本节重点介绍 :
- 引入k8s sdk获取k8s 的node和pod信息
- 定义相关metrics
- 初始化k8s-client
- 使用k8s-client get node
- 使用k8s-client get pod
- 打点
k8s中关注四大块指标总结
- 之前在k8s中关注4块指标有过总结
| 指标类型 | 采集源 | 应用举例 | 发现类型 | grafana截图 |
|---|---|---|---|---|
| 容器基础资源指标 | kubelet 内置cadvisor metrics接口 | 查看容器cpu、mem利用率等 | k8s_sd node级别直接访问node_ip | 容器基础资源 |
| k8s对象资源指标 | kube-stats-metrics (简称ksm) | 具体可以看 看pod状态如pod waiting状态的原因 数个数如：查看node pod按namespace分布情况 | 通过coredns访问域名 | k8s对象资源指标 |
| k8s服务组件指标 | 服务组件 metrics接口 | 查看apiserver 、scheduler、etc、coredns请求延迟等 | k8s_sd endpoint级别 | k8s服务组件指标 |
| 部署在pod中业务埋点指标 | pod 的metrics接口 | 依据业务指标场景 | k8s_sd pod级别，访问pod ip的metricspath |
使用golang引入sdk编写一个项目跑在k8s中
需求分析
- 编写一个go的项目，引用k8s的sdk 获取节点信息，获取pod信息
- 将获取到的信息通过prometheus sdk打点打出来
- 编写dockerfile 将该项目打成镜像
- 编写k8s 的yaml运行改项目
- prometheus采集该项目的pod指标
新建项目 ink8s-pod-metrics
- go 1.16以上，初始化项目
go mod init ink8s-pod-metrics
编写go代码
1. 定义相关metrics
const (namespace = "ink8s_pod_metrics"getNode = "get_node"getPod = "get_pod"
)var (// 将每个node的信息打印出来k8sNodeDetail = prometheus.NewGaugeVec(prometheus.GaugeOpts{Name: prometheus.BuildFQName(namespace, getNode, "detail"),Help: "k8s node detail each",}, []string{"ip", "hostname", "containerRuntimeVersion", "kubeletVersion"})// 计算获取节点的耗时getNodeDuration = prometheus.NewGauge(prometheus.GaugeOpts{Name: prometheus.BuildFQName(namespace, getNode, "last_duration_seconds"),Help: "get node last duration seconds",})// 将每个控制平面的pod信息打印出来k8sPodDetail = prometheus.NewGaugeVec(prometheus.GaugeOpts{Name: prometheus.BuildFQName(namespace, getPod, "control_plane_pod_detail"),Help: "k8s pod detail of control plane",}, []string{"ip", "pod_name", "component"})// 计算获取pod的耗时getPodDuration = prometheus.NewGauge(prometheus.GaugeOpts{Name: prometheus.BuildFQName(namespace, getPod, "last_duration_seconds"),Help: "get pod last duration seconds",})
)- metrics讲解
- k8sNodeDetail 将每个node的信息打印出来
- getNodeDuration 计算获取节点的耗时
- k8sPodDetail 将每个控制平面的pod信息打印出来
- getPodDuration 计算获取pod的耗时
prometheus.BuildFQName(namespace, getNode, "detail")代表使用共同前缀，namespace + subsystem
2. 注册metrics
func newMetrics() {prometheus.DefaultRegisterer.MustRegister(k8sNodeDetail)prometheus.DefaultRegisterer.MustRegister(k8sPodDetail)prometheus.DefaultRegisterer.MustRegister(getNodeDuration)prometheus.DefaultRegisterer.MustRegister(getPodDuration)
}
3. 初始化k8s-client
- 使用包 “k8s.io/client-go/kubernetes”
- 使用包 “k8s.io/client-go/rest”
- 配合后面的serviceaccount +clusterrole+clusterrolebinding
- 封装一个getK8sClient 方法
func getK8sClient() *kubernetes.Clientset {// creates the in-cluster configconfig, err := rest.InClusterConfig()if err != nil {logger.Errorf("[create_k8s_InClusterConfig_err][err:%v]", err)return nil}// creates the clientsetclientset, err := kubernetes.NewForConfig(config)if err != nil {logger.Errorf("[create_the_clientset_error][err:%v]", err)return nil}return clientset
