apm_tracing.go 26 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827
  1. package tracing
  2. import (
  3. "context"
  4. "fmt"
  5. "os"
  6. "strconv"
  7. "strings"
  8. "sync"
  9. "sync/atomic"
  10. "time"
  11. klog "github.com/sirupsen/logrus"
  12. "github.com/coroot/coroot-node-agent/ebpftracer"
  13. "github.com/coroot/coroot-node-agent/ebpftracer/l7"
  14. "github.com/coroot/coroot-node-agent/utils"
  15. . "github.com/coroot/coroot-node-agent/utils/modelse"
  16. "go.opentelemetry.io/otel/attribute"
  17. "go.opentelemetry.io/otel/codes"
  18. semconv "go.opentelemetry.io/otel/semconv/v1.18.0"
  19. "go.opentelemetry.io/otel/trace"
  20. "inet.af/netaddr"
  21. )
  22. /**
  23. * Trace
  24. */
  25. func (t *Trace) setContext(ctx context.Context) {
  26. t.lock.Lock()
  27. defer t.lock.Unlock()
  28. t.ctx = ctx
  29. }
  30. func (t *Trace) setSpan(span trace.Span) {
  31. t.lock.Lock()
  32. defer t.lock.Unlock()
  33. t.span = span
  34. }
  35. type TimeMap struct {
  36. Time uint64
  37. Type int
  38. Map *ebpftracer.StackFunEvent
  39. }
  40. //func (t *Trace) buildFun() {
  41. // mapSlice := []TimeMap{}
  42. // for i, v := range t.stack {
  43. // timeStartMap := TimeMap{}
  44. // if v.StackEvent.Location == 0 {
  45. // timeStartMap = TimeMap{
  46. // Time: v.StackEvent.TimeNsStart,
  47. // Type: 0,
  48. // Map: &t.stack[i],
  49. // }
  50. // } else {
  51. // timeStartMap = TimeMap{
  52. // Time: v.StackEvent.TimeNsEnd,
  53. // Type: 1,
  54. // Map: &t.stack[i],
  55. // }
  56. // }
  57. // mapSlice = append(mapSlice, timeStartMap)
  58. // }
  59. // sort.Slice(mapSlice, func(i, j int) bool {
  60. // return mapSlice[i].Time < mapSlice[j].Time
  61. // })
  62. //
  63. // funStack := []TimeMap{}
  64. //
  65. // currentfunNum := 1
  66. //
  67. // // for k, v := range mapSlice {
  68. // // fmt.Println("---SliceSliceindex", k, "value", v.Time, v.Type, v.Map.Uprobe.Funcname, v.Map.StackEvent.Nid)
  69. // // }
  70. //
  71. // mapSliceLen := len(mapSlice)
  72. // for k, v := range mapSlice {
  73. // // fmt.Println("SliceSliceindex", k, "value", v.Time, v.Type, v.Map.Uprobe.Funcname, v.Map.StackEvent.Nid)
  74. // if v.Type == 0 {
  75. // // 函数入口
  76. // funStack = append(funStack, v)
  77. // } else if v.Type == 1 {
  78. // // 函数出口
  79. // len := len(funStack)
  80. // if len < 1 {
  81. // fmt.Printf("buildFun ErrorError return before enter: %v\n", v)
  82. // continue
  83. // }
  84. // currnt := funStack[len-1]
  85. // if currnt.Map.StackEvent.Location != 0 {
  86. // fmt.Printf("currnt StackEvent ErrorError is not enter: %v\n", v)
  87. // continue
  88. // }
  89. // if k < mapSliceLen-1 && len >= 2 {
  90. // nextfun := mapSlice[k+1]
  91. // preCurrnt := funStack[len-2]
  92. // // // 处理 .netcore 多次 returun
  93. // // 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
  94. // if nextfun.Map.StackEvent.Location == 1 && nextfun.Map.Uprobe.Funcname == currnt.Map.Uprobe.Funcname && preCurrnt.Map.Uprobe.Funcname != currnt.Map.Uprobe.Funcname {
  95. // currentfunNum++
  96. // continue
  97. // }
  98. // }
  99. // funStack = funStack[:len-1]
