Client-go中的watch接口的resultChan會自動關閉
問題描述
在client-go工程中,有時候需要用到watch接口,實際場景如下:
namespacesWatch, err := clientSet.CoreV1().Namespaces().Watch(metav1.ListOptions{})
if err != nil {
klog.Errorf("create watch error, error is %s, program exit!", err.Error())
panic(err)
}
for {
select e, ok := <-namespacesWatch.ResultChan()
if e.Object == nil {
// 這個時候一般是chan已經被關閉了,可以順便開看下ok是不是flase。
} else {
// 正常處理邏輯
}
}
watch的resultChan會週期性的關閉,我不知道這個週期是不是可以設置,但是我在github的issue中看到,有人5分鐘就自動關閉了,我的集羣是大概40分鐘左右,進一步的追究需要深入,我這邊就說我遇到的問題以及解決方法吧。問題就是resultChan會定期自動關閉。github上client-go項目對於該問題的Issues: https://github.com/kubernetes/client-go/issues/623, 上面有大佬回覆:No, the server will close watch connections regularly. Re-establishing a watch at the last-received resourceVersion is a normal part of maintaining a watch as a client. There are helpers to do this for you in https://github.com/kubernetes/client-go/tree/master/tools/watch。是有官方解決方法的去client-go的tools目錄下的watch文件夾下看下。
resultChan會自動關閉的原因
看一下watch.go中的部分代碼:
// Interface can be implemented by anything that knows how to watch and report changes.
type Interface interface {
// Stops watching. Will close the channel returned by ResultChan(). Releases
// any resources used by the watch.
Stop()
// Returns a chan which will receive all the events. If an error occurs
// or Stop() is called, this channel will be closed, in which case the
// watch should be completely cleaned up. !!!明確說了在出現錯誤或者被調用Stop時,通道會自動關閉的
ResultChan() <-chan Event
}
我們接着看下有哪些error和哪些情況會調用stop,我就開始追watch方法的那個對象,最後追到這裏streamwatcher.go,Ok,這裏就可以看到watch實際對象了,struct的中參數可以看下,有result的通道,互斥鎖和Stoppe標誌接收是否已經結束了的標誌位。看下NewStreamWatcher方法,關注其中的go sw.receive(),在返回對象前就起了協程在接收數據了,那接着去看receive函數,看一下receive那邊的註釋,我寫的中文註釋,就是在解碼遇到錯誤是就會選擇return,return前看下defer sw.Stop()等清理操作。這樣就跟上面對上了!
// StreamWatcher turns any stream for which you can write a Decoder interface
// into a watch.Interface.
type StreamWatcher struct {
sync.Mutex
source Decoder
reporter Reporter
result chan Event
stopped bool
}
// NewStreamWatcher creates a StreamWatcher from the given decoder.
func NewStreamWatcher(d Decoder, r Reporter) *StreamWatcher {
sw := &StreamWatcher{
source: d,
reporter: r,
// It's easy for a consumer to add buffering via an extra
// goroutine/channel, but impossible for them to remove it,
// so nonbuffered is better.
result: make(chan Event),
}
go sw.receive()// !!!看這裏,新建完對象就開始接收數據了
return sw
}
// ResultChan implements Interface.
func (sw *StreamWatcher) ResultChan() <-chan Event {
return sw.result
}
// Stop implements Interface.
func (sw *StreamWatcher) Stop() {
// Call Close() exactly once by locking and setting a flag.
sw.Lock()
defer sw.Unlock()
if !sw.stopped {
sw.stopped = true
sw.source.Close()
}
}
// stopping returns true if Stop() was called previously.
func (sw *StreamWatcher) stopping() bool {
sw.Lock()
defer sw.Unlock()
return sw.stopped
}
// receive reads result from the decoder in a loop and sends down the result channel.
