apm_tracing.go 29 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936
  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) MQConsumerTraceStartEvent(sn string, sport uint16, r *l7.RequestData, appInfo AppInfo, container_id string) {
  142. t.span.SetAttributes(attribute.String("rpc.uri", string(r.Payload)))
  143. r.Protocol = l7.ProtocolKafka
  144. t.commonAttrs = []attribute.KeyValue{
  145. semconv.NetPeerName(sn),
  146. semconv.NetPeerPort(int(sport)),
  147. // buildAppMapFromEvent
  148. attribute.Int("server.code_type", appInfo.CodeType.Int()),
  149. attribute.String("server.app_name", appInfo.AppName),
  150. attribute.String("server.service_name", r.Protocol.ServiceNameString()),
  151. attribute.String("service_type", "MQ"),
  152. attribute.String("oper_type", "CONSUMER"),
  153. attribute.String("server.service_type", "TASK"),
  154. attribute.Int64("server.app_id", appInfo.AppIdHash.IntVal),
  155. attribute.Int64("server.agent_id", appInfo.AgentId),
  156. attribute.Int64("server.instance_id", appInfo.InstanceIdHash.IntVal),
  157. attribute.String("server.container_id", container_id),
  158. attribute.String("http.uri", "/"+r.MQTopic),
  159. attribute.Int("server.pid", int(r.Pid)),
  160. attribute.String("mq.ip", sn),
  161. attribute.Int("mq.port", int(sport)),
  162. }
  163. t.span.SetAttributes(t.commonAttrs...)
  164. t.startReady()
  165. }
  166. func (t *Trace) TraceStartEvent(method, path, sn, ua string, sport uint16, status l7.Status, addr netaddr.IPPort, pid uint32, appInfo AppInfo, container_id string) {
  167. t.span.SetAttributes(semconv.HTTPURL(fmt.Sprintf("http://%s:%d%s", sn, sport, path)),
  168. semconv.HTTPMethod(method),
  169. attribute.String("http.uri", path))
  170. if status > 399 {
  171. t.span.SetStatus(codes.Error, "")
  172. }
  173. //host := fmt.Sprintf("%s:%s", sn, sport)
  174. t.destination = addr
  175. t.commonAttrs = []attribute.KeyValue{
  176. semconv.NetPeerName(sn),
  177. semconv.NetPeerPort(int(sport)),
  178. // buildAppMapFromEvent
  179. attribute.Int("server.code_type", appInfo.CodeType.Int()),
  180. attribute.String("server.app_name", appInfo.AppName),
  181. attribute.String("server.service_name", appInfo.ServiceType),
  182. attribute.Int64("server.app_id", appInfo.AppIdHash.IntVal),
  183. attribute.Int64("server.agent_id", appInfo.AgentId),
  184. attribute.Int64("server.instance_id", appInfo.InstanceIdHash.IntVal),
  185. attribute.Int("server.pid", int(pid)),
  186. attribute.String("server.container_id", container_id),
  187. attribute.String("server.user_agent", ua),
  188. }
  189. t.span.SetAttributes(t.commonAttrs...)
  190. t.startReady()
  191. }
  192. // set context span
  193. func (t *Trace) CreateRootSpan(traceId uint64) {
  194. traceIdStr := strconv.Itoa(int(traceId))
  195. ctx, span := tracer(t.containerId).Start(context.Background(), traceIdStr, trace.WithSpanKind(trace.SpanKindClient))
  196. t.setContext(ctx)
  197. t.setSpan(span)
  198. }
  199. // Deprecated: use TraceStartEvent instead.
  200. func (t *Trace) TraceStart(method, path string, status l7.Status, duration time.Duration) {
  201. if t == nil || method == "" {
  202. return
  203. }
  204. t.createParentSpan("APPLICATION", duration, status >= 400,
  205. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  206. semconv.HTTPMethod(method),
  207. //semconv.HTTPStatusCode(int(status)),
  208. attribute.String("http.uri", path),
  209. )
  210. }
  211. // Deprecated: use TraceEndEvent instead.
