1. cAdvisor简介
cAdvisor对Node机器上的资源及容器进行实时监控和性能数据采集，包括CPU使用情况、内存使用情况、网络吞吐量及文件系统使用情况，cAdvisor集成在Kubelet中，当kubelet启动时会自动启动cAdvisor，即一个cAdvisor仅对一台Node机器进行监控。kubelet的启动参数–cadvisor-port可以定义cAdvisor对外提供服务的端口，默认为4194。可以通过浏览器访问。项目主页：http://github.com/google/cadvisor 。
2. cAdvisor结构图
3. Metrics
分类 字段 描述
1.cAvisor简介:
cAdvisor是Google开源的容器资源监控和性能分析工具，它是专门为容器而生，在Kubernetes中，我们不需要单独去安装，cAdvisor作为kubelet内置的一部分程序可以直接使用，也就是我们可以直接使用cadvisor采集数据，可以采集到和容器运行相关的所有指标，单独安装cAdvisor时的数据路径为/api/v1/nodes/[节点名称]/proxy/metrics/cadvisor，如果cadvisor集成到kubelet，采集数据的路径是https://127.0.0.1:10250/metrics/cadvisor
2.查看cadvisor监控指标，在k8s-master节点操作
kubectl create ns monitor-sa #创建一个monitor-sa的名称空间
kubectl create serviceaccount monitor -n monitor-sa #创建一个sa账号
kubectl create clusterrolebinding monitor-clusterrolebinding -n monitor-sa --clusterrole=cluster-admin --serviceaccount=monitor-sa:monitor
#把sa账号monitor通过clusterrolebing绑定到clusterrole上
kubectl get secret -n monitor-sa #查看monitor-sa名称空间下的secret密钥
kubectl describe secret monitor-token-j4jwf -n monitor-sa #可看到token相关的内容如下所示
eyJhbGciOiJSUzI1NiIsImtpZCI6IkV5VUZuUmlPa0pMSF9sSFdUYktjdWdGVk9CR3owMlZhUDg4UzdVQWtveEEifQ.eyJpc3MiOiJrdWJlcm5ldGVzL3NlcnZpY2VhY2NvdW50Iiwia3ViZXJuZXRlcy5pby9zZXJ2aWNlYWNjb3VudC9uYW1lc3BhY2UiOiJtb25pdG9yLXNhIiwia3ViZXJuZXRlcy5pby9zZXJ2aWNlYWNjb3VudC9zZWNyZXQubmFtZSI6Im1vbml0b3ItdG9rZW4tajRqd2YiLCJrdWJlcm5ldGVzLmlvL3NlcnZpY2VhY2NvdW50L3NlcnZpY2UtYWNjb3VudC5uYW1lIjoibW9uaXRvciIsImt1YmVybmV0ZXMuaW8vc2VydmljZWFjY291bnQvc2VydmljZS1hY2NvdW50LnVpZCI6ImQ5NzJiNDA1LWEzZTYtNDJiYS04YzU3LTA2MjE2YmE3Nzk1MCIsInN1YiI6InN5c3RlbTpzZXJ2aWNlYWNjb3VudDptb25pdG9yLXNhOm1vbml0b3IifQ.U0fMb34xlIcMrC5g_v3jeTMwxg3L3VkAD6lUa84Ke3kor3aB9tT092PM4N5_8cVPRJkHkh5UXx3A7mWOErjftgux41azA2N1Zkuqt-7VXkvvBCOBAmv-95mRz9FPEbzbR9gG5EudcCFeJypYOO3n7Oipr1MS4YxGLYVjUTQ46f5GIMJli9Uw6MYkij9HwuoD8qbLulAq6W540qvJfK4Bd20kvjqzZQveD2Ej-hmUlHR2cqshgD64VgBOIAJJir4bQ04JthLqgpC9peTTYo2hJ8XK-Y5OCx2v419syb0xPC2jrzwcZabvTBG_QCB4Ly8BRAxjEDB4ox3R6EMw8Ie68A
通过下面命令可以获取到cadvisor采集的指标数据
curl https://127.0.0.1:10250/metrics/cadvisor -k -H "Authorization: Bearer eyJhbGciOiJSUzI1NiIsImtpZCI6IkV5VUZuUmlPa0pMSF9sSFdUYktjdWdGVk9CR3owMlZhUDg4UzdVQWtveEEifQ.eyJpc3MiOiJrdWJlcm5ldGVzL3NlcnZpY2VhY2NvdW50Iiwia3ViZXJuZXRlcy5pby9zZXJ2aWNlYWNjb3VudC9uYW1lc3BhY2UiOiJtb25pdG9yLXNhIiwia3ViZXJuZXRlcy5pby9zZXJ2aWNlYWNjb3VudC9zZWNyZXQubmFtZSI6Im1vbml0b3ItdG9rZW4tajRqd2YiLCJrdWJlcm5ldGVzLmlvL3NlcnZpY2VhY2NvdW50L3NlcnZpY2UtYWNjb3VudC5uYW1lIjoibW9uaXRvciIsImt1YmVybmV0ZXMuaW8vc2VydmljZWFjY291bnQvc2VydmljZS1hY2NvdW50LnVpZCI6ImQ5NzJiNDA1LWEzZTYtNDJiYS04YzU3LTA2MjE2YmE3Nzk1MCIsInN1YiI6InN5c3RlbTpzZXJ2aWNlYWNjb3VudDptb25pdG9yLXNhOm1vbml0b3IifQ.U0fMb34xlIcMrC5g_v3jeTMwxg3L3VkAD6lUa84Ke3kor3aB9tT092PM4N5_8cVPRJkHkh5UXx3A7mWOErjftgux41azA2N1Zkuqt-7VXkvvBCOBAmv-95mRz9FPEbzbR9gG5EudcCFeJypYOO3n7Oipr1MS4YxGLYVjUTQ46f5GIMJli9Uw6MYkij9HwuoD8qbLulAq6W540qvJfK4Bd20kvjqzZQveD2Ej-hmUlHR2cqshgD64VgBOIAJJir4bQ04JthLqgpC9peTTYo2hJ8XK-Y5OCx2v419syb0xPC2jrzwcZabvTBG_QCB4Ly8BRAxjEDB4ox3R6EMw8Ie68A"
3.cadvisor中获取到的典型监控指标如下：
指标名称 类型 含义
container_cpu_load_average_10s gauge 过去10秒容器CPU的平均负载