  100. // duration := v.Map.StackEvent.TimeNsEnd - currnt.Map.StackEvent.TimeNsStart
  101. // t.FuncTraceQuery(currnt.Map.Uprobe.Funcname, time.Duration(duration), currnt.Map.StackEvent.TimeNsStart, v.Map.StackEvent.TimeNsEnd, currentfunNum)
  102. // currentfunNum = 1
  103. // }
  104. // }
  105. //}
  106. func (t *Trace) startReady() {
  107. t.lock.Lock()
  108. defer t.lock.Unlock()
  109. t.startEventReady = true
  110. }
  111. func (t *Trace) endReadyEvent(needCount uint32) {
  112. t.lock.Lock()
  113. defer t.lock.Unlock()
  114. t.endEventReady = true
  115. t.needEventCount = needCount
  116. }
  117. func (t *Trace) AllEventReady(traceID uint64) bool {
  118. klog.Debugf("[AllEventReady (current/need)|start|end|traceid](%d/%d)%v|%v|%d", *t.currenEventCount, t.needEventCount, t.startEventReady, t.endEventReady, traceID)
  119. return t.startEventReady && t.endEventReady && *t.currenEventCount >= t.needEventCount
  120. }
  121. func (t *Trace) GrpcServerTraceStartEvent(sn string, sport uint16, r *l7.RequestData, appInfo AppInfo, container_id string) {
  122. t.span.SetAttributes(attribute.String("rpc.uri", string(r.Payload)))
  123. t.commonAttrs = []attribute.KeyValue{
  124. semconv.NetPeerName(sn),
  125. semconv.NetPeerPort(int(sport)),
  126. // buildAppMapFromEvent
  127. attribute.Int("server.code_type", appInfo.CodeType.Int()),
  128. attribute.String("server.app_name", appInfo.AppName),
  129. attribute.String("server.service_name", "GRPC"),
  130. attribute.String("rpc.service_type", "RPC"),
  131. attribute.String("rpc.oper_type", "PROVIDER"),
  132. attribute.Int64("server.app_id", appInfo.AppIdHash.IntVal),
  133. attribute.Int64("server.agent_id", appInfo.AgentId),
  134. attribute.Int64("server.instance_id", appInfo.InstanceIdHash.IntVal),
  135. attribute.String("server.container_id", container_id),
  136. }
  137. t.span.SetAttributes(t.commonAttrs...)
  138. t.startReady()
  139. }
  140. func (t *Trace) TraceStartEvent(method, path, sn, ua string, sport uint16, status l7.Status, addr netaddr.IPPort, pid uint32, appInfo AppInfo, container_id string) {
  141. t.span.SetAttributes(semconv.HTTPURL(fmt.Sprintf("http://%s:%d%s", sn, sport, path)),
  142. semconv.HTTPMethod(method),
  143. attribute.String("http.uri", path))
  144. if status > 399 {
  145. t.span.SetStatus(codes.Error, "")
  146. }
  147. //host := fmt.Sprintf("%s:%s", sn, sport)
  148. t.destination = addr
  149. t.commonAttrs = []attribute.KeyValue{
  150. semconv.NetPeerName(sn),
  151. semconv.NetPeerPort(int(sport)),
  152. // buildAppMapFromEvent
  153. attribute.Int("server.code_type", appInfo.CodeType.Int()),
  154. attribute.String("server.app_name", appInfo.AppName),
  155. attribute.String("server.service_name", appInfo.ServiceType),
  156. attribute.Int64("server.app_id", appInfo.AppIdHash.IntVal),
  157. attribute.Int64("server.agent_id", appInfo.AgentId),
  158. attribute.Int64("server.instance_id", appInfo.InstanceIdHash.IntVal),
  159. attribute.Int("server.pid", int(pid)),
  160. attribute.String("server.container_id", container_id),
  161. attribute.String("server.user_agent", ua),
  162. }
  163. t.span.SetAttributes(t.commonAttrs...)