  212. func (t *Trace) TraceEnd(r *l7.RequestData) {
  213. if t == nil {
  214. return
  215. }
  216. t.span.SetAttributes(
  217. semconv.HTTPStatusCode(int(r.Status)),
  218. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  219. )
  220. CalledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  221. if err == nil && CalledId != 0 {
  222. t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  223. }
  224. InstanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  225. if err == nil && InstanceIdFrom != 0 {
  226. t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  227. }
  228. AppIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  229. if err == nil && AppIdFrom != 0 {
  230. t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  231. }
  232. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  233. t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  234. }
  235. // klog.Debugf("r.ParentSpanContext.TypeFrom tpp %s", r.ParentSpanContext.TypeFrom)
  236. if r.ParentSpanContext.TypeFrom == "10" {
  237. t.span.SetAttributes(attribute.String("server.type_from", "SDK"))
  238. t.span.SetAttributes(attribute.String("server.app_id_from", "-1"))
  239. t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  240. t.span.SetAttributes(attribute.String("server.instance_id_from", "-1"))
  241. t.span.SetAttributes(attribute.String("server.call_id", "-1"))
  242. }
  243. // for _, v := range t.stack {
  244. // fmt.Printf("TraceEndTraceEndTraceEnd%s\n", v)
  245. // }
  246. //t.buildFun()
  247. t.span.End(trace.WithTimestamp(time.Now()))
  248. }
  249. // 新增结束事件
  250. func (t *Trace) TraceEndEvent(r *l7.RequestData) {
  251. if t == nil {
  252. return
  253. }
  254. var attr []attribute.KeyValue
  255. attr = append(attr,
  256. semconv.HTTPStatusCode(int(r.Status)),
  257. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  258. )
  259. //t.span.SetAttributes(
  260. // semconv.HTTPStatusCode(int(r.Status)),
  261. // attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  262. //)
  263. calledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  264. if err == nil && calledId != 0 {
  265. attr = append(attr, attribute.Int64("server.called_id", calledId))
  266. //t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  267. }
  268. instanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  269. if err == nil && instanceIdFrom != 0 {
  270. attr = append(attr, attribute.Int64("server.instance_id_from", instanceIdFrom))
  271. //t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  272. }
  273. appIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  274. if err == nil && appIdFrom != 0 {
  275. attr = append(attr, attribute.Int64("server.app_id_from", appIdFrom))
  276. //t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  277. }
  278. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  279. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  280. //t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  281. }
  282. // klog.Debugf("r.ParentSpanContext.TypeFrom event tpp %s", r.ParentSpanContext.TypeFrom)
  283. if r.ParentSpanContext.TypeFrom == "10" {
  284. attr = append(attr, attribute.String("server.type_from", "SDK"))
  285. attr = append(attr, attribute.String("server.app_id_from", "-1"))
  286. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  287. attr = append(attr, attribute.String("server.instance_id_from", "-1"))
  288. attr = append(attr, attribute.String("server.call_id", "-1"))
  289. } else if r.ParentSpanContext.TypeFrom == "02" {
  290. attr = append(attr, attribute.String("server.type_from", "Nodejs"))
  291. }
  292. attr = append(attr, attribute.String("server.src_addr", r.DAddr.String()))
  293. attr = append(attr, attribute.String("server.dst_addr", r.SAddr.String()))
  294. klog.Debugf("TraceEndEvent SAddr.String %s", r.SAddr.String())
  295. klog.Debugf("TraceEndEvent DAddr.String %s", r.DAddr.String())
  296. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  297. t.span.SetAttributes(attr...)
  298. t.endReadyEvent(r.EventCount)
  299. }
  300. func (t *Trace) appendTimestamp(attr *[]attribute.KeyValue, s, e uint64, d int64) {
  301. *attr = append(*attr,
  302. attribute.Int64("time.start_at", utils.KtimeToTimestamp(s)),
  303. attribute.Int64("time.end_at", utils.KtimeToTimestamp(e)),
  304. attribute.Int64("time.duration", d),
  305. )
  306. }
  307. func (t *Trace) createParentSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  308. end := time.Now()
  309. start := end.Add(-duration)
  310. ctx, span := tracer(t.containerId).Start(context.Background(), name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  311. span.SetAttributes(attrs...)
  312. span.SetAttributes(t.commonAttrs...)
