main.go 7.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278
  1. package main
  2. import (
  3. "bytes"
  4. "fmt"
  5. "net/http"
  6. _ "net/http/pprof"
  7. "os"
  8. "path"
  9. "runtime"
  10. "strings"
  11. "encoding/json"
  12. "github.com/coroot/coroot-node-agent/common"
  13. "github.com/coroot/coroot-node-agent/config"
  14. "github.com/coroot/coroot-node-agent/containers"
  15. "github.com/coroot/coroot-node-agent/flags"
  16. "github.com/coroot/coroot-node-agent/logs"
  17. "github.com/coroot/coroot-node-agent/node"
  18. "github.com/coroot/coroot-node-agent/prom"
  19. "github.com/coroot/coroot-node-agent/tracing"
  20. "github.com/prometheus/client_golang/prometheus"
  21. "github.com/prometheus/client_golang/prometheus/promhttp"
  22. dto "github.com/prometheus/client_model/go"
  23. "golang.org/x/mod/semver"
  24. "golang.org/x/sys/unix"
  25. "golang.org/x/time/rate"
  26. "k8s.io/klog/v2"
  27. )
  28. var (
  29. version = "unknown"
  30. )
  31. const minSupportedKernelVersion = "4.16"
  32. func uname() (string, string, error) {
  33. runtime.LockOSThread()
  34. defer runtime.UnlockOSThread()
  35. f, err := os.Open("/proc/1/ns/uts")
  36. if err != nil {
  37. return "", "", err
  38. }
  39. defer f.Close()
  40. self, err := os.Open("/proc/self/ns/uts")
  41. if err != nil {
  42. return "", "", err
  43. }
  44. defer self.Close()
  45. defer func() {
  46. unix.Setns(int(self.Fd()), unix.CLONE_NEWUTS)
  47. }()
  48. err = unix.Setns(int(f.Fd()), unix.CLONE_NEWUTS)
  49. if err != nil {
  50. return "", "", err
  51. }
  52. var utsname unix.Utsname
  53. if err := unix.Uname(&utsname); err != nil {
  54. return "", "", err
  55. }
  56. hostname := string(bytes.Split(utsname.Nodename[:], []byte{0})[0])
  57. kernelVersion := string(bytes.Split(utsname.Release[:], []byte{0})[0])
  58. return hostname, kernelVersion, nil
  59. }
  60. func machineID() string {
  61. for _, p := range []string{"sys/devices/virtual/dmi/id/product_uuid", "etc/machine-id", "var/lib/dbus/machine-id"} {
  62. payload, err := os.ReadFile(path.Join("/proc/1/root", p))
  63. if err != nil {
  64. klog.Warningln("failed to read machine-id:", err)
  65. continue
  66. }
  67. id := strings.TrimSpace(strings.Replace(string(payload), "-", "", -1))
  68. klog.Infoln("machine-id: ", id)
  69. return id
  70. }
  71. return ""
  72. }
  73. func whitelistNodeExternalNetworks() {
  74. netdevs, err := node.NetDevices()
  75. if err != nil {
  76. klog.Warningln("failed to get network interfaces:", err)
  77. return
  78. }
  79. for _, iface := range netdevs {
  80. for _, p := range iface.IPPrefixes {
  81. if p.IP().IsLoopback() || common.IsIpPrivate(p.IP()) {
  82. continue
  83. }
  84. // if the node has an external network IP, whitelist that network // IK: 如果节点具有外部网络IP,则将该网络列入白名单...
  85. common.ConnectionFilter.WhitelistPrefix(p)
  86. }
  87. }
  88. }
  89. type MetricItemData struct {
  90. Label map[string]string `json:"metric_tags"`
  91. Value any `json:"value"`
  92. }
  93. type MetricData struct {
  94. MetricKey string `json:"metric_key"`
  95. Metric []MetricItemData `json:"metric"`
  96. }
  97. func main() {
  98. // IK: 增加配置文件
  99. agentConfig, err := config.LoadConfig()
  100. if err != nil {
  101. fmt.Println(err)
  102. return
  103. }
  104. fmt.Printf("License=%+v\r\n", agentConfig)
  105. // IK: 限制资源
  106. maxProcs := agentConfig.ResourceConfig.ResourceMaxProcs
  107. if maxProcs <= 0 {
  108. maxProcs = 1
  109. }
  110. runtime.GOMAXPROCS(maxProcs)
  111. // IK: 初始化日志组件, klog是将日志以http方式发送到server端, 支持配置LogPerSecond等配置.
  112. klog.LogToStderr(false)
  113. klog.SetOutput(&RateLimitedLogOutput{limiter: rate.NewLimiter(rate.Limit(*flags.LogPerSecond), *flags.LogBurst)})
  114. klog.Infoln("agent version:", version)
  115. hostname, kv, err := uname() // IK: 获取主机名称,内核版本...
  116. if err != nil {
  117. klog.Exitln("failed to get uname:", err)
  118. }
  119. klog.Infoln("hostname:", hostname)
  120. klog.Infoln("kernel version:", kv)
  121. ver := common.KernelMajorMinor(kv)
  122. if ver == "" {
  123. klog.Exitln("invalid kernel version:", kv)
  124. }
  125. if semver.Compare("v"+ver, "v"+minSupportedKernelVersion) == -1 {
  126. klog.Exitf("the minimum Linux kernel version required is %s or later", minSupportedKernelVersion)
  127. }
  128. whitelistNodeExternalNetworks() // IK:TODO: 将外部网络放到白名单???
