apm_tracing.go 30 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944
  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, r *l7.RequestData, 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 r.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. if r.SysvcFrom != "" {
  190. t.commonAttrs = append(t.commonAttrs, attribute.String("server.sysvc_from", r.SysvcFrom))
  191. }
  192. t.span.SetAttributes(t.commonAttrs...)
  193. t.startReady()
  194. }
  195. // set context span
  196. func (t *Trace) CreateRootSpan(traceId uint64) {
  197. traceIdStr := strconv.Itoa(int(traceId))
  198. ctx, span := tracer(t.containerId).Start(context.Background(), traceIdStr, trace.WithSpanKind(trace.SpanKindClient))
  199. t.setContext(ctx)
  200. t.setSpan(span)
  201. }
  202. // Deprecated: use TraceStartEvent instead.
  203. func (t *Trace) TraceStart(method, path string, status l7.Status, duration time.Duration) {
  204. if t == nil || method == "" {
  205. return
  206. }
  207. t.createParentSpan("APPLICATION", duration, status >= 400,
  208. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  209. semconv.HTTPMethod(method),
  210. //semconv.HTTPStatusCode(int(status)),
  211. attribute.String("http.uri", path),
  212. )
  213. }
  214. // Deprecated: use TraceEndEvent instead.
  215. //func (t *Trace) TraceEnd(r *l7.RequestData) {
  216. // if t == nil {
  217. // return
  218. // }
  219. // t.span.SetAttributes(
  220. // semconv.HTTPStatusCode(int(r.Status)),
  221. // attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  222. // )
  223. //
  224. // CalledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  225. // if err == nil && CalledId != 0 {
  226. // t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  227. // }
  228. //
  229. // InstanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  230. // if err == nil && InstanceIdFrom != 0 {
  231. // t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  232. // }
  233. //
  234. // AppIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  235. // if err == nil && AppIdFrom != 0 {
  236. // t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  237. // }
  238. // if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  239. // t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  240. // }
  241. // // klog.Debugf("r.ParentSpanContext.TypeFrom tpp %s", r.ParentSpanContext.TypeFrom)
  242. // if r.ParentSpanContext.TypeFrom == "10" {
  243. // t.span.SetAttributes(attribute.String("server.type_from", "SDK"))
  244. // t.span.SetAttributes(attribute.String("server.app_id_from", "-1"))
  245. // t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  246. // t.span.SetAttributes(attribute.String("server.instance_id_from", "-1"))
  247. // t.span.SetAttributes(attribute.String("server.call_id", "-1"))
  248. // }
  249. // // for _, v := range t.stack {
  250. // // fmt.Printf("TraceEndTraceEndTraceEnd%s\n", v)
  251. // // }
  252. // //t.buildFun()
  253. // t.span.End(trace.WithTimestamp(time.Now()))
  254. //}
  255. // 新增结束事件
  256. func (t *Trace) TraceEndEvent(r *l7.RequestData) {
  257. if t == nil {
  258. return
  259. }
  260. var attr []attribute.KeyValue
  261. attr = append(attr,
  262. semconv.HTTPStatusCode(int(r.Status)),
  263. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  264. )
  265. //t.span.SetAttributes(
  266. // semconv.HTTPStatusCode(int(r.Status)),
  267. // attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  268. //)
  269. calledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  270. if err == nil && calledId != 0 {
  271. attr = append(attr, attribute.Int64("server.called_id", calledId))
  272. //t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  273. }
  274. instanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  275. if err == nil && instanceIdFrom != 0 {
  276. attr = append(attr, attribute.Int64("server.instance_id_from", instanceIdFrom))
  277. //t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  278. }
  279. appIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  280. if err == nil && appIdFrom != 0 {
  281. attr = append(attr, attribute.Int64("server.app_id_from", appIdFrom))
  282. // Add type_from attribute for all cases
  283. attr = append(attr, attribute.String("server.type_from", r.ParentSpanContext.TypeFrom.String()))
  284. }
  285. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  286. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  287. //t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  288. }
  289. // klog.Debugf("r.ParentSpanContext.TypeFrom event tpp %s", r.ParentSpanContext.TypeFrom)
  290. if r.ParentSpanContext.TypeFrom == "10" {
  291. attr = append(attr, attribute.String("server.app_id_from", "-1"))
  292. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  293. attr = append(attr, attribute.String("server.instance_id_from", "-1"))
  294. attr = append(attr, attribute.String("server.call_id", "-1"))
  295. }
  296. attr = append(attr, attribute.String("server.src_addr", r.DAddr.String()))
  297. attr = append(attr, attribute.String("server.dst_addr", r.SAddr.String()))
  298. klog.Debugf("TraceEndEvent SAddr.String %s", r.SAddr.String())
  299. klog.Debugf("TraceEndEvent DAddr.String %s", r.DAddr.String())
  300. // 添加 SysvcFrom 到 attributes
  301. if r.SysvcFrom != "" {
  302. attr = append(attr, attribute.String("server.sysvc_from", r.SysvcFrom))
  303. }
  304. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  305. t.span.SetAttributes(attr...)
