metrics.go 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360
  1. package metrics
  2. import (
  3. "bytes"
  4. "encoding/json"
  5. "net/http"
  6. "runtime"
  7. "strings"
  8. "time"
  9. "github.com/grafana/grafana/pkg/bus"
  10. "github.com/grafana/grafana/pkg/models"
  11. "github.com/grafana/grafana/pkg/plugins"
  12. "github.com/grafana/grafana/pkg/setting"
  13. "github.com/prometheus/client_golang/prometheus"
  14. )
  15. var (
  16. M_Instance_Start prometheus.Counter
  17. M_Page_Status *prometheus.CounterVec
  18. M_Api_Status *prometheus.CounterVec
  19. M_Proxy_Status *prometheus.CounterVec
  20. M_Http_Request_Total *prometheus.CounterVec
  21. M_Http_Request_Summary *prometheus.SummaryVec
  22. M_Api_User_SignUpStarted prometheus.Counter
  23. M_Api_User_SignUpCompleted prometheus.Counter
  24. M_Api_User_SignUpInvite prometheus.Counter
  25. M_Api_Dashboard_Save prometheus.Summary
  26. M_Api_Dashboard_Get prometheus.Summary
  27. M_Api_Dashboard_Search prometheus.Summary
  28. M_Api_Admin_User_Create prometheus.Counter
  29. M_Api_Login_Post prometheus.Counter
  30. M_Api_Login_OAuth prometheus.Counter
  31. M_Api_Org_Create prometheus.Counter
  32. M_Api_Dashboard_Snapshot_Create prometheus.Counter
  33. M_Api_Dashboard_Snapshot_External prometheus.Counter
  34. M_Api_Dashboard_Snapshot_Get prometheus.Counter
  35. M_Api_Dashboard_Insert prometheus.Counter
  36. M_Alerting_Result_State *prometheus.CounterVec
  37. M_Alerting_Notification_Sent *prometheus.CounterVec
  38. M_Aws_CloudWatch_GetMetricStatistics prometheus.Counter
  39. M_Aws_CloudWatch_ListMetrics prometheus.Counter
  40. M_DB_DataSource_QueryById prometheus.Counter
  41. // Timers
  42. M_DataSource_ProxyReq_Timer prometheus.Summary
  43. M_Alerting_Execution_Time prometheus.Summary
  44. // StatTotals
  45. M_Alerting_Active_Alerts prometheus.Gauge
  46. M_StatTotal_Dashboards prometheus.Gauge
  47. M_StatTotal_Users prometheus.Gauge
  48. M_StatTotal_Orgs prometheus.Gauge
  49. M_StatTotal_Playlists prometheus.Gauge
  50. )
  51. func init() {
  52. M_Instance_Start = prometheus.NewCounter(prometheus.CounterOpts{
  53. Name: "instance_start_total",
  54. Help: "counter for started instances",
  55. })
  56. M_Page_Status = prometheus.NewCounterVec(
  57. prometheus.CounterOpts{
  58. Name: "page_response_status_total",
  59. Help: "page http response status",
  60. },
  61. []string{"code"},
  62. )
  63. M_Api_Status = prometheus.NewCounterVec(
  64. prometheus.CounterOpts{
  65. Name: "api_response_status_total",
  66. Help: "api http response status",
  67. },
  68. []string{"code"},
  69. )
  70. M_Proxy_Status = prometheus.NewCounterVec(
  71. prometheus.CounterOpts{
  72. Name: "proxy_response_status_total",
  73. Help: "proxy http response status",
  74. },
  75. []string{"code"},
  76. )
  77. M_Http_Request_Total = prometheus.NewCounterVec(
  78. prometheus.CounterOpts{
  79. Name: "http_request_total",
  80. Help: "http request counter",
  81. },
  82. []string{"code", "method"},
  83. )
  84. M_Http_Request_Summary = prometheus.NewSummaryVec(
  85. prometheus.SummaryOpts{
  86. Name: "http_request_duration",
  87. Help: "http request summary",
  88. },
  89. []string{"code", "method"},
  90. )
  91. M_Api_User_SignUpStarted = prometheus.NewCounter(prometheus.CounterOpts{
  92. Name: "api_user_signup_started_total",
  93. Help: "amount of users who started the signup flow",
  94. })
  95. M_Api_User_SignUpCompleted = prometheus.NewCounter(prometheus.CounterOpts{
  96. Name: "api_user_signup_completed_total",
  97. Help: "amount of users who completed the signup flow",
  98. })
  99. M_Api_User_SignUpInvite = prometheus.NewCounter(prometheus.CounterOpts{
  100. Name: "api_user_signup_invite_total",
  101. Help: "amount of users who have been invited",
  102. })
  103. M_Api_Dashboard_Save = prometheus.NewSummary(prometheus.SummaryOpts{
  104. Name: "api_dashboard_save",
  105. Help: "summary for dashboard save duration",
  106. })
  107. M_Api_Dashboard_Get = prometheus.NewSummary(prometheus.SummaryOpts{
  108. Name: "api_dashboard_get",
  109. Help: "summary for dashboard get duration",
