apm_tracing.go 26 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817
  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 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. }
  162. t.span.SetAttributes(t.commonAttrs...)
  163. t.startReady()
  164. }
  165. // set context span
  166. func (t *Trace) CreateRootSpan(traceId uint64) {
  167. traceIdStr := strconv.Itoa(int(traceId))
  168. ctx, span := tracer(t.containerId).Start(context.Background(), traceIdStr, trace.WithSpanKind(trace.SpanKindClient))
  169. t.setContext(ctx)
  170. t.setSpan(span)
  171. }
  172. // Deprecated: use TraceStartEvent instead.
  173. func (t *Trace) TraceStart(method, path string, status l7.Status, duration time.Duration) {
  174. if t == nil || method == "" {
  175. return
  176. }
  177. t.createParentSpan("APPLICATION", duration, status >= 400,
  178. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  179. semconv.HTTPMethod(method),
  180. //semconv.HTTPStatusCode(int(status)),
  181. attribute.String("http.uri", path),
  182. )
  183. }
  184. // Deprecated: use TraceEndEvent instead.
  185. func (t *Trace) TraceEnd(r *l7.RequestData) {
  186. if t == nil {
  187. return
  188. }
  189. t.span.SetAttributes(
  190. semconv.HTTPStatusCode(int(r.Status)),
  191. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  192. )
  193. CalledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  194. if err == nil && CalledId != 0 {
  195. t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  196. }
  197. InstanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  198. if err == nil && InstanceIdFrom != 0 {
  199. t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  200. }
  201. AppIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  202. if err == nil && AppIdFrom != 0 {
  203. t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  204. }
  205. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  206. t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  207. }
  208. // klog.Debugf("r.ParentSpanContext.TypeFrom tpp %s", r.ParentSpanContext.TypeFrom)
  209. if r.ParentSpanContext.TypeFrom == "10" {
  210. t.span.SetAttributes(attribute.String("server.type_from", "SDK"))
  211. t.span.SetAttributes(attribute.String("server.app_id_from", "-1"))
  212. t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  213. t.span.SetAttributes(attribute.String("server.instance_id_from", "-1"))
  214. t.span.SetAttributes(attribute.String("server.call_id", "-1"))
  215. }
  216. // for _, v := range t.stack {
  217. // fmt.Printf("TraceEndTraceEndTraceEnd%s\n", v)
  218. // }
  219. //t.buildFun()
  220. t.span.End(trace.WithTimestamp(time.Now()))
  221. }
  222. // 新增结束事件
  223. func (t *Trace) TraceEndEvent(r *l7.RequestData) {
  224. if t == nil {
  225. return
  226. }
  227. var attr []attribute.KeyValue
  228. attr = append(attr,
  229. semconv.HTTPStatusCode(int(r.Status)),
  230. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  231. )
  232. //t.span.SetAttributes(
  233. // semconv.HTTPStatusCode(int(r.Status)),
  234. // attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  235. //)
  236. calledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  237. if err == nil && calledId != 0 {
  238. attr = append(attr, attribute.Int64("server.called_id", calledId))
  239. //t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  240. }
  241. instanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  242. if err == nil && instanceIdFrom != 0 {
  243. attr = append(attr, attribute.Int64("server.instance_id_from", instanceIdFrom))
  244. //t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  245. }
  246. appIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  247. if err == nil && appIdFrom != 0 {
  248. attr = append(attr, attribute.Int64("server.app_id_from", appIdFrom))
  249. //t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  250. }
  251. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  252. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  253. //t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  254. }
  255. // klog.Debugf("r.ParentSpanContext.TypeFrom event tpp %s", r.ParentSpanContext.TypeFrom)
  256. if r.ParentSpanContext.TypeFrom == "10" {
  257. attr = append(attr, attribute.String("server.type_from", "SDK"))
  258. attr = append(attr, attribute.String("server.app_id_from", "-1"))
  259. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  260. attr = append(attr, attribute.String("server.instance_id_from", "-1"))
  261. attr = append(attr, attribute.String("server.call_id", "-1"))
  262. } else if r.ParentSpanContext.TypeFrom == "02" {
  263. attr = append(attr, attribute.String("server.type_from", "Nodejs"))
  264. }
  265. attr = append(attr, attribute.String("server.src_addr", r.DAddr.String()))
  266. attr = append(attr, attribute.String("server.dst_addr", r.SAddr.String()))
  267. klog.Debugf("TraceEndEvent SAddr.String %s", r.SAddr.String())
  268. klog.Debugf("TraceEndEvent DAddr.String %s", r.DAddr.String())
  269. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  270. t.span.SetAttributes(attr...)
