apm_tracing.go 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489
  1. package tracing
  2. import (
  3. "context"
  4. "fmt"
  5. "github.com/coroot/coroot-node-agent/ebpftracer"
  6. "github.com/coroot/coroot-node-agent/ebpftracer/l7"
  7. "github.com/coroot/coroot-node-agent/utils"
  8. . "github.com/coroot/coroot-node-agent/utils/modelse"
  9. klog "github.com/sirupsen/logrus"
  10. "go.opentelemetry.io/otel/attribute"
  11. "go.opentelemetry.io/otel/codes"
  12. semconv "go.opentelemetry.io/otel/semconv/v1.18.0"
  13. "go.opentelemetry.io/otel/trace"
  14. "inet.af/netaddr"
  15. "strconv"
  16. "sync/atomic"
  17. "time"
  18. )
  19. /**
  20. * Trace
  21. */
  22. func (t *Trace) setContext(ctx context.Context) {
  23. t.lock.Lock()
  24. defer t.lock.Unlock()
  25. t.ctx = ctx
  26. }
  27. func (t *Trace) setSpan(span trace.Span) {
  28. t.lock.Lock()
  29. defer t.lock.Unlock()
  30. t.span = span
  31. }
  32. type TimeMap struct {
  33. Time uint64
  34. Type int
  35. Map *ebpftracer.StackFunEvent
  36. }
  37. //func (t *Trace) buildFun() {
  38. // mapSlice := []TimeMap{}
  39. // for i, v := range t.stack {
  40. // timeStartMap := TimeMap{}
  41. // if v.StackEvent.Location == 0 {
  42. // timeStartMap = TimeMap{
  43. // Time: v.StackEvent.TimeNsStart,
  44. // Type: 0,
  45. // Map: &t.stack[i],
  46. // }
  47. // } else {
  48. // timeStartMap = TimeMap{
  49. // Time: v.StackEvent.TimeNsEnd,
  50. // Type: 1,
  51. // Map: &t.stack[i],
  52. // }
  53. // }
  54. // mapSlice = append(mapSlice, timeStartMap)
  55. // }
  56. // sort.Slice(mapSlice, func(i, j int) bool {
  57. // return mapSlice[i].Time < mapSlice[j].Time
  58. // })
  59. //
  60. // funStack := []TimeMap{}
  61. //
  62. // currentfunNum := 1
  63. //
  64. // // for k, v := range mapSlice {
  65. // // fmt.Println("---SliceSliceindex", k, "value", v.Time, v.Type, v.Map.Uprobe.Funcname, v.Map.StackEvent.Nid)
  66. // // }
  67. //
  68. // mapSliceLen := len(mapSlice)
  69. // for k, v := range mapSlice {
  70. // // fmt.Println("SliceSliceindex", k, "value", v.Time, v.Type, v.Map.Uprobe.Funcname, v.Map.StackEvent.Nid)
  71. // if v.Type == 0 {
  72. // // 函数入口
  73. // funStack = append(funStack, v)
  74. // } else if v.Type == 1 {
  75. // // 函数出口
  76. // len := len(funStack)
  77. // if len < 1 {
  78. // fmt.Printf("buildFun ErrorError return before enter: %v\n", v)
  79. // continue
  80. // }
  81. // currnt := funStack[len-1]
  82. // if currnt.Map.StackEvent.Location != 0 {
  83. // fmt.Printf("currnt StackEvent ErrorError is not enter: %v\n", v)
  84. // continue
  85. // }
  86. // if k < mapSliceLen-1 && len >= 2 {
  87. // nextfun := mapSlice[k+1]
  88. // preCurrnt := funStack[len-2]
  89. // // // 处理 .netcore 多次 returun
  90. // // 下一个事件是 return 并且函数名跟当前事件是一样的,且上一个函数不是当前函数
  91. // if nextfun.Map.StackEvent.Location == 1 && nextfun.Map.Uprobe.Funcname == currnt.Map.Uprobe.Funcname && preCurrnt.Map.Uprobe.Funcname != currnt.Map.Uprobe.Funcname {
  92. // currentfunNum++
  93. // continue
  94. // }
  95. // }
  96. // funStack = funStack[:len-1]
  97. // duration := v.Map.StackEvent.TimeNsEnd - currnt.Map.StackEvent.TimeNsStart
  98. // t.FuncTraceQuery(currnt.Map.Uprobe.Funcname, time.Duration(duration), currnt.Map.StackEvent.TimeNsStart, v.Map.StackEvent.TimeNsEnd, currentfunNum)
  99. // currentfunNum = 1
  100. // }
  101. // }
  102. //}
  103. func (t *Trace) startReady() {
  104. t.lock.Lock()
  105. defer t.lock.Unlock()
  106. t.startEventReady = true
  107. }
  108. func (t *Trace) endReadyEvent(needCount uint32) {
  109. t.lock.Lock()
  110. defer t.lock.Unlock()
  111. t.endEventReady = true
  112. t.needEventCount = needCount
  113. }
  114. func (t *Trace) AllEventReady(traceID uint64) bool {
  115. klog.Infof("[AllEventReady (current/need)|start|end|traceid](%d/%d)%v|%v|%d", *t.currenEventCount, t.needEventCount, t.startEventReady, t.endEventReady, traceID)
  116. return t.startEventReady && t.endEventReady && *t.currenEventCount >= t.needEventCount
  117. }
  118. func (t *Trace) TraceStartEvent(method, path string, status l7.Status, addr netaddr.IPPort, pid uint32, appInfo AppInfo) {
  119. t.span.SetAttributes(semconv.HTTPURL(fmt.Sprintf("http://%s%s", addr.String(), path)),
  120. semconv.HTTPMethod(method),
  121. attribute.String("http.uri", path))
  122. if status > 399 {
