apm_tracing.go 29 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941
  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. // 添加 SysvcFrom 到 attributes
  297. if r.SysvcFrom != "" {
  298. attr = append(attr, attribute.String("server.sysvc_from", r.SysvcFrom))
  299. }
  300. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  301. t.span.SetAttributes(attr...)
  302. t.endReadyEvent(r.EventCount)
  303. }
  304. func (t *Trace) appendTimestamp(attr *[]attribute.KeyValue, s, e uint64, d int64) {
  305. *attr = append(*attr,
  306. attribute.Int64("time.start_at", utils.KtimeToTimestamp(s)),
  307. attribute.Int64("time.end_at", utils.KtimeToTimestamp(e)),
  308. attribute.Int64("time.duration", d),
  309. )
  310. }
  311. func (t *Trace) createParentSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  312. end := time.Now()
  313. start := end.Add(-duration)
  314. ctx, span := tracer(t.containerId).Start(context.Background(), name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  315. span.SetAttributes(attrs...)
  316. span.SetAttributes(t.commonAttrs...)
  317. if error {
  318. span.SetStatus(codes.Error, "")
  319. }
  320. t.setContext(ctx)
  321. t.setSpan(span)
  322. }
  323. func (t *Trace) SendEvent() {
  324. t.span.End()
  325. }
  326. func (t *Trace) GetSpan() trace.Span {
  327. return t.span
  328. }
  329. func (t *Trace) createTraceEvent(protocol l7.Protocol, eventType int, attrs ...attribute.KeyValue) {
  330. t.span.AddEventApm(protocol.String(), eventType, protocol.Int(), trace.WithAttributes(attrs...))
  331. //atomic.AddUint32(t.currenEventCount, 1)
  332. }
  333. func (t *Trace) createFuncTraceEvent(name string, eventType int, attrs ...attribute.KeyValue) {
  334. t.span.AddEventApm(name, eventType, 0, trace.WithAttributes(attrs...))
  335. }
  336. func (t *Trace) addEvent() {
  337. atomic.AddUint32(t.currenEventCount, 1)
  338. }
  339. func (t *Trace) createTraceSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  340. end := time.Now()
  341. start := end.Add(-duration)
  342. //fmt.Println("createTraceSpan:", t.ctx)
  343. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  344. span.SetAttributes(t.commonAttrs...)
  345. span.SetAttributes(attrs...)
  346. if error {
  347. span.SetStatus(codes.Error, "")
  348. }
  349. span.End(trace.WithTimestamp(end))
  350. }
  351. func (t *Trace) MysqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  352. if t == nil || query == "" {
  353. return
  354. }
  355. t.createTraceSpan(l7.ProtocolMysql.String(), duration, error,
  356. semconv.DBSystemMySQL,
  357. semconv.DBStatement(query),
  358. semconv.NetPeerName(destination.IP().String()),
  359. semconv.NetPeerPort(int(destination.Port())),
  360. )
  361. }
  362. // 内置允许操作(始终放行)
  363. var builtinOps = map[string]struct{}{
  364. "SELECT": {},
  365. "INSERT": {},
  366. "UPDATE": {},
  367. "DELETE": {},
  368. "WITH": {},
  369. }
  370. var (
  371. allowedOps map[string]struct{}
  372. initOnce sync.Once
  373. )
  374. // 解析并缓存环境变量中的前缀
  375. func initAllowedOps() {
  376. allowedOps = make(map[string]struct{})
  377. env := os.Getenv("ALLOWED_SQL_PREFIXES")
  378. for _, s := range strings.Split(env, ",") {
  379. if trimmed := strings.ToUpper(strings.TrimSpace(s)); trimmed != "" {
  380. allowedOps[trimmed] = struct{}{}
  381. }
  382. }
  383. }
  384. func isCURDOperation(q string) bool {
  385. if len(q) < 6 {
  386. return false
  387. }
  388. // 提取前缀(考虑最长10个字符)
  389. prefix := strings.ToUpper(strings.TrimSpace(q))
  390. if len(prefix) > 10 {
  391. prefix = prefix[:10]
  392. }
  393. // 提取第一个单词作为操作关键字
  394. spaceIdx := strings.IndexAny(prefix, " \t\n")
  395. if spaceIdx > 0 {
  396. prefix = prefix[:spaceIdx]
  397. }
  398. // 先查固定允许的操作
  399. if _, ok := builtinOps[prefix]; ok {
  400. return true
  401. } else {
  402. // 尝试清理所有非字母字符(移除控制字符、BOM、数字、标点等):
  403. //`SELECT` 长度 6,字节:[83 69 76 69 67 84]
  404. //prefix中的值长度 8,字节:[0 1 83 69 76 69 67 84]
  405. prefixTmp := strings.Map(func(r rune) rune {
  406. // 只保留大写字母 A-Z
  407. if r >= 'A' && r <= 'Z' {
  408. return r
  409. }
  410. return -1 // 删除其他所有字符
  411. }, prefix)
  412. if _, ok := builtinOps[prefixTmp]; ok {
  413. return true
  414. }
  415. }
  416. // 加载环境变量控制的操作(只执行一次)
  417. initOnce.Do(initAllowedOps)
