apm_tracing.go 28 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910
  1. package tracing
  2. import (
  3. "context"
  4. "fmt"
  5. "net"
  6. "os"
  7. "strconv"
  8. "strings"
  9. "sync"
  10. "sync/atomic"
  11. "time"
  12. klog "github.com/sirupsen/logrus"
  13. "github.com/coroot/coroot-node-agent/ebpftracer"
  14. "github.com/coroot/coroot-node-agent/ebpftracer/l7"
  15. "github.com/coroot/coroot-node-agent/utils"
  16. . "github.com/coroot/coroot-node-agent/utils/modelse"
  17. "go.opentelemetry.io/otel/attribute"
  18. "go.opentelemetry.io/otel/codes"
  19. semconv "go.opentelemetry.io/otel/semconv/v1.18.0"
  20. "go.opentelemetry.io/otel/trace"
  21. "inet.af/netaddr"
  22. )
  23. /**
  24. * Trace
  25. */
  26. func (t *Trace) setContext(ctx context.Context) {
  27. t.lock.Lock()
  28. defer t.lock.Unlock()
  29. t.ctx = ctx
  30. }
  31. func (t *Trace) setSpan(span trace.Span) {
  32. t.lock.Lock()
  33. defer t.lock.Unlock()
  34. t.span = span
  35. }
  36. type TimeMap struct {
  37. Time uint64
  38. Type int
  39. Map *ebpftracer.StackFunEvent
  40. }
  41. //func (t *Trace) buildFun() {
  42. // mapSlice := []TimeMap{}
  43. // for i, v := range t.stack {
  44. // timeStartMap := TimeMap{}
  45. // if v.StackEvent.Location == 0 {
  46. // timeStartMap = TimeMap{
  47. // Time: v.StackEvent.TimeNsStart,
  48. // Type: 0,
  49. // Map: &t.stack[i],
  50. // }
  51. // } else {
  52. // timeStartMap = TimeMap{
  53. // Time: v.StackEvent.TimeNsEnd,
  54. // Type: 1,
  55. // Map: &t.stack[i],
  56. // }
  57. // }
  58. // mapSlice = append(mapSlice, timeStartMap)
  59. // }
  60. // sort.Slice(mapSlice, func(i, j int) bool {
  61. // return mapSlice[i].Time < mapSlice[j].Time
  62. // })
  63. //
  64. // funStack := []TimeMap{}
  65. //
  66. // currentfunNum := 1
  67. //
  68. // // for k, v := range mapSlice {
  69. // // fmt.Println("---SliceSliceindex", k, "value", v.Time, v.Type, v.Map.Uprobe.Funcname, v.Map.StackEvent.Nid)
  70. // // }
  71. //
  72. // mapSliceLen := len(mapSlice)
  73. // for k, v := range mapSlice {
  74. // // fmt.Println("SliceSliceindex", k, "value", v.Time, v.Type, v.Map.Uprobe.Funcname, v.Map.StackEvent.Nid)
  75. // if v.Type == 0 {
  76. // // 函数入口
  77. // funStack = append(funStack, v)
  78. // } else if v.Type == 1 {
  79. // // 函数出口
  80. // len := len(funStack)
  81. // if len < 1 {
  82. // fmt.Printf("buildFun ErrorError return before enter: %v\n", v)
  83. // continue
  84. // }
  85. // currnt := funStack[len-1]
  86. // if currnt.Map.StackEvent.Location != 0 {
  87. // fmt.Printf("currnt StackEvent ErrorError is not enter: %v\n", v)
  88. // continue
  89. // }
  90. // if k < mapSliceLen-1 && len >= 2 {
  91. // nextfun := mapSlice[k+1]
  92. // preCurrnt := funStack[len-2]
  93. // // // 处理 .netcore 多次 returun
  94. // // 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
  95. // if nextfun.Map.StackEvent.Location == 1 && nextfun.Map.Uprobe.Funcname == currnt.Map.Uprobe.Funcname && preCurrnt.Map.Uprobe.Funcname != currnt.Map.Uprobe.Funcname {
  96. // currentfunNum++
  97. // continue
  98. // }
  99. // }
  100. // funStack = funStack[:len-1]
  101. // duration := v.Map.StackEvent.TimeNsEnd - currnt.Map.StackEvent.TimeNsStart
  102. // t.FuncTraceQuery(currnt.Map.Uprobe.Funcname, time.Duration(duration), currnt.Map.StackEvent.TimeNsStart, v.Map.StackEvent.TimeNsEnd, currentfunNum)
