apm_tracing.go 29 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939
  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. //
  221. // CalledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  222. // if err == nil && CalledId != 0 {
  223. // t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  224. // }
  225. //
  226. // InstanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  227. // if err == nil && InstanceIdFrom != 0 {
  228. // t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  229. // }
  230. //
  231. // AppIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  232. // if err == nil && AppIdFrom != 0 {
  233. // t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  234. // }
  235. // if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  236. // t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  237. // }
  238. // // klog.Debugf("r.ParentSpanContext.TypeFrom tpp %s", r.ParentSpanContext.TypeFrom)
  239. // if r.ParentSpanContext.TypeFrom == "10" {
  240. // t.span.SetAttributes(attribute.String("server.type_from", "SDK"))
  241. // t.span.SetAttributes(attribute.String("server.app_id_from", "-1"))
  242. // t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  243. // t.span.SetAttributes(attribute.String("server.instance_id_from", "-1"))
  244. // t.span.SetAttributes(attribute.String("server.call_id", "-1"))
  245. // }
  246. // // for _, v := range t.stack {
  247. // // fmt.Printf("TraceEndTraceEndTraceEnd%s\n", v)
  248. // // }
  249. // //t.buildFun()
  250. // t.span.End(trace.WithTimestamp(time.Now()))
  251. //}
  252. // 新增结束事件
  253. func (t *Trace) TraceEndEvent(r *l7.RequestData) {
  254. if t == nil {
  255. return
  256. }
  257. var attr []attribute.KeyValue
  258. attr = append(attr,
  259. semconv.HTTPStatusCode(int(r.Status)),
  260. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  261. )
  262. //t.span.SetAttributes(
  263. // semconv.HTTPStatusCode(int(r.Status)),
  264. // attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  265. //)
  266. calledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  267. if err == nil && calledId != 0 {
  268. attr = append(attr, attribute.Int64("server.called_id", calledId))
  269. //t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  270. }
  271. instanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  272. if err == nil && instanceIdFrom != 0 {
  273. attr = append(attr, attribute.Int64("server.instance_id_from", instanceIdFrom))
  274. //t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  275. }
  276. appIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  277. if err == nil && appIdFrom != 0 {
  278. attr = append(attr, attribute.Int64("server.app_id_from", appIdFrom))
  279. // Add type_from attribute for all cases
  280. attr = append(attr, attribute.String("server.type_from", r.ParentSpanContext.TypeFrom.String()))
  281. }
  282. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  283. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  284. //t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  285. }
  286. // klog.Debugf("r.ParentSpanContext.TypeFrom event tpp %s", r.ParentSpanContext.TypeFrom)
  287. if r.ParentSpanContext.TypeFrom == "10" {
  288. attr = append(attr, attribute.String("server.app_id_from", "-1"))
  289. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  290. attr = append(attr, attribute.String("server.instance_id_from", "-1"))
  291. attr = append(attr, attribute.String("server.call_id", "-1"))
  292. }
  293. attr = append(attr, attribute.String("server.src_addr", r.DAddr.String()))
  294. attr = append(attr, attribute.String("server.dst_addr", r.SAddr.String()))
  295. klog.Debugf("TraceEndEvent SAddr.String %s", r.SAddr.String())
  296. klog.Debugf("TraceEndEvent DAddr.String %s", r.DAddr.String())
  297. // 添加 SysvcFrom 到 attributes
  298. if r.SysvcFrom != "" {
  299. attr = append(attr, attribute.String("server.sysvc_from", r.SysvcFrom))
  300. }
  301. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  302. t.span.SetAttributes(attr...)
  303. t.endReadyEvent(r.EventCount)
  304. }
  305. func (t *Trace) appendTimestamp(attr *[]attribute.KeyValue, s, e uint64, d int64) {
  306. *attr = append(*attr,
  307. attribute.Int64("time.start_at", utils.KtimeToTimestamp(s)),
  308. attribute.Int64("time.end_at", utils.KtimeToTimestamp(e)),
  309. attribute.Int64("time.duration", d),
  310. )
  311. }
  312. func (t *Trace) createParentSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  313. end := time.Now()
  314. start := end.Add(-duration)
  315. ctx, span := tracer(t.containerId).Start(context.Background(), name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  316. span.SetAttributes(attrs...)
  317. span.SetAttributes(t.commonAttrs...)