  271. t.endReadyEvent(r.EventCount)
  272. }
  273. func (t *Trace) appendTimestamp(attr *[]attribute.KeyValue, s, e uint64, d int64) {
  274. *attr = append(*attr,
  275. attribute.Int64("time.start_at", utils.KtimeToTimestamp(s)),
  276. attribute.Int64("time.end_at", utils.KtimeToTimestamp(e)),
  277. attribute.Int64("time.duration", d),
  278. )
  279. }
  280. func (t *Trace) createParentSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  281. end := time.Now()
  282. start := end.Add(-duration)
  283. ctx, span := tracer(t.containerId).Start(context.Background(), name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  284. span.SetAttributes(attrs...)
  285. span.SetAttributes(t.commonAttrs...)
  286. if error {
  287. span.SetStatus(codes.Error, "")
  288. }
  289. t.setContext(ctx)
  290. t.setSpan(span)
  291. }
  292. func (t *Trace) SendEvent() {
  293. t.span.End()
  294. }
  295. func (t *Trace) GetSpan() trace.Span {
  296. return t.span
  297. }
  298. func (t *Trace) createTraceEvent(protocol l7.Protocol, eventType int, attrs ...attribute.KeyValue) {
  299. t.span.AddEventApm(protocol.String(), eventType, protocol.Int(), trace.WithAttributes(attrs...))
  300. //atomic.AddUint32(t.currenEventCount, 1)
  301. }
  302. func (t *Trace) createFuncTraceEvent(name string, eventType int, attrs ...attribute.KeyValue) {
  303. t.span.AddEventApm(name, eventType, 0, trace.WithAttributes(attrs...))
  304. }
  305. func (t *Trace) addEvent() {
  306. atomic.AddUint32(t.currenEventCount, 1)
  307. }
  308. func (t *Trace) createTraceSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  309. end := time.Now()
  310. start := end.Add(-duration)
  311. //fmt.Println("createTraceSpan:", t.ctx)
  312. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  313. span.SetAttributes(t.commonAttrs...)
  314. span.SetAttributes(attrs...)