container_cpu_usage_seconds_total counter 容器在每个CPU内核上的累积占用时间 (单位：秒)
container_cpu_system_seconds_total counter System CPU累积占用时间（单位：秒）
container_cpu_user_seconds_total counter User CPU累积占用时间（单位：秒）
container_fs_usage_bytes gauge 容器中文件系统的使用量(单位：字节)
container_fs_limit_bytes gauge 容器可以使用的文件系统总量(单位：字节)
container_fs_reads_bytes_total counter 容器累积读取数据的总量(单位：字节)
container_fs_writes_bytes_total counter 容器累积写入数据的总量(单位：字节)
container_memory_max_usage_bytes gauge 容器的最大内存使用量（单位：字节）
container_memory_usage_bytes gauge 容器当前的内存使用量（单位：字节
container_spec_memory_limit_bytes gauge 容器的内存使用量限制
machine_memory_bytes gauge 当前主机的内存总量
container_network_receive_bytes_total counter 容器网络累积接收数据总量（单位：字节）
container_network_transmit_bytes_total counter 容器网络累积传输数据总量（单位：字节）
4.当能够正常采集到cAdvisor的样本数据后，可以通过以下表达式计算容器的CPU使用率：
（1）sum(irate(container_cpu_usage_seconds_total{image!=""}[1m])) without (cpu)
容器CPU使用率
（2）container_memory_usage_bytes{image!=""}
查询容器内存使用量（单位：字节）:
（3）sum(rate(container_network_receive_bytes_total{image!=""}[1m])) without (interface)
查询容器网络接收量（速率）（单位：字节/秒）：
（4）sum(rate(container_network_transmit_bytes_total{image!=""}[1m])) without (interface)
容器网络传输量 字节/秒
（5）sum(rate(container_fs_reads_bytes_total{image!=""}[1m])) without (device)
容器文件系统读取速率 字节/秒
（6）sum(rate(container_fs_writes_bytes_total{image!=""}[1m])) without (device)
容器文件系统写入速率 字节/秒
5.cadvisor 常用容器监控指标
（1）网络流量
sum(rate(container_network_receive_bytes_total{name=~".+"}[1m])) by (name)
##容器网络接收的字节数（1分钟内），根据名称查询 name=~".+"
sum(rate(container_network_transmit_bytes_total{name=~".+"}[1m])) by (name)
##容器网络传输的字节数（1分钟内），根据名称查询 name=~".+"
（2）容器 CPU相关
sum(rate(container_cpu_system_seconds_total[1m]))
###所用容器system cpu的累计使用时间（1min钟内）
sum(irate(container_cpu_system_seconds_total{image!=""}[1m])) without (cpu)
###每个容器system cpu的使用时间（1min钟内）
sum(rate(container_cpu_usage_seconds_total{name=~".+"}[1m])) by (name) * 100
#每个容器的cpu使用率
sum(sum(rate(container_cpu_usage_seconds_total{name=~".+"}[1m])) by (name) * 100)
#总容器的cpu使用率
4. cAdvisor源码
4.1. cAdvisor入口函数
cadvisor.go
func main() {
defer glog.Flush()
flag.Parse()
if *versionFlag {
fmt.Printf("cAdvisor version %s (%s)\n", version.Info["version"], version.Info["revision"])
os.Exit(0)
}
setMaxProcs()
memoryStorage, err := NewMemoryStorage()
if err != nil {
glog.Fatalf("Failed to initialize storage driver: %s", err)
}
sysFs, err := sysfs.NewRealSysFs()
if err != nil {
glog.Fatalf("Failed to create a system interface: %s", err)
}
collectorHttpClient := createCollectorHttpClient(*collectorCert, *collectorKey)
containerManager, err := manager.New(memoryStorage, sysFs, *maxHousekeepingInterval, *allowDynamicHousekeeping, ignoreMetrics.MetricSet, &collectorHttpClient)
if err != nil {
glog.Fatalf("Failed to create a Container Manager: %s", err)
}
mux := http.NewServeMux()
if *enableProfiling {
mux.HandleFunc("/debug/pprof/", pprof.Index)
mux.HandleFunc("/debug/pprof/cmdline", pprof.Cmdline)
mux.HandleFunc("/debug/pprof/profile", pprof.Profile)
mux.HandleFunc("/debug/pprof/symbol", pprof.Symbol)
}
// Register all HTTP handlers.
