我們通過 kubectl describe [資源]
包，我們完全可以在寫入自定義的event。
現在讓我們來一步步揭開event的面紗。
Event定義
其實event也是一個資源對象，并且通過apiserver将event存儲在etcd中，是以我們也可以通過
kubectl get event
指令檢視對應的event對象。
以下是一個event的yaml檔案：
apiVersion: v1
count: 1
eventTime: null
firstTimestamp: "2020-03-02T13:08:22Z"
involvedObject:
apiVersion: v1
kind: Pod
name: example-foo-d75d8587c-xsf64
namespace: default
resourceVersion: "429837"
uid: ce611c62-6c1a-4bd8-9029-136a1adf7de4
kind: Event
lastTimestamp: "2020-03-02T13:08:22Z"
message: Pod sandbox changed, it will be killed and re-created.
metadata:
creationTimestamp: "2020-03-02T13:08:30Z"
name: example-foo-d75d8587c-xsf64.15f87ea1df862b64
namespace: default
resourceVersion: "479466"
selfLink: /api/v1/namespaces/default/events/example-foo-d75d8587c-xsf64.15f87ea1df862b64
uid: 9fe6f72a-341d-4c49-960b-e185982d331a
reason: SandboxChanged
reportingComponent: ""
reportingInstance: ""
source:
component: kubelet
host: minikube
type: Normal 主要字段說明：**
- involvedObject： 觸發event的資源類型
- lastTimestamp：最後一次觸發的時間
- message：事件說明
- metadata :event的元資訊，name，namespace等
- reason：event的原因
- source：上報事件的來源，比如kubelet中的某個節點
- type:事件類型，Normal或Warning
event字段定義可以看這裡：
types.go#L5078接下來我們來看看，整個event是如何下入的。
寫入事件
1、這裡以kubelet為例，看看是如何進行事件寫入的
2、文中代碼以Kubernetes 1.17.3為例進行分析
先以一幅圖來看下整個的處理流程
建立操作事件的用戶端：
kubelet/app/server.go#L461// makeEventRecorder sets up kubeDeps.Recorder if it's nil. It's a no-op otherwise.
func makeEventRecorder(kubeDeps *kubelet.Dependencies, nodeName types.NodeName) {
if kubeDeps.Recorder != nil {
return
}
//事件廣播
eventBroadcaster := record.NewBroadcaster()
//建立EventRecorder
kubeDeps.Recorder = eventBroadcaster.NewRecorder(legacyscheme.Scheme, v1.EventSource{Component: componentKubelet, Host: string(nodeName)})
//發送event至log輸出
eventBroadcaster.StartLogging(klog.V(3).Infof)
if kubeDeps.EventClient != nil {
klog.V(4).Infof("Sending events to api server.")
//發送event至apiserver
eventBroadcaster.StartRecordingToSink(&v1core.EventSinkImpl{Interface: kubeDeps.EventClient.Events("")})
} else {
klog.Warning("No api server defined - no events will be sent to API server.")
}
} 通過
makeEventRecorder
建立了
EventRecorder
執行個體，這是一個事件廣播器，通過它提供了StartLogging和StartRecordingToSink兩個事件處理函數，分别将event發送給log和apiserver。
NewRecorder
EventRecorder
的執行個體，它提供了
Event
，
Eventf
等方法供事件記錄。
EventBroadcaster
我們來看下EventBroadcaster接口定義：
event.go#L113// EventBroadcaster knows how to receive events and send them to any EventSink, watcher, or log.
type EventBroadcaster interface {
//
StartEventWatcher(eventHandler func(*v1.Event)) watch.Interface
StartRecordingToSink(sink EventSink) watch.Interface
StartLogging(logf func(format string, args ...interface{})) watch.Interface
NewRecorder(scheme *runtime.Scheme, source v1.EventSource) EventRecorder
Shutdown()
} 具體實作是通過
eventBroadcasterImplstruct來實作了各個方法。
其中StartLogging 和 StartRecordingToSink 其實就是完成了對事件的消費，EventRecorder實作對事件的寫入，中間通過channel實作了生産者消費者模型。
EventRecorder
EventRecorder
我們先來看下
EventRecorder
接口定義：
event.go#L88，提供了一下4個方法
// EventRecorder knows how to record events on behalf of an EventSource.
type EventRecorder interface {
// Event constructs an event from the given information and puts it in the queue for sending.
// 'object' is the object this event is about. Event will make a reference-- or you may also
// pass a reference to the object directly.
// 'type' of this event, and can be one of Normal, Warning. New types could be added in future
// 'reason' is the reason this event is generated. 'reason' should be short and unique; it
// should be in UpperCamelCase format (starting with a capital letter). "reason" will be used
// to automate handling of events, so imagine people writing switch statements to handle them.