  313. if error {
  314. span.SetStatus(codes.Error, "")
  315. }
  316. t.setContext(ctx)
  317. t.setSpan(span)
  318. }
  319. func (t *Trace) SendEvent() {
  320. t.span.End()
  321. }
  322. func (t *Trace) GetSpan() trace.Span {
  323. return t.span
  324. }
  325. func (t *Trace) createTraceEvent(protocol l7.Protocol, eventType int, attrs ...attribute.KeyValue) {
  326. t.span.AddEventApm(protocol.String(), eventType, protocol.Int(), trace.WithAttributes(attrs...))
  327. //atomic.AddUint32(t.currenEventCount, 1)
  328. }
  329. func (t *Trace) createFuncTraceEvent(name string, eventType int, attrs ...attribute.KeyValue) {
  330. t.span.AddEventApm(name, eventType, 0, trace.WithAttributes(attrs...))
  331. }
  332. func (t *Trace) addEvent() {
  333. atomic.AddUint32(t.currenEventCount, 1)
  334. }
  335. func (t *Trace) createTraceSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  336. end := time.Now()
  337. start := end.Add(-duration)
  338. //fmt.Println("createTraceSpan:", t.ctx)
  339. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  340. span.SetAttributes(t.commonAttrs...)
  341. span.SetAttributes(attrs...)
  342. if error {
  343. span.SetStatus(codes.Error, "")
  344. }
  345. span.End(trace.WithTimestamp(end))
  346. }
  347. func (t *Trace) MysqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  348. if t == nil || query == "" {
  349. return
  350. }
  351. t.createTraceSpan(l7.ProtocolMysql.String(), duration, error,
  352. semconv.DBSystemMySQL,
  353. semconv.DBStatement(query),
  354. semconv.NetPeerName(destination.IP().String()),
  355. semconv.NetPeerPort(int(destination.Port())),
  356. )
  357. }
  358. // 内置允许操作(始终放行)
  359. var builtinOps = map[string]struct{}{
  360. "SELECT": {},
  361. "INSERT": {},
  362. "UPDATE": {},
  363. "DELETE": {},
  364. "WITH": {},
  365. }
  366. var (
  367. allowedOps map[string]struct{}
  368. initOnce sync.Once
  369. )
  370. // 解析并缓存环境变量中的前缀
  371. func initAllowedOps() {
  372. allowedOps = make(map[string]struct{})
  373. env := os.Getenv("ALLOWED_SQL_PREFIXES")
  374. for _, s := range strings.Split(env, ",") {
  375. if trimmed := strings.ToUpper(strings.TrimSpace(s)); trimmed != "" {
  376. allowedOps[trimmed] = struct{}{}
  377. }
  378. }
  379. }
  380. func isCURDOperation(q string) bool {
  381. if len(q) < 6 {
  382. return false
  383. }
  384. // 提取前缀(考虑最长10个字符)
  385. prefix := strings.ToUpper(strings.TrimSpace(q))
  386. if len(prefix) > 10 {
  387. prefix = prefix[:10]
  388. }
  389. // 提取第一个单词作为操作关键字
  390. spaceIdx := strings.IndexAny(prefix, " \t\n")
  391. if spaceIdx > 0 {
  392. prefix = prefix[:spaceIdx]
  393. }
  394. // 先查固定允许的操作
  395. if _, ok := builtinOps[prefix]; ok {
  396. return true
  397. } else {
  398. // 尝试清理所有非字母字符(移除控制字符、BOM、数字、标点等):
  399. //`SELECT` 长度 6,字节:[83 69 76 69 67 84]
  400. //prefix中的值长度 8,字节:[0 1 83 69 76 69 67 84]
  401. prefixTmp := strings.Map(func(r rune) rune {
  402. // 只保留大写字母 A-Z
  403. if r >= 'A' && r <= 'Z' {
  404. return r
  405. }
  406. return -1 // 删除其他所有字符
  407. }, prefix)
  408. if _, ok := builtinOps[prefixTmp]; ok {
  409. return true
  410. }
  411. }
  412. // 加载环境变量控制的操作(只执行一次)
  413. initOnce.Do(initAllowedOps)
  414. // 查额外允许操作
  415. _, ok := allowedOps[prefix]
  416. return ok
  417. }
  418. func (t *Trace) SQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, query string, r *l7.RequestData, destination netaddr.IPPort) {
  419. //fmt.Println("query", query)
  420. if t == nil {
  421. return
  422. }
  423. t.addEvent()
  424. // 只保留增删改查基本操作
  425. if !isCURDOperation(query) {
  426. return
  427. }
  428. // 拼接 destination 的 IP 和端口为 ip:port 格式
  429. destAddr := fmt.Sprintf("%s:%d", destination.IP().String(), destination.Port())