  418. // 查额外允许操作
  419. _, ok := allowedOps[prefix]
  420. return ok
  421. }
  422. func (t *Trace) SQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, query string, r *l7.RequestData, destination netaddr.IPPort) {
  423. //fmt.Println("query", query)
  424. if t == nil {
  425. return
  426. }
  427. t.addEvent()
  428. // 只保留增删改查基本操作
  429. if !isCURDOperation(query) {
  430. return
  431. }
  432. // 拼接 destination 的 IP 和端口为 ip:port 格式
  433. destAddr := fmt.Sprintf("%s:%d", destination.IP().String(), destination.Port())
  434. var attr []attribute.KeyValue
  435. attr = append(attr,
  436. semconvVal,
  437. semconv.DBStatement(query),
  438. semconv.NetPeerName(destination.IP().String()),
  439. semconv.NetPeerPort(int(destination.Port())),
  440. attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  441. attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  442. attribute.String("sql.dbn", destAddr),
  443. )
  444. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  445. attr = append(attr,
  446. attribute.Bool("sql.exception", r.Status.Error()),
  447. )
  448. if r.Status.Error() {
  449. attr = append(attr,
  450. attribute.String("sql.exception_msg", r.ErrorMsg),
  451. )
  452. }
  453. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  454. }
  455. //func (t *Trace) MysqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  456. // fmt.Println("query", query)
  457. // if t == nil {
  458. // return
  459. // }
  460. //
  461. // var attr []attribute.KeyValue
  462. // attr = append(attr,
  463. // semconv.DBSystemMySQL,
  464. // semconv.DBStatement(query),
  465. // semconv.NetPeerName(destination.IP().String()),
  466. // semconv.NetPeerPort(int(destination.Port())),
  467. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  468. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  469. // )
  470. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  471. //
  472. // attr = append(attr,
  473. // attribute.Bool("sql.exception", r.Status.Error()),
  474. // )
  475. // t.createTraceEvent(l7.ProtocolMysql.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolMysql), attr...)
  476. //}
  477. //func (t *Trace) PostGreSqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  478. // if t == nil || query == "" {
  479. // return
  480. // }
  481. // t.createTraceSpan(l7.ProtocolPostgres.String(), duration, error,
  482. // semconv.DBSystemPostgreSQL,
  483. // semconv.DBStatement(query),
  484. // semconv.NetPeerName(destination.IP().String()),
  485. // semconv.NetPeerPort(int(destination.Port())),
  486. // )
  487. //}
  488. //func (t *Trace) PostGreSqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  489. // fmt.Println("query", query)
  490. // if t == nil {
  491. // return
  492. // }
  493. //
  494. // var attr []attribute.KeyValue
  495. // attr = append(attr,
  496. // semconv.DBSystemPostgreSQL,
  497. // semconv.DBStatement(query),
  498. // semconv.NetPeerName(destination.IP().String()),
  499. // semconv.NetPeerPort(int(destination.Port())),
  500. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  501. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  502. // )
  503. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  504. // attr = append(attr,
  505. // attribute.Bool("sql.exception", r.Status.Error()),
  506. // )
  507. // t.createTraceEvent(l7.ProtocolPostgres.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolPostgres), attr...)
  508. //}
  509. //func (t *Trace) DmTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  510. // if t == nil || query == "" {
  511. // return
  512. // }
  513. // l7Type := int(l7.ProtocolDM)
  514. // var attr []attribute.KeyValue
  515. // attr = append(attr,
  516. // semconv.DBSystemDaMengDB,
  517. // semconv.DBStatement(query),
  518. // semconv.NetPeerName(destination.IP().String()),
  519. // semconv.NetPeerPort(int(destination.Port())),
  520. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  521. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  522. // )
  523. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  524. //
  525. // attr = append(attr,
  526. // attribute.Bool("sql.exception", r.Status.Error()),
  527. // )
  528. // t.createTraceEvent(l7.ProtocolHTTP.String(), int(ebpftracer.EventTypeL7Request), l7Type, attr...)