  123. t.span.SetStatus(codes.Error, "")
  124. }
  125. t.destination = addr
  126. t.commonAttrs = []attribute.KeyValue{
  127. semconv.NetPeerName(addr.IP().String()),
  128. semconv.NetPeerPort(int(addr.Port())),
  129. // buildAppMapFromEvent
  130. attribute.Int("server.code_type", appInfo.CodeType.Int()),
  131. attribute.String("server.app_name", appInfo.AppName),
  132. attribute.String("server.service_name", appInfo.ServiceName),
  133. attribute.Int64("server.app_id", appInfo.AppIdHash.IntVal),
  134. attribute.Int64("server.agent_id", appInfo.AgentId),
  135. attribute.Int64("server.instance_id", appInfo.InstanceIdHash.IntVal),
  136. }
  137. t.span.SetAttributes(t.commonAttrs...)
  138. t.startReady()
  139. }
  140. // set context span
  141. func (t *Trace) CreateRootSpan(traceId uint64) {
  142. traceIdStr := strconv.Itoa(int(traceId))
  143. ctx, span := tracer(t.containerId).Start(context.Background(), traceIdStr, trace.WithSpanKind(trace.SpanKindClient))
  144. t.setContext(ctx)
  145. t.setSpan(span)
  146. }
  147. func (t *Trace) TraceStart(method, path string, status l7.Status, duration time.Duration) {
  148. if t == nil || method == "" {
  149. return
  150. }
  151. t.createParentSpan("APPLICATION", duration, status >= 400,
  152. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  153. semconv.HTTPMethod(method),
  154. //semconv.HTTPStatusCode(int(status)),
  155. attribute.String("http.uri", path),
  156. )
  157. }
  158. func (t *Trace) TraceEnd(r *l7.RequestData) {
  159. if t == nil {
  160. return
  161. }
  162. t.span.SetAttributes(
  163. semconv.HTTPStatusCode(int(r.Status)),
  164. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  165. )
  166. CalledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  167. if err == nil && CalledId != 0 {
  168. t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  169. }
  170. InstanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  171. if err == nil && InstanceIdFrom != 0 {
  172. t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  173. }
  174. AppIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  175. if err == nil && AppIdFrom != 0 {
  176. t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  177. }
  178. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  179. t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  180. }
  181. // for _, v := range t.stack {
  182. // fmt.Printf("TraceEndTraceEndTraceEnd%s\n", v)
  183. // }
  184. //t.buildFun()
  185. t.span.End(trace.WithTimestamp(time.Now()))
  186. }
  187. // 新增结束事件
  188. func (t *Trace) TraceEndEvent(r *l7.RequestData) {
  189. if t == nil {
  190. return
  191. }
  192. var attr []attribute.KeyValue
  193. attr = append(attr,
  194. semconv.HTTPStatusCode(int(r.Status)),
  195. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  196. )
  197. //t.span.SetAttributes(
  198. // semconv.HTTPStatusCode(int(r.Status)),
  199. // attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  200. //)
  201. calledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  202. if err == nil && calledId != 0 {
  203. attr = append(attr, attribute.Int64("server.called_id", calledId))
  204. //t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  205. }
  206. instanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  207. if err == nil && instanceIdFrom != 0 {
  208. attr = append(attr, attribute.Int64("server.instance_id_from", instanceIdFrom))
  209. //t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  210. }
  211. appIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  212. if err == nil && appIdFrom != 0 {
  213. attr = append(attr, attribute.Int64("server.app_id_from", appIdFrom))
  214. //t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  215. }
  216. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  217. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  218. //t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  219. }
  220. attr = append(attr, attribute.String("server.src_addr", r.SAddr.String()))
  221. attr = append(attr, attribute.String("server.dst_addr", r.DAddr.String()))
  222. klog.Infof("TraceEndEvent SAddr.String %s", r.SAddr.String())
  223. klog.Infof("TraceEndEvent DAddr.String %s", r.DAddr.String())
  224. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  225. t.span.SetAttributes(attr...)