  103. // currentfunNum = 1
  104. // }
  105. // }
  106. //}
  107. func (t *Trace) startReady() {
  108. t.lock.Lock()
  109. defer t.lock.Unlock()
  110. t.startEventReady = true
  111. }
  112. func (t *Trace) endReadyEvent(needCount uint32) {
  113. t.lock.Lock()
  114. defer t.lock.Unlock()
  115. t.endEventReady = true
  116. t.needEventCount = needCount
  117. }
  118. func (t *Trace) AllEventReady(traceID uint64) bool {
  119. klog.Debugf("[AllEventReady (current/need)|start|end|traceid](%d/%d)%v|%v|%d", *t.currenEventCount, t.needEventCount, t.startEventReady, t.endEventReady, traceID)
  120. return t.startEventReady && t.endEventReady && *t.currenEventCount >= t.needEventCount
  121. }
  122. func (t *Trace) GrpcServerTraceStartEvent(sn string, sport uint16, r *l7.RequestData, appInfo AppInfo, container_id string) {
  123. t.span.SetAttributes(attribute.String("rpc.uri", string(r.Payload)))
  124. t.commonAttrs = []attribute.KeyValue{
  125. semconv.NetPeerName(sn),
  126. semconv.NetPeerPort(int(sport)),
  127. // buildAppMapFromEvent
  128. attribute.Int("server.code_type", appInfo.CodeType.Int()),
  129. attribute.String("server.app_name", appInfo.AppName),
  130. attribute.String("server.service_name", "GRPC"),
  131. attribute.String("rpc.service_type", "RPC"),
  132. attribute.String("rpc.oper_type", "PROVIDER"),
  133. attribute.Int64("server.app_id", appInfo.AppIdHash.IntVal),
  134. attribute.Int64("server.agent_id", appInfo.AgentId),
  135. attribute.Int64("server.instance_id", appInfo.InstanceIdHash.IntVal),
  136. attribute.String("server.container_id", container_id),
  137. }
  138. t.span.SetAttributes(t.commonAttrs...)
  139. t.startReady()
  140. }
  141. func (t *Trace) TraceStartEvent(method, path, sn, ua string, sport uint16, status l7.Status, addr netaddr.IPPort, pid uint32, appInfo AppInfo, container_id string) {
  142. t.span.SetAttributes(semconv.HTTPURL(fmt.Sprintf("http://%s:%d%s", sn, sport, path)),
  143. semconv.HTTPMethod(method),
  144. attribute.String("http.uri", path))
  145. if status > 399 {
  146. t.span.SetStatus(codes.Error, "")
  147. }
  148. //host := fmt.Sprintf("%s:%s", sn, sport)
  149. t.destination = addr
  150. t.commonAttrs = []attribute.KeyValue{
  151. semconv.NetPeerName(sn),
  152. semconv.NetPeerPort(int(sport)),
  153. // buildAppMapFromEvent
  154. attribute.Int("server.code_type", appInfo.CodeType.Int()),
  155. attribute.String("server.app_name", appInfo.AppName),
  156. attribute.String("server.service_name", appInfo.ServiceType),
  157. attribute.Int64("server.app_id", appInfo.AppIdHash.IntVal),
  158. attribute.Int64("server.agent_id", appInfo.AgentId),
  159. attribute.Int64("server.instance_id", appInfo.InstanceIdHash.IntVal),
  160. attribute.Int("server.pid", int(pid)),
  161. attribute.String("server.container_id", container_id),
  162. attribute.String("server.user_agent", ua),
  163. }
  164. t.span.SetAttributes(t.commonAttrs...)
  165. t.startReady()
  166. }
  167. // set context span
  168. func (t *Trace) CreateRootSpan(traceId uint64) {
  169. traceIdStr := strconv.Itoa(int(traceId))
  170. ctx, span := tracer(t.containerId).Start(context.Background(), traceIdStr, trace.WithSpanKind(trace.SpanKindClient))
  171. t.setContext(ctx)
  172. t.setSpan(span)
  173. }
  174. // Deprecated: use TraceStartEvent instead.
