分析kubernetes中的事件機制

咱們經過 kubectl describe [資源] 命令，能夠在看到Event輸出，而且常常依賴event進行問題定位，從event中能夠分析整個POD的運行軌跡，爲服務的客觀測性提供數據來源，因而可知，event在Kubernetes中起着舉足輕重的做用。

event並不僅是kubelet中都有的，關於event的操做被封裝在client-go/tools/record包，咱們徹底能夠在寫入自定義的event。

如今讓咱們來一步步揭開event的面紗。

Event定義

其實event也是一個資源對象，而且經過apiserver將event存儲在etcd中，因此咱們也能夠經過 kubectl get event 命令查看對應的event對象。

如下是一個event的yaml文件：

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的元信息，name，namespace等
• reason：event的緣由
• source：上報事件的來源，好比kubelet中的某個節點
• type:事件類型，Normal或Warning

event字段定義能夠看這裏：types.go#L5078

接下來咱們來看看，整個event是如何下入的。

寫入事件

一、這裏以kubelet爲例，看看是如何進行事件寫入的

二、文中代碼以Kubernetes 1.17.3爲例進行分析

先以一幅圖來看下整個的處理流程

建立操做事件的客戶端：
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.")
    }
}

經過 makeEventRecorder 建立了 EventRecorder 實例，這是一個事件廣播器，經過它提供了StartLogging和StartRecordingToSink兩個事件處理函數，分別將event發送給log和apiserver。
NewRecorder建立了 EventRecorder 的實例，它提供了 Event ，Eventf 等方法供事件記錄。

EventBroadcaster

咱們來看下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()
}

具體實現是經過 eventBroadcasterImpl struct來實現了各個方法。

其中StartLogging 和 StartRecordingToSink 其實就是完成了對事件的消費，EventRecorder實現對事件的寫入，中間經過channel實現了生產者消費者模型。

EventRecorder

咱們先來看下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 對應event資源定義中的 involvedObject
• eventtype 對應event資源定義中的type，可選Normal，Warning.
• reason ：事件緣由
• message ：事件消息

咱們來看下當咱們調用 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#L316

func (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 中經過調用 recorder.Action 進入處理，這裏保證了每次調用event方法都是非阻塞的。
其中 makeEvent 的做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：對於一些非namespace資源產生的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}
}

將event寫入到了一個channel裏面。
注意：
這個Action方式是apimachinery包中的方法，由於實現的sturt recorderImpl
將 *watch.Broadcaster 做爲一個匿名struct，而且在 NewRecorder 進行 Broadcaster 賦值，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 中的 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直接將event輸出到日誌
• StartRecordingToSink將事件寫入到apiserver

兩個方法內部都調用了 StartEventWatcher 方法，而且傳入一個 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
}

其中 watcher.ResultChan 方法就拿到了事件，這裏是在一個goroutine中經過func (m *Broadcaster) loop() ==>func (m *Broadcaster) distribute(event Event) 方法調用將event又寫入了broadcasterWatcher.result

主要看下 StartRecordingToSink 提供的的eventHandler， recordToSink 方法：

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.
        // 第一次重試增長隨機性，防止 apiserver 重啓的時候全部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被通過了一個 eventCorrelator.EventCorrelate(event) 方法作預處理，主要是聚合相同的事件（避免產生的事件過多，增長 etcd 和 apiserver 的壓力，也會致使查看 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 方法才真正將event寫入到了apiserver。

事件處理

咱們來看下EventCorrelate方法：

// 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
}

分別調用了 aggregator.EventAggregate ， logger.eventObserve ， filterFunc 三個方法，分別做用是：

1. aggregator.EventAggregate：聚合event，若是在最近 10 分鐘出現過 10 個類似的事件（除了 message 和時間戳以外其餘關鍵字段都相同的事件），aggregator 會把它們的 message 設置爲 (combined from similar events)+event.Message
2. logger.eventObserve：它會把相同的事件以及包含 aggregator 被聚合了的類似的事件，經過增長 Count 字段來記錄事件發生了多少次。
3. filterFunc: 這裏實現了一個基於令牌桶的限流算法，若是超過設定的速率則丟棄，保證了apiserver的安全。

咱們主要來看下aggregator.EventAggregate方法：

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 是將除了時間戳外全部字段結合在一塊兒
    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塊兒，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，若是是相同或者相相似的事件會被放入cache中
    value, found := e.cache.Get(aggregateKey)
    if found {
        record = value.(aggregateRecord)
    }

    //判斷上次事件產生的時間是否超過10分鐘，如何操做則從新生成一個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
    //將locakKey也就是message放入集合中，若是message相同就是覆蓋了
    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集合中存放的相似事件是否超過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,
        //這裏會對message加個前綴：(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經過cache和localKeys判斷事件是否類似，若是最近 10 分鐘出現過 10 個類似的事件就合併並加上前綴，後續經過logger.eventObserve方法進行count累加，若是message也相同，確定就是直接count++。

總結

好了，event處理的整個流程基本就是這樣，咱們能夠歸納一下，能夠結合文中的圖對比一塊兒看下：

1. 建立 EventRecorder 對象，經過其提供的 Event 等方法，建立好event對象
2. 將建立出來的對象發送給 EventBroadcaster 中的channel中
3. EventBroadcaster 經過後臺運行的goroutine，從管道中取出事件，並廣播給提早註冊好的handler處理
4. 當輸出log的handler收到事件就直接打印事件
5. 當 EventSink handler收處處理事件就經過預處理以後將事件發送給apiserver
6. 其中預處理包含三個動做，一、限流 二、聚合 三、計數
7. apiserver收到事件處理以後就存儲在etcd中

回顧event的整個流程，能夠看到event並非保證100%事件寫入（從預處理的過程來看），這樣作是爲了後端服務etcd的可用性，由於event事件在整個集羣中產生是很是頻繁的，尤爲在服務不穩定的時候，而相比Deployment,Pod等其餘資源，又沒那麼的重要。因此這裏作了個取捨。

參考文檔：

• https://cizixs.com/2017/06/22/kubelet-source-code-analysis-part4-event/

原文地址：https://silenceper.com/blog/2...

關注"學點程序"公衆號，瞭解更多幹貨內容 !