  430. var attr []attribute.KeyValue
  431. attr = append(attr,
  432. semconvVal,
  433. semconv.DBStatement(query),
  434. semconv.NetPeerName(destination.IP().String()),
  435. semconv.NetPeerPort(int(destination.Port())),
  436. attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  437. attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  438. attribute.String("sql.dbn", destAddr),
  439. )
  440. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  441. attr = append(attr,
  442. attribute.Bool("sql.exception", r.Status.Error()),
  443. )
  444. if r.Status.Error() {
  445. attr = append(attr,
  446. attribute.String("sql.exception_msg", r.ErrorMsg),
  447. )
  448. }
  449. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  450. }
  451. //func (t *Trace) MysqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  452. // fmt.Println("query", query)
  453. // if t == nil {
  454. // return
  455. // }
  456. //
  457. // var attr []attribute.KeyValue
  458. // attr = append(attr,
  459. // semconv.DBSystemMySQL,
  460. // semconv.DBStatement(query),
  461. // semconv.NetPeerName(destination.IP().String()),
  462. // semconv.NetPeerPort(int(destination.Port())),
  463. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  464. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  465. // )
  466. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  467. //
  468. // attr = append(attr,
  469. // attribute.Bool("sql.exception", r.Status.Error()),
  470. // )
  471. // t.createTraceEvent(l7.ProtocolMysql.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolMysql), attr...)
  472. //}
  473. //func (t *Trace) PostGreSqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  474. // if t == nil || query == "" {
  475. // return
  476. // }
  477. // t.createTraceSpan(l7.ProtocolPostgres.String(), duration, error,
  478. // semconv.DBSystemPostgreSQL,
  479. // semconv.DBStatement(query),
  480. // semconv.NetPeerName(destination.IP().String()),
  481. // semconv.NetPeerPort(int(destination.Port())),
  482. // )
  483. //}
  484. //func (t *Trace) PostGreSqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  485. // fmt.Println("query", query)
  486. // if t == nil {
  487. // return
  488. // }
  489. //
  490. // var attr []attribute.KeyValue
  491. // attr = append(attr,
  492. // semconv.DBSystemPostgreSQL,
  493. // semconv.DBStatement(query),
  494. // semconv.NetPeerName(destination.IP().String()),
  495. // semconv.NetPeerPort(int(destination.Port())),
  496. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  497. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  498. // )
  499. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  500. // attr = append(attr,
  501. // attribute.Bool("sql.exception", r.Status.Error()),
  502. // )
  503. // t.createTraceEvent(l7.ProtocolPostgres.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolPostgres), attr...)
  504. //}
  505. //func (t *Trace) DmTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  506. // if t == nil || query == "" {
  507. // return
  508. // }
  509. // l7Type := int(l7.ProtocolDM)
  510. // var attr []attribute.KeyValue
  511. // attr = append(attr,
  512. // semconv.DBSystemDaMengDB,
  513. // semconv.DBStatement(query),
  514. // semconv.NetPeerName(destination.IP().String()),
  515. // semconv.NetPeerPort(int(destination.Port())),
  516. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  517. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  518. // )
  519. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  520. //
  521. // attr = append(attr,
  522. // attribute.Bool("sql.exception", r.Status.Error()),
  523. // )
  524. // t.createTraceEvent(l7.ProtocolHTTP.String(), int(ebpftracer.EventTypeL7Request), l7Type, attr...)
