList.Items)
conditions := len(job.Status.Conditions)
if job.Status.StartTime == nil {
now := unversioned.Now()
job.Status.StartTime = &now
}
// if job was finished previously, we don't want to redo the termination
if isJobFinished(&job) {
return nil
}
if pastActiveDeadline(&job) {
// TODO: below code should be replaced with pod termination resulting in
// pod failures, rather than killing pods. Unfortunately none such solution
// exists ATM. There's an open discussion in the topic in
// https://github.com/kubernetes/kubernetes/issues/14602 which might give
// some sort of solution to above problem.
// kill remaining active pods
wait := sync.WaitGroup{}
wait.Add(int(active))
for i := int32(0); i < active; i++ {
go func(ix int32) {
defer wait.Done()
if err := jm.podControl.DeletePod(job.Namespace, activePods[ix].Name, &job); err != nil {
defer utilruntime.HandleError(err)
}
}(i)
}
wait.Wait()
// update status values accordingly
failed += active
active = 0
job.Status.Conditions = append(job.Status.Conditions, newCondition(batch.JobFailed, "DeadlineExceeded", "Job was active longer than specified deadline"))
jm.recorder.Event(&job, api.EventTypeNormal, "DeadlineExceeded", "Job was active longer than specified deadline")
} else {
if jobNeedsSync {
active = jm.manageJob(activePods, succeeded, &job)
}
completions := succeeded
complete := false
if job.Spec.Completions == nil {
// This type of job is complete when any pod exits with success.
// Each pod is capable of
// determining whether or not the entire Job is done. Subsequent pods are
// not expected to fail, but if they do, the failure is ignored. Once any
// pod succeeds, the controller waits for remaining pods to finish, and
// then the job is complete.
if succeeded > 0 && active == 0 {
complete = true
}
} else {
// Job specifies a number of completions. This type of job signals
// success by having that number of successes. Since we do not
// start more pods than there are remaining completions, there should
// not be any remaining active pods once this count is reached.
if completions >= *job.Spec.Completions {
complete = true
if active > 0 {
jm.recorder.Event(&job, api.EventTypeWarning, "TooManyActivePods", "Too many active pods running after completion count reached")
}
if completions > *job.Spec.Completions {
jm.recorder.Event(&job, api.EventTypeWarning, "TooManySucceededPods", "Too many succeeded pods running after completion count reached")
}
}
}
if complete {
job.Status.Conditions = append(job.Status.Conditions, newCondition(batch.JobComplete, "", ""))
now := unversioned.Now()
job.Status.CompletionTime = &now
}
}
// no need to update the job if the status hasn't changed since last time
if job.Status.Active != active || job.Status.Succeeded != succeeded || job.Status.Failed != failed || len(job.Status.Conditions) != conditions {
job.Status.Active = active
job.Status.Succeeded = succeeded
job.Status.Failed = failed
if err := jm.updateHandler(&job); err != nil {
glog.Errorf("Failed to update job %v, requeuing. Error: %v", job.Name, err)
jm.enqueueController(&job)
}
}
return nil
}
// monitorNodeStatus verifies node status are constantly updated by kubelet, and if not,
// post "NodeReady==ConditionUnknown". It also evicts all pods if node is not ready or
// not reachable for a long period of time.
func (nc *NodeController) monitorNodeStatus() error {
nodes, err := nc.kubeClient.Core().Nodes().List(api.ListOptions{})
if err != nil {
return err
}
for _, node := range nodes.Items {
if !nc.knownNodeSet.Has(node.Name) {
glog.V(1).Infof("NodeController observed a new Node: %#v", node)
nc.recordNodeEvent(node.Name, api.EventTypeNormal, "RegisteredNode", fmt.Sprintf("Registered Node %v in NodeController", node.Name))
nc.cancelPodEviction(node.Name)
nc.knownNodeSet.Insert(node.Name)
}
}
// If there's a difference between lengths of known Nodes and observed nodes
// we must have removed some Node.
