Koordinator-NodeMetricInformer

初始化

我们看下nodeMetricInformer是干嘛用的。

nodeMetricInformer在stateInformer一开始的时候作为一个Plugin进行了初始化。

var DefaultPluginRegistry = map[PluginName]informerPlugin{
    nodeSLOInformerName:    NewNodeSLOInformer(),
    pvcInformerName:        NewPVCInformer(),
    nodeTopoInformerName:   NewNodeTopoInformer(),
    nodeInformerName:       NewNodeInformer(),
    podsInformerName:       NewPodsInformer(),
    nodeMetricInformerName: NewNodeMetricInformer(),
}

func NewNodeMetricInformer() *nodeMetricInformer {
    return &nodeMetricInformer{}
}

继续往下看，stateInformer启动plugin

func (s *statesInformer) startPlugins(stopCh <-chan struct{}) {
    for name, p := range s.states.informerPlugins {
        klog.V(4).Infof("starting informer plugin %v", name)
        go p.Start(stopCh)
    }
}

启动Start()

nodeMetricInformer也在这里进行了Start()

• nodeMetricInformer.Run(stopCh) 就不用说了，正常的informer流程。
• syncNodeMetricWorker(stopCh) 从名称上来看是进行同步nodeMetric数据，怎样同步？同步到哪？

func (r *nodeMetricInformer) Start(stopCh <-chan struct{}) {
    defer utilruntime.HandleCrash()
    klog.Infof("starting nodeMetricInformer")
    if !r.reportEnabled {
        klog.Infof("node metric report is disabled.")
        return
    }
    go r.nodeMetricInformer.Run(stopCh)
    if !cache.WaitForCacheSync(stopCh, r.nodeMetricInformer.HasSynced, r.podsInformer.HasSynced) {
        klog.Errorf("timed out waiting for node metric caches to sync")
    }
    go r.syncNodeMetricWorker(stopCh)
    klog.Info("start nodeMetricInformer successfully")
    <-stopCh
    klog.Info("shutting down nodeMetricInformer daemon")
}

syncNodeMetricWorker()

• 先获取下上报的interval，然后进入一个for循环。
• stopCh有数据则跳过for循环，否则每过interval后进行调用sync()方法

func (r *nodeMetricInformer) syncNodeMetricWorker(stopCh <-chan struct{}) {
    reportInterval := r.getNodeMetricReportInterval()
    for {
        select {
        case <-stopCh:
            return
        case <-time.After(reportInterval):
            r.sync()
            reportInterval = r.getNodeMetricReportInterval()
        }
    }
}

sync()

• 核心方法collectMetrics()，采集nodeMetricInfo, podMetricInfo, prodReclaimableMetric

• 把采集的数据放入nodeMetricStatus，node对应的更新NodeMetric CR

• • 这个有点要注意，若没有找到对应NodeMetric直接返回。

func (r *nodeMetricInformer) sync() {
    if !r.isNodeMetricInited() {
        klog.Warningf("node metric has not initialized, skip this round.")
        return
    }
    nodeMetricInfo, podMetricInfo, prodReclaimableMetric := r.collectMetric()
    if nodeMetricInfo == nil {
        klog.Warningf("node metric is not ready, skip this round.")
        return
    }
    newStatus := &slov1alpha1.NodeMetricStatus{
        UpdateTime:            &metav1.Time{Time: time.Now()},
        NodeMetric:            nodeMetricInfo,
        PodsMetric:            podMetricInfo,
        ProdReclaimableMetric: prodReclaimableMetric,
    }
    retErr := retry.RetryOnConflict(retry.DefaultBackoff, func() error {
        nodeMetric, err := r.nodeMetricLister.Get(r.nodeName)
        if errors.IsNotFound(err) {
            klog.Warningf("nodeMetric %v not found, skip", r.nodeName)
            return nil
        } else if err != nil {
            klog.Warningf("failed to get %s nodeMetric: %v", r.nodeName, err)
            return err
        }
        err = r.statusUpdater.updateStatus(nodeMetric, newStatus)
        return err
    })
    if retErr != nil {
        klog.Warningf("update node metric status failed, status %v, err %v", util.DumpJSON(newStatus), retErr)
    } else {
        klog.V(4).Infof("update node metric status success, detail: %v", util.DumpJSON(newStatus))
    }
}

再看下nodeMetric的数据结构

type NodeMetric struct {
	metav1.TypeMeta   `json:",inline"`
	metav1.ObjectMeta `json:"metadata,omitempty"`

	Spec   NodeMetricSpec   `json:"spec,omitempty"`
	Status NodeMetricStatus `json:"status,omitempty"`
}
type NodeMetricStatus struct {
	// UpdateTime is the last time this NodeMetric was updated.
	UpdateTime *metav1.Time `json:"updateTime,omitempty"`

	// NodeMetric contains the metrics for this node.
	NodeMetric *NodeMetricInfo `json:"nodeMetric,omitempty"`

	// PodsMetric contains the metrics for pods belong to this node.
	PodsMetric []*PodMetricInfo `json:"podsMetric,omitempty"`

	// HostApplicationMetric contains the metrics of out-out-band applications on node.
	HostApplicationMetric []*HostApplicationMetricInfo `json:"hostApplicationMetric,omitempty"`

	// ProdReclaimableMetric is the indicator statistics of Prod type resources reclaimable
	ProdReclaimableMetric *ReclaimableMetric `json:"prodReclaimableMetric,omitempty"`
}