func (sw *StreamWatcher) receive() {
defer close(sw.result)
defer sw.Stop()// 注意看這裏,這個方法退出前就會調用stop函數
defer utilruntime.HandleCrash()
for {// for循環,一直接收
action, obj, err := sw.source.Decode()
if err != nil {//以下是接收到的錯,反正有錯誤就會return
// Ignore expected error.
if sw.stopping() {
return
}
switch err {
case io.EOF:
// watch closed normally
case io.ErrUnexpectedEOF:
klog.V(1).Infof("Unexpected EOF during watch stream event decoding: %v", err)
default:
if net.IsProbableEOF(err) {
klog.V(5).Infof("Unable to decode an event from the watch stream: %v", err)
} else {
sw.result <- Event{
Type: Error,
Object: sw.reporter.AsObject(fmt.Errorf("unable to decode an event from the watch stream: %v", err)),
}
}
}
return //只要是錯誤就是停止接收了!!!
}
sw.result <- Event{
Type: action,
Object: obj,
}// 沒錯誤就往result裏塞數據
}
}
在這部分源代碼中,我學到一個點就是在要從一個對象中持續的處理數據時,開通到,並且在新建完返回對象前就可以開始傳數據了,接收到只要使用對象中的chan就可以拿到數據,這樣就不用手動開啓reecive了,很省事,很安全,所有操作(除了Stop和ResultChan方法)都是內部做好了,不容許外部調用者做額外的操作干擾我正常的邏輯。
解決辦法
首先說下,這個問題是有解決方法的:
No, the server will close watch connections regularly. Re-establishing a watch at the last-received resourceVersion is a normal part of maintaining a watch as a client. There are helpers to do this for you in https://github.com/kubernetes/client-go/tree/master/tools/watch
這串是大佬的回覆。
先說下我自己挫比解決方案,代碼如下,在我知道有watch接口的resultChan會自動關閉後,第一時間想到的就是,關閉了,我重新起不就可以了嗎,所以我寫了這兩段for循環,檢測到通道close掉後跳出內層循環,再次創建就可以了。挫吧!當時還挺有用,作爲上線時候的代碼使用了。(我是真不專業!當時應該想到會有官方解決方案的)
for {
klog.Info("start watch")
config, err := rest.InClusterConfig()
clientSet, err := kubernetes.NewForConfig(config)
namespacesWatch, err := clientSet.CoreV1().Namespaces().Watch(metav1.ListOptions{})
if err != nil {
klog.Errorf("create watch error, error is %s, program exit!", err.Error())
panic(err)
}
loopier:
for {
select {
case e, ok := <-namespacesWatch.ResultChan():
if !ok {
// 說明該通道已經被close掉了
klog.Warning("!!!!!namespacesWatch chan has been close!!!!")
klog.Info("clean chan over!")
time.Sleep(time.Second * 5)
break loopier
}
if e.Object != nil {
// 業務邏輯
}
}
}
}
下面就是官方解決方案,在client工程下有個tools文件夾中有個watch文件夾,裏面有個retrywatcher.go:我是把這個文件下所有的源碼複製出來了,下面會細講,你先看看!
// resourceVersionGetter is an interface used to get resource version from events.
// We can't reuse an interface from meta otherwise it would be a cyclic dependency and we need just this one method
type resourceVersionGetter interface {
GetResourceVersion() string
}
// RetryWatcher will make sure that in case the underlying watcher is closed (e.g. due to API timeout or etcd timeout)
// it will get restarted from the last point without the consumer even knowing about it.
// RetryWatcher does that by inspecting events and keeping track of resourceVersion.
// Especially useful when using watch.UntilWithoutRetry where premature termination is causing issues and flakes.
// Please note that this is not resilient to etcd cache not having the resource version anymore - you would need to
// use Informers for that.
type RetryWatcher struct {
lastResourceVersion string
watcherClient cache.Watcher
resultChan chan watch.Event
stopChan chan struct{}
doneChan chan struct{}
minRestartDelay time.Duration
}
// NewRetryWatcher creates a new RetryWatcher.