  529. //}
  530. //func (t *Trace) RedisTraceQuery(cmd, args string, error bool, duration time.Duration) {
  531. // if t == nil || cmd == "" {
  532. // return
  533. // }
  534. // statement := cmd
  535. // if args != "" {
  536. // statement += " " + args
  537. // }
  538. // t.createTraceSpan(l7.ProtocolRedis.String(), duration, error,
  539. // semconv.DBSystemRedis,
  540. // semconv.DBOperation(cmd),
  541. // semconv.DBStatement(statement),
  542. // )
  543. //}
  544. func (t *Trace) NoSQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, operation, statement string, r *l7.RequestData, connSrc, connDest netaddr.IPPort) {
  545. if t == nil {
  546. return
  547. }
  548. t.addEvent()
  549. if statement == "" {
  550. return
  551. }
  552. // 拼接 destination 的 IP 和端口为 ip:port 格式
  553. srcAddr := r.ComponentSAddr.String()
  554. if r.ComponentSAddr.Port() == 0 {
  555. srcAddr = connSrc.String()
  556. }
  557. destinationAddr := r.ComponentDAddr.String()
  558. if r.ComponentDAddr.Port() == 0 {
  559. destinationAddr = connDest.String()
  560. }
  561. var attr []attribute.KeyValue
  562. attr = append(attr,
  563. semconvVal,
  564. semconv.DBStatement(statement),
  565. semconv.NetPeerName(connDest.IP().String()),
  566. semconv.NetPeerPort(int(connDest.Port())),
  567. attribute.String("nosql.src_addr", srcAddr),
  568. attribute.String("nosql.destination_addr", destinationAddr),
  569. attribute.String("nosql.dbn", connDest.String()),
  570. )
  571. // 如果 operation 不为空,添加 DBOperation 属性
  572. if operation != "" {
  573. attr = append(attr, semconv.DBOperation(operation))
  574. }
  575. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  576. attr = append(attr,
  577. attribute.Bool("nosql.exception", r.Status.Error()),
  578. )
  579. if r.Status.Error() {
  580. attr = append(attr,
  581. attribute.String("nosql.exception_msg", r.ErrorMsg),
  582. )
  583. }
  584. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  585. }
  586. func (t *Trace) MQTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, operation, statement string, r *l7.RequestData, connSrc, connDest netaddr.IPPort) {
  587. if t == nil {
  588. return
  589. }
  590. t.addEvent()
  591. if r.MQTopic != "" || r.MQKey != "" {
  592. statement = fmt.Sprintf("PRODUCE topic:%s key:%s", r.MQTopic, r.MQKey)
  593. }
  594. if statement == "" {
  595. return
  596. }
  597. srcAddr := r.ComponentSAddr.String()
  598. if r.ComponentSAddr.Port() == 0 {
  599. srcAddr = connSrc.String()
  600. }
  601. // 尝试解析目标地址,如果失败则使用 connDest 作为备用
  602. dAddr, err := net.ResolveTCPAddr("tcp", r.DestAddrString)
  603. var ip string
  604. port := 0
  605. if err != nil {
  606. klog.Errorln(err)
  607. } else {
  608. ip = dAddr.IP.String()
  609. port = dAddr.Port
  610. }
  611. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  612. if err != nil {
  613. assumedAppID = 0
  614. }
  615. var attr []attribute.KeyValue
  616. attr = append(attr,
  617. semconvVal,
  618. semconv.NetPeerName(ip),
  619. semconv.NetPeerPort(port),
  620. attribute.String("mq.info", statement),
  621. attribute.String("mq.src_addr", srcAddr),
  622. attribute.String("mq.destination_addr", r.DestAddrString),
  623. attribute.String("mq.topic", r.MQTopic),
  624. attribute.String("mq.key", r.MQKey),
  625. attribute.Int64("mq.assumed_app_id", assumedAppID),
  626. attribute.String("mq.span_id", r.SpanId),
  627. attribute.String("mq.oper_type", "PRODUCER"),
  628. )
  629. // 如果 operation 不为空,添加 DBOperation 属性
  630. if operation != "" {
  631. attr = append(attr, semconv.DBOperation(operation))
  632. }
  633. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  634. attr = append(attr,
  635. attribute.Bool("mq.exception", r.Status.Error()),
  636. )
  637. if r.Status.Error() {
  638. attr = append(attr,