  164. t.startReady()
  165. }
  166. // set context span
  167. func (t *Trace) CreateRootSpan(traceId uint64) {
  168. traceIdStr := strconv.Itoa(int(traceId))
  169. ctx, span := tracer(t.containerId).Start(context.Background(), traceIdStr, trace.WithSpanKind(trace.SpanKindClient))
  170. t.setContext(ctx)
  171. t.setSpan(span)
  172. }
  173. // Deprecated: use TraceStartEvent instead.
  174. func (t *Trace) TraceStart(method, path string, status l7.Status, duration time.Duration) {
  175. if t == nil || method == "" {
  176. return
  177. }
  178. t.createParentSpan("APPLICATION", duration, status >= 400,
  179. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  180. semconv.HTTPMethod(method),
  181. //semconv.HTTPStatusCode(int(status)),
  182. attribute.String("http.uri", path),
  183. )
  184. }
  185. // Deprecated: use TraceEndEvent instead.
  186. func (t *Trace) TraceEnd(r *l7.RequestData) {
  187. if t == nil {
  188. return
  189. }
  190. t.span.SetAttributes(
  191. semconv.HTTPStatusCode(int(r.Status)),
  192. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  193. )
  194. CalledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  195. if err == nil && CalledId != 0 {
  196. t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  197. }
  198. InstanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  199. if err == nil && InstanceIdFrom != 0 {
  200. t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  201. }
  202. AppIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  203. if err == nil && AppIdFrom != 0 {
  204. t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  205. }
  206. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  207. t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  208. }
  209. // klog.Debugf("r.ParentSpanContext.TypeFrom tpp %s", r.ParentSpanContext.TypeFrom)
  210. if r.ParentSpanContext.TypeFrom == "10" {
  211. t.span.SetAttributes(attribute.String("server.type_from", "SDK"))
  212. t.span.SetAttributes(attribute.String("server.app_id_from", "-1"))
  213. t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  214. t.span.SetAttributes(attribute.String("server.instance_id_from", "-1"))
  215. t.span.SetAttributes(attribute.String("server.call_id", "-1"))
  216. }
  217. // for _, v := range t.stack {
  218. // fmt.Printf("TraceEndTraceEndTraceEnd%s\n", v)
  219. // }
  220. //t.buildFun()
  221. t.span.End(trace.WithTimestamp(time.Now()))
  222. }
  223. // 新增结束事件
  224. func (t *Trace) TraceEndEvent(r *l7.RequestData) {
  225. if t == nil {
  226. return
  227. }
  228. var attr []attribute.KeyValue
  229. attr = append(attr,
  230. semconv.HTTPStatusCode(int(r.Status)),
  231. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  232. )
  233. //t.span.SetAttributes(
  234. // semconv.HTTPStatusCode(int(r.Status)),
  235. // attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  236. //)
  237. calledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  238. if err == nil && calledId != 0 {
  239. attr = append(attr, attribute.Int64("server.called_id", calledId))
  240. //t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  241. }
  242. instanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  243. if err == nil && instanceIdFrom != 0 {
  244. attr = append(attr, attribute.Int64("server.instance_id_from", instanceIdFrom))
  245. //t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  246. }
  247. appIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  248. if err == nil && appIdFrom != 0 {
  249. attr = append(attr, attribute.Int64("server.app_id_from", appIdFrom))
  250. //t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  251. }
  252. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  253. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  254. //t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  255. }
  256. // klog.Debugf("r.ParentSpanContext.TypeFrom event tpp %s", r.ParentSpanContext.TypeFrom)
  257. if r.ParentSpanContext.TypeFrom == "10" {
  258. attr = append(attr, attribute.String("server.type_from", "SDK"))
  259. attr = append(attr, attribute.String("server.app_id_from", "-1"))
  260. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  261. attr = append(attr, attribute.String("server.instance_id_from", "-1"))
  262. attr = append(attr, attribute.String("server.call_id", "-1"))
  263. } else if r.ParentSpanContext.TypeFrom == "02" {
  264. attr = append(attr, attribute.String("server.type_from", "Nodejs"))
  265. }
  266. attr = append(attr, attribute.String("server.src_addr", r.DAddr.String()))
  267. attr = append(attr, attribute.String("server.dst_addr", r.SAddr.String()))
  268. klog.Debugf("TraceEndEvent SAddr.String %s", r.SAddr.String())
  269. klog.Debugf("TraceEndEvent DAddr.String %s", r.DAddr.String())
  270. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  271. t.span.SetAttributes(attr...)
