##cAdvisor监控数据分析
入口
- 代码入口：
github.com/google/cadvisor/cadvisor.go - API handler:
github.com/google/cadvisor/api/handler.go - 基于 tag： v0.19.2
# github.com/google/cadvisor/cadvisor.go
... ...
73 containerManager, err := manager.New(memoryStorage, sysFs, *maxHousekeepingInterval, *allowDynamicHousekeeping)
74 if err != nil { 75 glog.Fatalf("Failed to create a Container Manager: %s", err) 76 } 77 78 mux := http.DefaultServeMux 79 80 // Register all HTTP handlers. 81 err = cadvisorHttp.RegisterHandlers(mux, containerManager, *httpAuthFile, *httpAuthRealm, *httpDigestFile, *httpDigestRealm) 82 if err != nil { 83 glog.Fatalf("Failed to register HTTP handlers: %v", err) 84 } 85 86 cadvisorHttp.RegisterPrometheusHandler(mux, containerManager, *prometheusEndpoint, nil) 87 88 // Start the manager. 89 if err := containerManager.Start(); err != nil { 90 glog.Fatalf("Failed to start container manager: %v", err) 91 } ... ... 从入口可以得知真正获取 container 和 machine 信息的 interface 为 Manager，github.com/google/cadvisor/manager/manager.go
# github.com/google/cadvisor/manager/manager.go
52 // The Manager interface defines operations for starting a manager and getting
53 // container and machine information. 54 type Manager interface { 55 // Start the manager. Calling other manager methods before this returns 56 // may produce undefined behavior. 57 Start() error 58 59 // Stops the manager. 60 Stop() error 61 62 // Get information about a container. 63 GetContainerInfo(containerName string, query *info.ContainerInfoRequest) (*info.ContainerInfo, error) ... ... 以下对 cadvisor 具体如何获取到 machine 和 docker container 的信息结合代码具体讲解。
Machine Info
# github.com/google/cadvisor/manager/manager.go
158 machineInfo, err := getMachineInfo(sysfs, fsInfo) // 此处通过 getMachineInfo() 函数获取 machine 信息
159 if err != nil { 160 return nil, err 161 } 162 newManager.machineInfo = *machineInfo 163 glog.Infof("Machine: %+v", newManager.machineInfo) 获取 machineinfo 的具体函数 getMachineInfo()
# github.com/google/cadvisor/manager/machine.go
52 func getMachineInfo(sysFs sysfs.SysFs, fsInfo fs.FsInfo) (*info.MachineInfo, error) { 53 cpuinfo, err := ioutil.ReadFile("/proc/cpuinfo") 54 clockSpeed, err := machine.GetClockSpeed(cpuinfo) 55 if err != nil { 56 return nil, err 57 } 58 ... ... MachineInfo 数据结构：
# github.com/google/cadvisor/info/v1/machine.go
131 type MachineInfo struct {
132 // The number of cores in this machine. 133 NumCores int `json:"num_cores"` 134 135 // Maximum clock speed for the cores, in KHz. 136 CpuFrequency uint64 `json:"cpu_frequency_khz"` 137 138 // The amount of memory (in bytes) in this machine 139 MemoryCapacity int64 `json:"memory_capacity"` 140 141 // The machine id 142 MachineID string `json:"machine_id"` 143 144 // The system uuid 145 SystemUUID string `json:"system_uuid"` 146 147 // The boot id 148 BootID string `json:"boot_id"` 149 150 // Filesystems on this machine. 151 Filesystems []FsInfo `json:"filesystems"` 152 153 // Disk map 154 DiskMap map[string]DiskInfo `json:"disk_map"` 155 156 // Network devices 157 NetworkDevices []NetInfo `json:"network_devices"` 158 159 // Machine Topology 160 // Describes cpu/memory layout and hierarchy. 161 Topology []Node `json:"topology"` 162 163 // Cloud provider the machine belongs to. 164 CloudProvider CloudProvider `json:"cloud_provider"` 165 166 // Type of cloud instance (e.g. GCE standard) the machine is. 167 InstanceType InstanceType `json:"instance_type"` 168 } - cpu
/proc/cpuinfo - mem
/proc/meminfo
# github.com/google/cadvisor/utils/machine/machine.go
76 func GetMachineMemoryCapacity() (int64, error) { 77 out, err := ioutil.ReadFile("/proc/meminfo") 78 if err != nil { 79 return 0, err 80 } 81 82 memoryCapacity, err := parseCapacity(out, memoryCapacityRegexp) 83 if err != nil { 84 return 0, err 85 } 86 return memoryCapacity, err 87 } - machine_id
/var/lib/dbus/machine-id - boot_id
/proc/sys/kernel/random/boot_id
➜ ~ cat /var/lib/dbus/machine-id
a8919ec7c9ee43bd1bcaf6cd55e1a7bc
➜ ~ cat /proc/sys/kernel/random/boot_id
2f6bac7f-738a-4d85-8b58-53e71296119e - disk
/proc/diskstats
# github.com/google/cadvisor/fs/fs.go
189 func (self *RealFsInfo) GetFsInfoForPath(mountSet map[string]struct{}) ([]Fs, error) { 190 filesystems := make([]Fs, 0) 191 deviceSet := make(map[string]struct{}) 192 diskStatsMap, err := getDiskStatsMap("/proc/diskstats") // 磁盘状态通过 /proc/diskstats 获取 193 if err != nil { 194 return nil, err 195 } ... ... 对于分配给容器的 pool 值，如果是 device mapper 类型的，则通过 dmsetup status pool_device_name 获取设备使用情况
# github.com/google/cadvisor/fs/fs.go
... ...