  639. attribute.String("mq.exception_msg", r.ErrorMsg),
  640. )
  641. }
  642. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  643. }
  644. func (t *Trace) GrpcServerTraceQueryEvent(r *l7.RequestData, appInfo AppInfo) {
  645. if t == nil {
  646. return
  647. }
  648. t.addEvent()
  649. // if method == "" {
  650. // return
  651. // }
  652. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  653. if err != nil {
  654. assumedAppID = 0
  655. }
  656. var attr []attribute.KeyValue
  657. attr = append(attr,
  658. semconv.RPCSystemGRPC,
  659. attribute.String("rpc.uri", string(r.Payload)),
  660. attribute.String("rpc.schema", "grpc"),
  661. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  662. attribute.String("rpc.span_id", r.SpanId),
  663. attribute.Int("rpc.port", appInfo.Sport),
  664. attribute.String("rpc.ip", appInfo.Sn),
  665. attribute.String("rpc.oper_type", "PROVIDER"),
  666. attribute.String("rpc.method", "test_method"),
  667. )
  668. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  669. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  670. }
  671. func (t *Trace) GrpcClientTraceQueryEvent(r *l7.RequestData) {
  672. if t == nil {
  673. return
  674. }
  675. t.addEvent()
  676. // if method == "" {
  677. // return
  678. // }
  679. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  680. if err != nil {
  681. assumedAppID = 0
  682. }
  683. // 解析 RPCTarget 字符串,格式为 "ip:port"
  684. klog.Infof("r.RPCTarget is %s", r.DestAddrString)
  685. var rpcIP, rpcPort string
  686. if r.DestAddrString != "" {
  687. // 清理空字节,只保留有效字符
  688. cleanTarget := strings.TrimRight(r.DestAddrString, "\x00")
  689. cleanTarget = strings.TrimSpace(cleanTarget)
  690. klog.Infof("cleanTarget is %s", cleanTarget)
  691. if colonIndex := strings.LastIndex(cleanTarget, ":"); colonIndex != -1 {
  692. rpcIP = cleanTarget[:colonIndex]
  693. rpcPort = cleanTarget[colonIndex+1:]
  694. klog.Infof("rpcIP is %s, rpcPort is %s", rpcIP, rpcPort)
  695. } else {
  696. // 如果没有找到冒号,可能是只有IP或只有端口
  697. rpcIP = cleanTarget
  698. }
  699. }
  700. // 设置默认值
  701. if rpcIP == "" {
  702. rpcIP = "127.0.0.1"
  703. }
  704. if rpcPort == "" {
  705. rpcPort = "80"
  706. }
  707. // 转换端口为整数
  708. portInt, err := strconv.Atoi(rpcPort)
  709. if err != nil {
  710. klog.Infof("r.RPCTarget convert port err is %s", err)
  711. portInt = 80 // 默认端口
  712. } else {
  713. klog.Infof("r.RPCTarget convert port success, portInt is %d", portInt)
  714. }
  715. var attr []attribute.KeyValue
  716. attr = append(attr,
  717. semconv.RPCSystemGRPC,
  718. attribute.String("rpc.uri", string(r.Payload)),
  719. attribute.String("rpc.schema", "grpc"),
  720. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  721. attribute.String("rpc.span_id", r.SpanId),
  722. attribute.Int("rpc.port", portInt),
  723. attribute.String("rpc.ip", rpcIP),
  724. attribute.String("rpc.oper_type", "CONSUMER"),
  725. attribute.String("rpc.method", string(r.Payload)),
  726. )
  727. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  728. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  729. }
  730. func (t *Trace) DNSTraceQueryEvent(r *l7.RequestData, _type string, fqdn string, ttl uint32, ips []netaddr.IP) {
  731. if t == nil {
  732. return
  733. }
  734. t.addEvent()
  735. var attr []attribute.KeyValue
  736. var ipList []string
  737. for _, ip := range ips {
  738. ipList = append(ipList, ip.String())
  739. }
  740. ip := strings.Join(ipList, ",")
  741. attr = append(attr,
  742. attribute.String("dns.type", _type),
  743. attribute.String("dns.fqdn", fqdn),
  744. attribute.Int("dns.ttl", int(ttl)),
  745. attribute.String("dns.ips", ip),
  746. )
  747. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  748. t.createTraceEvent(l7.ProtocolDNS, ebpftracer.EventTypeL7Request.Int(), attr...)