  175. func (t *Trace) TraceStart(method, path string, status l7.Status, duration time.Duration) {
  176. if t == nil || method == "" {
  177. return
  178. }
  179. t.createParentSpan("APPLICATION", duration, status >= 400,
  180. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  181. semconv.HTTPMethod(method),
  182. //semconv.HTTPStatusCode(int(status)),
  183. attribute.String("http.uri", path),
  184. )
  185. }
  186. // Deprecated: use TraceEndEvent instead.
  187. func (t *Trace) TraceEnd(r *l7.RequestData) {
  188. if t == nil {
  189. return
  190. }
  191. t.span.SetAttributes(
  192. semconv.HTTPStatusCode(int(r.Status)),
  193. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  194. )
  195. CalledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  196. if err == nil && CalledId != 0 {
  197. t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  198. }
  199. InstanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  200. if err == nil && InstanceIdFrom != 0 {
  201. t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  202. }
  203. AppIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  204. if err == nil && AppIdFrom != 0 {
  205. t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  206. }
  207. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  208. t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  209. }
  210. // klog.Debugf("r.ParentSpanContext.TypeFrom tpp %s", r.ParentSpanContext.TypeFrom)
  211. if r.ParentSpanContext.TypeFrom == "10" {
  212. t.span.SetAttributes(attribute.String("server.type_from", "SDK"))
  213. t.span.SetAttributes(attribute.String("server.app_id_from", "-1"))
  214. t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  215. t.span.SetAttributes(attribute.String("server.instance_id_from", "-1"))
  216. t.span.SetAttributes(attribute.String("server.call_id", "-1"))
  217. }
  218. // for _, v := range t.stack {
  219. // fmt.Printf("TraceEndTraceEndTraceEnd%s\n", v)
  220. // }
  221. //t.buildFun()
  222. t.span.End(trace.WithTimestamp(time.Now()))
  223. }
  224. // 新增结束事件
  225. func (t *Trace) TraceEndEvent(r *l7.RequestData) {
  226. if t == nil {
  227. return
  228. }
  229. var attr []attribute.KeyValue
  230. attr = append(attr,
  231. semconv.HTTPStatusCode(int(r.Status)),
  232. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  233. )
  234. //t.span.SetAttributes(
  235. // semconv.HTTPStatusCode(int(r.Status)),
  236. // attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  237. //)
  238. calledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  239. if err == nil && calledId != 0 {
  240. attr = append(attr, attribute.Int64("server.called_id", calledId))
  241. //t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  242. }
  243. instanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  244. if err == nil && instanceIdFrom != 0 {
  245. attr = append(attr, attribute.Int64("server.instance_id_from", instanceIdFrom))
  246. //t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  247. }
  248. appIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  249. if err == nil && appIdFrom != 0 {
  250. attr = append(attr, attribute.Int64("server.app_id_from", appIdFrom))
  251. //t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  252. }
  253. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  254. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  255. //t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  256. }
  257. // klog.Debugf("r.ParentSpanContext.TypeFrom event tpp %s", r.ParentSpanContext.TypeFrom)
  258. if r.ParentSpanContext.TypeFrom == "10" {
  259. attr = append(attr, attribute.String("server.type_from", "SDK"))
  260. attr = append(attr, attribute.String("server.app_id_from", "-1"))
  261. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  262. attr = append(attr, attribute.String("server.instance_id_from", "-1"))
  263. attr = append(attr, attribute.String("server.call_id", "-1"))
  264. } else if r.ParentSpanContext.TypeFrom == "02" {
  265. attr = append(attr, attribute.String("server.type_from", "Nodejs"))
  266. }
  267. attr = append(attr, attribute.String("server.src_addr", r.DAddr.String()))
  268. attr = append(attr, attribute.String("server.dst_addr", r.SAddr.String()))
  269. klog.Debugf("TraceEndEvent SAddr.String %s", r.SAddr.String())
  270. klog.Debugf("TraceEndEvent DAddr.String %s", r.DAddr.String())
  271. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  272. t.span.SetAttributes(attr...)
  273. t.endReadyEvent(r.EventCount)
  274. }
  275. func (t *Trace) appendTimestamp(attr *[]attribute.KeyValue, s, e uint64, d int64) {
  276. *attr = append(*attr,
  277. attribute.Int64("time.start_at", utils.KtimeToTimestamp(s)),
  278. attribute.Int64("time.end_at", utils.KtimeToTimestamp(e)),
  279. attribute.Int64("time.duration", d),
  280. )
  281. }
  282. func (t *Trace) createParentSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  283. end := time.Now()
  284. start := end.Add(-duration)
  285. ctx, span := tracer(t.containerId).Start(context.Background(), name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  286. span.SetAttributes(attrs...)
