apm_tracing.go 26 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819
  1. package tracing
  2. import (
  3. "context"
  4. "fmt"
  5. klog "github.com/sirupsen/logrus"
  6. "os"
  7. "strconv"
  8. "strings"
  9. "sync"
  10. "sync/atomic"
  11. "time"
  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(name string, eventType int, l7Type int, attrs ...attribute.KeyValue) {
  299. t.span.AddEventApm(name, eventType, l7Type, trace.WithAttributes(attrs...))
  300. //atomic.AddUint32(t.currenEventCount, 1)
  301. }
  302. func (t *Trace) addEvent() {
  303. atomic.AddUint32(t.currenEventCount, 1)
  304. }
  305. func (t *Trace) createTraceSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  306. end := time.Now()
  307. start := end.Add(-duration)
  308. //fmt.Println("createTraceSpan:", t.ctx)
  309. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  310. span.SetAttributes(t.commonAttrs...)
  311. span.SetAttributes(attrs...)
  312. if error {
  313. span.SetStatus(codes.Error, "")
  314. }
  315. span.End(trace.WithTimestamp(end))
  316. }
  317. func (t *Trace) MysqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  318. if t == nil || query == "" {
  319. return
  320. }
  321. t.createTraceSpan(l7.ProtocolMysql.String(), duration, error,
  322. semconv.DBSystemMySQL,
  323. semconv.DBStatement(query),
  324. semconv.NetPeerName(destination.IP().String()),
  325. semconv.NetPeerPort(int(destination.Port())),
  326. )
  327. }
  328. // 内置允许操作(始终放行)
  329. var builtinOps = map[string]struct{}{
  330. "SELECT": {},
  331. "INSERT": {},
  332. "UPDATE": {},
  333. "DELETE": {},
  334. "WITH": {},
  335. }
  336. var (
  337. allowedOps map[string]struct{}
  338. initOnce sync.Once
  339. )
  340. // 解析并缓存环境变量中的前缀
  341. func initAllowedOps() {
  342. allowedOps = make(map[string]struct{})
  343. env := os.Getenv("ALLOWED_SQL_PREFIXES")
  344. for _, s := range strings.Split(env, ",") {
  345. if trimmed := strings.ToUpper(strings.TrimSpace(s)); trimmed != "" {
  346. allowedOps[trimmed] = struct{}{}
  347. }
  348. }
  349. }
  350. func isCURDOperation(q string) bool {
  351. if len(q) < 6 {
  352. return false
  353. }
  354. // 提取前缀(考虑最长10个字符)
  355. prefix := strings.ToUpper(strings.TrimSpace(q))
  356. if len(prefix) > 10 {
  357. prefix = prefix[:10]
  358. }
  359. // 提取第一个单词作为操作关键字
  360. spaceIdx := strings.IndexAny(prefix, " \t\n")
  361. if spaceIdx > 0 {
  362. prefix = prefix[:spaceIdx]
  363. }
  364. // 先查固定允许的操作
  365. if _, ok := builtinOps[prefix]; ok {
  366. return true
  367. }
  368. // 加载环境变量控制的操作(只执行一次)
  369. initOnce.Do(initAllowedOps)
  370. // 查额外允许操作
  371. _, ok := allowedOps[prefix]
  372. return ok
  373. }
  374. func (t *Trace) SQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, query string, r *l7.RequestData, destination netaddr.IPPort) {
  375. //fmt.Println("query", query)
  376. if t == nil {
  377. return
  378. }
  379. t.addEvent()
  380. // 只保留增删改查基本操作
  381. if !isCURDOperation(query) {
  382. return
  383. }
  384. // 拼接 destination 的 IP 和端口为 ip:port 格式
  385. destAddr := fmt.Sprintf("%s:%d", destination.IP().String(), destination.Port())
  386. var attr []attribute.KeyValue
  387. attr = append(attr,
  388. semconvVal,
  389. semconv.DBStatement(query),
  390. semconv.NetPeerName(destination.IP().String()),
  391. semconv.NetPeerPort(int(destination.Port())),
  392. attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  393. attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  394. attribute.String("sql.dbn", destAddr),
  395. )
  396. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  397. attr = append(attr,
  398. attribute.Bool("sql.exception", r.Status.Error()),
  399. )
  400. if r.Status.Error() {
  401. attr = append(attr,
  402. attribute.String("sql.exception_msg", r.ErrorMsg),
  403. )
  404. }
  405. t.createTraceEvent(l7Type.String(), ebpftracer.EventTypeL7Request.Int(), l7Type.Int(), attr...)