  306. t.endReadyEvent(r.EventCount)
  307. }
  308. func (t *Trace) appendTimestamp(attr *[]attribute.KeyValue, s, e uint64, d int64) {
  309. *attr = append(*attr,
  310. attribute.Int64("time.start_at", utils.KtimeToTimestamp(s)),
  311. attribute.Int64("time.end_at", utils.KtimeToTimestamp(e)),
  312. attribute.Int64("time.duration", d),
  313. )
  314. }
  315. func (t *Trace) createParentSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  316. end := time.Now()
  317. start := end.Add(-duration)
  318. ctx, span := tracer(t.containerId).Start(context.Background(), name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  319. span.SetAttributes(attrs...)
  320. span.SetAttributes(t.commonAttrs...)
  321. if error {
  322. span.SetStatus(codes.Error, "")
  323. }
  324. t.setContext(ctx)
  325. t.setSpan(span)
  326. }
  327. func (t *Trace) SendEvent() {
  328. t.span.End()
  329. }
  330. func (t *Trace) GetSpan() trace.Span {
  331. return t.span
  332. }
  333. func (t *Trace) createTraceEvent(protocol l7.Protocol, eventType int, attrs ...attribute.KeyValue) {
  334. t.span.AddEventApm(protocol.String(), eventType, protocol.Int(), trace.WithAttributes(attrs...))
  335. //atomic.AddUint32(t.currenEventCount, 1)
  336. }
  337. func (t *Trace) createFuncTraceEvent(name string, eventType int, attrs ...attribute.KeyValue) {
  338. t.span.AddEventApm(name, eventType, 0, trace.WithAttributes(attrs...))
  339. }
  340. func (t *Trace) addEvent() {
  341. atomic.AddUint32(t.currenEventCount, 1)
  342. }
  343. func (t *Trace) createTraceSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  344. end := time.Now()
  345. start := end.Add(-duration)
  346. //fmt.Println("createTraceSpan:", t.ctx)
  347. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  348. span.SetAttributes(t.commonAttrs...)
  349. span.SetAttributes(attrs...)
  350. if error {
  351. span.SetStatus(codes.Error, "")
  352. }
  353. span.End(trace.WithTimestamp(end))
  354. }
  355. func (t *Trace) MysqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  356. if t == nil || query == "" {
  357. return
  358. }
  359. t.createTraceSpan(l7.ProtocolMysql.String(), duration, error,
  360. semconv.DBSystemMySQL,
  361. semconv.DBStatement(query),
  362. semconv.NetPeerName(destination.IP().String()),
  363. semconv.NetPeerPort(int(destination.Port())),
  364. )
  365. }
  366. // 内置允许操作(始终放行)
  367. var builtinOps = map[string]struct{}{
  368. "SELECT": {},
  369. "INSERT": {},
  370. "UPDATE": {},
  371. "DELETE": {},
  372. "WITH": {},
  373. }
  374. var (
  375. allowedOps map[string]struct{}
  376. initOnce sync.Once
  377. )
  378. // 解析并缓存环境变量中的前缀
  379. func initAllowedOps() {
  380. allowedOps = make(map[string]struct{})
  381. env := os.Getenv("ALLOWED_SQL_PREFIXES")
  382. for _, s := range strings.Split(env, ",") {
  383. if trimmed := strings.ToUpper(strings.TrimSpace(s)); trimmed != "" {
  384. allowedOps[trimmed] = struct{}{}
  385. }
  386. }
  387. }
  388. func isCURDOperation(q string) bool {
  389. if len(q) < 6 {
  390. return false
  391. }
  392. // 提取前缀(考虑最长10个字符)
  393. prefix := strings.ToUpper(strings.TrimSpace(q))
  394. if len(prefix) > 10 {
  395. prefix = prefix[:10]
  396. }
  397. // 提取第一个单词作为操作关键字
  398. spaceIdx := strings.IndexAny(prefix, " \t\n")
  399. if spaceIdx > 0 {
  400. prefix = prefix[:spaceIdx]
  401. }
  402. // 先查固定允许的操作
  403. if _, ok := builtinOps[prefix]; ok {
  404. return true
  405. } else {
  406. // 尝试清理所有非字母字符(移除控制字符、BOM、数字、标点等):
  407. //`SELECT` 长度 6,字节:[83 69 76 69 67 84]
  408. //prefix中的值长度 8,字节:[0 1 83 69 76 69 67 84]
  409. prefixTmp := strings.Map(func(r rune) rune {
  410. // 只保留大写字母 A-Z
  411. if r >= 'A' && r <= 'Z' {
  412. return r
  413. }
  414. return -1 // 删除其他所有字符
  415. }, prefix)
  416. if _, ok := builtinOps[prefixTmp]; ok {
  417. return true
  418. }
  419. }
  420. // 加载环境变量控制的操作(只执行一次)