205 case "devicemapper": 206 total, free, avail, err = getDMStats(device, partition.blockSize) .... - 获取磁盘设备信息
/sys/block
# github.com/google/cadvisor/utils/sysinfo/sysinfo.go
31 func GetBlockDeviceInfo(sysfs sysfs.SysFs) (map[string]info.DiskInfo, error) { 32 disks, err := sysfs.GetBlockDevices() // cadvisor/utils/sysfs/sysfs.go GetBlockDevices() 可知，通过 /sys/block 获取磁盘设备名，其它信息雷同 33 if err != nil { 34 return nil, err 35 } ... ... ➜ ~ cat /sys/block/sda/dev # 主次设备号
8:0
➜ ~ cat /sys/block/sda/size # 磁盘设备大小
976773168
➜ ~ cat /sys/block/sda/queue/scheduler # 磁盘调度类型
noop deadline [cfq]
➜ ~
- netDevices
/sys/class/net
# github.com/google/cadvisor/utils/sysinfo/sysinfo.go
87 func GetNetworkDevices(sysfs sysfs.SysFs) ([]info.NetInfo, error) { 88 devs, err := sysfs.GetNetworkDevices() 89 if err != nil { 90 return nil, err 91 } 92 netDevices := []info.NetInfo{} ... ... ➜ ~ cat /sys/class/net/eth0/address f0:76:1c:6d:26:d6 ➜ ~ cat /sys/class/net/eth0/mtu 1500 ➜ ~ cat /sys/class/net/eth0/speed 10 - system UUID
/sys/class/dmi/id/product_uuid
➜ ~ sudo cat /sys/class/dmi/id/product_uuid
F27316D5-4DB3-E411-A26A-F0761C6D26D6
➜ ~
Docker Info
# github.com/google/cadvisor/manager/manager.go
201 // Start the container manager.
202 func (self *manager) Start() error { 203 // Register Docker container factory. 204 err := docker.Register(self, self.fsInfo) 205 if err != nil { 206 glog.Errorf("Docker container factory registration failed: %v.", err) 207 } ... ... cAdvisor 启动 container manager，通过调用 Linux 系统 inotify 监测 cgroup docker 目录的文件变化判断 docker 的创建和删除。
Getprocesslist
cAdvisor 首先判断自己是否运行在 container 中，如果是则 chroot /rootfs （cAdvisor 在 container 中运行会通过 volume 方式把系统 / mount 到 /rootfs 下）再执行 ps -e -o user,pid,ppid,stime,pcpu,pmem,rss,vsz,stat,time,comm,cgrou，否则直接执行获取。(github.com/google/cadvisor/manager/container.go GetProcessList() 函数实现)
# github.com/google/cadvisor/manager/manager.go
687 func (m *manager) GetProcessList(containerName string, options v2.RequestOptions) ([]v2.ProcessInfo, error) { 688 // override recursive. Only support single container listing. 689 options.Recursive = false 690 conts, err := m.getRequestedContainers(containerName, options) 691 if err != nil { 692 return nil, err 693 } 694 if len(conts) != 1 { 695 return nil, fmt.Errorf("Expected the request to match only one container") 696 } ... ... 如果是获取全局而非某个 container 的 processlist，则 containername 默认为 "/"，判断逻辑如下：
# github.com/google/cadvisor/manager/container.go
273 if isRoot || c.info.Name == cgroup {
274 processes = append(processes, v2.ProcessInfo{ 275 User: fields[0], 276 Pid: pid, 277 Ppid: ppid, 278 StartTime: fields[3], 279 PercentCpu: float32(percentCpu), ... ... ContainerInfo
获取指定 container 的 info 数据：
# github.com/google/cadvisor manager/manager.go
531 func (self *manager) DockerContainer(containerName string, query *info.ContainerInfoRequest) (info.ContainerInfo, error) { 532 container, err := self.getDockerContainer(containerName) 533 if err != nil { 534 return info.ContainerInfo{}, err 535 } 536 ... ... 具体获取函数：
# github.com/google/cadvisor/manager/container.go
102 func (c *containerData) GetInfo() (*containerInfo, error) { 103 // Get spec and subcontainers. 104 if time.Since(c.lastUpdatedTime) > 5*time.Second { // 可以看出当上一次更新时间超过 5s 的时候才会更新 info 数据 5 105 err := c.updateSpec() 106 if err != nil { 107 return nil, err 108 } ... ... - ContainerSpec
cAdvisor 获取 ContainerSpec 总体的逻辑是找出对应 container 对应的系统 Spec json 文件。
container info handler：