  406. }
  407. //func (t *Trace) MysqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  408. // fmt.Println("query", query)
  409. // if t == nil {
  410. // return
  411. // }
  412. //
  413. // var attr []attribute.KeyValue
  414. // attr = append(attr,
  415. // semconv.DBSystemMySQL,
  416. // semconv.DBStatement(query),
  417. // semconv.NetPeerName(destination.IP().String()),
  418. // semconv.NetPeerPort(int(destination.Port())),
  419. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  420. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  421. // )
  422. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  423. //
  424. // attr = append(attr,
  425. // attribute.Bool("sql.exception", r.Status.Error()),
  426. // )
  427. // t.createTraceEvent(l7.ProtocolMysql.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolMysql), attr...)
  428. //}
  429. //func (t *Trace) PostGreSqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  430. // if t == nil || query == "" {
  431. // return
  432. // }
  433. // t.createTraceSpan(l7.ProtocolPostgres.String(), duration, error,
  434. // semconv.DBSystemPostgreSQL,
  435. // semconv.DBStatement(query),
  436. // semconv.NetPeerName(destination.IP().String()),
  437. // semconv.NetPeerPort(int(destination.Port())),
  438. // )
  439. //}
  440. //func (t *Trace) PostGreSqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  441. // fmt.Println("query", query)
  442. // if t == nil {
  443. // return
  444. // }
  445. //
  446. // var attr []attribute.KeyValue
  447. // attr = append(attr,
  448. // semconv.DBSystemPostgreSQL,
  449. // semconv.DBStatement(query),
  450. // semconv.NetPeerName(destination.IP().String()),
  451. // semconv.NetPeerPort(int(destination.Port())),
  452. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  453. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  454. // )
  455. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  456. // attr = append(attr,
  457. // attribute.Bool("sql.exception", r.Status.Error()),
  458. // )
  459. // t.createTraceEvent(l7.ProtocolPostgres.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolPostgres), attr...)
  460. //}
  461. //func (t *Trace) DmTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  462. // if t == nil || query == "" {
  463. // return
  464. // }
  465. // l7Type := int(l7.ProtocolDM)
  466. // var attr []attribute.KeyValue
  467. // attr = append(attr,
  468. // semconv.DBSystemDaMengDB,
  469. // semconv.DBStatement(query),
  470. // semconv.NetPeerName(destination.IP().String()),
  471. // semconv.NetPeerPort(int(destination.Port())),
  472. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  473. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  474. // )
  475. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  476. //
  477. // attr = append(attr,
  478. // attribute.Bool("sql.exception", r.Status.Error()),
  479. // )
  480. // t.createTraceEvent(l7.ProtocolHTTP.String(), int(ebpftracer.EventTypeL7Request), l7Type, attr...)
  481. //}
  482. //func (t *Trace) RedisTraceQuery(cmd, args string, error bool, duration time.Duration) {
  483. // if t == nil || cmd == "" {
  484. // return
  485. // }
  486. // statement := cmd
  487. // if args != "" {
  488. // statement += " " + args
  489. // }
  490. // t.createTraceSpan(l7.ProtocolRedis.String(), duration, error,
  491. // semconv.DBSystemRedis,
  492. // semconv.DBOperation(cmd),
  493. // semconv.DBStatement(statement),
  494. // )
  495. //}
  496. func (t *Trace) RedisTraceQueryEvent(cmd, args string, r *l7.RequestData, destination netaddr.IPPort) {
  497. if t == nil {
  498. return
  499. }
  500. t.addEvent()
  501. if cmd == "" {
  502. return
  503. }
  504. statement := cmd
  505. if args != "" {
  506. statement += " " + args
  507. }
  508. var attr []attribute.KeyValue
  509. attr = append(attr,
  510. semconv.DBSystemRedis,
  511. semconv.DBOperation(cmd),
  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. )
  518. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  519. t.createTraceEvent(l7.ProtocolRedis.String(), ebpftracer.EventTypeL7Request.Int(), l7.ProtocolRedis.Int(), attr...)
  520. }
  521. func (t *Trace) GrpcServerTraceQueryEvent(r *l7.RequestData, appInfo AppInfo) {
  522. if t == nil {
  523. return
  524. }
  525. t.addEvent()
  526. // if method == "" {
  527. // return
  528. // }
  529. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  530. if err != nil {
  531. assumedAppID = 0
  532. }
  533. var attr []attribute.KeyValue
  534. attr = append(attr,
  535. semconv.RPCSystemGRPC,
  536. attribute.String("rpc.uri", string(r.Payload)),
  537. attribute.String("rpc.schema", "grpc"),
  538. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  539. attribute.String("rpc.span_id", r.SpanId),
  540. attribute.Int("rpc.port", appInfo.Sport),
  541. attribute.String("rpc.ip", appInfo.Sn),
  542. attribute.String("rpc.oper_type", "PROVIDER"),
  543. attribute.String("rpc.method", "test_method"),
  544. )
  545. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  546. t.createTraceEvent(l7.ProtocolGrpc.String(), ebpftracer.EventTypeL7Request.Int(), l7.ProtocolGrpc.Int(), attr...)