  272. t.endReadyEvent(r.EventCount)
  273. }
  274. func (t *Trace) appendTimestamp(attr *[]attribute.KeyValue, s, e uint64, d int64) {
  275. *attr = append(*attr,
  276. attribute.Int64("time.start_at", utils.KtimeToTimestamp(s)),
  277. attribute.Int64("time.end_at", utils.KtimeToTimestamp(e)),
  278. attribute.Int64("time.duration", d),
  279. )
  280. }
  281. func (t *Trace) createParentSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  282. end := time.Now()
  283. start := end.Add(-duration)
  284. ctx, span := tracer(t.containerId).Start(context.Background(), name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  285. span.SetAttributes(attrs...)
  286. span.SetAttributes(t.commonAttrs...)
  287. if error {
  288. span.SetStatus(codes.Error, "")
  289. }
  290. t.setContext(ctx)
  291. t.setSpan(span)
  292. }
  293. func (t *Trace) SendEvent() {
  294. t.span.End()
  295. }
  296. func (t *Trace) GetSpan() trace.Span {
  297. return t.span
  298. }
  299. func (t *Trace) createTraceEvent(protocol l7.Protocol, eventType int, attrs ...attribute.KeyValue) {
  300. t.span.AddEventApm(protocol.String(), eventType, protocol.Int(), trace.WithAttributes(attrs...))
  301. //atomic.AddUint32(t.currenEventCount, 1)
  302. }
  303. func (t *Trace) createFuncTraceEvent(name string, eventType int, attrs ...attribute.KeyValue) {
  304. t.span.AddEventApm(name, eventType, 0, trace.WithAttributes(attrs...))
  305. }
  306. func (t *Trace) addEvent() {
  307. atomic.AddUint32(t.currenEventCount, 1)
  308. }
  309. func (t *Trace) createTraceSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  310. end := time.Now()
  311. start := end.Add(-duration)
  312. //fmt.Println("createTraceSpan:", t.ctx)
  313. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  314. span.SetAttributes(t.commonAttrs...)
  315. span.SetAttributes(attrs...)
  316. if error {
  317. span.SetStatus(codes.Error, "")
  318. }
  319. span.End(trace.WithTimestamp(end))
  320. }
  321. func (t *Trace) MysqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  322. if t == nil || query == "" {
  323. return
  324. }
  325. t.createTraceSpan(l7.ProtocolMysql.String(), duration, error,
  326. semconv.DBSystemMySQL,
  327. semconv.DBStatement(query),
  328. semconv.NetPeerName(destination.IP().String()),
  329. semconv.NetPeerPort(int(destination.Port())),
  330. )
  331. }
  332. // 内置允许操作(始终放行)
  333. var builtinOps = map[string]struct{}{
  334. "SELECT": {},
  335. "INSERT": {},
  336. "UPDATE": {},
  337. "DELETE": {},
  338. "WITH": {},
  339. }
  340. var (
  341. allowedOps map[string]struct{}
  342. initOnce sync.Once
  343. )
  344. // 解析并缓存环境变量中的前缀
  345. func initAllowedOps() {
  346. allowedOps = make(map[string]struct{})
  347. env := os.Getenv("ALLOWED_SQL_PREFIXES")
  348. for _, s := range strings.Split(env, ",") {
  349. if trimmed := strings.ToUpper(strings.TrimSpace(s)); trimmed != "" {
  350. allowedOps[trimmed] = struct{}{}
  351. }
  352. }
  353. }
  354. func isCURDOperation(q string) bool {
  355. if len(q) < 6 {
  356. return false
  357. }
  358. // 提取前缀(考虑最长10个字符)
  359. prefix := strings.ToUpper(strings.TrimSpace(q))
  360. if len(prefix) > 10 {
  361. prefix = prefix[:10]
  362. }
  363. // 提取第一个单词作为操作关键字
  364. spaceIdx := strings.IndexAny(prefix, " \t\n")
  365. if spaceIdx > 0 {
  366. prefix = prefix[:spaceIdx]
  367. }
  368. // 先查固定允许的操作
  369. if _, ok := builtinOps[prefix]; ok {
  370. return true
  371. }
  372. // 加载环境变量控制的操作(只执行一次)