  287. span.SetAttributes(t.commonAttrs...)
  288. if error {
  289. span.SetStatus(codes.Error, "")
  290. }
  291. t.setContext(ctx)
  292. t.setSpan(span)
  293. }
  294. func (t *Trace) SendEvent() {
  295. t.span.End()
  296. }
  297. func (t *Trace) GetSpan() trace.Span {
  298. return t.span
  299. }
  300. func (t *Trace) createTraceEvent(protocol l7.Protocol, eventType int, attrs ...attribute.KeyValue) {
  301. t.span.AddEventApm(protocol.String(), eventType, protocol.Int(), trace.WithAttributes(attrs...))
  302. //atomic.AddUint32(t.currenEventCount, 1)
  303. }
  304. func (t *Trace) createFuncTraceEvent(name string, eventType int, attrs ...attribute.KeyValue) {
  305. t.span.AddEventApm(name, eventType, 0, trace.WithAttributes(attrs...))
  306. }
  307. func (t *Trace) addEvent() {
  308. atomic.AddUint32(t.currenEventCount, 1)
  309. }
  310. func (t *Trace) createTraceSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  311. end := time.Now()
  312. start := end.Add(-duration)
  313. //fmt.Println("createTraceSpan:", t.ctx)
  314. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  315. span.SetAttributes(t.commonAttrs...)
  316. span.SetAttributes(attrs...)
  317. if error {
  318. span.SetStatus(codes.Error, "")
  319. }
  320. span.End(trace.WithTimestamp(end))
  321. }
  322. func (t *Trace) MysqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  323. if t == nil || query == "" {
  324. return
  325. }
  326. t.createTraceSpan(l7.ProtocolMysql.String(), duration, error,
  327. semconv.DBSystemMySQL,
  328. semconv.DBStatement(query),
  329. semconv.NetPeerName(destination.IP().String()),
  330. semconv.NetPeerPort(int(destination.Port())),
  331. )
  332. }
  333. // 内置允许操作(始终放行)
  334. var builtinOps = map[string]struct{}{
  335. "SELECT": {},
  336. "INSERT": {},
  337. "UPDATE": {},
  338. "DELETE": {},
  339. "WITH": {},
  340. }
  341. var (
  342. allowedOps map[string]struct{}
  343. initOnce sync.Once
  344. )
  345. // 解析并缓存环境变量中的前缀
  346. func initAllowedOps() {
  347. allowedOps = make(map[string]struct{})
  348. env := os.Getenv("ALLOWED_SQL_PREFIXES")
  349. for _, s := range strings.Split(env, ",") {
  350. if trimmed := strings.ToUpper(strings.TrimSpace(s)); trimmed != "" {
  351. allowedOps[trimmed] = struct{}{}
  352. }
  353. }
  354. }
  355. func isCURDOperation(q string) bool {
  356. if len(q) < 6 {
  357. return false
  358. }
  359. // 提取前缀(考虑最长10个字符)
  360. prefix := strings.ToUpper(strings.TrimSpace(q))
  361. if len(prefix) > 10 {
  362. prefix = prefix[:10]
  363. }
  364. // 提取第一个单词作为操作关键字
  365. spaceIdx := strings.IndexAny(prefix, " \t\n")
  366. if spaceIdx > 0 {
  367. prefix = prefix[:spaceIdx]
  368. }
  369. // 先查固定允许的操作
  370. if _, ok := builtinOps[prefix]; ok {
  371. return true
  372. } else {
  373. // 尝试清理所有非字母字符(移除控制字符、BOM、数字、标点等):
  374. //`SELECT` 长度 6,字节:[83 69 76 69 67 84]
  375. //prefix中的值长度 8,字节:[0 1 83 69 76 69 67 84]
  376. prefixTmp := strings.Map(func(r rune) rune {
  377. // 只保留大写字母 A-Z
  378. if r >= 'A' && r <= 'Z' {
  379. return r
  380. }
  381. return -1 // 删除其他所有字符
  382. }, prefix)
  383. if _, ok := builtinOps[prefixTmp]; ok {
  384. return true
  385. }
  386. }
  387. // 加载环境变量控制的操作(只执行一次)
  388. initOnce.Do(initAllowedOps)
  389. // 查额外允许操作
  390. _, ok := allowedOps[prefix]
  391. return ok
  392. }
  393. func (t *Trace) SQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, query string, r *l7.RequestData, destination netaddr.IPPort) {
  394. //fmt.Println("query", query)
  395. if t == nil {
  396. return
  397. }
  398. t.addEvent()
  399. // 只保留增删改查基本操作
  400. if !isCURDOperation(query) {
  401. return
  402. }
  403. // 拼接 destination 的 IP 和端口为 ip:port 格式
  404. destAddr := fmt.Sprintf("%s:%d", destination.IP().String(), destination.Port())