// It will make sure that watches gets restarted in case of recoverable errors.
// The initialResourceVersion will be given to watch method when first called.
func NewRetryWatcher(initialResourceVersion string, watcherClient cache.Watcher) (*RetryWatcher, error) {
return newRetryWatcher(initialResourceVersion, watcherClient, 1*time.Second)
}
func newRetryWatcher(initialResourceVersion string, watcherClient cache.Watcher, minRestartDelay time.Duration) (*RetryWatcher, error) {
switch initialResourceVersion {
case "", "0":
// TODO: revisit this if we ever get WATCH v2 where it means start "now"
// without doing the synthetic list of objects at the beginning (see #74022)
return nil, fmt.Errorf("initial RV %q is not supported due to issues with underlying WATCH", initialResourceVersion)
default:
break
}
rw := &RetryWatcher{
lastResourceVersion: initialResourceVersion,
watcherClient: watcherClient,
stopChan: make(chan struct{}),
doneChan: make(chan struct{}),
resultChan: make(chan watch.Event, 0),
minRestartDelay: minRestartDelay,
}
go rw.receive()
return rw, nil
}
func (rw *RetryWatcher) send(event watch.Event) bool {
// Writing to an unbuffered channel is blocking operation
// and we need to check if stop wasn't requested while doing so.
select {
case rw.resultChan <- event:
return true
case <-rw.stopChan:
return false
}
}
// doReceive returns true when it is done, false otherwise.
// If it is not done the second return value holds the time to wait before calling it again.
func (rw *RetryWatcher) doReceive() (bool, time.Duration) {
watcher, err := rw.watcherClient.Watch(metav1.ListOptions{
ResourceVersion: rw.lastResourceVersion,
})
// We are very unlikely to hit EOF here since we are just establishing the call,
// but it may happen that the apiserver is just shutting down (e.g. being restarted)
// This is consistent with how it is handled for informers
switch err {
case nil:
break
case io.EOF:
// watch closed normally
return false, 0
case io.ErrUnexpectedEOF:
klog.V(1).Infof("Watch closed with unexpected EOF: %v", err)
return false, 0
default:
msg := "Watch failed: %v"
if net.IsProbableEOF(err) {
klog.V(5).Infof(msg, err)
// Retry
return false, 0
}
klog.Errorf(msg, err)
// Retry
return false, 0
}
if watcher == nil {
klog.Error("Watch returned nil watcher")
// Retry
return false, 0
}
ch := watcher.ResultChan()
defer watcher.Stop()
for {
select {
case <-rw.stopChan:
klog.V(4).Info("Stopping RetryWatcher.")
return true, 0
case event, ok := <-ch:
if !ok {
klog.V(4).Infof("Failed to get event! Re-creating the watcher. Last RV: %s", rw.lastResourceVersion)
return false, 0
}
// We need to inspect the event and get ResourceVersion out of it
switch event.Type {
case watch.Added, watch.Modified, watch.Deleted, watch.Bookmark:
metaObject, ok := event.Object.(resourceVersionGetter)
if !ok {
_ = rw.send(watch.Event{
Type: watch.Error,
Object: &apierrors.NewInternalError(errors.New("retryWatcher: doesn't support resourceVersion")).ErrStatus,
})
// We have to abort here because this might cause lastResourceVersion inconsistency by skipping a potential RV with valid data!
return true, 0
}
resourceVersion := metaObject.GetResourceVersion()
if resourceVersion == "" {
_ = rw.send(watch.Event{
Type: watch.Error,
Object: &apierrors.NewInternalError(fmt.Errorf("retryWatcher: object %#v doesn't support resourceVersion", event.Object)).ErrStatus,
})
// We have to abort here because this might cause lastResourceVersion inconsistency by skipping a potential RV with valid data!