  749. }
  750. func (t *Trace) HttpTraceRequest(method, path, ip string, port uint16, r *l7.RequestData) {
  751. if t == nil || method == "" {
  752. return
  753. }
  754. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  755. if err != nil {
  756. assumedAppID = 0
  757. }
  758. status := r.Status
  759. duration := r.Duration
  760. t.createTraceSpan(l7.ProtocolHTTP.String(), duration, status >= 400,
  761. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  762. semconv.HTTPMethod(method),
  763. semconv.HTTPStatusCode(int(status)),
  764. attribute.String("http.uri", path),
  765. attribute.String("http.ip", ip),
  766. attribute.Int64("http.assumed_app_id", assumedAppID),
  767. attribute.String("http.span_id", r.SpanId),
  768. attribute.Int("http.port", int(port)),
  769. )
  770. }
  771. // 新增事件处理
  772. func (t *Trace) HttpTraceRequestEvent(method, path, ip string, port uint16, r *l7.RequestData) {
  773. if t == nil {
  774. return
  775. }
  776. t.addEvent()
  777. if method == "" {
  778. return
  779. }
  780. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  781. if err != nil {
  782. assumedAppID = 0
  783. }
  784. status := r.Status
  785. var attr []attribute.KeyValue
  786. attr = append(attr,
  787. semconv.HTTPURL(fmt.Sprintf("http://%s:%d%s", ip, port, path)),
  788. semconv.HTTPMethod(method),
  789. semconv.HTTPStatusCode(int(status)),
  790. attribute.Bool("http.status_error", status > 399),
  791. attribute.String("http.uri", path),
  792. attribute.String("http.ip", ip),
  793. attribute.Int64("http.assumed_app_id", assumedAppID),
  794. attribute.String("http.span_id", r.SpanId),
  795. attribute.Int("http.port", int(port)),
  796. attribute.String("http.src_addr", r.ComponentSAddr.String()),
  797. attribute.String("http.destination_addr", r.ComponentDAddr.String()),
  798. attribute.Bool("http.is_tls", r.IsTls),
  799. )
  800. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  801. t.createTraceEvent(l7.ProtocolHTTP, ebpftracer.EventTypeL7Request.Int(), attr...)
  802. }
  803. func (t *Trace) FuncTraceQuery(funcname string, duration time.Duration, start uint64, end uint64) {
  804. if t == nil {
  805. return
  806. }
  807. t.addEvent()
  808. if funcname == "" {
  809. return
  810. }
  811. t.createTraceSpanNoTime2(funcname, duration, false, start, end)
  812. //t.createTraceSpanNoTime2(funcname, duration, false, start, end, attribute.Int("num", num))
  813. }
  814. func (t *Trace) createTraceSpanNoTime(name string, duration time.Duration, error bool, start uint64, end uint64, attrs ...attribute.KeyValue) {
  815. // end := time.Now()
  816. // start := end.Add(-duration)
  817. startTime := time.Unix(0, int64(start))
  818. endTime := time.Unix(0, int64(end))
  819. //fmt.Println("createTraceSpan:", t.ctx)
  820. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  821. span.SetAttributes(t.commonAttrs...)
  822. span.SetAttributes(attrs...)
  823. if error {
  824. span.SetStatus(codes.Error, "")
  825. }
  826. span.End(trace.WithTimestamp(endTime))
  827. }
  828. func (t *Trace) createTraceSpanNoTime2(name string, duration time.Duration, error bool, start uint64, end uint64) {
  829. // end := time.Now()
  830. // start := end.Add(-duration)
  831. //startTime := time.Unix(0, int64(start))
  832. //endTime := time.Unix(0, int64(end))
  833. ////fmt.Println("createTraceSpan:", t.ctx)
  834. //_, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  835. //span.SetAttributes(t.commonAttrs...)
  836. //span.SetAttributes(attrs...)
  837. //if error {
  838. // span.SetStatus(codes.Error, "")
  839. //}
  840. //span.End(trace.WithTimestamp(endTime))
  841. //attrs = append([]attribute.KeyValue{
  842. // attribute.Int64("startAt", int64(start)),
  843. // attribute.Int64("endAt", int64(end)),
  844. //})
  845. var attr []attribute.KeyValue
  846. t.appendTimestamp(&attr, start, end, int64(end-start))
  847. t.createFuncTraceEvent(name, ebpftracer.EventTypeFunEnt.Int(), attr...)
  848. }