  373. initOnce.Do(initAllowedOps)
  374. // 查额外允许操作
  375. _, ok := allowedOps[prefix]
  376. return ok
  377. }
  378. func (t *Trace) SQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, query string, r *l7.RequestData, destination netaddr.IPPort) {
  379. //fmt.Println("query", query)
  380. if t == nil {
  381. return
  382. }
  383. t.addEvent()
  384. // 只保留增删改查基本操作
  385. if !isCURDOperation(query) {
  386. return
  387. }
  388. // 拼接 destination 的 IP 和端口为 ip:port 格式
  389. destAddr := fmt.Sprintf("%s:%d", destination.IP().String(), destination.Port())
  390. var attr []attribute.KeyValue
  391. attr = append(attr,
  392. semconvVal,
  393. semconv.DBStatement(query),
  394. semconv.NetPeerName(destination.IP().String()),
  395. semconv.NetPeerPort(int(destination.Port())),
  396. attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  397. attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  398. attribute.String("sql.dbn", destAddr),
  399. )
  400. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  401. attr = append(attr,
  402. attribute.Bool("sql.exception", r.Status.Error()),
  403. )
  404. if r.Status.Error() {
  405. attr = append(attr,
  406. attribute.String("sql.exception_msg", r.ErrorMsg),
  407. )
  408. }
  409. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  410. }
  411. //func (t *Trace) MysqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  412. // fmt.Println("query", query)
  413. // if t == nil {
  414. // return
  415. // }
  416. //
  417. // var attr []attribute.KeyValue
  418. // attr = append(attr,
  419. // semconv.DBSystemMySQL,
  420. // semconv.DBStatement(query),
  421. // semconv.NetPeerName(destination.IP().String()),
  422. // semconv.NetPeerPort(int(destination.Port())),
  423. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  424. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  425. // )
  426. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  427. //
  428. // attr = append(attr,
  429. // attribute.Bool("sql.exception", r.Status.Error()),
  430. // )
  431. // t.createTraceEvent(l7.ProtocolMysql.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolMysql), attr...)
  432. //}
  433. //func (t *Trace) PostGreSqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  434. // if t == nil || query == "" {
  435. // return
  436. // }
  437. // t.createTraceSpan(l7.ProtocolPostgres.String(), duration, error,
  438. // semconv.DBSystemPostgreSQL,
  439. // semconv.DBStatement(query),
  440. // semconv.NetPeerName(destination.IP().String()),
  441. // semconv.NetPeerPort(int(destination.Port())),
  442. // )
  443. //}
  444. //func (t *Trace) PostGreSqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  445. // fmt.Println("query", query)
  446. // if t == nil {
  447. // return
  448. // }
  449. //
  450. // var attr []attribute.KeyValue
  451. // attr = append(attr,
  452. // semconv.DBSystemPostgreSQL,
  453. // semconv.DBStatement(query),
  454. // semconv.NetPeerName(destination.IP().String()),
  455. // semconv.NetPeerPort(int(destination.Port())),
  456. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  457. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  458. // )
  459. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  460. // attr = append(attr,
  461. // attribute.Bool("sql.exception", r.Status.Error()),
  462. // )
  463. // t.createTraceEvent(l7.ProtocolPostgres.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolPostgres), attr...)