  110. })
  111. M_Api_Dashboard_Search = prometheus.NewSummary(prometheus.SummaryOpts{
  112. Name: "api_dashboard_search",
  113. Help: "summary for dashboard search duration",
  114. })
  115. M_Api_Admin_User_Create = prometheus.NewCounter(prometheus.CounterOpts{
  116. Name: "api_admin_user_created_total",
  117. Help: "api admin user created counter",
  118. })
  119. M_Api_Login_Post = prometheus.NewCounter(prometheus.CounterOpts{
  120. Name: "api_login_post_total",
  121. Help: "api login post counter",
  122. })
  123. M_Api_Login_OAuth = prometheus.NewCounter(prometheus.CounterOpts{
  124. Name: "api_login_oauth_total",
  125. Help: "api login oauth counter",
  126. })
  127. M_Api_Org_Create = prometheus.NewCounter(prometheus.CounterOpts{
  128. Name: "api_org_create_total",
  129. Help: "api org created counter",
  130. })
  131. M_Api_Dashboard_Snapshot_Create = prometheus.NewCounter(prometheus.CounterOpts{
  132. Name: "api_dashboard_snapshot_create_total",
  133. Help: "dashboard snapshots created",
  134. })
  135. M_Api_Dashboard_Snapshot_External = prometheus.NewCounter(prometheus.CounterOpts{
  136. Name: "api_dashboard_snapshot_external_total",
  137. Help: "external dashboard snapshots created",
  138. })
  139. M_Api_Dashboard_Snapshot_Get = prometheus.NewCounter(prometheus.CounterOpts{
  140. Name: "api_dashboard_snapshot_get_total",
  141. Help: "loaded dashboards",
  142. })
  143. M_Api_Dashboard_Insert = prometheus.NewCounter(prometheus.CounterOpts{
  144. Name: "api_models_dashboard_insert_total",
  145. Help: "dashboards inserted ",
  146. })
  147. M_Alerting_Result_State = prometheus.NewCounterVec(prometheus.CounterOpts{
  148. Name: "alerting_result_total",
  149. Help: "alert execution result counter",
  150. }, []string{"state"})
  151. M_Alerting_Notification_Sent = prometheus.NewCounterVec(prometheus.CounterOpts{
  152. Name: "alerting_notification_sent_total",
  153. Help: "counter for how many alert notifications been sent",
  154. }, []string{"type"})
  155. M_Aws_CloudWatch_GetMetricStatistics = prometheus.NewCounter(prometheus.CounterOpts{
  156. Name: "aws_cloudwatch_get_metric_statistics_total",
  157. Help: "counter for getting metric statistics from aws",
  158. })
  159. M_Aws_CloudWatch_ListMetrics = prometheus.NewCounter(prometheus.CounterOpts{
  160. Name: "aws_cloudwatch_list_metrics_total",
  161. Help: "counter for getting list of metrics from aws",
  162. })
  163. M_DB_DataSource_QueryById = prometheus.NewCounter(prometheus.CounterOpts{
  164. Name: "db_datasource_query_by_id_total",
  165. Help: "counter for getting datasource by id",
  166. })
  167. M_DataSource_ProxyReq_Timer = prometheus.NewSummary(prometheus.SummaryOpts{
  168. Name: "api_dataproxy_request_all",
  169. Help: "summary for dashboard search duration",
  170. })
  171. M_Alerting_Execution_Time = prometheus.NewSummary(prometheus.SummaryOpts{
  172. Name: "alerting_execution_time_seconds",
  173. Help: "summary of alert exeuction duration",
  174. })
  175. M_Alerting_Active_Alerts = prometheus.NewGauge(prometheus.GaugeOpts{
  176. Name: "alerting_active_alerts",
  177. Help: "amount of active alerts",
  178. })
  179. M_StatTotal_Dashboards = prometheus.NewGauge(prometheus.GaugeOpts{
  180. Name: "stat_totals_dashboard",
  181. Help: "total amount of dashboards",
  182. })
  183. M_StatTotal_Users = prometheus.NewGauge(prometheus.GaugeOpts{
  184. Name: "stat_total_users",
  185. Help: "total amount of users",
  186. })
  187. M_StatTotal_Orgs = prometheus.NewGauge(prometheus.GaugeOpts{
  188. Name: "stat_total_orgs",
  189. Help: "total amount of orgs",
  190. })
  191. M_StatTotal_Playlists = prometheus.NewGauge(prometheus.GaugeOpts{
  192. Name: "stat_total_playlists",
  193. Help: "total amount of playlists",
  194. })
  195. }
  196. func initMetricVars(settings *MetricSettings) {
  197. prometheus.MustRegister(
  198. M_Instance_Start,
  199. M_Page_Status,
  200. M_Api_Status,
  201. M_Proxy_Status,
  202. M_Http_Request_Total,
  203. M_Http_Request_Summary,
  204. M_Api_User_SignUpStarted,
  205. M_Api_User_SignUpCompleted,
  206. M_Api_User_SignUpInvite,