if len(nc.knownNodeSet) != len(nodes.Items) {
observedSet := make(sets.String)
for _, node := range nodes.Items {
observedSet.Insert(node.Name)
}
deleted := nc.knownNodeSet.Difference(observedSet)
for nodeName := range deleted {
glog.V(1).Infof("NodeController observed a Node deletion: %v", nodeName)
nc.recordNodeEvent(nodeName, api.EventTypeNormal, "RemovingNode", fmt.Sprintf("Removing Node %v from NodeController", nodeName))
nc.evictPods(nodeName)
nc.knownNodeSet.Delete(nodeName)
}
}
seenReady := false
for i := range nodes.Items {
var gracePeriod time.Duration
var observedReadyCondition api.NodeCondition
var currentReadyCondition *api.NodeCondition
node := &nodes.Items[i]
for rep := 0; rep < nodeStatusUpdateRetry; rep++ {
gracePeriod, observedReadyCondition, currentReadyCondition, err = nc.tryUpdateNodeStatus(node)
if err == nil {
break
}
name := node.Name
node, err = nc.kubeClient.Core().Nodes().Get(name)
if err != nil {
glog.Errorf("Failed while getting a Node to retry updating NodeStatus. Probably Node %s was deleted.", name)
break
}
}
if err != nil {
glog.Errorf("Update status of Node %v from NodeController exceeds retry count."+
"Skipping - no pods will be evicted.", node.Name)
continue
}
decisionTimestamp := nc.now()
if currentReadyCondition != nil {
// Check eviction timeout against decisionTimestamp
if observedReadyCondition.Status == api.ConditionFalse &&
decisionTimestamp.After(nc.nodeStatusMap[node.Name].readyTransitionTimestamp.Add(nc.podEvictionTimeout)) {
if nc.evictPods(node.Name) {
glog.V(4).Infof("Evicting pods on node %s: %v is later than %v + %v", node.Name, decisionTimestamp, nc.nodeStatusMap[node.Name].readyTransitionTimestamp, nc.podEvictionTimeout)
}
}
if observedReadyCondition.Status == api.ConditionUnknown &&
decisionTimestamp.After(nc.nodeStatusMap[node.Name].probeTimestamp.Add(nc.podEvictionTimeout)) {
if nc.evictPods(node.Name) {
glog.V(4).Infof("Evicting pods on node %s: %v is later than %v + %v", node.Name, decisionTimestamp, nc.nodeStatusMap[node.Name].readyTransitionTimestamp, nc.podEvictionTimeout-gracePeriod)
}
}
if observedReadyCondition.Status == api.ConditionTrue {
// We do not treat a master node as a part of the cluster for network segmentation checking.
if !system.IsMasterNode(node) {
seenReady = true
}
if nc.cancelPodEviction(node.Name) {
glog.V(2).Infof("Node %s is ready again, cancelled pod eviction", node.Name)
}
}
// Report node event.
if currentReadyCondition.Status != api.ConditionTrue && observedReadyCondition.Status == api.ConditionTrue {
nc.recordNodeStatusChange(node, "NodeNotReady")
if err = nc.markAllPodsNotReady(node.Name); err != nil {
utilruntime.HandleError(fmt.Errorf("Unable to mark all pods NotReady on node %v: %v", node.Name, err))
}
}
// Check with the cloud provider to see if the node still exists. If it
// doesn't, delete the node immediately.
if currentReadyCondition.Status != api.ConditionTrue && nc.cloud != nil {
exists, err := nc.nodeExistsInCloudProvider(node.Name)
if err != nil {
glog.Errorf("Error determining if node %v exists in cloud: %v", node.Name, err)
continue
}
if !exists {
glog.V(2).Infof("Deleting node (no longer present in cloud provider): %s", node.Name)
nc.recordNodeEvent(node.Name, api.EventTypeNormal, "DeletingNode", fmt.Sprintf("Deleting Node %v because it's not present according to cloud provider", node.Name))
go func(nodeName string) {
defer utilruntime.HandleCrash()
// Kubelet is not reporting and Cloud Provider says node
// is gone. Delete it without worrying about grace
// periods.
if err := nc.forcefullyDeleteNode(nodeName); err != nil {
glog.Errorf("Unable to forcefully delete node %q: %v", nodeName, err)
}
}(node.Name)
continue
}
}
}
}
// NC don't see any Ready Node. We assume that the network is segmented and Nodes cannot connect to API server and
// update their statuses. NC enteres network segmentation mode and cancels all evictions in progress.
if !seenReady {
nc.networkSegmentationMode = true
nc.stopAllPodEvictions()
glog.V(2).Info("NodeController is entering network segmentation mode.")
} else {
if nc.networkSegmentationMode {
nc.forceUpdateAllProbeTimes()
nc.networkSegmentationMode = false
glog.V(2).Info("NodeController exited network segmentation mode.")
}
}
return nil
}