err = cadvisorhttp.RegisterHandlers(mux, containerManager, *httpAuthFile, *httpAuthRealm, *httpDigestFile, *httpDigestRealm)
if err != nil {
glog.Fatalf("Failed to register HTTP handlers: %v", err)
}
cadvisorhttp.RegisterPrometheusHandler(mux, containerManager, *prometheusEndpoint, nil)
// Start the manager.
if err := containerManager.Start(); err != nil {
glog.Fatalf("Failed to start container manager: %v", err)
}
// Install signal handler.
installSignalHandler(containerManager)
glog.Infof("Starting cAdvisor version: %s-%s on port %d", version.Info["version"], version.Info["revision"], *argPort)
addr := fmt.Sprintf("%s:%d", *argIp, *argPort)
glog.Fatal(http.ListenAndServe(addr, mux))
}
核心代码：
memoryStorage, err := NewMemoryStorage()
sysFs, err := sysfs.NewRealSysFs()
#创建containerManager
containerManager, err := manager.New(memoryStorage, sysFs, *maxHousekeepingInterval, *allowDynamicHousekeeping, ignoreMetrics.MetricSet, &collectorHttpClient)
#启动containerManager
err := containerManager.Start()
4.2. cAdvisor Client的使用
import "github.com/google/cadvisor/client"
func main(){
client, err := client.NewClient("http://192.168.19.30:4194/") //http://<host-ip>:<port>/
}
4.2.1 client定义
cadvisor/client/client.go
// Client represents the base URL for a cAdvisor client.
type Client struct {
baseUrl string
}
// NewClient returns a new v1.3 client with the specified base URL.
func NewClient(url string) (*Client, error) {
if !strings.HasSuffix(url, "/") {
url += "/"
}
return &Client{
baseUrl: fmt.Sprintf("%sapi/v1.3/", url),
}, nil
}
4.2.2. client方法
1）MachineInfo
// MachineInfo returns the JSON machine information for this client.
// A non-nil error result indicates a problem with obtaining
// the JSON machine information data.
func (self *Client) MachineInfo() (minfo *v1.MachineInfo, err error) {
u := self.machineInfoUrl()
ret := new(v1.MachineInfo)
if err = self.httpGetJsonData(ret, nil, u, "machine info"); err != nil {
return
}
minfo = ret
return
}
2）ContainerInfo
// ContainerInfo returns the JSON container information for the specified
// container and request.
func (self *Client) ContainerInfo(name string, query *v1.ContainerInfoRequest) (cinfo *v1.ContainerInfo, err error) {
u := self.containerInfoUrl(name)
ret := new(v1.ContainerInfo)
if err = self.httpGetJsonData(ret, query, u, fmt.Sprintf("container info for %q", name)); err != nil {
return
}
cinfo = ret
return
}
3）DockerContainer
// Returns the JSON container information for the specified
// Docker container and request.
func (self *Client) DockerContainer(name string, query *v1.ContainerInfoRequest) (cinfo v1.ContainerInfo, err error) {
u := self.dockerInfoUrl(name)
ret := make(map[string]v1.ContainerInfo)
if err = self.httpGetJsonData(&ret, query, u, fmt.Sprintf("Docker container info for %q", name)); err != nil {
return
}
if len(ret) != 1 {
err = fmt.Errorf("expected to only receive 1 Docker container: %+v", ret)
return
}
for _, cont := range ret {
cinfo = cont
}
return
}
4）AllDockerContainers
// Returns the JSON container information for all Docker containers.
func (self *Client) AllDockerContainers(query *v1.ContainerInfoRequest) (cinfo []v1.ContainerInfo, err error) {
u := self.dockerInfoUrl("/")
ret := make(map[string]v1.ContainerInfo)
if err = self.httpGetJsonData(&ret, query, u, "all Docker containers info"); err != nil {
return
}
cinfo = make([]v1.ContainerInfo, 0, len(ret))
for _, cont := range ret {
cinfo = append(cinfo, cont)
}
return
}
文章参考：http://blog.opskumu.com/cadvisor.html