  318. if error {
  319. span.SetStatus(codes.Error, "")
  320. }
  321. t.setContext(ctx)
  322. t.setSpan(span)
  323. }
  324. func (t *Trace) SendEvent() {
  325. t.span.End()
  326. }
  327. func (t *Trace) GetSpan() trace.Span {
  328. return t.span
  329. }
  330. func (t *Trace) createTraceEvent(protocol l7.Protocol, eventType int, attrs ...attribute.KeyValue) {
  331. t.span.AddEventApm(protocol.String(), eventType, protocol.Int(), trace.WithAttributes(attrs...))
  332. //atomic.AddUint32(t.currenEventCount, 1)
  333. }
  334. func (t *Trace) createFuncTraceEvent(name string, eventType int, attrs ...attribute.KeyValue) {
  335. t.span.AddEventApm(name, eventType, 0, trace.WithAttributes(attrs...))
  336. }
  337. func (t *Trace) addEvent() {
  338. atomic.AddUint32(t.currenEventCount, 1)
  339. }
  340. func (t *Trace) createTraceSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  341. end := time.Now()
  342. start := end.Add(-duration)
  343. //fmt.Println("createTraceSpan:", t.ctx)
  344. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  345. span.SetAttributes(t.commonAttrs...)
  346. span.SetAttributes(attrs...)
  347. if error {
  348. span.SetStatus(codes.Error, "")
  349. }
  350. span.End(trace.WithTimestamp(end))
  351. }
  352. func (t *Trace) MysqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  353. if t == nil || query == "" {
  354. return
  355. }
  356. t.createTraceSpan(l7.ProtocolMysql.String(), duration, error,
  357. semconv.DBSystemMySQL,
  358. semconv.DBStatement(query),
  359. semconv.NetPeerName(destination.IP().String()),
  360. semconv.NetPeerPort(int(destination.Port())),
  361. )
  362. }
  363. // 内置允许操作(始终放行)
  364. var builtinOps = map[string]struct{}{
  365. "SELECT": {},
  366. "INSERT": {},
  367. "UPDATE": {},
  368. "DELETE": {},
  369. "WITH": {},
  370. }
  371. var (
  372. allowedOps map[string]struct{}
  373. initOnce sync.Once
  374. )
  375. // 解析并缓存环境变量中的前缀
  376. func initAllowedOps() {
  377. allowedOps = make(map[string]struct{})
  378. env := os.Getenv("ALLOWED_SQL_PREFIXES")
  379. for _, s := range strings.Split(env, ",") {
  380. if trimmed := strings.ToUpper(strings.TrimSpace(s)); trimmed != "" {
  381. allowedOps[trimmed] = struct{}{}
  382. }
  383. }
  384. }
  385. func isCURDOperation(q string) bool {
  386. if len(q) < 6 {
  387. return false
  388. }
  389. // 提取前缀(考虑最长10个字符)
  390. prefix := strings.ToUpper(strings.TrimSpace(q))
  391. if len(prefix) > 10 {
  392. prefix = prefix[:10]
  393. }
  394. // 提取第一个单词作为操作关键字
  395. spaceIdx := strings.IndexAny(prefix, " \t\n")
  396. if spaceIdx > 0 {
  397. prefix = prefix[:spaceIdx]
  398. }
  399. // 先查固定允许的操作
  400. if _, ok := builtinOps[prefix]; ok {
  401. return true
  402. } else {
  403. // 尝试清理所有非字母字符(移除控制字符、BOM、数字、标点等):
  404. //`SELECT` 长度 6,字节:[83 69 76 69 67 84]
  405. //prefix中的值长度 8,字节:[0 1 83 69 76 69 67 84]
  406. prefixTmp := strings.Map(func(r rune) rune {
  407. // 只保留大写字母 A-Z
  408. if r >= 'A' && r <= 'Z' {
  409. return r
  410. }
  411. return -1 // 删除其他所有字符
  412. }, prefix)
  413. if _, ok := builtinOps[prefixTmp]; ok {
  414. return true
  415. }
  416. }
  417. // 加载环境变量控制的操作(只执行一次)
  418. initOnce.Do(initAllowedOps)
  419. // 查额外允许操作
  420. _, ok := allowedOps[prefix]
  421. return ok
  422. }
  423. func (t *Trace) SQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, query string, r *l7.RequestData, destination netaddr.IPPort) {
  424. //fmt.Println("query", query)
  425. if t == nil {
  426. return
  427. }
  428. t.addEvent()
  429. // 只保留增删改查基本操作
  430. if !isCURDOperation(query) {
  431. return
  432. }
  433. // 拼接 destination 的 IP 和端口为 ip:port 格式
  434. destAddr := fmt.Sprintf("%s:%d", destination.IP().String(), destination.Port())