  315. if error {
  316. span.SetStatus(codes.Error, "")
  317. }
  318. span.End(trace.WithTimestamp(end))
  319. }
  320. func (t *Trace) MysqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  321. if t == nil || query == "" {
  322. return
  323. }
  324. t.createTraceSpan(l7.ProtocolMysql.String(), duration, error,
  325. semconv.DBSystemMySQL,
  326. semconv.DBStatement(query),
  327. semconv.NetPeerName(destination.IP().String()),
  328. semconv.NetPeerPort(int(destination.Port())),
  329. )
  330. }
  331. // 内置允许操作(始终放行)
  332. var builtinOps = map[string]struct{}{
  333. "SELECT": {},
  334. "INSERT": {},
  335. "UPDATE": {},
  336. "DELETE": {},
  337. "WITH": {},
  338. }
  339. var (
  340. allowedOps map[string]struct{}
  341. initOnce sync.Once
  342. )
  343. // 解析并缓存环境变量中的前缀
  344. func initAllowedOps() {
  345. allowedOps = make(map[string]struct{})
  346. env := os.Getenv("ALLOWED_SQL_PREFIXES")
  347. for _, s := range strings.Split(env, ",") {
  348. if trimmed := strings.ToUpper(strings.TrimSpace(s)); trimmed != "" {
  349. allowedOps[trimmed] = struct{}{}
  350. }
  351. }
  352. }
  353. func isCURDOperation(q string) bool {
  354. if len(q) < 6 {
  355. return false
  356. }
  357. // 提取前缀(考虑最长10个字符)
  358. prefix := strings.ToUpper(strings.TrimSpace(q))
  359. if len(prefix) > 10 {
  360. prefix = prefix[:10]
  361. }
  362. // 提取第一个单词作为操作关键字
  363. spaceIdx := strings.IndexAny(prefix, " \t\n")
  364. if spaceIdx > 0 {
  365. prefix = prefix[:spaceIdx]
  366. }
  367. // 先查固定允许的操作
  368. if _, ok := builtinOps[prefix]; ok {
  369. return true
  370. }
  371. // 加载环境变量控制的操作(只执行一次)
  372. initOnce.Do(initAllowedOps)
  373. // 查额外允许操作
  374. _, ok := allowedOps[prefix]
  375. return ok
  376. }
  377. func (t *Trace) SQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, query string, r *l7.RequestData, destination netaddr.IPPort) {
  378. //fmt.Println("query", query)
  379. if t == nil {
  380. return
  381. }
  382. t.addEvent()
  383. // 只保留增删改查基本操作
  384. if !isCURDOperation(query) {
  385. return
  386. }
  387. // 拼接 destination 的 IP 和端口为 ip:port 格式
  388. destAddr := fmt.Sprintf("%s:%d", destination.IP().String(), destination.Port())
  389. var attr []attribute.KeyValue
  390. attr = append(attr,
  391. semconvVal,
  392. semconv.DBStatement(query),
  393. semconv.NetPeerName(destination.IP().String()),
  394. semconv.NetPeerPort(int(destination.Port())),
  395. attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  396. attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  397. attribute.String("sql.dbn", destAddr),
  398. )
  399. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  400. attr = append(attr,
  401. attribute.Bool("sql.exception", r.Status.Error()),
  402. )
  403. if r.Status.Error() {
  404. attr = append(attr,
  405. attribute.String("sql.exception_msg", r.ErrorMsg),
  406. )
  407. }
  408. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  409. }
  410. //func (t *Trace) MysqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  411. // fmt.Println("query", query)
  412. // if t == nil {
  413. // return
  414. // }
  415. //
  416. // var attr []attribute.KeyValue
  417. // attr = append(attr,
  418. // semconv.DBSystemMySQL,
  419. // semconv.DBStatement(query),
  420. // semconv.NetPeerName(destination.IP().String()),
  421. // semconv.NetPeerPort(int(destination.Port())),
  422. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  423. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  424. // )
  425. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  426. //
  427. // attr = append(attr,
  428. // attribute.Bool("sql.exception", r.Status.Error()),
  429. // )
  430. // t.createTraceEvent(l7.ProtocolMysql.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolMysql), attr...)
  431. //}
  432. //func (t *Trace) PostGreSqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  433. // if t == nil || query == "" {
  434. // return
  435. // }
  436. // t.createTraceSpan(l7.ProtocolPostgres.String(), duration, error,
  437. // semconv.DBSystemPostgreSQL,
  438. // semconv.DBStatement(query),
  439. // semconv.NetPeerName(destination.IP().String()),
  440. // semconv.NetPeerPort(int(destination.Port())),
  441. // )
  442. //}
  443. //func (t *Trace) PostGreSqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  444. // fmt.Println("query", query)
  445. // if t == nil {
  446. // return
  447. // }
  448. //
  449. // var attr []attribute.KeyValue
  450. // attr = append(attr,
  451. // semconv.DBSystemPostgreSQL,
  452. // semconv.DBStatement(query),
  453. // semconv.NetPeerName(destination.IP().String()),
  454. // semconv.NetPeerPort(int(destination.Port())),
  455. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  456. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  457. // )
  458. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  459. // attr = append(attr,
  460. // attribute.Bool("sql.exception", r.Status.Error()),
  461. // )
  462. // t.createTraceEvent(l7.ProtocolPostgres.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolPostgres), attr...)