  207. M_Api_Dashboard_Save,
  208. M_Api_Dashboard_Get,
  209. M_Api_Dashboard_Search,
  210. M_DataSource_ProxyReq_Timer,
  211. M_Alerting_Execution_Time,
  212. M_Api_Admin_User_Create,
  213. M_Api_Login_Post,
  214. M_Api_Login_OAuth,
  215. M_Api_Org_Create,
  216. M_Api_Dashboard_Snapshot_Create,
  217. M_Api_Dashboard_Snapshot_External,
  218. M_Api_Dashboard_Snapshot_Get,
  219. M_Api_Dashboard_Insert,
  220. M_Alerting_Result_State,
  221. M_Alerting_Notification_Sent,
  222. M_Aws_CloudWatch_GetMetricStatistics,
  223. M_Aws_CloudWatch_ListMetrics,
  224. M_DB_DataSource_QueryById,
  225. M_Alerting_Active_Alerts,
  226. M_StatTotal_Dashboards,
  227. M_StatTotal_Users,
  228. M_StatTotal_Orgs,
  229. M_StatTotal_Playlists)
  230. go instrumentationLoop(settings)
  231. }
  232. func instrumentationLoop(settings *MetricSettings) chan struct{} {
  233. M_Instance_Start.Inc()
  234. onceEveryDayTick := time.NewTicker(time.Hour * 24)
  235. secondTicker := time.NewTicker(time.Second * time.Duration(settings.IntervalSeconds))
  236. for {
  237. select {
  238. case <-onceEveryDayTick.C:
  239. sendUsageStats()
  240. case <-secondTicker.C:
  241. updateTotalStats()
  242. }
  243. }
  244. }
  245. var metricPublishCounter int64 = 0
  246. func updateTotalStats() {
  247. metricPublishCounter++
  248. if metricPublishCounter == 1 || metricPublishCounter%10 == 0 {
  249. statsQuery := models.GetSystemStatsQuery{}
  250. if err := bus.Dispatch(&statsQuery); err != nil {
  251. metricsLogger.Error("Failed to get system stats", "error", err)
  252. return
  253. }
  254. M_StatTotal_Dashboards.Set(float64(statsQuery.Result.Dashboards))
  255. M_StatTotal_Users.Set(float64(statsQuery.Result.Users))
  256. M_StatTotal_Playlists.Set(float64(statsQuery.Result.Playlists))
  257. M_StatTotal_Orgs.Set(float64(statsQuery.Result.Orgs))
  258. }
  259. }
  260. func sendUsageStats() {
  261. if !setting.ReportingEnabled {
  262. return
  263. }
  264. metricsLogger.Debug("Sending anonymous usage stats to stats.grafana.org")
  265. version := strings.Replace(setting.BuildVersion, ".", "_", -1)
  266. metrics := map[string]interface{}{}
  267. report := map[string]interface{}{
  268. "version": version,
  269. "metrics": metrics,
  270. "os": runtime.GOOS,
  271. "arch": runtime.GOARCH,
  272. }
  273. statsQuery := models.GetSystemStatsQuery{}
  274. if err := bus.Dispatch(&statsQuery); err != nil {
  275. metricsLogger.Error("Failed to get system stats", "error", err)
  276. return
  277. }
  278. metrics["stats.dashboards.count"] = statsQuery.Result.Dashboards
  279. metrics["stats.users.count"] = statsQuery.Result.Users
  280. metrics["stats.orgs.count"] = statsQuery.Result.Orgs
  281. metrics["stats.playlist.count"] = statsQuery.Result.Playlists
  282. metrics["stats.plugins.apps.count"] = len(plugins.Apps)
  283. metrics["stats.plugins.panels.count"] = len(plugins.Panels)
  284. metrics["stats.plugins.datasources.count"] = len(plugins.DataSources)
  285. metrics["stats.alerts.count"] = statsQuery.Result.Alerts
  286. metrics["stats.active_users.count"] = statsQuery.Result.ActiveUsers
  287. metrics["stats.datasources.count"] = statsQuery.Result.Datasources
  288. dsStats := models.GetDataSourceStatsQuery{}
  289. if err := bus.Dispatch(&dsStats); err != nil {
  290. metricsLogger.Error("Failed to get datasource stats", "error", err)
  291. return
  292. }
  293. // send counters for each data source
  294. // but ignore any custom data sources
  295. // as sending that name could be sensitive information
  296. dsOtherCount := 0
  297. for _, dsStat := range dsStats.Result {
  298. if models.IsKnownDataSourcePlugin(dsStat.Type) {
  299. metrics["stats.ds."+dsStat.Type+".count"] = dsStat.Count
  300. } else {
  301. dsOtherCount += dsStat.Count
  302. }
  303. }
  304. metrics["stats.ds.other.count"] = dsOtherCount
  305. out, _ := json.MarshalIndent(report, "", " ")
  306. data := bytes.NewBuffer(out)
  307. client := http.Client{Timeout: time.Duration(5 * time.Second)}
  308. go client.Post("https://stats.grafana.org/grafana-usage-report", "application/json", data)
  309. }