}4. 使用k8s-client get node
clientset.CoreV1().Nodes().List代表 get node- 遍历nodes
- 获取ip地址
p.Status.Addresses中的类型为 apiv1.NodeInternalIP 就是内网ip - containerRuntimeVersion和kubeletVersion信息在 p.Status.NodeInfo中
- 获取ip地址
- 在结尾的时候打印个日志，记录下节点数和耗时，并把耗时打个metrics上报
- 完整代码如下
func doGetNode() {start := time.Now()clientset := getK8sClient()if clientset == nil {return}nodes, err := clientset.CoreV1().Nodes().List(context.TODO(), metav1.ListOptions{})if err != nil {logger.Errorf("list_kube-system_pod_error:%v", err)return}if len(nodes.Items) == 0 {return}for _, p := range nodes.Items {var ip stringaddr := p.Status.Addressesif len(addr) == 0 {continue}for _, a := range addr {if a.Type == apiv1.NodeInternalIP {ip = a.Address}}k8sNodeDetail.With(prometheus.Labels{"ip": ip,"hostname": p.Name,"containerRuntimeVersion": p.Status.NodeInfo.ContainerRuntimeVersion,"kubeletVersion": p.Status.NodeInfo.KubeletVersion,}).Set(1)}timeTook := time.Since(start).Seconds()getNodeDuration.Set(timeTook)logger.Infof("server_node_ips_result][num_node:%v][time_took_seconds:%v]", len(nodes.Items), timeTook)}5. 使用k8s-client get pod
-
clientset.CoreV1().Pods("kube-system").List代表获取kube-system namespace下面的pods -
遍历pods
- 控制平面中的pod 都会有 tie=control-plane的标签
- 打点即可
-
完整代码如下
func doGetPod() {start := time.Now()clientset := getK8sClient()if clientset == nil {return}pods, err := clientset.CoreV1().Pods("kube-system").List(context.TODO(), metav1.ListOptions{})if err != nil {logger.Errorf("list_kube-system_pod_error:%v", err)return}if len(pods.Items) == 0 {return}for _, p := range pods.Items {logger.Infof("[pod.label:%v]", p.Labels)if p.Labels["tier"] == "control-plane" {ip := p.Status.PodIPcomponent := p.Labels["component"]k8sPodDetail.With(prometheus.Labels{"ip": ip,"pod_name": p.Name,"component": component,}).Set(1)}}timeTook := time.Since(start).Seconds()getPodDuration.Set(timeTook)logger.Infof("server_pod_ips_result][num_pod:%v][time_took_seconds:%v]", len(pods.Items), timeTook)}
6. 编写运行的ticker函数
- 每隔10秒就执行一下getnode 和getpod上报数据
- 外部的ctx被cancel会导致for退出
func getK8sObjTicker(ctx context.Context) {ticker := time.NewTicker(time.Second * 10)logger.Infof("GetK8sObjTicker start....")defer ticker.Stop()for {select {case <-ticker.C:go doGetNode()go doGetPod()case <-ctx.Done():return}}}7. 编写main函数
- newMetrics 注册metrics
go getK8sObjTicker(ctx)开启获取 k8s对象的协程http.Handle("/metrics", promhttp.Handler())开启prometheus metric path
func main() {// 注册metricsnewMetrics()ctx := context.Background()// 开启获取 k8s对象的协程go getK8sObjTicker(ctx)// 开启prometheus metric pathhttp.Handle("/metrics", promhttp.Handler())err := http.ListenAndServe(":8080", nil)if err != nil {logger.Errorf("failed to start prometheus metrics web :%v", err)}
}本节重点总结 :
- 引入k8s sdk获取k8s 的node和pod信息
- 定义相关metrics
- 初始化k8s-client
- 使用k8s-client get node
- 使用k8s-client get pod
- 打点