  464. //}
  465. //func (t *Trace) DmTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  466. // if t == nil || query == "" {
  467. // return
  468. // }
  469. // l7Type := int(l7.ProtocolDM)
  470. // var attr []attribute.KeyValue
  471. // attr = append(attr,
  472. // semconv.DBSystemDaMengDB,
  473. // semconv.DBStatement(query),
  474. // semconv.NetPeerName(destination.IP().String()),
  475. // semconv.NetPeerPort(int(destination.Port())),
  476. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  477. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  478. // )
  479. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  480. //
  481. // attr = append(attr,
  482. // attribute.Bool("sql.exception", r.Status.Error()),
  483. // )
  484. // t.createTraceEvent(l7.ProtocolHTTP.String(), int(ebpftracer.EventTypeL7Request), l7Type, attr...)
  485. //}
  486. //func (t *Trace) RedisTraceQuery(cmd, args string, error bool, duration time.Duration) {
  487. // if t == nil || cmd == "" {
  488. // return
  489. // }
  490. // statement := cmd
  491. // if args != "" {
  492. // statement += " " + args
  493. // }
  494. // t.createTraceSpan(l7.ProtocolRedis.String(), duration, error,
  495. // semconv.DBSystemRedis,
  496. // semconv.DBOperation(cmd),
  497. // semconv.DBStatement(statement),
  498. // )
  499. //}
  500. func (t *Trace) NoSQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, operation, statement string, r *l7.RequestData, connSrc, connDest netaddr.IPPort) {
  501. if t == nil {
  502. return
  503. }
  504. t.addEvent()
  505. if statement == "" {
  506. return
  507. }
  508. // 拼接 destination 的 IP 和端口为 ip:port 格式
  509. srcAddr := r.ComponentSAddr.String()
  510. if r.ComponentSAddr.Port() == 0 {
  511. srcAddr = connSrc.String()
  512. }
  513. destinationAddr := r.ComponentDAddr.String()
  514. if r.ComponentDAddr.Port() == 0 {
  515. destinationAddr = connDest.String()
  516. }
  517. var attr []attribute.KeyValue
  518. attr = append(attr,
  519. semconvVal,
  520. semconv.DBStatement(statement),
  521. semconv.NetPeerName(connDest.IP().String()),
  522. semconv.NetPeerPort(int(connDest.Port())),
  523. attribute.String("nosql.src_addr", srcAddr),
  524. attribute.String("nosql.destination_addr", destinationAddr),
  525. attribute.String("nosql.dbn", connDest.String()),
  526. )
  527. // 如果 operation 不为空,添加 DBOperation 属性
  528. if operation != "" {
  529. attr = append(attr, semconv.DBOperation(operation))
  530. }
  531. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  532. attr = append(attr,
  533. attribute.Bool("nosql.exception", r.Status.Error()),
  534. )
  535. if r.Status.Error() {
  536. attr = append(attr,
  537. attribute.String("nosql.exception_msg", r.ErrorMsg),
  538. )
  539. }
  540. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  541. }
  542. func (t *Trace) GrpcServerTraceQueryEvent(r *l7.RequestData, appInfo AppInfo) {
  543. if t == nil {
  544. return
  545. }
  546. t.addEvent()
  547. // if method == "" {
  548. // return
  549. // }
  550. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  551. if err != nil {
  552. assumedAppID = 0
  553. }
  554. var attr []attribute.KeyValue
  555. attr = append(attr,
  556. semconv.RPCSystemGRPC,