  525. //}
  526. //func (t *Trace) RedisTraceQuery(cmd, args string, error bool, duration time.Duration) {
  527. // if t == nil || cmd == "" {
  528. // return
  529. // }
  530. // statement := cmd
  531. // if args != "" {
  532. // statement += " " + args
  533. // }
  534. // t.createTraceSpan(l7.ProtocolRedis.String(), duration, error,
  535. // semconv.DBSystemRedis,
  536. // semconv.DBOperation(cmd),
  537. // semconv.DBStatement(statement),
  538. // )
  539. //}
  540. func (t *Trace) NoSQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, operation, statement string, r *l7.RequestData, connSrc, connDest netaddr.IPPort) {
  541. if t == nil {
  542. return
  543. }
  544. t.addEvent()
  545. if statement == "" {
  546. return
  547. }
  548. // 拼接 destination 的 IP 和端口为 ip:port 格式
  549. srcAddr := r.ComponentSAddr.String()
  550. if r.ComponentSAddr.Port() == 0 {
  551. srcAddr = connSrc.String()
  552. }
  553. destinationAddr := r.ComponentDAddr.String()
  554. if r.ComponentDAddr.Port() == 0 {
  555. destinationAddr = connDest.String()
  556. }
  557. var attr []attribute.KeyValue
  558. attr = append(attr,
  559. semconvVal,
  560. semconv.DBStatement(statement),
  561. semconv.NetPeerName(connDest.IP().String()),
  562. semconv.NetPeerPort(int(connDest.Port())),
  563. attribute.String("nosql.src_addr", srcAddr),
  564. attribute.String("nosql.destination_addr", destinationAddr),
  565. attribute.String("nosql.dbn", connDest.String()),
  566. )
  567. // 如果 operation 不为空,添加 DBOperation 属性
  568. if operation != "" {
  569. attr = append(attr, semconv.DBOperation(operation))
  570. }
  571. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  572. attr = append(attr,
  573. attribute.Bool("nosql.exception", r.Status.Error()),
  574. )
  575. if r.Status.Error() {
  576. attr = append(attr,
  577. attribute.String("nosql.exception_msg", r.ErrorMsg),
  578. )
  579. }
  580. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  581. }
  582. func (t *Trace) MQTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, operation, statement string, r *l7.RequestData, connSrc, connDest netaddr.IPPort) {
  583. if t == nil {
  584. return
  585. }
  586. t.addEvent()
  587. if r.MQTopic != "" || r.MQKey != "" {
  588. statement = fmt.Sprintf("PRODUCE topic:%s key:%s", r.MQTopic, r.MQKey)
  589. }
  590. if statement == "" {
  591. return
  592. }
  593. srcAddr := r.ComponentSAddr.String()
  594. if r.ComponentSAddr.Port() == 0 {
  595. srcAddr = connSrc.String()
  596. }
  597. // 尝试解析目标地址,如果失败则使用 connDest 作为备用
  598. dAddr, err := net.ResolveTCPAddr("tcp", r.DestAddrString)
  599. var ip string
  600. port := 0
  601. if err != nil {
  602. klog.Errorln(err)
  603. } else {
  604. ip = dAddr.IP.String()
  605. port = dAddr.Port
  606. }
  607. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  608. if err != nil {
  609. assumedAppID = 0
  610. }
  611. var attr []attribute.KeyValue
  612. attr = append(attr,
  613. semconvVal,
  614. semconv.NetPeerName(ip),
  615. semconv.NetPeerPort(port),
  616. attribute.String("mq.info", statement),
  617. attribute.String("mq.src_addr", srcAddr),
  618. attribute.String("mq.destination_addr", r.DestAddrString),
  619. attribute.String("mq.topic", r.MQTopic),
  620. attribute.String("mq.key", r.MQKey),
  621. attribute.Int64("mq.assumed_app_id", assumedAppID),
  622. attribute.String("mq.span_id", r.SpanId),
  623. attribute.String("mq.oper_type", "PRODUCER"),
  624. )
  625. // 如果 operation 不为空,添加 DBOperation 属性
  626. if operation != "" {
  627. attr = append(attr, semconv.DBOperation(operation))
  628. }
  629. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  630. attr = append(attr,
  631. attribute.Bool("mq.exception", r.Status.Error()),
  632. )
  633. if r.Status.Error() {
  634. attr = append(attr,