// You want to make that easy.
// 'message' is intended to be human readable.
//
// The resulting event will be created in the same namespace as the reference object.
Event(object runtime.Object, eventtype, reason, message string)
// Eventf is just like Event, but with Sprintf for the message field.
Eventf(object runtime.Object, eventtype, reason, messageFmt string, args ...interface{})
// PastEventf is just like Eventf, but with an option to specify the event's 'timestamp' field.
PastEventf(object runtime.Object, timestamp metav1.Time, eventtype, reason, messageFmt string, args ...interface{})
// AnnotatedEventf is just like eventf, but with annotations attached
AnnotatedEventf(object runtime.Object, annotations map[string]string, eventtype, reason, messageFmt string, args ...interface{})
}
主要參數說明：
-
object
involvedObject
-
eventtype
-
reason
-
message
我們來看下當我們調用
Event(object runtime.Object, eventtype, reason, message string)
的整個過程。
發現最終都調用到了
generateEvent
方法：
event.go#L316func (recorder *recorderImpl) generateEvent(object runtime.Object, annotations map[string]string, timestamp metav1.Time, eventtype, reason, message string) {
.....
event := recorder.makeEvent(ref, annotations, eventtype, reason, message)
event.Source = recorder.source
go func() {
// NOTE: events should be a non-blocking operation
defer utilruntime.HandleCrash()
recorder.Action(watch.Added, event)
}()
} 最終事件在一個
goroutine
中通過調用
recorder.Action
進入處理，這裡保證了每次調用event方法都是非阻塞的。
其中
makeEvent
的作用主要是構造了一個event對象，事件name根據InvolvedObject中的name加上時間戳生成：
注意看：對于一些非namespace資源産生的event，event的namespace是default
func (recorder *recorderImpl) makeEvent(ref *v1.ObjectReference, annotations map[string]string, eventtype, reason, message string) *v1.Event {
t := metav1.Time{Time: recorder.clock.Now()}
namespace := ref.Namespace
if namespace == "" {
namespace = metav1.NamespaceDefault
}
return &v1.Event{
ObjectMeta: metav1.ObjectMeta{
Name: fmt.Sprintf("%v.%x", ref.Name, t.UnixNano()),
Namespace: namespace,
Annotations: annotations,
},
InvolvedObject: *ref,
Reason: reason,
Message: message,
FirstTimestamp: t,
LastTimestamp: t,
Count: 1,
Type: eventtype,
}
} 進一步跟蹤
Action
方法，
apimachinery/blob/master/pkg/watch/mux.go#L188:23// Action distributes the given event among all watchers.
func (m *Broadcaster) Action(action EventType, obj runtime.Object) {
m.incoming <- Event{action, obj}
} 将event寫入到了一個channel裡面。
注意：
這個Action方式是
apimachinery包中的方法，因為實作的sturt
recorderImpl
将
*watch.Broadcaster
作為一個匿名struct，并且在
NewRecorder
進行
Broadcaster
指派，這個
Broadcaster
其實就是
eventBroadcasterImpl
中的
Broadcaster
。
到此，基本清楚了event最終被寫入到了
Broadcaster
incoming
channel中，下面看下是怎麼進行消費的。
消費事件
在
makeEventRecorder
調用的
StartLogging
和
StartRecordingToSink
其實就是完成了對事件的消費。
-
StartLogging
-
StartRecordingToSink
兩個方法内部都調用了
StartEventWatcher
方法，并且傳入一個
eventHandler
方法對event進行處理
func (e *eventBroadcasterImpl) StartEventWatcher(eventHandler func(*v1.Event)) watch.Interface {
watcher := e.Watch()
go func() {
defer utilruntime.HandleCrash()
for watchEvent := range watcher.ResultChan() {
event, ok := watchEvent.Object.(*v1.Event)
if !ok {
// This is all local, so there's no reason this should
// ever happen.
continue
}
eventHandler(event)
}
}()
return watcher
} watcher.ResultChan
方法就拿到了事件，這裡是在一個goroutine中通過
func (m *Broadcaster) loop()==>
func (m *Broadcaster) distribute(event Event)方法調用将event又寫入了
broadcasterWatcher.result主要看下
StartRecordingToSink
提供的的
eventHandler
recordToSink
func recordToSink(sink EventSink, event *v1.Event, eventCorrelator *EventCorrelator, 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 {
klog.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.
