apm_tracing.go 25 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801
  1. package tracing
  2. import (
  3. "context"
  4. "fmt"
  5. klog "github.com/sirupsen/logrus"
  6. "os"
  7. "strconv"
  8. "strings"
  9. "sync"
  10. "sync/atomic"
  11. "time"
  12. "github.com/coroot/coroot-node-agent/ebpftracer"
  13. "github.com/coroot/coroot-node-agent/ebpftracer/l7"
  14. "github.com/coroot/coroot-node-agent/utils"
  15. . "github.com/coroot/coroot-node-agent/utils/modelse"
  16. "go.opentelemetry.io/otel/attribute"
  17. "go.opentelemetry.io/otel/codes"
  18. semconv "go.opentelemetry.io/otel/semconv/v1.18.0"
  19. "go.opentelemetry.io/otel/trace"
  20. "inet.af/netaddr"
  21. )
  22. /**
  23. * Trace
  24. */
  25. func (t *Trace) setContext(ctx context.Context) {
  26. t.lock.Lock()
  27. defer t.lock.Unlock()
  28. t.ctx = ctx
  29. }
  30. func (t *Trace) setSpan(span trace.Span) {
  31. t.lock.Lock()
  32. defer t.lock.Unlock()
  33. t.span = span
  34. }
  35. type TimeMap struct {
  36. Time uint64
  37. Type int
  38. Map *ebpftracer.StackFunEvent
  39. }
  40. //func (t *Trace) buildFun() {
  41. // mapSlice := []TimeMap{}
  42. // for i, v := range t.stack {
  43. // timeStartMap := TimeMap{}
  44. // if v.StackEvent.Location == 0 {
  45. // timeStartMap = TimeMap{
  46. // Time: v.StackEvent.TimeNsStart,
  47. // Type: 0,
  48. // Map: &t.stack[i],
  49. // }
  50. // } else {
  51. // timeStartMap = TimeMap{
  52. // Time: v.StackEvent.TimeNsEnd,
  53. // Type: 1,
  54. // Map: &t.stack[i],
  55. // }
  56. // }
  57. // mapSlice = append(mapSlice, timeStartMap)
  58. // }
  59. // sort.Slice(mapSlice, func(i, j int) bool {
  60. // return mapSlice[i].Time < mapSlice[j].Time
  61. // })
  62. //
  63. // funStack := []TimeMap{}
  64. //
  65. // currentfunNum := 1
  66. //
  67. // // for k, v := range mapSlice {
  68. // // fmt.Println("---SliceSliceindex", k, "value", v.Time, v.Type, v.Map.Uprobe.Funcname, v.Map.StackEvent.Nid)
  69. // // }
  70. //
  71. // mapSliceLen := len(mapSlice)
  72. // for k, v := range mapSlice {
  73. // // fmt.Println("SliceSliceindex", k, "value", v.Time, v.Type, v.Map.Uprobe.Funcname, v.Map.StackEvent.Nid)
  74. // if v.Type == 0 {
  75. // // 函数入口
  76. // funStack = append(funStack, v)
  77. // } else if v.Type == 1 {
  78. // // 函数出口
  79. // len := len(funStack)
  80. // if len < 1 {
  81. // fmt.Printf("buildFun ErrorError return before enter: %v\n", v)
  82. // continue
  83. // }
  84. // currnt := funStack[len-1]
  85. // if currnt.Map.StackEvent.Location != 0 {
  86. // fmt.Printf("currnt StackEvent ErrorError is not enter: %v\n", v)
  87. // continue
  88. // }
  89. // if k < mapSliceLen-1 && len >= 2 {
  90. // nextfun := mapSlice[k+1]
  91. // preCurrnt := funStack[len-2]
  92. // // // 处理 .netcore 多次 returun
  93. // // 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
  94. // if nextfun.Map.StackEvent.Location == 1 && nextfun.Map.Uprobe.Funcname == currnt.Map.Uprobe.Funcname && preCurrnt.Map.Uprobe.Funcname != currnt.Map.Uprobe.Funcname {
  95. // currentfunNum++
  96. // continue
  97. // }
  98. // }
  99. // funStack = funStack[:len-1]
  100. // duration := v.Map.StackEvent.TimeNsEnd - currnt.Map.StackEvent.TimeNsStart
  101. // t.FuncTraceQuery(currnt.Map.Uprobe.Funcname, time.Duration(duration), currnt.Map.StackEvent.TimeNsStart, v.Map.StackEvent.TimeNsEnd, currentfunNum)
  102. // currentfunNum = 1
  103. // }
  104. // }
  105. //}
  106. func (t *Trace) startReady() {
  107. t.lock.Lock()
  108. defer t.lock.Unlock()
  109. t.startEventReady = true
  110. }
  111. func (t *Trace) endReadyEvent(needCount uint32) {
  112. t.lock.Lock()
  113. defer t.lock.Unlock()
  114. t.endEventReady = true
  115. t.needEventCount = needCount
  116. }
  117. func (t *Trace) AllEventReady(traceID uint64) bool {
  118. klog.Debugf("[AllEventReady (current/need)|start|end|traceid](%d/%d)%v|%v|%d", *t.currenEventCount, t.needEventCount, t.startEventReady, t.endEventReady, traceID)
  119. return t.startEventReady && t.endEventReady && *t.currenEventCount >= t.needEventCount
  120. }
  121. func (t *Trace) GrpcServerTraceStartEvent(r *l7.RequestData, appInfo AppInfo) {
  122. t.span.SetAttributes(attribute.String("rpc.uri", string(r.Payload)))
  123. t.commonAttrs = []attribute.KeyValue{
  124. // buildAppMapFromEvent
  125. attribute.Int("server.code_type", appInfo.CodeType.Int()),
  126. attribute.String("server.app_name", appInfo.AppName),
  127. attribute.String("server.service_name", "GRPC"),
  128. attribute.String("rpc.service_type", "RPC"),
  129. attribute.String("rpc.oper_type", "PROVIDER"),
  130. attribute.Int64("server.app_id", appInfo.AppIdHash.IntVal),
  131. attribute.Int64("server.agent_id", appInfo.AgentId),
  132. attribute.Int64("server.instance_id", appInfo.InstanceIdHash.IntVal),
  133. }
  134. t.span.SetAttributes(t.commonAttrs...)