  421. initOnce.Do(initAllowedOps)
  422. // 查额外允许操作
  423. _, ok := allowedOps[prefix]
  424. return ok
  425. }
  426. func (t *Trace) SQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, query string, r *l7.RequestData, destination netaddr.IPPort) {
  427. //fmt.Println("query", query)
  428. if t == nil {
  429. return
  430. }
  431. t.addEvent()
  432. // 只保留增删改查基本操作
  433. if !isCURDOperation(query) {
  434. return
  435. }
  436. // 拼接 destination 的 IP 和端口为 ip:port 格式
  437. destAddr := fmt.Sprintf("%s:%d", destination.IP().String(), destination.Port())
  438. var attr []attribute.KeyValue
  439. attr = append(attr,
  440. semconvVal,
  441. semconv.DBStatement(query),
  442. semconv.NetPeerName(destination.IP().String()),
  443. semconv.NetPeerPort(int(destination.Port())),
  444. attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  445. attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  446. attribute.String("sql.dbn", destAddr),
  447. )
  448. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  449. attr = append(attr,
  450. attribute.Bool("sql.exception", r.Status.Error()),
  451. )
  452. if r.Status.Error() {
  453. attr = append(attr,
  454. attribute.String("sql.exception_msg", r.ErrorMsg),
  455. )
  456. }
  457. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  458. }
  459. //func (t *Trace) MysqlTraceQueryEvent(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.DBSystemMySQL,
  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. //
  476. // attr = append(attr,
  477. // attribute.Bool("sql.exception", r.Status.Error()),
  478. // )
  479. // t.createTraceEvent(l7.ProtocolMysql.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolMysql), attr...)
  480. //}
  481. //func (t *Trace) PostGreSqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  482. // if t == nil || query == "" {
  483. // return
  484. // }
  485. // t.createTraceSpan(l7.ProtocolPostgres.String(), duration, error,
  486. // semconv.DBSystemPostgreSQL,
  487. // semconv.DBStatement(query),
  488. // semconv.NetPeerName(destination.IP().String()),
  489. // semconv.NetPeerPort(int(destination.Port())),
  490. // )
  491. //}
  492. //func (t *Trace) PostGreSqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  493. // fmt.Println("query", query)
  494. // if t == nil {
  495. // return
  496. // }
  497. //
  498. // var attr []attribute.KeyValue
  499. // attr = append(attr,
  500. // semconv.DBSystemPostgreSQL,
  501. // semconv.DBStatement(query),
  502. // semconv.NetPeerName(destination.IP().String()),
  503. // semconv.NetPeerPort(int(destination.Port())),
  504. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  505. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  506. // )
  507. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  508. // attr = append(attr,
  509. // attribute.Bool("sql.exception", r.Status.Error()),
  510. // )
  511. // t.createTraceEvent(l7.ProtocolPostgres.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolPostgres), attr...)
  512. //}
  513. //func (t *Trace) DmTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  514. // if t == nil || query == "" {
  515. // return
  516. // }
  517. // l7Type := int(l7.ProtocolDM)
  518. // var attr []attribute.KeyValue
  519. // attr = append(attr,
  520. // semconv.DBSystemDaMengDB,
  521. // semconv.DBStatement(query),
  522. // semconv.NetPeerName(destination.IP().String()),
  523. // semconv.NetPeerPort(int(destination.Port())),
  524. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  525. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  526. // )
  527. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  528. //
  529. // attr = append(attr,
  530. // attribute.Bool("sql.exception", r.Status.Error()),
  531. // )
  532. // t.createTraceEvent(l7.ProtocolHTTP.String(), int(ebpftracer.EventTypeL7Request), l7Type, attr...)