  635. attribute.String("mq.exception_msg", r.ErrorMsg),
  636. )
  637. }
  638. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  639. }
  640. func (t *Trace) GrpcServerTraceQueryEvent(r *l7.RequestData, appInfo AppInfo) {
  641. if t == nil {
  642. return
  643. }
  644. t.addEvent()
  645. // if method == "" {
  646. // return
  647. // }
  648. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  649. if err != nil {
  650. assumedAppID = 0
  651. }
  652. var attr []attribute.KeyValue
  653. attr = append(attr,
  654. semconv.RPCSystemGRPC,
  655. attribute.String("rpc.uri", string(r.Payload)),
  656. attribute.String("rpc.schema", "grpc"),
  657. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  658. attribute.String("rpc.span_id", r.SpanId),
  659. attribute.Int("rpc.port", appInfo.Sport),
  660. attribute.String("rpc.ip", appInfo.Sn),
  661. attribute.String("rpc.oper_type", "PROVIDER"),
  662. attribute.String("rpc.method", "test_method"),
  663. )
  664. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  665. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  666. }
  667. func (t *Trace) GrpcClientTraceQueryEvent(r *l7.RequestData) {
  668. if t == nil {
  669. return
  670. }
  671. t.addEvent()
  672. // if method == "" {
  673. // return
  674. // }
  675. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  676. if err != nil {
  677. assumedAppID = 0
  678. }
  679. // 解析 RPCTarget 字符串,格式为 "ip:port"
  680. klog.Infof("r.RPCTarget is %s", r.DestAddrString)
  681. var rpcIP, rpcPort string
  682. if r.DestAddrString != "" {
  683. // 清理空字节,只保留有效字符
  684. cleanTarget := strings.TrimRight(r.DestAddrString, "\x00")
  685. cleanTarget = strings.TrimSpace(cleanTarget)
  686. klog.Infof("cleanTarget is %s", cleanTarget)
  687. if colonIndex := strings.LastIndex(cleanTarget, ":"); colonIndex != -1 {
  688. rpcIP = cleanTarget[:colonIndex]
  689. rpcPort = cleanTarget[colonIndex+1:]
  690. klog.Infof("rpcIP is %s, rpcPort is %s", rpcIP, rpcPort)
  691. } else {
  692. // 如果没有找到冒号,可能是只有IP或只有端口
  693. rpcIP = cleanTarget
  694. }
  695. }
  696. // 设置默认值
  697. if rpcIP == "" {
  698. rpcIP = "127.0.0.1"
  699. }
  700. if rpcPort == "" {
  701. rpcPort = "80"
  702. }
  703. // 转换端口为整数
  704. portInt, err := strconv.Atoi(rpcPort)
  705. if err != nil {
  706. klog.Infof("r.RPCTarget convert port err is %s", err)
  707. portInt = 80 // 默认端口
  708. } else {
  709. klog.Infof("r.RPCTarget convert port success, portInt is %d", portInt)
  710. }
  711. var attr []attribute.KeyValue
  712. attr = append(attr,
  713. semconv.RPCSystemGRPC,
  714. attribute.String("rpc.uri", string(r.Payload)),
  715. attribute.String("rpc.schema", "grpc"),
  716. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  717. attribute.String("rpc.span_id", r.SpanId),
  718. attribute.Int("rpc.port", portInt),
  719. attribute.String("rpc.ip", rpcIP),
  720. attribute.String("rpc.oper_type", "CONSUMER"),
  721. attribute.String("rpc.method", string(r.Payload)),
  722. )
  723. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  724. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  725. }
  726. func (t *Trace) DNSTraceQueryEvent(r *l7.RequestData, _type string, fqdn string, ttl uint32, ips []netaddr.IP) {
  727. if t == nil {
  728. return
  729. }
  730. t.addEvent()
  731. var attr []attribute.KeyValue
  732. var ipList []string
  733. for _, ip := range ips {
  734. ipList = append(ipList, ip.String())
  735. }
  736. ip := strings.Join(ipList, ",")
  737. attr = append(attr,
  738. attribute.String("dns.type", _type),
  739. attribute.String("dns.fqdn", fqdn),
  740. attribute.Int("dns.ttl", int(ttl)),
  741. attribute.String("dns.ips", ip),
  742. )
  743. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  744. t.createTraceEvent(l7.ProtocolDNS, ebpftracer.EventTypeL7Request.Int(), attr...)