  547. }
  548. func (t *Trace) GrpcClientTraceQueryEvent(r *l7.RequestData) {
  549. if t == nil {
  550. return
  551. }
  552. t.addEvent()
  553. // if method == "" {
  554. // return
  555. // }
  556. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  557. if err != nil {
  558. assumedAppID = 0
  559. }
  560. // 解析 RPCTarget 字符串,格式为 "ip:port"
  561. klog.Infof("r.RPCTarget is %s", r.RPCTarget)
  562. var rpcIP, rpcPort string
  563. if r.RPCTarget != "" {
  564. // 清理空字节,只保留有效字符
  565. cleanTarget := strings.TrimRight(r.RPCTarget, "\x00")
  566. cleanTarget = strings.TrimSpace(cleanTarget)
  567. klog.Infof("cleanTarget is %s", cleanTarget)
  568. if colonIndex := strings.LastIndex(cleanTarget, ":"); colonIndex != -1 {
  569. rpcIP = cleanTarget[:colonIndex]
  570. rpcPort = cleanTarget[colonIndex+1:]
  571. klog.Infof("rpcIP is %s, rpcPort is %s", rpcIP, rpcPort)
  572. } else {
  573. // 如果没有找到冒号,可能是只有IP或只有端口
  574. rpcIP = cleanTarget
  575. }
  576. }
  577. // 设置默认值
  578. if rpcIP == "" {
  579. rpcIP = "127.0.0.1"
  580. }
  581. if rpcPort == "" {
  582. rpcPort = "80"
  583. }
  584. // 转换端口为整数
  585. portInt, err := strconv.Atoi(rpcPort)
  586. if err != nil {
  587. klog.Infof("r.RPCTarget convert port err is %s", err)
  588. portInt = 80 // 默认端口
  589. } else {
  590. klog.Infof("r.RPCTarget convert port success, portInt is %d", portInt)
  591. }
  592. var attr []attribute.KeyValue
  593. attr = append(attr,
  594. semconv.RPCSystemGRPC,
  595. attribute.String("rpc.uri", string(r.Payload)),
  596. attribute.String("rpc.schema", "grpc"),
  597. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  598. attribute.String("rpc.span_id", r.SpanId),
  599. attribute.Int("rpc.port", portInt),
  600. attribute.String("rpc.ip", rpcIP),
  601. attribute.String("rpc.oper_type", "CONSUMER"),
  602. attribute.String("rpc.method", string(r.Payload)),
  603. )
  604. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  605. t.createTraceEvent(l7.ProtocolGrpc.String(), ebpftracer.EventTypeL7Request.Int(), l7.ProtocolGrpc.Int(), attr...)
  606. }
  607. func (t *Trace) MongoTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  608. if t == nil {
  609. return
  610. }
  611. t.addEvent()
  612. if query == "" {
  613. return
  614. }
  615. var attr []attribute.KeyValue
  616. attr = append(attr,
  617. semconv.DBSystemMongoDB,
  618. semconv.DBStatement(query),
  619. semconv.NetPeerName(destination.IP().String()),
  620. semconv.NetPeerPort(int(destination.Port())),
  621. attribute.String("nosql.src_addr", r.ComponentSAddr.String()),
  622. attribute.String("nosql.destination_addr", r.ComponentDAddr.String()),
  623. )
  624. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  625. t.createTraceEvent(l7.ProtocolMongo.String(), ebpftracer.EventTypeL7Request.Int(), l7.ProtocolMongo.Int(), attr...)
  626. }
  627. func (t *Trace) DNSTraceQueryEvent(r *l7.RequestData, _type string, fqdn string, ttl uint32, ips []netaddr.IP) {
  628. if t == nil {
  629. return
  630. }
  631. t.addEvent()
  632. var attr []attribute.KeyValue
  633. var ipList []string
  634. for _, ip := range ips {
  635. ipList = append(ipList, ip.String())
  636. }
  637. ip := strings.Join(ipList, ",")
  638. attr = append(attr,
  639. attribute.String("dns.type", _type),
  640. attribute.String("dns.fqdn", fqdn),
  641. attribute.Int("dns.ttl", int(ttl)),
  642. attribute.String("dns.ips", ip),
  643. )
  644. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  645. t.createTraceEvent(l7.ProtocolDNS.String(), ebpftracer.EventTypeL7Request.Int(), l7.ProtocolDNS.Int(), attr...)