  533. //}
  534. //func (t *Trace) RedisTraceQuery(cmd, args string, error bool, duration time.Duration) {
  535. // if t == nil || cmd == "" {
  536. // return
  537. // }
  538. // statement := cmd
  539. // if args != "" {
  540. // statement += " " + args
  541. // }
  542. // t.createTraceSpan(l7.ProtocolRedis.String(), duration, error,
  543. // semconv.DBSystemRedis,
  544. // semconv.DBOperation(cmd),
  545. // semconv.DBStatement(statement),
  546. // )
  547. //}
  548. func (t *Trace) NoSQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, operation, statement string, r *l7.RequestData, connSrc, connDest netaddr.IPPort) {
  549. if t == nil {
  550. return
  551. }
  552. t.addEvent()
  553. if statement == "" {
  554. return
  555. }
  556. // 拼接 destination 的 IP 和端口为 ip:port 格式
  557. srcAddr := r.ComponentSAddr.String()
  558. if r.ComponentSAddr.Port() == 0 {
  559. srcAddr = connSrc.String()
  560. }
  561. destinationAddr := r.ComponentDAddr.String()
  562. if r.ComponentDAddr.Port() == 0 {
  563. destinationAddr = connDest.String()
  564. }
  565. var attr []attribute.KeyValue
  566. attr = append(attr,
  567. semconvVal,
  568. semconv.DBStatement(statement),
  569. semconv.NetPeerName(connDest.IP().String()),
  570. semconv.NetPeerPort(int(connDest.Port())),
  571. attribute.String("nosql.src_addr", srcAddr),
  572. attribute.String("nosql.destination_addr", destinationAddr),
  573. attribute.String("nosql.dbn", connDest.String()),
  574. )
  575. // 如果 operation 不为空,添加 DBOperation 属性
  576. if operation != "" {
  577. attr = append(attr, semconv.DBOperation(operation))
  578. }
  579. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  580. attr = append(attr,
  581. attribute.Bool("nosql.exception", r.Status.Error()),
  582. )
  583. if r.Status.Error() {
  584. attr = append(attr,
  585. attribute.String("nosql.exception_msg", r.ErrorMsg),
  586. )
  587. }
  588. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  589. }
  590. func (t *Trace) MQTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, operation, statement string, r *l7.RequestData, connSrc, connDest netaddr.IPPort) {
  591. if t == nil {
  592. return
  593. }
  594. t.addEvent()
  595. if r.MQTopic != "" || r.MQKey != "" {
  596. statement = fmt.Sprintf("PRODUCE topic:%s key:%s", r.MQTopic, r.MQKey)
  597. }
  598. if statement == "" {
  599. return
  600. }
  601. srcAddr := r.ComponentSAddr.String()
  602. if r.ComponentSAddr.Port() == 0 {
  603. srcAddr = connSrc.String()
  604. }
  605. // 尝试解析目标地址,如果失败则使用 connDest 作为备用
  606. dAddr, err := net.ResolveTCPAddr("tcp", r.DestAddrString)
  607. var ip string
  608. port := 0
  609. if err != nil {
  610. klog.Errorln(err)
  611. } else {
  612. ip = dAddr.IP.String()
  613. port = dAddr.Port
  614. }
  615. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  616. if err != nil {
  617. assumedAppID = 0
  618. }
  619. var attr []attribute.KeyValue
  620. attr = append(attr,
  621. semconvVal,
  622. semconv.NetPeerName(ip),
  623. semconv.NetPeerPort(port),
  624. attribute.String("mq.info", statement),
  625. attribute.String("mq.src_addr", srcAddr),
  626. attribute.String("mq.destination_addr", r.DestAddrString),
  627. attribute.String("mq.topic", r.MQTopic),
  628. attribute.String("mq.key", r.MQKey),
  629. attribute.Int64("mq.assumed_app_id", assumedAppID),
  630. attribute.String("mq.span_id", r.SpanId),
  631. attribute.String("mq.oper_type", "PRODUCER"),
  632. )
  633. // 如果 operation 不为空,添加 DBOperation 属性
  634. if operation != "" {
  635. attr = append(attr, semconv.DBOperation(operation))
  636. }
  637. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  638. attr = append(attr,
  639. attribute.Bool("mq.exception", r.Status.Error()),
  640. )
  641. if r.Status.Error() {
  642. attr = append(attr,