  557. attribute.String("rpc.uri", string(r.Payload)),
  558. attribute.String("rpc.schema", "grpc"),
  559. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  560. attribute.String("rpc.span_id", r.SpanId),
  561. attribute.Int("rpc.port", appInfo.Sport),
  562. attribute.String("rpc.ip", appInfo.Sn),
  563. attribute.String("rpc.oper_type", "PROVIDER"),
  564. attribute.String("rpc.method", "test_method"),
  565. )
  566. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  567. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  568. }
  569. func (t *Trace) GrpcClientTraceQueryEvent(r *l7.RequestData) {
  570. if t == nil {
  571. return
  572. }
  573. t.addEvent()
  574. // if method == "" {
  575. // return
  576. // }
  577. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  578. if err != nil {
  579. assumedAppID = 0
  580. }
  581. // 解析 RPCTarget 字符串,格式为 "ip:port"
  582. klog.Infof("r.RPCTarget is %s", r.RPCTarget)
  583. var rpcIP, rpcPort string
  584. if r.RPCTarget != "" {
  585. // 清理空字节,只保留有效字符
  586. cleanTarget := strings.TrimRight(r.RPCTarget, "\x00")
  587. cleanTarget = strings.TrimSpace(cleanTarget)
  588. klog.Infof("cleanTarget is %s", cleanTarget)
  589. if colonIndex := strings.LastIndex(cleanTarget, ":"); colonIndex != -1 {
  590. rpcIP = cleanTarget[:colonIndex]
  591. rpcPort = cleanTarget[colonIndex+1:]
  592. klog.Infof("rpcIP is %s, rpcPort is %s", rpcIP, rpcPort)
  593. } else {
  594. // 如果没有找到冒号,可能是只有IP或只有端口
  595. rpcIP = cleanTarget
  596. }
  597. }
  598. // 设置默认值
  599. if rpcIP == "" {
  600. rpcIP = "127.0.0.1"
  601. }
  602. if rpcPort == "" {
  603. rpcPort = "80"
  604. }
  605. // 转换端口为整数
  606. portInt, err := strconv.Atoi(rpcPort)
  607. if err != nil {
  608. klog.Infof("r.RPCTarget convert port err is %s", err)
  609. portInt = 80 // 默认端口
  610. } else {
  611. klog.Infof("r.RPCTarget convert port success, portInt is %d", portInt)
  612. }
  613. var attr []attribute.KeyValue
  614. attr = append(attr,
  615. semconv.RPCSystemGRPC,
  616. attribute.String("rpc.uri", string(r.Payload)),
  617. attribute.String("rpc.schema", "grpc"),
  618. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  619. attribute.String("rpc.span_id", r.SpanId),
  620. attribute.Int("rpc.port", portInt),
  621. attribute.String("rpc.ip", rpcIP),
  622. attribute.String("rpc.oper_type", "CONSUMER"),
  623. attribute.String("rpc.method", string(r.Payload)),
  624. )
  625. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  626. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  627. }
  628. func (t *Trace) DNSTraceQueryEvent(r *l7.RequestData, _type string, fqdn string, ttl uint32, ips []netaddr.IP) {
  629. if t == nil {
  630. return
  631. }
  632. t.addEvent()
  633. var attr []attribute.KeyValue
  634. var ipList []string
  635. for _, ip := range ips {
  636. ipList = append(ipList, ip.String())
  637. }
  638. ip := strings.Join(ipList, ",")
  639. attr = append(attr,
  640. attribute.String("dns.type", _type),
  641. attribute.String("dns.fqdn", fqdn),
  642. attribute.Int("dns.ttl", int(ttl)),
  643. attribute.String("dns.ips", ip),
  644. )
  645. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  646. t.createTraceEvent(l7.ProtocolDNS, ebpftracer.EventTypeL7Request.Int(), attr...)