  646. }
  647. func (t *Trace) HttpTraceRequest(method, path, ip string, port uint16, r *l7.RequestData) {
  648. if t == nil || method == "" {
  649. return
  650. }
  651. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  652. if err != nil {
  653. assumedAppID = 0
  654. }
  655. status := r.Status
  656. duration := r.Duration
  657. t.createTraceSpan(l7.ProtocolHTTP.String(), duration, status >= 400,
  658. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  659. semconv.HTTPMethod(method),
  660. semconv.HTTPStatusCode(int(status)),
  661. attribute.String("http.uri", path),
  662. attribute.String("http.ip", ip),
  663. attribute.Int64("http.assumed_app_id", assumedAppID),
  664. attribute.String("http.span_id", r.SpanId),
  665. attribute.Int("http.port", int(port)),
  666. )
  667. }
  668. // 新增事件处理
  669. func (t *Trace) HttpTraceRequestEvent(method, path, ip string, port uint16, r *l7.RequestData) {
  670. if t == nil {
  671. return
  672. }
  673. t.addEvent()
  674. if method == "" {
  675. return
  676. }
  677. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  678. if err != nil {
  679. assumedAppID = 0
  680. }
  681. status := r.Status
  682. var attr []attribute.KeyValue
  683. attr = append(attr,
  684. semconv.HTTPURL(fmt.Sprintf("http://%s:%d%s", ip, port, path)),
  685. semconv.HTTPMethod(method),
  686. semconv.HTTPStatusCode(int(status)),
  687. attribute.Bool("http.status_error", status > 399),
  688. attribute.String("http.uri", path),
  689. attribute.String("http.ip", ip),
  690. attribute.Int64("http.assumed_app_id", assumedAppID),
  691. attribute.String("http.span_id", r.SpanId),
  692. attribute.Int("http.port", int(port)),
  693. attribute.String("http.src_addr", r.ComponentSAddr.String()),
  694. attribute.String("http.destination_addr", r.ComponentDAddr.String()),
  695. attribute.Bool("http.is_tls", r.IsTls),
  696. )
  697. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  698. t.createTraceEvent(l7.ProtocolHTTP.String(), ebpftracer.EventTypeL7Request.Int(), l7.ProtocolHTTP.Int(), attr...)
  699. }
  700. func (t *Trace) FuncTraceQuery(funcname string, duration time.Duration, start uint64, end uint64) {
  701. if t == nil {
  702. return
  703. }
  704. t.addEvent()
  705. if funcname == "" {
  706. return
  707. }
  708. t.createTraceSpanNoTime2(funcname, duration, false, start, end)
  709. //t.createTraceSpanNoTime2(funcname, duration, false, start, end, attribute.Int("num", num))
  710. }
  711. func (t *Trace) createTraceSpanNoTime(name string, duration time.Duration, error bool, start uint64, end uint64, attrs ...attribute.KeyValue) {
  712. // end := time.Now()
  713. // start := end.Add(-duration)
  714. startTime := time.Unix(0, int64(start))
  715. endTime := time.Unix(0, int64(end))
  716. //fmt.Println("createTraceSpan:", t.ctx)
  717. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  718. span.SetAttributes(t.commonAttrs...)
  719. span.SetAttributes(attrs...)
  720. if error {
  721. span.SetStatus(codes.Error, "")
  722. }
  723. span.End(trace.WithTimestamp(endTime))
  724. }
  725. func (t *Trace) createTraceSpanNoTime2(name string, duration time.Duration, error bool, start uint64, end uint64) {
  726. // end := time.Now()
  727. // start := end.Add(-duration)
  728. //startTime := time.Unix(0, int64(start))
  729. //endTime := time.Unix(0, int64(end))
  730. ////fmt.Println("createTraceSpan:", t.ctx)
  731. //_, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  732. //span.SetAttributes(t.commonAttrs...)
  733. //span.SetAttributes(attrs...)
  734. //if error {
  735. // span.SetStatus(codes.Error, "")
  736. //}
  737. //span.End(trace.WithTimestamp(endTime))
  738. //attrs = append([]attribute.KeyValue{
  739. // attribute.Int64("startAt", int64(start)),
  740. // attribute.Int64("endAt", int64(end)),
  741. //})
  742. var attr []attribute.KeyValue
  743. t.appendTimestamp(&attr, start, end, int64(end-start))
  744. t.createTraceEvent(name, ebpftracer.EventTypeFunEnt.Int(), 0, attr...)
  745. }