  463. //}
  464. //func (t *Trace) DmTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  465. // if t == nil || query == "" {
  466. // return
  467. // }
  468. // l7Type := int(l7.ProtocolDM)
  469. // var attr []attribute.KeyValue
  470. // attr = append(attr,
  471. // semconv.DBSystemDaMengDB,
  472. // semconv.DBStatement(query),
  473. // semconv.NetPeerName(destination.IP().String()),
  474. // semconv.NetPeerPort(int(destination.Port())),
  475. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  476. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  477. // )
  478. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  479. //
  480. // attr = append(attr,
  481. // attribute.Bool("sql.exception", r.Status.Error()),
  482. // )
  483. // t.createTraceEvent(l7.ProtocolHTTP.String(), int(ebpftracer.EventTypeL7Request), l7Type, attr...)
  484. //}
  485. //func (t *Trace) RedisTraceQuery(cmd, args string, error bool, duration time.Duration) {
  486. // if t == nil || cmd == "" {
  487. // return
  488. // }
  489. // statement := cmd
  490. // if args != "" {
  491. // statement += " " + args
  492. // }
  493. // t.createTraceSpan(l7.ProtocolRedis.String(), duration, error,
  494. // semconv.DBSystemRedis,
  495. // semconv.DBOperation(cmd),
  496. // semconv.DBStatement(statement),
  497. // )
  498. //}
  499. func (t *Trace) NoSQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, operation, statement string, r *l7.RequestData, destination netaddr.IPPort) {
  500. if t == nil {
  501. return
  502. }
  503. t.addEvent()
  504. if statement == "" {
  505. return
  506. }
  507. // 拼接 destination 的 IP 和端口为 ip:port 格式
  508. destAddr := fmt.Sprintf("%s:%d", destination.IP().String(), destination.Port())
  509. var attr []attribute.KeyValue
  510. attr = append(attr,
  511. semconvVal,
  512. semconv.DBStatement(statement),
  513. semconv.NetPeerName(destination.IP().String()),
  514. semconv.NetPeerPort(int(destination.Port())),
  515. attribute.String("nosql.src_addr", r.ComponentSAddr.String()),
  516. attribute.String("nosql.destination_addr", r.ComponentDAddr.String()),
  517. attribute.String("nosql.dbn", destAddr),
  518. )
  519. // 如果 operation 不为空,添加 DBOperation 属性
  520. if operation != "" {
  521. attr = append(attr, semconv.DBOperation(operation))
  522. }
  523. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  524. attr = append(attr,
  525. attribute.Bool("nosql.exception", r.Status.Error()),
  526. )
  527. if r.Status.Error() {
  528. attr = append(attr,
  529. attribute.String("nosql.exception_msg", r.ErrorMsg),
  530. )
  531. }
  532. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  533. }
  534. func (t *Trace) GrpcServerTraceQueryEvent(r *l7.RequestData, appInfo AppInfo) {
  535. if t == nil {
  536. return
  537. }
  538. t.addEvent()
  539. // if method == "" {
  540. // return
  541. // }
  542. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  543. if err != nil {
  544. assumedAppID = 0
  545. }
  546. var attr []attribute.KeyValue
  547. attr = append(attr,
  548. semconv.RPCSystemGRPC,
  549. attribute.String("rpc.uri", string(r.Payload)),
  550. attribute.String("rpc.schema", "grpc"),