  647. }
  648. func (t *Trace) HttpTraceRequest(method, path, ip string, port uint16, r *l7.RequestData) {
  649. if t == nil || method == "" {
  650. return
  651. }
  652. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  653. if err != nil {
  654. assumedAppID = 0
  655. }
  656. status := r.Status
  657. duration := r.Duration
  658. t.createTraceSpan(l7.ProtocolHTTP.String(), duration, status >= 400,
  659. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  660. semconv.HTTPMethod(method),
  661. semconv.HTTPStatusCode(int(status)),
  662. attribute.String("http.uri", path),
  663. attribute.String("http.ip", ip),
  664. attribute.Int64("http.assumed_app_id", assumedAppID),
  665. attribute.String("http.span_id", r.SpanId),
  666. attribute.Int("http.port", int(port)),
  667. )
  668. }
  669. // 新增事件处理
  670. func (t *Trace) HttpTraceRequestEvent(method, path, ip string, port uint16, r *l7.RequestData) {
  671. if t == nil {
  672. return
  673. }
  674. t.addEvent()
  675. if method == "" {
  676. return
  677. }
  678. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  679. if err != nil {
  680. assumedAppID = 0
  681. }
  682. status := r.Status
  683. var attr []attribute.KeyValue
  684. attr = append(attr,
  685. semconv.HTTPURL(fmt.Sprintf("http://%s:%d%s", ip, port, path)),
  686. semconv.HTTPMethod(method),
  687. semconv.HTTPStatusCode(int(status)),
  688. attribute.Bool("http.status_error", status > 399),
  689. attribute.String("http.uri", path),
  690. attribute.String("http.ip", ip),
  691. attribute.Int64("http.assumed_app_id", assumedAppID),
  692. attribute.String("http.span_id", r.SpanId),
  693. attribute.Int("http.port", int(port)),
  694. attribute.String("http.src_addr", r.ComponentSAddr.String()),
  695. attribute.String("http.destination_addr", r.ComponentDAddr.String()),
  696. attribute.Bool("http.is_tls", r.IsTls),
  697. )
  698. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  699. t.createTraceEvent(l7.ProtocolHTTP, ebpftracer.EventTypeL7Request.Int(), attr...)
  700. }
  701. func (t *Trace) FuncTraceQuery(funcname string, duration time.Duration, start uint64, end uint64) {
  702. if t == nil {
  703. return
  704. }
  705. t.addEvent()
  706. if funcname == "" {
  707. return
  708. }
  709. t.createTraceSpanNoTime2(funcname, duration, false, start, end)
  710. //t.createTraceSpanNoTime2(funcname, duration, false, start, end, attribute.Int("num", num))
  711. }
  712. func (t *Trace) createTraceSpanNoTime(name string, duration time.Duration, error bool, start uint64, end uint64, attrs ...attribute.KeyValue) {
  713. // end := time.Now()
  714. // start := end.Add(-duration)
  715. startTime := time.Unix(0, int64(start))
  716. endTime := time.Unix(0, int64(end))
  717. //fmt.Println("createTraceSpan:", t.ctx)
  718. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  719. span.SetAttributes(t.commonAttrs...)
  720. span.SetAttributes(attrs...)
  721. if error {
  722. span.SetStatus(codes.Error, "")
  723. }
  724. span.End(trace.WithTimestamp(endTime))
  725. }
  726. func (t *Trace) createTraceSpanNoTime2(name string, duration time.Duration, error bool, start uint64, end uint64) {
  727. // end := time.Now()
  728. // start := end.Add(-duration)
  729. //startTime := time.Unix(0, int64(start))
  730. //endTime := time.Unix(0, int64(end))
  731. ////fmt.Println("createTraceSpan:", t.ctx)
  732. //_, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  733. //span.SetAttributes(t.commonAttrs...)
  734. //span.SetAttributes(attrs...)
  735. //if error {
  736. // span.SetStatus(codes.Error, "")
  737. //}
  738. //span.End(trace.WithTimestamp(endTime))
  739. //attrs = append([]attribute.KeyValue{
  740. // attribute.Int64("startAt", int64(start)),
  741. // attribute.Int64("endAt", int64(end)),
  742. //})
  743. var attr []attribute.KeyValue
  744. t.appendTimestamp(&attr, start, end, int64(end-start))
  745. t.createFuncTraceEvent(name, ebpftracer.EventTypeFunEnt.Int(), attr...)
  746. }