return true, 0
}
// All is fine; send the event and update lastResourceVersion
ok = rw.send(event)
if !ok {
return true, 0
}
rw.lastResourceVersion = resourceVersion
continue
case watch.Error:
// This round trip allows us to handle unstructured status
errObject := apierrors.FromObject(event.Object)
statusErr, ok := errObject.(*apierrors.StatusError)
if !ok {
klog.Error(spew.Sprintf("Received an error which is not *metav1.Status but %#+v", event.Object))
// Retry unknown errors
return false, 0
}
status := statusErr.ErrStatus
statusDelay := time.Duration(0)
if status.Details != nil {
statusDelay = time.Duration(status.Details.RetryAfterSeconds) * time.Second
}
switch status.Code {
case http.StatusGone:
// Never retry RV too old errors
_ = rw.send(event)
return true, 0
case http.StatusGatewayTimeout, http.StatusInternalServerError:
// Retry
return false, statusDelay
default:
// We retry by default. RetryWatcher is meant to proceed unless it is certain
// that it can't. If we are not certain, we proceed with retry and leave it
// up to the user to timeout if needed.
// Log here so we have a record of hitting the unexpected error
// and we can whitelist some error codes if we missed any that are expected.
klog.V(5).Info(spew.Sprintf("Retrying after unexpected error: %#+v", event.Object))
// Retry
return false, statusDelay
}
default:
klog.Errorf("Failed to recognize Event type %q", event.Type)
_ = rw.send(watch.Event{
Type: watch.Error,
Object: &apierrors.NewInternalError(fmt.Errorf("retryWatcher failed to recognize Event type %q", event.Type)).ErrStatus,
})
// We are unable to restart the watch and have to stop the loop or this might cause lastResourceVersion inconsistency by skipping a potential RV with valid data!
return true, 0
}
}
}
}
// receive reads the result from a watcher, restarting it if necessary.
func (rw *RetryWatcher) receive() {
defer close(rw.doneChan)
defer close(rw.resultChan)
klog.V(4).Info("Starting RetryWatcher.")
defer klog.V(4).Info("Stopping RetryWatcher.")
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
go func() {
select {
case <-rw.stopChan:
cancel()
return
case <-ctx.Done():
return
}
}()
// We use non sliding until so we don't introduce delays on happy path when WATCH call
// timeouts or gets closed and we need to reestablish it while also avoiding hot loops.
wait.NonSlidingUntilWithContext(ctx, func(ctx context.Context) {
done, retryAfter := rw.doReceive()
if done {
cancel()
return
}
time.Sleep(retryAfter)
klog.V(4).Infof("Restarting RetryWatcher at RV=%q", rw.lastResourceVersion)
}, rw.minRestartDelay)
}
// ResultChan implements Interface.
func (rw *RetryWatcher) ResultChan() <-chan watch.Event {
return rw.resultChan
}
// Stop implements Interface.
func (rw *RetryWatcher) Stop() {
close(rw.stopChan)
}
// Done allows the caller to be notified when Retry watcher stops.