  745. }
  746. func (t *Trace) HttpTraceRequest(method, path, ip string, port uint16, r *l7.RequestData) {
  747. if t == nil || method == "" {
  748. return
  749. }
  750. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  751. if err != nil {
  752. assumedAppID = 0
  753. }
  754. status := r.Status
  755. duration := r.Duration
  756. t.createTraceSpan(l7.ProtocolHTTP.String(), duration, status >= 400,
  757. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  758. semconv.HTTPMethod(method),
  759. semconv.HTTPStatusCode(int(status)),
  760. attribute.String("http.uri", path),
  761. attribute.String("http.ip", ip),
  762. attribute.Int64("http.assumed_app_id", assumedAppID),
  763. attribute.String("http.span_id", r.SpanId),
  764. attribute.Int("http.port", int(port)),
  765. )
  766. }
  767. // 新增事件处理
  768. func (t *Trace) HttpTraceRequestEvent(method, path, ip string, port uint16, r *l7.RequestData) {
  769. if t == nil {
  770. return
  771. }
  772. t.addEvent()
  773. if method == "" {
  774. return
  775. }
  776. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  777. if err != nil {
  778. assumedAppID = 0
  779. }
  780. status := r.Status
  781. var attr []attribute.KeyValue
  782. attr = append(attr,
  783. semconv.HTTPURL(fmt.Sprintf("http://%s:%d%s", ip, port, path)),
  784. semconv.HTTPMethod(method),
  785. semconv.HTTPStatusCode(int(status)),
  786. attribute.Bool("http.status_error", status > 399),
  787. attribute.String("http.uri", path),
  788. attribute.String("http.ip", ip),
  789. attribute.Int64("http.assumed_app_id", assumedAppID),
  790. attribute.String("http.span_id", r.SpanId),
  791. attribute.Int("http.port", int(port)),
  792. attribute.String("http.src_addr", r.ComponentSAddr.String()),
  793. attribute.String("http.destination_addr", r.ComponentDAddr.String()),
  794. attribute.Bool("http.is_tls", r.IsTls),
  795. )
  796. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  797. t.createTraceEvent(l7.ProtocolHTTP, ebpftracer.EventTypeL7Request.Int(), attr...)
  798. }
  799. func (t *Trace) FuncTraceQuery(funcname string, duration time.Duration, start uint64, end uint64) {
  800. if t == nil {
  801. return
  802. }
  803. t.addEvent()
  804. if funcname == "" {
  805. return
  806. }
  807. t.createTraceSpanNoTime2(funcname, duration, false, start, end)
  808. //t.createTraceSpanNoTime2(funcname, duration, false, start, end, attribute.Int("num", num))
  809. }
  810. func (t *Trace) createTraceSpanNoTime(name string, duration time.Duration, error bool, start uint64, end uint64, attrs ...attribute.KeyValue) {
  811. // end := time.Now()
  812. // start := end.Add(-duration)
  813. startTime := time.Unix(0, int64(start))
  814. endTime := time.Unix(0, int64(end))
  815. //fmt.Println("createTraceSpan:", t.ctx)
  816. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  817. span.SetAttributes(t.commonAttrs...)
  818. span.SetAttributes(attrs...)
  819. if error {
  820. span.SetStatus(codes.Error, "")
  821. }
  822. span.End(trace.WithTimestamp(endTime))
  823. }
  824. func (t *Trace) createTraceSpanNoTime2(name string, duration time.Duration, error bool, start uint64, end uint64) {
  825. // end := time.Now()
  826. // start := end.Add(-duration)
  827. //startTime := time.Unix(0, int64(start))
  828. //endTime := time.Unix(0, int64(end))
  829. ////fmt.Println("createTraceSpan:", t.ctx)
  830. //_, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  831. //span.SetAttributes(t.commonAttrs...)
  832. //span.SetAttributes(attrs...)
  833. //if error {
  834. // span.SetStatus(codes.Error, "")
  835. //}
  836. //span.End(trace.WithTimestamp(endTime))
  837. //attrs = append([]attribute.KeyValue{
  838. // attribute.Int64("startAt", int64(start)),
  839. // attribute.Int64("endAt", int64(end)),
  840. //})
  841. var attr []attribute.KeyValue
  842. t.appendTimestamp(&attr, start, end, int64(end-start))
  843. t.createFuncTraceEvent(name, ebpftracer.EventTypeFunEnt.Int(), attr...)
  844. }