  405. var attr []attribute.KeyValue
  406. attr = append(attr,
  407. semconvVal,
  408. semconv.DBStatement(query),
  409. semconv.NetPeerName(destination.IP().String()),
  410. semconv.NetPeerPort(int(destination.Port())),
  411. attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  412. attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  413. attribute.String("sql.dbn", destAddr),
  414. )
  415. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  416. attr = append(attr,
  417. attribute.Bool("sql.exception", r.Status.Error()),
  418. )
  419. if r.Status.Error() {
  420. attr = append(attr,
  421. attribute.String("sql.exception_msg", r.ErrorMsg),
  422. )
  423. }
  424. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  425. }
  426. //func (t *Trace) MysqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  427. // fmt.Println("query", query)
  428. // if t == nil {
  429. // return
  430. // }
  431. //
  432. // var attr []attribute.KeyValue
  433. // attr = append(attr,
  434. // semconv.DBSystemMySQL,
  435. // semconv.DBStatement(query),
  436. // semconv.NetPeerName(destination.IP().String()),
  437. // semconv.NetPeerPort(int(destination.Port())),
  438. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  439. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  440. // )
  441. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  442. //
  443. // attr = append(attr,
  444. // attribute.Bool("sql.exception", r.Status.Error()),
  445. // )
  446. // t.createTraceEvent(l7.ProtocolMysql.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolMysql), attr...)
  447. //}
  448. //func (t *Trace) PostGreSqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  449. // if t == nil || query == "" {
  450. // return
  451. // }
  452. // t.createTraceSpan(l7.ProtocolPostgres.String(), duration, error,
  453. // semconv.DBSystemPostgreSQL,
  454. // semconv.DBStatement(query),
  455. // semconv.NetPeerName(destination.IP().String()),
  456. // semconv.NetPeerPort(int(destination.Port())),
  457. // )
  458. //}
  459. //func (t *Trace) PostGreSqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  460. // fmt.Println("query", query)
  461. // if t == nil {
  462. // return
  463. // }
  464. //
  465. // var attr []attribute.KeyValue
  466. // attr = append(attr,
  467. // semconv.DBSystemPostgreSQL,
  468. // semconv.DBStatement(query),
  469. // semconv.NetPeerName(destination.IP().String()),
  470. // semconv.NetPeerPort(int(destination.Port())),
  471. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  472. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  473. // )
  474. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  475. // attr = append(attr,
  476. // attribute.Bool("sql.exception", r.Status.Error()),
  477. // )
  478. // t.createTraceEvent(l7.ProtocolPostgres.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolPostgres), attr...)
  479. //}
  480. //func (t *Trace) DmTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  481. // if t == nil || query == "" {
  482. // return
  483. // }
  484. // l7Type := int(l7.ProtocolDM)
  485. // var attr []attribute.KeyValue
  486. // attr = append(attr,
  487. // semconv.DBSystemDaMengDB,
  488. // semconv.DBStatement(query),
  489. // semconv.NetPeerName(destination.IP().String()),
  490. // semconv.NetPeerPort(int(destination.Port())),
  491. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  492. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  493. // )
  494. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  495. //
  496. // attr = append(attr,
  497. // attribute.Bool("sql.exception", r.Status.Error()),
  498. // )
  499. // t.createTraceEvent(l7.ProtocolHTTP.String(), int(ebpftracer.EventTypeL7Request), l7Type, attr...)