  435. var attr []attribute.KeyValue
  436. attr = append(attr,
  437. semconvVal,
  438. semconv.DBStatement(query),
  439. semconv.NetPeerName(destination.IP().String()),
  440. semconv.NetPeerPort(int(destination.Port())),
  441. attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  442. attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  443. attribute.String("sql.dbn", destAddr),
  444. )
  445. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  446. attr = append(attr,
  447. attribute.Bool("sql.exception", r.Status.Error()),
  448. )
  449. if r.Status.Error() {
  450. attr = append(attr,
  451. attribute.String("sql.exception_msg", r.ErrorMsg),
  452. )
  453. }
  454. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  455. }
  456. //func (t *Trace) MysqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  457. // fmt.Println("query", query)
  458. // if t == nil {
  459. // return
  460. // }
  461. //
  462. // var attr []attribute.KeyValue
  463. // attr = append(attr,
  464. // semconv.DBSystemMySQL,
  465. // semconv.DBStatement(query),
  466. // semconv.NetPeerName(destination.IP().String()),
  467. // semconv.NetPeerPort(int(destination.Port())),
  468. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  469. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  470. // )
  471. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  472. //
  473. // attr = append(attr,
  474. // attribute.Bool("sql.exception", r.Status.Error()),
  475. // )
  476. // t.createTraceEvent(l7.ProtocolMysql.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolMysql), attr...)
  477. //}
  478. //func (t *Trace) PostGreSqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  479. // if t == nil || query == "" {
  480. // return
  481. // }
  482. // t.createTraceSpan(l7.ProtocolPostgres.String(), duration, error,
  483. // semconv.DBSystemPostgreSQL,
  484. // semconv.DBStatement(query),
  485. // semconv.NetPeerName(destination.IP().String()),
  486. // semconv.NetPeerPort(int(destination.Port())),
  487. // )
  488. //}
  489. //func (t *Trace) PostGreSqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  490. // fmt.Println("query", query)
  491. // if t == nil {
  492. // return
  493. // }
  494. //
  495. // var attr []attribute.KeyValue
  496. // attr = append(attr,
  497. // semconv.DBSystemPostgreSQL,
  498. // semconv.DBStatement(query),
  499. // semconv.NetPeerName(destination.IP().String()),
  500. // semconv.NetPeerPort(int(destination.Port())),
  501. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  502. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  503. // )
  504. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  505. // attr = append(attr,
  506. // attribute.Bool("sql.exception", r.Status.Error()),
  507. // )
  508. // t.createTraceEvent(l7.ProtocolPostgres.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolPostgres), attr...)
  509. //}
  510. //func (t *Trace) DmTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  511. // if t == nil || query == "" {
  512. // return
  513. // }
  514. // l7Type := int(l7.ProtocolDM)
  515. // var attr []attribute.KeyValue
  516. // attr = append(attr,
  517. // semconv.DBSystemDaMengDB,
  518. // semconv.DBStatement(query),
  519. // semconv.NetPeerName(destination.IP().String()),
  520. // semconv.NetPeerPort(int(destination.Port())),
  521. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  522. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  523. // )
  524. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  525. //
  526. // attr = append(attr,
  527. // attribute.Bool("sql.exception", r.Status.Error()),
  528. // )
  529. // t.createTraceEvent(l7.ProtocolHTTP.String(), int(ebpftracer.EventTypeL7Request), l7Type, attr...)