  643. attribute.String("mq.exception_msg", r.ErrorMsg),
  644. )
  645. }
  646. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  647. }
  648. func (t *Trace) GrpcServerTraceQueryEvent(r *l7.RequestData, appInfo AppInfo) {
  649. if t == nil {
  650. return
  651. }
  652. t.addEvent()
  653. // if method == "" {
  654. // return
  655. // }
  656. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  657. if err != nil {
  658. assumedAppID = 0
  659. }
  660. var attr []attribute.KeyValue
  661. attr = append(attr,
  662. semconv.RPCSystemGRPC,
  663. attribute.String("rpc.uri", string(r.Payload)),
  664. attribute.String("rpc.schema", "grpc"),
  665. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  666. attribute.String("rpc.span_id", r.SpanId),
  667. attribute.Int("rpc.port", appInfo.Sport),
  668. attribute.String("rpc.ip", appInfo.Sn),
  669. attribute.String("rpc.oper_type", "PROVIDER"),
  670. attribute.String("rpc.method", "test_method"),
  671. )
  672. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  673. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  674. }
  675. func (t *Trace) GrpcClientTraceQueryEvent(r *l7.RequestData) {
  676. if t == nil {
  677. return
  678. }
  679. t.addEvent()
  680. // if method == "" {
  681. // return
  682. // }
  683. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  684. if err != nil {
  685. assumedAppID = 0
  686. }
  687. // 解析 RPCTarget 字符串,格式为 "ip:port"
  688. klog.Infof("r.RPCTarget is %s", r.DestAddrString)
  689. var rpcIP, rpcPort string
  690. if r.DestAddrString != "" {
  691. // 清理空字节,只保留有效字符
  692. cleanTarget := strings.TrimRight(r.DestAddrString, "\x00")
  693. cleanTarget = strings.TrimSpace(cleanTarget)
  694. klog.Infof("cleanTarget is %s", cleanTarget)
  695. if colonIndex := strings.LastIndex(cleanTarget, ":"); colonIndex != -1 {
  696. rpcIP = cleanTarget[:colonIndex]
  697. rpcPort = cleanTarget[colonIndex+1:]
  698. klog.Infof("rpcIP is %s, rpcPort is %s", rpcIP, rpcPort)
  699. } else {
  700. // 如果没有找到冒号,可能是只有IP或只有端口
  701. rpcIP = cleanTarget
  702. }
  703. }
  704. // 设置默认值
  705. if rpcIP == "" {
  706. rpcIP = "127.0.0.1"
  707. }
  708. if rpcPort == "" {
  709. rpcPort = "80"
  710. }
  711. // 转换端口为整数
  712. portInt, err := strconv.Atoi(rpcPort)
  713. if err != nil {
  714. klog.Infof("r.RPCTarget convert port err is %s", err)
  715. portInt = 80 // 默认端口
  716. } else {
  717. klog.Infof("r.RPCTarget convert port success, portInt is %d", portInt)
  718. }
  719. var attr []attribute.KeyValue
  720. attr = append(attr,
  721. semconv.RPCSystemGRPC,
  722. attribute.String("rpc.uri", string(r.Payload)),
  723. attribute.String("rpc.schema", "grpc"),
  724. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  725. attribute.String("rpc.span_id", r.SpanId),
  726. attribute.Int("rpc.port", portInt),
  727. attribute.String("rpc.ip", rpcIP),
  728. attribute.String("rpc.oper_type", "CONSUMER"),
  729. attribute.String("rpc.method", string(r.Payload)),
  730. )
  731. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  732. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  733. }
  734. func (t *Trace) DNSTraceQueryEvent(r *l7.RequestData, _type string, fqdn string, ttl uint32, ips []netaddr.IP) {
  735. if t == nil {
  736. return
  737. }
  738. t.addEvent()
  739. var attr []attribute.KeyValue
  740. var ipList []string
  741. for _, ip := range ips {
  742. ipList = append(ipList, ip.String())
  743. }
  744. ip := strings.Join(ipList, ",")
  745. attr = append(attr,
  746. attribute.String("dns.type", _type),
  747. attribute.String("dns.fqdn", fqdn),
  748. attribute.Int("dns.ttl", int(ttl)),
  749. attribute.String("dns.ips", ip),
  750. )
  751. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  752. t.createTraceEvent(l7.ProtocolDNS, ebpftracer.EventTypeL7Request.Int(), attr...)