  500. //}
  501. //func (t *Trace) RedisTraceQuery(cmd, args string, error bool, duration time.Duration) {
  502. // if t == nil || cmd == "" {
  503. // return
  504. // }
  505. // statement := cmd
  506. // if args != "" {
  507. // statement += " " + args
  508. // }
  509. // t.createTraceSpan(l7.ProtocolRedis.String(), duration, error,
  510. // semconv.DBSystemRedis,
  511. // semconv.DBOperation(cmd),
  512. // semconv.DBStatement(statement),
  513. // )
  514. //}
  515. func (t *Trace) NoSQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, operation, statement string, r *l7.RequestData, connSrc, connDest netaddr.IPPort) {
  516. if t == nil {
  517. return
  518. }
  519. t.addEvent()
  520. if statement == "" {
  521. return
  522. }
  523. // 拼接 destination 的 IP 和端口为 ip:port 格式
  524. srcAddr := r.ComponentSAddr.String()
  525. if r.ComponentSAddr.Port() == 0 {
  526. srcAddr = connSrc.String()
  527. }
  528. destinationAddr := r.ComponentDAddr.String()
  529. if r.ComponentDAddr.Port() == 0 {
  530. destinationAddr = connDest.String()
  531. }
  532. var attr []attribute.KeyValue
  533. attr = append(attr,
  534. semconvVal,
  535. semconv.DBStatement(statement),
  536. semconv.NetPeerName(connDest.IP().String()),
  537. semconv.NetPeerPort(int(connDest.Port())),
  538. attribute.String("nosql.src_addr", srcAddr),
  539. attribute.String("nosql.destination_addr", destinationAddr),
  540. attribute.String("nosql.dbn", connDest.String()),
  541. )
  542. // 如果 operation 不为空,添加 DBOperation 属性
  543. if operation != "" {
  544. attr = append(attr, semconv.DBOperation(operation))
  545. }
  546. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  547. attr = append(attr,
  548. attribute.Bool("nosql.exception", r.Status.Error()),
  549. )
  550. if r.Status.Error() {
  551. attr = append(attr,
  552. attribute.String("nosql.exception_msg", r.ErrorMsg),
  553. )
  554. }
  555. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  556. }
  557. func (t *Trace) MQTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, operation, statement string, r *l7.RequestData, connSrc, connDest netaddr.IPPort) {
  558. if t == nil {
  559. return
  560. }
  561. t.addEvent()
  562. if r.MQTopic != "" || r.MQKey != "" {
  563. statement = fmt.Sprintf("PRODUCE topic:%s key:%s", r.MQTopic, r.MQKey)
  564. }
  565. if statement == "" {
  566. return
  567. }
  568. srcAddr := r.ComponentSAddr.String()
  569. if r.ComponentSAddr.Port() == 0 {
  570. srcAddr = connSrc.String()
  571. }
  572. // 尝试解析目标地址,如果失败则使用 connDest 作为备用
  573. dAddr, err := net.ResolveTCPAddr("tcp", r.DestAddrString)
  574. var ip string
  575. port := 0
  576. if err != nil {
  577. klog.Errorln(err)
  578. } else {
  579. ip = dAddr.IP.String()
  580. port = dAddr.Port
  581. }
  582. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  583. if err != nil {
  584. assumedAppID = 0
  585. }
  586. var attr []attribute.KeyValue
  587. attr = append(attr,
  588. semconvVal,
  589. semconv.NetPeerName(ip),
  590. semconv.NetPeerPort(port),
  591. attribute.String("mq.info", statement),
  592. attribute.String("mq.src_addr", srcAddr),
  593. attribute.String("mq.destination_addr", r.DestAddrString),
  594. attribute.String("mq.topic", r.MQTopic),
  595. attribute.String("mq.key", r.MQKey),
  596. attribute.Int64("mq.assumed_app_id", assumedAppID),
  597. attribute.String("mq.span_id", r.SpanId),
  598. attribute.String("mq.oper_type", "PRODUCER"),
  599. )
  600. // 如果 operation 不为空,添加 DBOperation 属性
  601. if operation != "" {
  602. attr = append(attr, semconv.DBOperation(operation))
  603. }
  604. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  605. attr = append(attr,
  606. attribute.Bool("mq.exception", r.Status.Error()),
  607. )
  608. if r.Status.Error() {
  609. attr = append(attr,