  530. //}
  531. //func (t *Trace) RedisTraceQuery(cmd, args string, error bool, duration time.Duration) {
  532. // if t == nil || cmd == "" {
  533. // return
  534. // }
  535. // statement := cmd
  536. // if args != "" {
  537. // statement += " " + args
  538. // }
  539. // t.createTraceSpan(l7.ProtocolRedis.String(), duration, error,
  540. // semconv.DBSystemRedis,
  541. // semconv.DBOperation(cmd),
  542. // semconv.DBStatement(statement),
  543. // )
  544. //}
  545. func (t *Trace) NoSQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, operation, statement string, r *l7.RequestData, connSrc, connDest netaddr.IPPort) {
  546. if t == nil {
  547. return
  548. }
  549. t.addEvent()
  550. if statement == "" {
  551. return
  552. }
  553. // 拼接 destination 的 IP 和端口为 ip:port 格式
  554. srcAddr := r.ComponentSAddr.String()
  555. if r.ComponentSAddr.Port() == 0 {
  556. srcAddr = connSrc.String()
  557. }
  558. destinationAddr := r.ComponentDAddr.String()
  559. if r.ComponentDAddr.Port() == 0 {
  560. destinationAddr = connDest.String()
  561. }
  562. var attr []attribute.KeyValue
  563. attr = append(attr,
  564. semconvVal,
  565. semconv.DBStatement(statement),
  566. semconv.NetPeerName(connDest.IP().String()),
  567. semconv.NetPeerPort(int(connDest.Port())),
  568. attribute.String("nosql.src_addr", srcAddr),
  569. attribute.String("nosql.destination_addr", destinationAddr),
  570. attribute.String("nosql.dbn", connDest.String()),
  571. )
  572. // 如果 operation 不为空,添加 DBOperation 属性
  573. if operation != "" {
  574. attr = append(attr, semconv.DBOperation(operation))
  575. }
  576. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  577. attr = append(attr,
  578. attribute.Bool("nosql.exception", r.Status.Error()),
  579. )
  580. if r.Status.Error() {
  581. attr = append(attr,
  582. attribute.String("nosql.exception_msg", r.ErrorMsg),
  583. )
  584. }
  585. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  586. }
  587. func (t *Trace) MQTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, operation, statement string, r *l7.RequestData, connSrc, connDest netaddr.IPPort) {
  588. if t == nil {
  589. return
  590. }
  591. t.addEvent()
  592. if r.MQTopic != "" || r.MQKey != "" {
  593. statement = fmt.Sprintf("PRODUCE topic:%s key:%s", r.MQTopic, r.MQKey)
  594. }
  595. if statement == "" {
  596. return
  597. }
  598. srcAddr := r.ComponentSAddr.String()
  599. if r.ComponentSAddr.Port() == 0 {
  600. srcAddr = connSrc.String()
  601. }
  602. // 尝试解析目标地址,如果失败则使用 connDest 作为备用
  603. dAddr, err := net.ResolveTCPAddr("tcp", r.DestAddrString)
  604. var ip string
  605. port := 0
  606. if err != nil {
  607. klog.Errorln(err)
  608. } else {
  609. ip = dAddr.IP.String()
  610. port = dAddr.Port
  611. }
  612. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  613. if err != nil {
  614. assumedAppID = 0
  615. }
  616. var attr []attribute.KeyValue
  617. attr = append(attr,
  618. semconvVal,
  619. semconv.NetPeerName(ip),
  620. semconv.NetPeerPort(port),
  621. attribute.String("mq.info", statement),
  622. attribute.String("mq.src_addr", srcAddr),
  623. attribute.String("mq.destination_addr", r.DestAddrString),
  624. attribute.String("mq.topic", r.MQTopic),
  625. attribute.String("mq.key", r.MQKey),
  626. attribute.Int64("mq.assumed_app_id", assumedAppID),
  627. attribute.String("mq.span_id", r.SpanId),
  628. attribute.String("mq.oper_type", "PRODUCER"),
  629. )
  630. // 如果 operation 不为空,添加 DBOperation 属性
  631. if operation != "" {
  632. attr = append(attr, semconv.DBOperation(operation))
  633. }
  634. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  635. attr = append(attr,
  636. attribute.Bool("mq.exception", r.Status.Error()),
  637. )
  638. if r.Status.Error() {
  639. attr = append(attr,
  640. attribute.String("mq.exception_msg", r.ErrorMsg),
  641. )
  642. }
  643. t.createTraceEvent(l7Type, ebpftracer.EventTypeL7Request.Int(), attr...)