  226. t.endReadyEvent(r.EventCount)
  227. }
  228. func (t *Trace) appendTimestamp(attr *[]attribute.KeyValue, s, e uint64, d int64) {
  229. *attr = append(*attr,
  230. attribute.Int64("time.start_at", utils.KtimeToTimestamp(s)),
  231. attribute.Int64("time.end_at", utils.KtimeToTimestamp(e)),
  232. attribute.Int64("time.duration", d),
  233. )
  234. }
  235. func (t *Trace) createParentSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  236. end := time.Now()
  237. start := end.Add(-duration)
  238. ctx, span := tracer(t.containerId).Start(context.Background(), name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  239. span.SetAttributes(attrs...)
  240. span.SetAttributes(t.commonAttrs...)
  241. if error {
  242. span.SetStatus(codes.Error, "")
  243. }
  244. t.setContext(ctx)
  245. t.setSpan(span)
  246. }
  247. func (t *Trace) SendEvent() {
  248. t.span.End()
  249. }
  250. func (t *Trace) GetSpan() trace.Span {
  251. return t.span
  252. }
  253. func (t *Trace) createTraceEvent(name string, eventType int, l7Type int, attrs ...attribute.KeyValue) {
  254. t.span.AddEventApm(name, eventType, l7Type, trace.WithAttributes(attrs...))
  255. atomic.AddUint32(t.currenEventCount, 1)
  256. }
  257. func (t *Trace) createTraceSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  258. end := time.Now()
  259. start := end.Add(-duration)
  260. //fmt.Println("createTraceSpan:", t.ctx)
  261. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  262. span.SetAttributes(t.commonAttrs...)
  263. span.SetAttributes(attrs...)
  264. if error {
  265. span.SetStatus(codes.Error, "")
  266. }
  267. span.End(trace.WithTimestamp(end))
  268. }
  269. func (t *Trace) MysqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  270. if t == nil || query == "" {
  271. return
  272. }
  273. t.createTraceSpan(l7.ProtocolMysql.String(), duration, error,
  274. semconv.DBSystemMySQL,
  275. semconv.DBStatement(query),
  276. semconv.NetPeerName(destination.IP().String()),
  277. semconv.NetPeerPort(int(destination.Port())),
  278. )
  279. }
  280. func (t *Trace) MysqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  281. fmt.Println("query", query)
  282. if t == nil {
  283. return
  284. }
  285. var attr []attribute.KeyValue
  286. attr = append(attr,
  287. semconv.DBSystemMySQL,
  288. semconv.DBStatement(query),
  289. semconv.NetPeerName(destination.IP().String()),
  290. semconv.NetPeerPort(int(destination.Port())),
  291. )
  292. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  293. t.createTraceEvent(l7.ProtocolMysql.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolMysql), attr...)
  294. }
  295. func (t *Trace) DmTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  296. if t == nil || query == "" {
  297. return
  298. }
  299. l7Type := int(l7.ProtocolDM)
  300. var attr []attribute.KeyValue
  301. attr = append(attr,
  302. semconv.DBSystemDaMengDB,
  303. semconv.DBStatement(query),
  304. semconv.NetPeerName(destination.IP().String()),
  305. semconv.NetPeerPort(int(destination.Port())),
  306. )
  307. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  308. attr = append(attr,
  309. attribute.Bool("sql.exception", r.Status.Error()),
  310. )
  311. t.createTraceEvent(l7.ProtocolHTTP.String(), int(ebpftracer.EventTypeL7Request), l7Type, attr...)
  312. }
  313. func (t *Trace) RedisTraceQuery(cmd, args string, error bool, duration time.Duration) {
  314. if t == nil || cmd == "" {
  315. return
  316. }
  317. statement := cmd
  318. if args != "" {
  319. statement += " " + args
  320. }
  321. t.createTraceSpan(l7.ProtocolRedis.String(), duration, error,
  322. semconv.DBSystemRedis,
  323. semconv.DBOperation(cmd),
  324. semconv.DBStatement(statement),
  325. )
  326. }
  327. func (t *Trace) RedisTraceQueryEvent(cmd, args string, r *l7.RequestData, destination netaddr.IPPort) {
  328. if t == nil || cmd == "" {
  329. return
  330. }
  331. statement := cmd
  332. if args != "" {
  333. statement += " " + args
  334. }
  335. var attr []attribute.KeyValue
  336. attr = append(attr,
  337. semconv.DBSystemRedis,
  338. semconv.DBOperation(cmd),
  339. semconv.DBStatement(statement),
  340. semconv.NetPeerName(destination.IP().String()),
  341. semconv.NetPeerPort(int(destination.Port())),
  342. )
  343. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  344. t.createTraceEvent(l7.ProtocolRedis.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolRedis), attr...)