  551. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  552. attribute.String("rpc.span_id", r.SpanId),
  553. attribute.Int("rpc.port", appInfo.Sport),
  554. attribute.String("rpc.ip", appInfo.Sn),
  555. attribute.String("rpc.oper_type", "PROVIDER"),
  556. attribute.String("rpc.method", "test_method"),
  557. )
  558. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  559. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  560. }
  561. func (t *Trace) GrpcClientTraceQueryEvent(r *l7.RequestData) {
  562. if t == nil {
  563. return
  564. }
  565. t.addEvent()
  566. // if method == "" {
  567. // return
  568. // }
  569. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  570. if err != nil {
  571. assumedAppID = 0
  572. }
  573. // 解析 RPCTarget 字符串,格式为 "ip:port"
  574. klog.Infof("r.RPCTarget is %s", r.RPCTarget)
  575. var rpcIP, rpcPort string
  576. if r.RPCTarget != "" {
  577. // 清理空字节,只保留有效字符
  578. cleanTarget := strings.TrimRight(r.RPCTarget, "\x00")
  579. cleanTarget = strings.TrimSpace(cleanTarget)
  580. klog.Infof("cleanTarget is %s", cleanTarget)
  581. if colonIndex := strings.LastIndex(cleanTarget, ":"); colonIndex != -1 {
  582. rpcIP = cleanTarget[:colonIndex]
  583. rpcPort = cleanTarget[colonIndex+1:]
  584. klog.Infof("rpcIP is %s, rpcPort is %s", rpcIP, rpcPort)
  585. } else {
  586. // 如果没有找到冒号,可能是只有IP或只有端口
  587. rpcIP = cleanTarget
  588. }
  589. }
  590. // 设置默认值
  591. if rpcIP == "" {
  592. rpcIP = "127.0.0.1"
  593. }
  594. if rpcPort == "" {
  595. rpcPort = "80"
  596. }
  597. // 转换端口为整数
  598. portInt, err := strconv.Atoi(rpcPort)
  599. if err != nil {
  600. klog.Infof("r.RPCTarget convert port err is %s", err)
  601. portInt = 80 // 默认端口
  602. } else {
  603. klog.Infof("r.RPCTarget convert port success, portInt is %d", portInt)
  604. }
  605. var attr []attribute.KeyValue
  606. attr = append(attr,
  607. semconv.RPCSystemGRPC,
  608. attribute.String("rpc.uri", string(r.Payload)),
  609. attribute.String("rpc.schema", "grpc"),
  610. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  611. attribute.String("rpc.span_id", r.SpanId),
  612. attribute.Int("rpc.port", portInt),
  613. attribute.String("rpc.ip", rpcIP),
  614. attribute.String("rpc.oper_type", "CONSUMER"),
  615. attribute.String("rpc.method", string(r.Payload)),
  616. )
  617. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  618. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  619. }
  620. func (t *Trace) DNSTraceQueryEvent(r *l7.RequestData, _type string, fqdn string, ttl uint32, ips []netaddr.IP) {
  621. if t == nil {
  622. return
  623. }
  624. t.addEvent()
  625. var attr []attribute.KeyValue
  626. var ipList []string
  627. for _, ip := range ips {
  628. ipList = append(ipList, ip.String())
  629. }
  630. ip := strings.Join(ipList, ",")
  631. attr = append(attr,
  632. attribute.String("dns.type", _type),
  633. attribute.String("dns.fqdn", fqdn),
  634. attribute.Int("dns.ttl", int(ttl)),
  635. attribute.String("dns.ips", ip),
  636. )
  637. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  638. t.createTraceEvent(l7.ProtocolDNS, ebpftracer.EventTypeL7Request.Int(), attr...)