  135. t.startReady()
  136. }
  137. func (t *Trace) TraceStartEvent(method, path, sn string, sport uint16, status l7.Status, addr netaddr.IPPort, pid uint32, appInfo AppInfo) {
  138. t.span.SetAttributes(semconv.HTTPURL(fmt.Sprintf("http://%s:%d%s", sn, sport, path)),
  139. semconv.HTTPMethod(method),
  140. attribute.String("http.uri", path))
  141. if status > 399 {
  142. t.span.SetStatus(codes.Error, "")
  143. }
  144. //host := fmt.Sprintf("%s:%s", sn, sport)
  145. t.destination = addr
  146. t.commonAttrs = []attribute.KeyValue{
  147. semconv.NetPeerName(sn),
  148. semconv.NetPeerPort(int(sport)),
  149. // buildAppMapFromEvent
  150. attribute.Int("server.code_type", appInfo.CodeType.Int()),
  151. attribute.String("server.app_name", appInfo.AppName),
  152. attribute.String("server.service_name", appInfo.ServiceType),
  153. attribute.Int64("server.app_id", appInfo.AppIdHash.IntVal),
  154. attribute.Int64("server.agent_id", appInfo.AgentId),
  155. attribute.Int64("server.instance_id", appInfo.InstanceIdHash.IntVal),
  156. }
  157. t.span.SetAttributes(t.commonAttrs...)
  158. t.startReady()
  159. }
  160. // set context span
  161. func (t *Trace) CreateRootSpan(traceId uint64) {
  162. traceIdStr := strconv.Itoa(int(traceId))
  163. ctx, span := tracer(t.containerId).Start(context.Background(), traceIdStr, trace.WithSpanKind(trace.SpanKindClient))
  164. t.setContext(ctx)
  165. t.setSpan(span)
  166. }
  167. func (t *Trace) TraceStart(method, path string, status l7.Status, duration time.Duration) {
  168. if t == nil || method == "" {
  169. return
  170. }
  171. t.createParentSpan("APPLICATION", duration, status >= 400,
  172. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  173. semconv.HTTPMethod(method),
  174. //semconv.HTTPStatusCode(int(status)),
  175. attribute.String("http.uri", path),
  176. )
  177. }
  178. func (t *Trace) TraceEnd(r *l7.RequestData) {
  179. if t == nil {
  180. return
  181. }
  182. t.span.SetAttributes(
  183. semconv.HTTPStatusCode(int(r.Status)),
  184. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  185. )
  186. CalledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  187. if err == nil && CalledId != 0 {
  188. t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  189. }
  190. InstanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  191. if err == nil && InstanceIdFrom != 0 {
  192. t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  193. }
  194. AppIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  195. if err == nil && AppIdFrom != 0 {
  196. t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  197. }
  198. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  199. t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  200. }
  201. // klog.Debugf("r.ParentSpanContext.TypeFrom tpp %s", r.ParentSpanContext.TypeFrom)
  202. if r.ParentSpanContext.TypeFrom == "10" {
  203. t.span.SetAttributes(attribute.String("server.type_from", "SDK"))
  204. t.span.SetAttributes(attribute.String("server.app_id_from", "-1"))
  205. t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  206. t.span.SetAttributes(attribute.String("server.instance_id_from", "-1"))
  207. t.span.SetAttributes(attribute.String("server.call_id", "-1"))
  208. }
  209. // for _, v := range t.stack {
  210. // fmt.Printf("TraceEndTraceEndTraceEnd%s\n", v)
  211. // }
  212. //t.buildFun()
  213. t.span.End(trace.WithTimestamp(time.Now()))
  214. }
  215. // 新增结束事件
  216. func (t *Trace) TraceEndEvent(r *l7.RequestData) {
  217. if t == nil {
  218. return
  219. }
  220. var attr []attribute.KeyValue
  221. attr = append(attr,
  222. semconv.HTTPStatusCode(int(r.Status)),
  223. attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  224. )
  225. //t.span.SetAttributes(
  226. // semconv.HTTPStatusCode(int(r.Status)),
  227. // attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
  228. //)
  229. calledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
  230. if err == nil && calledId != 0 {
  231. attr = append(attr, attribute.Int64("server.called_id", calledId))
  232. //t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