  345. }
  346. func (t *Trace) DNSTraceQueryEvent(r *l7.RequestData, _type, fqdn string) {
  347. if t == nil {
  348. return
  349. }
  350. var attr []attribute.KeyValue
  351. attr = append(attr,
  352. attribute.String("dns.type", _type),
  353. attribute.String("dns.fqdn", fqdn),
  354. )
  355. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  356. t.createTraceEvent(l7.ProtocolDNS.String()+":"+_type+":"+fqdn, int(ebpftracer.EventTypeL7Request), int(l7.ProtocolDNS), attr...)
  357. }
  358. func (t *Trace) HttpTraceRequest(method, path, ip string, port uint16, r *l7.RequestData) {
  359. if t == nil || method == "" {
  360. return
  361. }
  362. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  363. if err != nil {
  364. assumedAppID = 0
  365. }
  366. status := r.Status
  367. duration := r.Duration
  368. t.createTraceSpan(l7.ProtocolHTTP.String(), duration, status >= 400,
  369. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  370. semconv.HTTPMethod(method),
  371. semconv.HTTPStatusCode(int(status)),
  372. attribute.String("http.uri", path),
  373. attribute.String("http.ip", ip),
  374. attribute.Int64("http.assumed_app_id", assumedAppID),
  375. attribute.String("http.span_id", r.SpanId),
  376. attribute.Int("http.port", int(port)),
  377. )
  378. }
  379. // 新增事件处理
  380. func (t *Trace) HttpTraceRequestEvent(method, path, ip string, port uint16, r *l7.RequestData) {
  381. if t == nil || method == "" {
  382. return
  383. }
  384. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  385. if err != nil {
  386. assumedAppID = 0
  387. }
  388. status := r.Status
  389. var attr []attribute.KeyValue
  390. attr = append(attr,
  391. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  392. semconv.HTTPMethod(method),
  393. semconv.HTTPStatusCode(int(status)),
  394. attribute.Bool("http.status_error", status > 399),
  395. attribute.String("http.uri", path),
  396. attribute.String("http.ip", ip),
  397. attribute.Int64("http.assumed_app_id", assumedAppID),
  398. attribute.String("http.span_id", r.SpanId),
  399. attribute.Int("http.port", int(port)),
  400. )
  401. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  402. t.createTraceEvent(l7.ProtocolHTTP.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolHTTP), attr...)
  403. }
  404. func (t *Trace) FuncTraceQuery(funcname string, duration time.Duration, start uint64, end uint64) {
  405. if t == nil || funcname == "" {
  406. return
  407. }
  408. t.createTraceSpanNoTime2(funcname, duration, false, start, end)
  409. //t.createTraceSpanNoTime2(funcname, duration, false, start, end, attribute.Int("num", num))
  410. }
  411. func (t *Trace) createTraceSpanNoTime(name string, duration time.Duration, error bool, start uint64, end uint64, attrs ...attribute.KeyValue) {
  412. // end := time.Now()
  413. // start := end.Add(-duration)
  414. startTime := time.Unix(0, int64(start))
  415. endTime := time.Unix(0, int64(end))
  416. //fmt.Println("createTraceSpan:", t.ctx)
  417. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  418. span.SetAttributes(t.commonAttrs...)
  419. span.SetAttributes(attrs...)
  420. if error {
  421. span.SetStatus(codes.Error, "")
  422. }
  423. span.End(trace.WithTimestamp(endTime))
  424. }
  425. func (t *Trace) createTraceSpanNoTime2(name string, duration time.Duration, error bool, start uint64, end uint64) {
  426. // end := time.Now()
  427. // start := end.Add(-duration)
  428. //startTime := time.Unix(0, int64(start))
  429. //endTime := time.Unix(0, int64(end))
  430. ////fmt.Println("createTraceSpan:", t.ctx)
  431. //_, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  432. //span.SetAttributes(t.commonAttrs...)
  433. //span.SetAttributes(attrs...)
  434. //if error {
  435. // span.SetStatus(codes.Error, "")
  436. //}
  437. //span.End(trace.WithTimestamp(endTime))
  438. //attrs = append([]attribute.KeyValue{
  439. // attribute.Int64("startAt", int64(start)),
  440. // attribute.Int64("endAt", int64(end)),
  441. //})
  442. var attr []attribute.KeyValue
  443. t.appendTimestamp(&attr, start, end, int64(end-start))
  444. t.createTraceEvent(name, int(ebpftracer.EventTypeFunEnt), 0, attr...)
  445. }