  610. attribute.String("mq.exception_msg", r.ErrorMsg),
  611. )
  612. }
  613. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  614. }
  615. func (t *Trace) GrpcServerTraceQueryEvent(r *l7.RequestData, appInfo AppInfo) {
  616. if t == nil {
  617. return
  618. }
  619. t.addEvent()
  620. // if method == "" {
  621. // return
  622. // }
  623. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  624. if err != nil {
  625. assumedAppID = 0
  626. }
  627. var attr []attribute.KeyValue
  628. attr = append(attr,
  629. semconv.RPCSystemGRPC,
  630. attribute.String("rpc.uri", string(r.Payload)),
  631. attribute.String("rpc.schema", "grpc"),
  632. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  633. attribute.String("rpc.span_id", r.SpanId),
  634. attribute.Int("rpc.port", appInfo.Sport),
  635. attribute.String("rpc.ip", appInfo.Sn),
  636. attribute.String("rpc.oper_type", "PROVIDER"),
  637. attribute.String("rpc.method", "test_method"),
  638. )
  639. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  640. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  641. }
  642. func (t *Trace) GrpcClientTraceQueryEvent(r *l7.RequestData) {
  643. if t == nil {
  644. return
  645. }
  646. t.addEvent()
  647. // if method == "" {
  648. // return
  649. // }
  650. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  651. if err != nil {
  652. assumedAppID = 0
  653. }
  654. // 解析 RPCTarget 字符串,格式为 "ip:port"
  655. klog.Infof("r.RPCTarget is %s", r.DestAddrString)
  656. var rpcIP, rpcPort string
  657. if r.DestAddrString != "" {
  658. // 清理空字节,只保留有效字符
  659. cleanTarget := strings.TrimRight(r.DestAddrString, "\x00")
  660. cleanTarget = strings.TrimSpace(cleanTarget)
  661. klog.Infof("cleanTarget is %s", cleanTarget)
  662. if colonIndex := strings.LastIndex(cleanTarget, ":"); colonIndex != -1 {
  663. rpcIP = cleanTarget[:colonIndex]
  664. rpcPort = cleanTarget[colonIndex+1:]
  665. klog.Infof("rpcIP is %s, rpcPort is %s", rpcIP, rpcPort)
  666. } else {
  667. // 如果没有找到冒号,可能是只有IP或只有端口
  668. rpcIP = cleanTarget
  669. }
  670. }
  671. // 设置默认值
  672. if rpcIP == "" {
  673. rpcIP = "127.0.0.1"
  674. }
  675. if rpcPort == "" {
  676. rpcPort = "80"
  677. }
  678. // 转换端口为整数
  679. portInt, err := strconv.Atoi(rpcPort)
  680. if err != nil {
  681. klog.Infof("r.RPCTarget convert port err is %s", err)
  682. portInt = 80 // 默认端口
  683. } else {
  684. klog.Infof("r.RPCTarget convert port success, portInt is %d", portInt)
  685. }
  686. var attr []attribute.KeyValue
  687. attr = append(attr,
  688. semconv.RPCSystemGRPC,
  689. attribute.String("rpc.uri", string(r.Payload)),
  690. attribute.String("rpc.schema", "grpc"),
  691. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  692. attribute.String("rpc.span_id", r.SpanId),
  693. attribute.Int("rpc.port", portInt),
  694. attribute.String("rpc.ip", rpcIP),
  695. attribute.String("rpc.oper_type", "CONSUMER"),
  696. attribute.String("rpc.method", string(r.Payload)),
  697. )
  698. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  699. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  700. }
  701. func (t *Trace) DNSTraceQueryEvent(r *l7.RequestData, _type string, fqdn string, ttl uint32, ips []netaddr.IP) {
  702. if t == nil {
  703. return
  704. }
  705. t.addEvent()
  706. var attr []attribute.KeyValue
  707. var ipList []string
  708. for _, ip := range ips {
  709. ipList = append(ipList, ip.String())
  710. }
  711. ip := strings.Join(ipList, ",")
  712. attr = append(attr,
  713. attribute.String("dns.type", _type),
  714. attribute.String("dns.fqdn", fqdn),
  715. attribute.Int("dns.ttl", int(ttl)),
  716. attribute.String("dns.ips", ip),
  717. )
  718. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  719. t.createTraceEvent(l7.ProtocolDNS, ebpftracer.EventTypeL7Request.Int(), attr...)