# github.com/google/cadvisor/container/docker/handler.go
220 func (self *dockerContainerHandler) GetSpec() (info.ContainerSpec, error) { 221 mi, err := self.machineInfoFactory.GetMachineInfo() 222 if err != nil { 223 return info.ContainerSpec{}, err 224 } /proc/self/mountinfo当前运行进程的挂载信息/proc/cgroups当前内核支持的 cgroup 类型- libcontainer configpath
/var/run/docker/execdriver/native/containerId
# github.com/google/cadvisor/container/libcontainer/compatibility.go 读取 libcontainer config 文件信息
150 func ReadConfig(dockerRoot, dockerRun, containerID string) (*configs.Config, error) { 151 // Try using the new config if it is available. 152 configPath := configPath(dockerRun, containerID) 153 if utils.FileExists(configPath) { 154 out, err := ioutil.ReadFile(configPath) 155 if err != nil { 156 return nil, err 157 } 158 159 var state libcontainer.State 160 err = json.Unmarshal(out, &state) 161 if err != nil { 162 return nil, err 163 } 164 return &state.Config, nil 165 } # github.com/google/cadvisor/container/docker/handler.go 磁盘使用率统计目前支持 aufs，这是目前 CentOS7 上没有 filesystem 大小的原因
... ...
232 // For now only enable for aufs filesystems 233 spec.HasFilesystem = self.storageDriver == aufsStorageDriver ... ... 241 func (self *dockerContainerHandler) getFsStats(stats *info.ContainerStats) error { 242 // No support for non-aufs storage drivers. 243 if self.storageDriver != aufsStorageDriver { 244 return nil 245 } ... ... -
文件系统相关
# github.com/google/cadvisor/fs/fs.go -
aufs storage dir:
/var/lib/docker/aufs/diff/containerId -
获取 cgroup 和 网络状态
# github.com/google/cadvisor/container/docker/handler.go
# cgroup 信息是调用 github.com/docker/libcontainer/cgroups 获取
276 func (self *dockerContainerHandler) GetStats() (*info.ContainerStats, error) { 277 stats, err := containerLibcontainer.GetStats(self.cgroupManager, self.rootFs, self.pid) 278 if err != nil { 279 return stats, err 280 } 281 // Clean up stats for containers that don't have their own network - this 282 // includes containers running in Kubernetes pods that use the network of the 283 // infrastructure container. This stops metrics being reported multiple times 284 // for each container in a pod. 如果 container 没有自己的网络，则返回空。对于 K8s pod 内部 container 都是通过 pause container 网络通信 285 if !hasNet(self.networkMode) { 286 stats.Network = info.NetworkStats{} 287 } 288 289 // Get filesystem stats. 290 err = self.getFsStats(stats) 291 if err != nil { 292 return stats, err 293 } 294 295 return stats, nil 296 } 具体获取 cgroup 和 network 的函数实现 GetStats()
- container 网络状态
/proc/<pid>/net/dev
# github.com/google/cadvisor/container/libcontainer/helpers.go
81 // Get cgroup and networking stats of the specified container
82 func GetStats(cgroupManager cgroups.Manager, rootFs string, pid int) (*info.ContainerStats, error) { 83 cgroupStats, err := cgroupManager.GetStats() 84 if err != nil { 85 return nil, err 86 } 87 libcontainerStats := &libcontainer.Stats{ 88 CgroupStats: cgroupStats, 89 } 90 stats := toContainerStats(libcontainerStats) 91 92 // If we know the pid then get network stats from /proc/<pid>/net/dev 93 if pid > 0 { 94 netStats, err := networkStatsFromProc(rootFs, pid) 95 if err != nil { 96 glog.V(2).Infof("Unable to get network stats from pid %d: %v", pid, err) 97 } else { 98 stats.Network.Interfaces = append(stats.Network.Interfaces, netStats...) 99 } ... ... cgroup status 包含信息如下(cadVisor 调用 licontainer/cgroups 获取 container 对应 cgroup 信息)：
# github.com/docker/libcontainer/cgroups/stats.go
88 func NewStats() *Stats { 89 memoryStats := MemoryStats{Stats: make(map[string]uint64)} 90 hugetlbStats := make(map[string]HugetlbStats) 91 return &Stats{MemoryStats: memoryStats, HugetlbStats: hugetlbStats} 92 }
194
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