// 第一次重試增加随機性，防止 apiserver 重新開機的時候所有的事件都在同一時間發送事件
if tries == 1 {
time.Sleep(time.Duration(float64(sleepDuration) * rand.Float64()))
} else {
time.Sleep(sleepDuration)
}
}
} 其中event被經過了一個
eventCorrelator.EventCorrelate(event)
方法做預處理，主要是聚合相同的事件（避免産生的事件過多，增加 etcd 和 apiserver 的壓力，也會導緻檢視 pod 事件很不清晰）
下面一個for循環就是在進行重試，最大重試次數是12次，調用
recordEvent
方法才真正将event寫入到了apiserver。
事件處理
我們來看下
EventCorrelate
// EventCorrelate filters, aggregates, counts, and de-duplicates all incoming events
func (c *EventCorrelator) EventCorrelate(newEvent *v1.Event) (*EventCorrelateResult, error) {
if newEvent == nil {
return nil, fmt.Errorf("event is nil")
}
aggregateEvent, ckey := c.aggregator.EventAggregate(newEvent)
observedEvent, patch, err := c.logger.eventObserve(aggregateEvent, ckey)
if c.filterFunc(observedEvent) {
return &EventCorrelateResult{Skip: true}, nil
}
return &EventCorrelateResult{Event: observedEvent, Patch: patch}, err
}
分别調用了
aggregator.EventAggregate
logger.eventObserve
filterFunc
三個方法，分别作用是：
-
aggregator.EventAggregate
(combined from similar events)+event.Message
-
logger.eventObserve
aggregator
Count
-
filterFunc
我們主要來看下
aggregator.EventAggregate
func (e *EventAggregator) EventAggregate(newEvent *v1.Event) (*v1.Event, string) {
now := metav1.NewTime(e.clock.Now())
var record aggregateRecord
// eventKey is the full cache key for this event
//eventKey 是将除了時間戳外所有字段結合在一起
eventKey := getEventKey(newEvent)
// aggregateKey is for the aggregate event, if one is needed.
//aggregateKey 是除了message和時間戳外的字段結合在一起，localKey 是message
aggregateKey, localKey := e.keyFunc(newEvent)
// Do we have a record of similar events in our cache?
e.Lock()
defer e.Unlock()
//從cache中根據aggregateKey查詢是否存在，如果是相同或者相類似的事件會被放入cache中
value, found := e.cache.Get(aggregateKey)
if found {
record = value.(aggregateRecord)
}
//判斷上次事件産生的時間是否超過10分鐘，如何操作則重新生成一個localKeys集合（集合中存放message）
maxInterval := time.Duration(e.maxIntervalInSeconds) * time.Second
interval := now.Time.Sub(record.lastTimestamp.Time)
if interval > maxInterval {
record = aggregateRecord{localKeys: sets.NewString()}
}
// Write the new event into the aggregation record and put it on the cache
//将locakKey也就是message放入集合中，如果message相同就是覆寫了
record.localKeys.Insert(localKey)
record.lastTimestamp = now
e.cache.Add(aggregateKey, record)
// If we are not yet over the threshold for unique events, don't correlate them
//判斷localKeys集合中存放的類似事件是否超過10個，
if uint(record.localKeys.Len()) < e.maxEvents {
return newEvent, eventKey
}
// do not grow our local key set any larger than max
record.localKeys.PopAny()
// create a new aggregate event, and return the aggregateKey as the cache key
// (so that it can be overwritten.)
eventCopy := &v1.Event{
ObjectMeta: metav1.ObjectMeta{
Name: fmt.Sprintf("%v.%x", newEvent.InvolvedObject.Name, now.UnixNano()),
Namespace: newEvent.Namespace,
},
Count: 1,
FirstTimestamp: now,
InvolvedObject: newEvent.InvolvedObject,
LastTimestamp: now,
//這裡會對message加個字首：(combined from similar events):
Message: e.messageFunc(newEvent),
Type: newEvent.Type,
Reason: newEvent.Reason,
Source: newEvent.Source,
}
return eventCopy, aggregateKey
} aggregator.EventAggregate
方法中其實就是判斷了通過cache和localKeys判斷事件是否相似，如果最近 10 分鐘出現過 10 個相似的事件就合并并加上字首，後續通過
logger.eventObserve
方法進行count累加，如果message也相同，肯定就是直接count++。
總結
好了，event處理的整個流程基本就是這樣，我們可以概括一下，可以結合文中的圖對比一起看下：
- 建立
EventRecorder
Event
- 将建立出來的對象發送給
EventBroadcaster
-
EventBroadcaster
- 當輸出log的handler收到事件就直接列印事件
- 當
EventSink
- 其中預處理包含三個動作，1、限流 2、聚合 3、計數
- apiserver收到事件處理之後就存儲在etcd中
回顧event的整個流程，可以看到event并不是保證100%事件寫入（從預處理的過程來看），這樣做是為了後端服務etcd的可用性，因為event事件在整個叢集中産生是非常頻繁的，尤其在服務不穩定的時候，而相比Deployment,Pod等其他資源，又沒那麼的重要。是以這裡做了個取舍。
參考文檔：
關注公衆号，擷取最新文章推送