  129. machineId := machineID()
  130. tracing.Init(machineId, hostname, version) // IK:TODO: 初始化发送???
  131. logs.Init(machineId, hostname, version)
  132. registry := prometheus.NewRegistry()
  133. registerer := prometheus.WrapRegistererWith(prometheus.Labels{"machine_id": machineId}, registry)
  134. registerer.MustRegister(info("node_agent_info", version))
  135. if err := registerer.Register(node.NewCollector(hostname, kv)); err != nil {
  136. klog.Exitln(err)
  137. }
  138. //processInfoCh := profiling.Init(machineId, hostname)
  139. cr, err := containers.NewRegistry(registerer, kv, agentConfig, nil) // IK: 核心代码
  140. if err != nil {
  141. klog.Exitln(err)
  142. }
  143. defer cr.Close()
  144. klog.Infoln("START_TRACE")
  145. //profiling.Start()
  146. //defer profiling.Stop()
  147. // 创建一个/metrics路由处理函数
  148. metricsHandler := func(w http.ResponseWriter, r *http.Request) {
  149. // 从注册表中获取指标数据
  150. metrics, err := registry.Gather()
  151. if err != nil {
  152. // 错误处理
  153. http.Error(w, err.Error(), http.StatusInternalServerError)
  154. return
  155. }
  156. var Data []MetricData
  157. for _, metric := range metrics {
  158. if metric.GetName() != "container_net_tcp_successful_connects_total" &&
  159. metric.GetName() != "container_net_tcp_failed_connects_total" &&
  160. metric.GetName() != "container_net_tcp_retransmits_total" &&
  161. metric.GetName() != "container_net_tcp_listen_info" &&
  162. metric.GetName() != "container_http_requests_total" &&
  163. metric.GetName() != "container_http_requests_duration_seconds_total" &&
  164. metric.GetName() != "container_application_type" {
  165. continue
  166. }
  167. var item MetricData
  168. var itemOther MetricData
  169. item.MetricKey = metric.GetName()
  170. for _, m := range metric.GetMetric() {
  171. metricItem := MetricItemData{}
  172. label := make(map[string]string)
  173. for _, l := range m.GetLabel() {
  174. label[l.GetName()] = l.GetValue()
  175. }
  176. metricItem.Label = label
  177. switch metric.GetType() {
  178. case dto.MetricType_COUNTER:
  179. metricItem.Value = m.GetCounter().GetValue()
  180. item.Metric = append(item.Metric, metricItem)
  181. case dto.MetricType_GAUGE:
  182. metricItem.Value = m.GetGauge().GetValue()
  183. item.Metric = append(item.Metric, metricItem)
  184. case dto.MetricType_HISTOGRAM:
  185. item.MetricKey = metric.GetName() + "_sum"
  186. metricItem.Value = m.GetHistogram().GetSampleSum()
  187. item.Metric = append(item.Metric, metricItem)
  188. metricItemOther := MetricItemData{}
  189. metricItemOther.Label = label
  190. itemOther.MetricKey = metric.GetName() + "_count"
  191. metricItemOther.Value = m.GetHistogram().GetSampleCount()
  192. itemOther.Metric = append(itemOther.Metric, metricItemOther)
  193. default:
  194. continue
  195. }
  196. }
  197. Data = append(Data, item)
  198. if metric.GetType() == dto.MetricType_HISTOGRAM {
  199. Data = append(Data, itemOther)
  200. }
  201. }
  202. // 将指标数据转换为JSON格式
  203. jsonData, err := json.Marshal(Data)
  204. //jsonData, err := json.Marshal(metrics)
  205. if err != nil {
  206. http.Error(w, err.Error(), http.StatusInternalServerError)
  207. return
  208. }
  209. w.Header().Set("Content-Type", "application/json")
  210. w.Write(jsonData)
  211. }
  212. if err := prom.StartAgent(machineId); err != nil {
  213. klog.Exitln(err)
  214. }
  215. http.Handle("/metrics", promhttp.HandlerFor(registry, promhttp.HandlerOpts{ErrorLog: logger{}, Registry: registerer}))
  216. http.HandleFunc("/metrics2", metricsHandler)
  217. klog.Infoln("listening on:", *flags.ListenAddress)
  218. klog.Errorln(http.ListenAndServe(*flags.ListenAddress, nil))
  219. }
  220. func info(name, version string) prometheus.Collector {
  221. g := prometheus.NewGauge(prometheus.GaugeOpts{
  222. Name: name,
  223. ConstLabels: prometheus.Labels{"version": version},
  224. })
  225. g.Set(1)
  226. return g
  227. }
  228. type logger struct{}
  229. func (l logger) Println(v ...interface{}) {
  230. klog.Errorln(v...)
  231. }
  232. type RateLimitedLogOutput struct {
  233. limiter *rate.Limiter
  234. }
  235. func (o *RateLimitedLogOutput) Write(data []byte) (int, error) {
  236. if !o.limiter.Allow() {
  237. return len(data), nil
  238. }
  239. return os.Stderr.Write(data)
  240. }