  720. }
  721. func (t *Trace) HttpTraceRequest(method, path, ip string, port uint16, r *l7.RequestData) {
  722. if t == nil || method == "" {
  723. return
  724. }
  725. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  726. if err != nil {
  727. assumedAppID = 0
  728. }
  729. status := r.Status
  730. duration := r.Duration
  731. t.createTraceSpan(l7.ProtocolHTTP.String(), duration, status >= 400,
  732. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  733. semconv.HTTPMethod(method),
  734. semconv.HTTPStatusCode(int(status)),
  735. attribute.String("http.uri", path),
  736. attribute.String("http.ip", ip),
  737. attribute.Int64("http.assumed_app_id", assumedAppID),
  738. attribute.String("http.span_id", r.SpanId),
  739. attribute.Int("http.port", int(port)),
  740. )
  741. }
  742. // 新增事件处理
  743. func (t *Trace) HttpTraceRequestEvent(method, path, ip string, port uint16, r *l7.RequestData) {
  744. if t == nil {
  745. return
  746. }
  747. t.addEvent()
  748. if method == "" {
  749. return
  750. }
  751. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  752. if err != nil {
  753. assumedAppID = 0
  754. }
  755. status := r.Status
  756. var attr []attribute.KeyValue
  757. attr = append(attr,
  758. semconv.HTTPURL(fmt.Sprintf("http://%s:%d%s", ip, port, path)),
  759. semconv.HTTPMethod(method),
  760. semconv.HTTPStatusCode(int(status)),
  761. attribute.Bool("http.status_error", status > 399),
  762. attribute.String("http.uri", path),
  763. attribute.String("http.ip", ip),
  764. attribute.Int64("http.assumed_app_id", assumedAppID),
  765. attribute.String("http.span_id", r.SpanId),
  766. attribute.Int("http.port", int(port)),
  767. attribute.String("http.src_addr", r.ComponentSAddr.String()),
  768. attribute.String("http.destination_addr", r.ComponentDAddr.String()),
  769. attribute.Bool("http.is_tls", r.IsTls),
  770. )
  771. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  772. t.createTraceEvent(l7.ProtocolHTTP, ebpftracer.EventTypeL7Request.Int(), attr...)
  773. }
  774. func (t *Trace) FuncTraceQuery(funcname string, duration time.Duration, start uint64, end uint64) {
  775. if t == nil {
  776. return
  777. }
  778. t.addEvent()
  779. if funcname == "" {
  780. return
  781. }
  782. t.createTraceSpanNoTime2(funcname, duration, false, start, end)
  783. //t.createTraceSpanNoTime2(funcname, duration, false, start, end, attribute.Int("num", num))
  784. }
  785. func (t *Trace) createTraceSpanNoTime(name string, duration time.Duration, error bool, start uint64, end uint64, attrs ...attribute.KeyValue) {
  786. // end := time.Now()
  787. // start := end.Add(-duration)
  788. startTime := time.Unix(0, int64(start))
  789. endTime := time.Unix(0, int64(end))
  790. //fmt.Println("createTraceSpan:", t.ctx)
  791. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  792. span.SetAttributes(t.commonAttrs...)
  793. span.SetAttributes(attrs...)
  794. if error {
  795. span.SetStatus(codes.Error, "")
  796. }
  797. span.End(trace.WithTimestamp(endTime))
  798. }
  799. func (t *Trace) createTraceSpanNoTime2(name string, duration time.Duration, error bool, start uint64, end uint64) {
  800. // end := time.Now()
  801. // start := end.Add(-duration)
  802. //startTime := time.Unix(0, int64(start))
  803. //endTime := time.Unix(0, int64(end))
  804. ////fmt.Println("createTraceSpan:", t.ctx)
  805. //_, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  806. //span.SetAttributes(t.commonAttrs...)
  807. //span.SetAttributes(attrs...)
  808. //if error {
  809. // span.SetStatus(codes.Error, "")
  810. //}
  811. //span.End(trace.WithTimestamp(endTime))
  812. //attrs = append([]attribute.KeyValue{
  813. // attribute.Int64("startAt", int64(start)),
  814. // attribute.Int64("endAt", int64(end)),
  815. //})
  816. var attr []attribute.KeyValue
  817. t.appendTimestamp(&attr, start, end, int64(end-start))
  818. t.createFuncTraceEvent(name, ebpftracer.EventTypeFunEnt.Int(), attr...)
  819. }