  639. }
  640. func (t *Trace) HttpTraceRequest(method, path, ip string, port uint16, r *l7.RequestData) {
  641. if t == nil || method == "" {
  642. return
  643. }
  644. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  645. if err != nil {
  646. assumedAppID = 0
  647. }
  648. status := r.Status
  649. duration := r.Duration
  650. t.createTraceSpan(l7.ProtocolHTTP.String(), duration, status >= 400,
  651. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  652. semconv.HTTPMethod(method),
  653. semconv.HTTPStatusCode(int(status)),
  654. attribute.String("http.uri", path),
  655. attribute.String("http.ip", ip),
  656. attribute.Int64("http.assumed_app_id", assumedAppID),
  657. attribute.String("http.span_id", r.SpanId),
  658. attribute.Int("http.port", int(port)),
  659. )
  660. }
  661. // 新增事件处理
  662. func (t *Trace) HttpTraceRequestEvent(method, path, ip string, port uint16, r *l7.RequestData) {
  663. if t == nil {
  664. return
  665. }
  666. t.addEvent()
  667. if method == "" {
  668. return
  669. }
  670. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  671. if err != nil {
  672. assumedAppID = 0
  673. }
  674. status := r.Status
  675. var attr []attribute.KeyValue
  676. attr = append(attr,
  677. semconv.HTTPURL(fmt.Sprintf("http://%s:%d%s", ip, port, path)),
  678. semconv.HTTPMethod(method),
  679. semconv.HTTPStatusCode(int(status)),
  680. attribute.Bool("http.status_error", status > 399),
  681. attribute.String("http.uri", path),
  682. attribute.String("http.ip", ip),
  683. attribute.Int64("http.assumed_app_id", assumedAppID),
  684. attribute.String("http.span_id", r.SpanId),
  685. attribute.Int("http.port", int(port)),
  686. attribute.String("http.src_addr", r.ComponentSAddr.String()),
  687. attribute.String("http.destination_addr", r.ComponentDAddr.String()),
  688. attribute.Bool("http.is_tls", r.IsTls),
  689. )
  690. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  691. t.createTraceEvent(l7.ProtocolHTTP, ebpftracer.EventTypeL7Request.Int(), attr...)
  692. }
  693. func (t *Trace) FuncTraceQuery(funcname string, duration time.Duration, start uint64, end uint64) {
  694. if t == nil {
  695. return
  696. }
  697. t.addEvent()
  698. if funcname == "" {
  699. return
  700. }
  701. t.createTraceSpanNoTime2(funcname, duration, false, start, end)
  702. //t.createTraceSpanNoTime2(funcname, duration, false, start, end, attribute.Int("num", num))
  703. }
  704. func (t *Trace) createTraceSpanNoTime(name string, duration time.Duration, error bool, start uint64, end uint64, attrs ...attribute.KeyValue) {
  705. // end := time.Now()
  706. // start := end.Add(-duration)
  707. startTime := time.Unix(0, int64(start))
  708. endTime := time.Unix(0, int64(end))
  709. //fmt.Println("createTraceSpan:", t.ctx)
  710. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  711. span.SetAttributes(t.commonAttrs...)
  712. span.SetAttributes(attrs...)
  713. if error {
  714. span.SetStatus(codes.Error, "")
  715. }
  716. span.End(trace.WithTimestamp(endTime))
  717. }
  718. func (t *Trace) createTraceSpanNoTime2(name string, duration time.Duration, error bool, start uint64, end uint64) {
  719. // end := time.Now()
  720. // start := end.Add(-duration)
  721. //startTime := time.Unix(0, int64(start))
  722. //endTime := time.Unix(0, int64(end))
  723. ////fmt.Println("createTraceSpan:", t.ctx)
  724. //_, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  725. //span.SetAttributes(t.commonAttrs...)
  726. //span.SetAttributes(attrs...)
  727. //if error {
  728. // span.SetStatus(codes.Error, "")
  729. //}
  730. //span.End(trace.WithTimestamp(endTime))
  731. //attrs = append([]attribute.KeyValue{
  732. // attribute.Int64("startAt", int64(start)),
  733. // attribute.Int64("endAt", int64(end)),
  734. //})
  735. var attr []attribute.KeyValue
  736. t.appendTimestamp(&attr, start, end, int64(end-start))
  737. t.createFuncTraceEvent(name, ebpftracer.EventTypeFunEnt.Int(), attr...)
  738. }