func (rw *RetryWatcher) Done() <-chan struct{} {
return rw.doneChan
}
看一下retrywatcher有哪些屬性,多了stopChan和doneChan以及minRestartDelay(重啓時的延遲時間可以設置)
type RetryWatcher struct {
lastResourceVersion string
watcherClient cache.Watcher
resultChan chan watch.Event
stopChan chan struct{}
doneChan chan struct{}
minRestartDelay time.Duration
}
看一下新建函數,NewRetryWatcher面向外部調用,主要還是看內部的newRetryWatcher,裏面設置了minRestartDelay的時間是1秒,詳細看下newRetryWatcher方法,最後調用了go rw.receive(),我們接着看receive()方法
func NewRetryWatcher(initialResourceVersion string, watcherClient cache.Watcher) (*RetryWatcher, error) {
return newRetryWatcher(initialResourceVersion, watcherClient, 1*time.Second)
}
func newRetryWatcher(initialResourceVersion string, watcherClient cache.Watcher, minRestartDelay time.Duration) (*RetryWatcher, error) {
switch initialResourceVersion {
case "", "0":
// TODO: revisit this if we ever get WATCH v2 where it means start "now"
// without doing the synthetic list of objects at the beginning (see #74022)
return nil, fmt.Errorf("initial RV %q is not supported due to issues with underlying WATCH", initialResourceVersion)
default:
break
}
rw := &RetryWatcher{
lastResourceVersion: initialResourceVersion,
watcherClient: watcherClient,
stopChan: make(chan struct{}),
doneChan: make(chan struct{}),
resultChan: make(chan watch.Event, 0),
minRestartDelay: minRestartDelay,
}
go rw.receive()
return rw, nil
}
看一下這段代碼ctx, cancel := context.WithCancel(context.Background()),看不懂的小夥伴建議去看下go語言的context包,WithCancel函數,傳遞一個父Context作爲參數,返回子Context,以及一個取消函數用來取消Context。context專門用來簡化對於處理單個請求的多個goroutine之間與請求域的數據、取消信號、截止時間等相關操作。意思就是你在子協程中需要關閉一連串相關協程時就用這個context,調用cancel函數即可。我們接着看下面wait.NonSlidingUntilWithContext方法,去看看這個方法的註釋,代碼在下方。意思是隻要context不被done,就將循環調用其中的匿名函數。看一下匿名函數做了些啥,調用了doReceive()函數,看下doReceive函數是幹嘛的,我們可以看到,他也是使用watch方法,watch方法報錯就返回,不報錯就繼續進for循環,select語句查看當前retrywatch是否被stop,或者ch中是否有數據,若ch被關閉,就return false,0。一旦返回就看外面的receive函數處理,receive函數會判斷如果返回是true就調用cancel就真的退出,不會重建watch。只有當返回false時纔會重新回到NonSlidingUntilWithContext,循環調用匿名函數進行偵聽。所以上面我的挫比解決方案,還是太粗了。這邊的主要亮點就是wait.NonSlidingUntilWithContext方法的妙用了,建議大家學一下,還有就是,重試機制需要明確定義哪些情況是需要重試,哪些情況不需要重試!
func (rw *RetryWatcher) receive() {
defer close(rw.doneChan)
defer close(rw.resultChan)
klog.V(4).Info("Starting RetryWatcher.")
defer klog.V(4).Info("Stopping RetryWatcher.")
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
go func() {
select {
case <-rw.stopChan:
cancel()
return
case <-ctx.Done():
return
}
}()
// We use non sliding until so we don't introduce delays on happy path when WATCH call
// timeouts or gets closed and we need to reestablish it while also avoiding hot loops.
wait.NonSlidingUntilWithContext(ctx, func(ctx context.Context) {
done, retryAfter := rw.doReceive()
if done {
cancel()
return
}
time.Sleep(retryAfter)
klog.V(4).Infof("Restarting RetryWatcher at RV=%q", rw.lastResourceVersion)
}, rw.minRestartDelay)
}
// NonSlidingUntilWithContext loops until context is done, running f every#意思是除非context調用了Done,不然就會循環調用f函數
// period.
//
// NonSlidingUntilWithContext is syntactic sugar on top of JitterUntilWithContext
// with zero jitter factor, with sliding = false (meaning the timer for period
// starts at the same time as the function starts).
