kubernetes 1.8 源码分析之kubelet资源上报

原文地址为： kubernetes 1.8 源码分析之kubelet资源上报

最近在给k8s 1.8添加磁盘调度的时候，又撸了一遍k8s的相关的源码，之前看过我源码分析的那个是k8s 1.4版本的，1.8版本还是有点变化的。
先看看kubelet上报机制，这个是kubelet的心跳机制，定时上报节点状态，默认是10s，时间可以设定
在kubelet创建的时候就已经创建了节点状态的收集方法
pkg/kubelet/kubelet.go里面NewMainKubelet方法

klet.setNodeStatusFuncs = klet.defaultNodeStatusFuncs()

那默认的方法有哪些呢？

// defaultNodeStatusFuncs is a factory that generates the default set of
// setNodeStatus funcs
func (kl *Kubelet) defaultNodeStatusFuncs() []func(*v1.Node) error {
    // initial set of node status update handlers, can be modified by Option's
    withoutError := func(f func(*v1.Node)) func(*v1.Node) error {
        return func(n *v1.Node) error {
            f(n)
            return nil
        }
    }
    return []func(*v1.Node) error{
        kl.setNodeAddress,
        withoutError(kl.setNodeStatusInfo),
        withoutError(kl.setNodeOODCondition),
        withoutError(kl.setNodeMemoryPressureCondition),
        withoutError(kl.setNodeDiskPressureCondition),
        withoutError(kl.setNodeDiskMountPressureCondition),
        withoutError(kl.setNodeReadyCondition),
        withoutError(kl.setNodeVolumesInUseStatus),
        withoutError(kl.recordNodeSchedulableEvent),
    }
}

注意上面的setNodeDiskMountPressureCondition这个是我自己添加上面的，不是k8s自带。这个里面包括节点信息、节点OOD信息、内存磁盘压力状态、节点监控状态等。
方法太多，那一个举例子吧setNodeDiskPressureCondition

func (kl *Kubelet) setNodeDiskPressureCondition(node *v1.Node) {
    currentTime := metav1.NewTime(kl.clock.Now())
    var condition *v1.NodeCondition

    // Check if NodeDiskPressure condition already exists and if it does, just pick it up for update.
    for i := range node.Status.Conditions {
        if node.Status.Conditions[i].Type == v1.NodeDiskPressure {
            condition = &node.Status.Conditions[i]
        }
    }

    newCondition := false
    // If the NodeDiskPressure condition doesn't exist, create one
    if condition == nil {
        condition = &v1.NodeCondition{
            Type:   v1.NodeDiskPressure,
            Status: v1.ConditionUnknown,
        }
        // cannot be appended to node.Status.Conditions here because it gets
        // copied to the slice. So if we append to the slice here none of the
        // updates we make below are reflected in the slice.
        newCondition = true
    }

    // Update the heartbeat time
    condition.LastHeartbeatTime = currentTime

    // Note: The conditions below take care of the case when a new NodeDiskPressure condition is
    // created and as well as the case when the condition already exists. When a new condition
    // is created its status is set to v1.ConditionUnknown which matches either
    // condition.Status != v1.ConditionTrue or
    // condition.Status != v1.ConditionFalse in the conditions below depending on whether
    // the kubelet is under disk pressure or not.
    if kl.evictionManager.IsUnderDiskPressure() {
        if condition.Status != v1.ConditionTrue {
            condition.Status = v1.ConditionTrue
            condition.Reason = "KubeletHasDiskPressure"
            condition.Message = "kubelet has disk pressure"
            condition.LastTransitionTime = currentTime
            kl.recordNodeStatusEvent(v1.EventTypeNormal, "NodeHasDiskPressure")
        }
    } else {
        if condition.Status != v1.ConditionFalse {
            condition.Status = v1.ConditionFalse
            condition.Reason = "KubeletHasNoDiskPressure"
            condition.Message = "kubelet has no disk pressure"
            condition.LastTransitionTime = currentTime
            kl.recordNodeStatusEvent(v1.EventTypeNormal, "NodeHasNoDiskPressure")
        }
    }

    if newCondition {
        node.Status.Conditions = append(node.Status.Conditions, *condition)
    }
}

方法很简单，核心就是IsUnderDiskPressure，如果是存在压力就把状态改成true，否则改为false。
看完上报的方法，那这个方法是什么是被调用的呢？本能就可以想到，一个是初始一个是定时的更新，下面看看具体是不是这样的，先看初始pkg/kubelet/kubelet_node_status.go里面的initialNode方法

kl.setNodeStatus(node)

这个方法就是调用刚才上面一系列收集状态的函数

func (kl *Kubelet) setNodeStatus(node *v1.Node) {
    for _, f := range kl.setNodeStatusFuncs {
        if err := f(node); err != nil {
            glog.Warningf("Failed to set some node status fields: %s", err)
        }
    }
}

然后就是定时上报，pkg/kubelet/kubelet_node_status.go里面tryUpdateNodeStatus方法，它是定时上报master的方法。

kl.setNodeStatus(node)

它同样调用setNodeStatus这个方法。顺便在多说一句，这个定时上报的方法什么时候被定时执行呢，答案在pkg/kubelet/kubelet.go的Run里面

if kl.kubeClient != nil {
        // Start syncing node status immediately, this may set up things the runtime needs to run.
        go wait.Until(kl.syncNodeStatus, kl.nodeStatusUpdateFrequency, wait.NeverStop)
    }

通过nodeStatusUpdateFrequency指定上报周期，好了关于资源上报的代码很简单，就不赘述了。

转载请注明本文地址： kubernetes 1.8 源码分析之kubelet资源上报