  644. }
  645. func (t *Trace) GrpcServerTraceQueryEvent(r *l7.RequestData, appInfo AppInfo) {
  646. if t == nil {
  647. return
  648. }
  649. t.addEvent()
  650. // if method == "" {
  651. // return
  652. // }
  653. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  654. if err != nil {
  655. assumedAppID = 0
  656. }
  657. var attr []attribute.KeyValue
  658. attr = append(attr,
  659. semconv.RPCSystemGRPC,
  660. attribute.String("rpc.uri", string(r.Payload)),
  661. attribute.String("rpc.schema", "grpc"),
  662. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  663. attribute.String("rpc.span_id", r.SpanId),
  664. attribute.Int("rpc.port", appInfo.Sport),
  665. attribute.String("rpc.ip", appInfo.Sn),
  666. attribute.String("rpc.oper_type", "PROVIDER"),
  667. attribute.String("rpc.method", "test_method"),
  668. )
  669. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  670. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  671. }
  672. func (t *Trace) GrpcClientTraceQueryEvent(r *l7.RequestData) {
  673. if t == nil {
  674. return
  675. }
  676. t.addEvent()
  677. // if method == "" {
  678. // return
  679. // }
  680. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  681. if err != nil {
  682. assumedAppID = 0
  683. }
  684. // 解析 RPCTarget 字符串,格式为 "ip:port"
  685. klog.Infof("r.RPCTarget is %s", r.DestAddrString)
  686. var rpcIP, rpcPort string
  687. if r.DestAddrString != "" {
  688. // 清理空字节,只保留有效字符
  689. cleanTarget := strings.TrimRight(r.DestAddrString, "\x00")
  690. cleanTarget = strings.TrimSpace(cleanTarget)
  691. klog.Infof("cleanTarget is %s", cleanTarget)
  692. if colonIndex := strings.LastIndex(cleanTarget, ":"); colonIndex != -1 {
  693. rpcIP = cleanTarget[:colonIndex]
  694. rpcPort = cleanTarget[colonIndex+1:]
  695. klog.Infof("rpcIP is %s, rpcPort is %s", rpcIP, rpcPort)
  696. } else {
  697. // 如果没有找到冒号,可能是只有IP或只有端口
  698. rpcIP = cleanTarget
  699. }
  700. }
  701. // 设置默认值
  702. if rpcIP == "" {
  703. rpcIP = "127.0.0.1"
  704. }
  705. if rpcPort == "" {
  706. rpcPort = "80"
  707. }
  708. // 转换端口为整数
  709. portInt, err := strconv.Atoi(rpcPort)
  710. if err != nil {
  711. klog.Infof("r.RPCTarget convert port err is %s", err)
  712. portInt = 80 // 默认端口
  713. } else {
  714. klog.Infof("r.RPCTarget convert port success, portInt is %d", portInt)
  715. }
  716. var attr []attribute.KeyValue
  717. attr = append(attr,
  718. semconv.RPCSystemGRPC,
  719. attribute.String("rpc.uri", string(r.Payload)),
  720. attribute.String("rpc.schema", "grpc"),
  721. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  722. attribute.String("rpc.span_id", r.SpanId),
  723. attribute.Int("rpc.port", portInt),
  724. attribute.String("rpc.ip", rpcIP),
  725. attribute.String("rpc.oper_type", "CONSUMER"),
  726. attribute.String("rpc.method", string(r.Payload)),
  727. )
  728. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  729. t.createTraceEvent(l7.ProtocolGrpc, ebpftracer.EventTypeL7Request.Int(), attr...)
  730. }
  731. func (t *Trace) DNSTraceQueryEvent(r *l7.RequestData, _type string, fqdn string, ttl uint32, ips []netaddr.IP) {
  732. if t == nil {
  733. return
  734. }
  735. t.addEvent()
  736. var attr []attribute.KeyValue
  737. var ipList []string
  738. for _, ip := range ips {
  739. ipList = append(ipList, ip.String())
  740. }
  741. ip := strings.Join(ipList, ",")
  742. attr = append(attr,
  743. attribute.String("dns.type", _type),
  744. attribute.String("dns.fqdn", fqdn),
  745. attribute.Int("dns.ttl", int(ttl)),
  746. attribute.String("dns.ips", ip),
  747. )
  748. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  749. t.createTraceEvent(l7.ProtocolDNS, ebpftracer.EventTypeL7Request.Int(), attr...)