func NonSlidingUntilWithContext(ctx context.Context, f func(context.Context), period time.Duration) {
JitterUntilWithContext(ctx, f, period, 0.0, false)
}
doReceive函數:
func (rw *RetryWatcher) doReceive() (bool, time.Duration) {
watcher, err := rw.watcherClient.Watch(metav1.ListOptions{
ResourceVersion: rw.lastResourceVersion,
})//打開watch
// We are very unlikely to hit EOF here since we are just establishing the call,
// but it may happen that the apiserver is just shutting down (e.g. being restarted)
// This is consistent with how it is handled for informers
switch err {
case nil:
break
case io.EOF:
// watch closed normally
return false, 0
case io.ErrUnexpectedEOF:
klog.V(1).Infof("Watch closed with unexpected EOF: %v", err)
return false, 0
default:
msg := "Watch failed: %v"
if net.IsProbableEOF(err) {
klog.V(5).Infof(msg, err)
// Retry
return false, 0
}
klog.Errorf(msg, err)
// Retry
return false, 0
}
if watcher == nil {
klog.Error("Watch returned nil watcher")
// Retry
return false, 0
}
ch := watcher.ResultChan()
defer watcher.Stop()
###########這裏很重要!這裏很重要,這裏很重要
for {
select {
case <-rw.stopChan://查看是否被停止了
klog.V(4).Info("Stopping RetryWatcher.")
return true, 0
case event, ok := <-ch://從通道拿出數據
if !ok {//通道是不是開着的,關閉的話,就返回
klog.V(4).Infof("Failed to get event! Re-creating the watcher. Last RV: %s", rw.lastResourceVersion)
return false, 0
}
// We need to inspect the event and get ResourceVersion out of it
switch event.Type {//下面是成功獲取到數據的邏輯
case watch.Added, watch.Modified, watch.Deleted, watch.Bookmark:
metaObject, ok := event.Object.(resourceVersionGetter)
if !ok {
_ = rw.send(watch.Event{
Type: watch.Error,
Object: &apierrors.NewInternalError(errors.New("retryWatcher: doesn't support resourceVersion")).ErrStatus,
})
// We have to abort here because this might cause lastResourceVersion inconsistency by skipping a potential RV with valid data!
return true, 0
}
resourceVersion := metaObject.GetResourceVersion()
if resourceVersion == "" {
_ = rw.send(watch.Event{
Type: watch.Error,
Object: &apierrors.NewInternalError(fmt.Errorf("retryWatcher: object %#v doesn't support resourceVersion", event.Object)).ErrStatus,
})
// We have to abort here because this might cause lastResourceVersion inconsistency by skipping a potential RV with valid data!
return true, 0
}
// All is fine; send the event and update lastResourceVersion
ok = rw.send(event)
if !ok {
return true, 0
}
rw.lastResourceVersion = resourceVersion
continue
case watch.Error:
// This round trip allows us to handle unstructured status
errObject := apierrors.FromObject(event.Object)
statusErr, ok := errObject.(*apierrors.StatusError)
if !ok {
klog.Error(spew.Sprintf("Received an error which is not *metav1.Status but %#+v", event.Object))
// Retry unknown errors
return false, 0
}
status := statusErr.ErrStatus
statusDelay := time.Duration(0)
if status.Details != nil {
statusDelay = time.Duration(status.Details.RetryAfterSeconds) * time.Second
}
switch status.Code {
case http.StatusGone:
// Never retry RV too old errors
_ = rw.send(event)
return true, 0
case http.StatusGatewayTimeout, http.StatusInternalServerError:
// Retry
return false, statusDelay
default:
// We retry by default. RetryWatcher is meant to proceed unless it is certain
// that it can't. If we are not certain, we proceed with retry and leave it
// up to the user to timeout if needed.
// Log here so we have a record of hitting the unexpected error
// and we can whitelist some error codes if we missed any that are expected.
klog.V(5).Info(spew.Sprintf("Retrying after unexpected error: %#+v", event.Object))
// Retry
return false, statusDelay
}
default:
klog.Errorf("Failed to recognize Event type %q", event.Type)
_ = rw.send(watch.Event{
Type: watch.Error,
Object: &apierrors.NewInternalError(fmt.Errorf("retryWatcher failed to recognize Event type %q", event.Type)).ErrStatus,
})
// We are unable to restart the watch and have to stop the loop or this might cause lastResourceVersion inconsistency by skipping a potential RV with valid data!
return true, 0
}
}
}
}