  753. }
  754. func (t *Trace) HttpTraceRequest(method, path, ip string, port uint16, r *l7.RequestData) {
  755. if t == nil || method == "" {
  756. return
  757. }
  758. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  759. if err != nil {
  760. assumedAppID = 0
  761. }
  762. status := r.Status
  763. duration := r.Duration
  764. t.createTraceSpan(l7.ProtocolHTTP.String(), duration, status >= 400,
  765. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  766. semconv.HTTPMethod(method),
  767. semconv.HTTPStatusCode(int(status)),
  768. attribute.String("http.uri", path),
  769. attribute.String("http.ip", ip),
  770. attribute.Int64("http.assumed_app_id", assumedAppID),
  771. attribute.String("http.span_id", r.SpanId),
  772. attribute.Int("http.port", int(port)),
  773. )
  774. }
  775. // 新增事件处理
  776. func (t *Trace) HttpTraceRequestEvent(method, path, ip string, port uint16, r *l7.RequestData) {
  777. if t == nil {
  778. return
  779. }
  780. t.addEvent()
  781. if method == "" {
  782. return
  783. }
  784. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  785. if err != nil {
  786. assumedAppID = 0
  787. }
  788. status := r.Status
  789. var attr []attribute.KeyValue
  790. attr = append(attr,
  791. semconv.HTTPURL(fmt.Sprintf("http://%s:%d%s", ip, port, path)),
  792. semconv.HTTPMethod(method),
  793. semconv.HTTPStatusCode(int(status)),
  794. attribute.Bool("http.status_error", status > 399),
  795. attribute.String("http.uri", path),
  796. attribute.String("http.ip", ip),
  797. attribute.Int64("http.assumed_app_id", assumedAppID),
  798. attribute.String("http.span_id", r.SpanId),
  799. attribute.Int("http.port", int(port)),
  800. attribute.String("http.src_addr", r.ComponentSAddr.String()),
  801. attribute.String("http.destination_addr", r.ComponentDAddr.String()),
  802. attribute.Bool("http.is_tls", r.IsTls),
  803. )
  804. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  805. t.createTraceEvent(l7.ProtocolHTTP, ebpftracer.EventTypeL7Request.Int(), attr...)
  806. }
  807. func (t *Trace) FuncTraceQuery(funcname string, duration time.Duration, start uint64, end uint64) {
  808. if t == nil {
  809. return
  810. }
  811. t.addEvent()
  812. if funcname == "" {
  813. return
  814. }
  815. t.createTraceSpanNoTime2(funcname, duration, false, start, end)
  816. //t.createTraceSpanNoTime2(funcname, duration, false, start, end, attribute.Int("num", num))
  817. }
  818. func (t *Trace) createTraceSpanNoTime(name string, duration time.Duration, error bool, start uint64, end uint64, attrs ...attribute.KeyValue) {
  819. // end := time.Now()
  820. // start := end.Add(-duration)
  821. startTime := time.Unix(0, int64(start))
  822. endTime := time.Unix(0, int64(end))
  823. //fmt.Println("createTraceSpan:", t.ctx)
  824. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  825. span.SetAttributes(t.commonAttrs...)
  826. span.SetAttributes(attrs...)
  827. if error {
  828. span.SetStatus(codes.Error, "")
  829. }
  830. span.End(trace.WithTimestamp(endTime))
  831. }
  832. func (t *Trace) createTraceSpanNoTime2(name string, duration time.Duration, error bool, start uint64, end uint64) {
  833. // end := time.Now()
  834. // start := end.Add(-duration)
  835. //startTime := time.Unix(0, int64(start))
  836. //endTime := time.Unix(0, int64(end))
  837. ////fmt.Println("createTraceSpan:", t.ctx)
  838. //_, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  839. //span.SetAttributes(t.commonAttrs...)
  840. //span.SetAttributes(attrs...)
  841. //if error {
  842. // span.SetStatus(codes.Error, "")
  843. //}
  844. //span.End(trace.WithTimestamp(endTime))
  845. //attrs = append([]attribute.KeyValue{
  846. // attribute.Int64("startAt", int64(start)),
  847. // attribute.Int64("endAt", int64(end)),
  848. //})
  849. var attr []attribute.KeyValue
  850. t.appendTimestamp(&attr, start, end, int64(end-start))
  851. t.createFuncTraceEvent(name, ebpftracer.EventTypeFunEnt.Int(), attr...)
  852. }