  750. }
  751. func (t *Trace) HttpTraceRequest(method, path, ip string, port uint16, r *l7.RequestData) {
  752. if t == nil || method == "" {
  753. return
  754. }
  755. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  756. if err != nil {
  757. assumedAppID = 0
  758. }
  759. status := r.Status
  760. duration := r.Duration
  761. t.createTraceSpan(l7.ProtocolHTTP.String(), duration, status >= 400,
  762. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  763. semconv.HTTPMethod(method),
  764. semconv.HTTPStatusCode(int(status)),
  765. attribute.String("http.uri", path),
  766. attribute.String("http.ip", ip),
  767. attribute.Int64("http.assumed_app_id", assumedAppID),
  768. attribute.String("http.span_id", r.SpanId),
  769. attribute.Int("http.port", int(port)),
  770. )
  771. }
  772. // 新增事件处理
  773. func (t *Trace) HttpTraceRequestEvent(method, path, ip string, port uint16, r *l7.RequestData) {
  774. if t == nil {
  775. return
  776. }
  777. t.addEvent()
  778. if method == "" {
  779. return
  780. }
  781. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  782. if err != nil {
  783. assumedAppID = 0
  784. }
  785. status := r.Status
  786. var attr []attribute.KeyValue
  787. attr = append(attr,
  788. semconv.HTTPURL(fmt.Sprintf("http://%s:%d%s", ip, port, path)),
  789. semconv.HTTPMethod(method),
  790. semconv.HTTPStatusCode(int(status)),
  791. attribute.Bool("http.status_error", status > 399),
  792. attribute.String("http.uri", path),
  793. attribute.String("http.ip", ip),
  794. attribute.Int64("http.assumed_app_id", assumedAppID),
  795. attribute.String("http.span_id", r.SpanId),
  796. attribute.Int("http.port", int(port)),
  797. attribute.String("http.src_addr", r.ComponentSAddr.String()),
  798. attribute.String("http.destination_addr", r.ComponentDAddr.String()),
  799. attribute.Bool("http.is_tls", r.IsTls),
  800. )
  801. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  802. t.createTraceEvent(l7.ProtocolHTTP, ebpftracer.EventTypeL7Request.Int(), attr...)
  803. }
  804. func (t *Trace) FuncTraceQuery(funcname string, duration time.Duration, start uint64, end uint64) {
  805. if t == nil {
  806. return
  807. }
  808. t.addEvent()
  809. if funcname == "" {
  810. return
  811. }
  812. t.createTraceSpanNoTime2(funcname, duration, false, start, end)
  813. //t.createTraceSpanNoTime2(funcname, duration, false, start, end, attribute.Int("num", num))
  814. }
  815. func (t *Trace) createTraceSpanNoTime(name string, duration time.Duration, error bool, start uint64, end uint64, attrs ...attribute.KeyValue) {
  816. // end := time.Now()
  817. // start := end.Add(-duration)
  818. startTime := time.Unix(0, int64(start))
  819. endTime := time.Unix(0, int64(end))
  820. //fmt.Println("createTraceSpan:", t.ctx)
  821. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  822. span.SetAttributes(t.commonAttrs...)
  823. span.SetAttributes(attrs...)
  824. if error {
  825. span.SetStatus(codes.Error, "")
  826. }
  827. span.End(trace.WithTimestamp(endTime))
  828. }
  829. func (t *Trace) createTraceSpanNoTime2(name string, duration time.Duration, error bool, start uint64, end uint64) {
  830. // end := time.Now()
  831. // start := end.Add(-duration)
  832. //startTime := time.Unix(0, int64(start))
  833. //endTime := time.Unix(0, int64(end))
  834. ////fmt.Println("createTraceSpan:", t.ctx)
  835. //_, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  836. //span.SetAttributes(t.commonAttrs...)
  837. //span.SetAttributes(attrs...)
  838. //if error {
  839. // span.SetStatus(codes.Error, "")
  840. //}
  841. //span.End(trace.WithTimestamp(endTime))
  842. //attrs = append([]attribute.KeyValue{
  843. // attribute.Int64("startAt", int64(start)),
  844. // attribute.Int64("endAt", int64(end)),
  845. //})
  846. var attr []attribute.KeyValue
  847. t.appendTimestamp(&attr, start, end, int64(end-start))
  848. t.createFuncTraceEvent(name, ebpftracer.EventTypeFunEnt.Int(), attr...)
  849. }