  233. }
  234. instanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
  235. if err == nil && instanceIdFrom != 0 {
  236. attr = append(attr, attribute.Int64("server.instance_id_from", instanceIdFrom))
  237. //t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
  238. }
  239. appIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
  240. if err == nil && appIdFrom != 0 {
  241. attr = append(attr, attribute.Int64("server.app_id_from", appIdFrom))
  242. //t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
  243. }
  244. if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
  245. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  246. //t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
  247. }
  248. // klog.Debugf("r.ParentSpanContext.TypeFrom event tpp %s", r.ParentSpanContext.TypeFrom)
  249. if r.ParentSpanContext.TypeFrom == "10" {
  250. attr = append(attr, attribute.String("server.type_from", "SDK"))
  251. attr = append(attr, attribute.String("server.app_id_from", "-1"))
  252. attr = append(attr, attribute.String("server.span_id_from", r.ParentSpanContext.TraceIdFrom))
  253. attr = append(attr, attribute.String("server.instance_id_from", "-1"))
  254. attr = append(attr, attribute.String("server.call_id", "-1"))
  255. }
  256. attr = append(attr, attribute.String("server.src_addr", r.DAddr.String()))
  257. attr = append(attr, attribute.String("server.dst_addr", r.SAddr.String()))
  258. klog.Debugf("TraceEndEvent SAddr.String %s", r.SAddr.String())
  259. klog.Debugf("TraceEndEvent DAddr.String %s", r.DAddr.String())
  260. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  261. t.span.SetAttributes(attr...)
  262. t.endReadyEvent(r.EventCount)
  263. }
  264. func (t *Trace) appendTimestamp(attr *[]attribute.KeyValue, s, e uint64, d int64) {
  265. *attr = append(*attr,
  266. attribute.Int64("time.start_at", utils.KtimeToTimestamp(s)),
  267. attribute.Int64("time.end_at", utils.KtimeToTimestamp(e)),
  268. attribute.Int64("time.duration", d),
  269. )
  270. }
  271. func (t *Trace) createParentSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  272. end := time.Now()
  273. start := end.Add(-duration)
  274. ctx, span := tracer(t.containerId).Start(context.Background(), name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  275. span.SetAttributes(attrs...)
  276. span.SetAttributes(t.commonAttrs...)
  277. if error {
  278. span.SetStatus(codes.Error, "")
  279. }
  280. t.setContext(ctx)
  281. t.setSpan(span)
  282. }
  283. func (t *Trace) SendEvent() {
  284. t.span.End()
  285. }
  286. func (t *Trace) GetSpan() trace.Span {
  287. return t.span
  288. }
  289. func (t *Trace) createTraceEvent(name string, eventType int, l7Type int, attrs ...attribute.KeyValue) {
  290. t.span.AddEventApm(name, eventType, l7Type, trace.WithAttributes(attrs...))
  291. //atomic.AddUint32(t.currenEventCount, 1)
  292. }
  293. func (t *Trace) addEvent() {
  294. atomic.AddUint32(t.currenEventCount, 1)
  295. }
  296. func (t *Trace) createTraceSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
  297. end := time.Now()
  298. start := end.Add(-duration)
  299. //fmt.Println("createTraceSpan:", t.ctx)
  300. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
  301. span.SetAttributes(t.commonAttrs...)
  302. span.SetAttributes(attrs...)
  303. if error {
  304. span.SetStatus(codes.Error, "")
  305. }
  306. span.End(trace.WithTimestamp(end))
  307. }
  308. func (t *Trace) MysqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  309. if t == nil || query == "" {
  310. return
  311. }
  312. t.createTraceSpan(l7.ProtocolMysql.String(), duration, error,
  313. semconv.DBSystemMySQL,
  314. semconv.DBStatement(query),
  315. semconv.NetPeerName(destination.IP().String()),
  316. semconv.NetPeerPort(int(destination.Port())),
  317. )
  318. }
  319. // 内置允许操作(始终放行)
  320. var builtinOps = map[string]struct{}{
  321. "SELECT": {},
  322. "INSERT": {},
  323. "UPDATE": {},
  324. "DELETE": {},
  325. "WITH": {},
  326. }
  327. var (
  328. allowedOps map[string]struct{}
  329. initOnce sync.Once
  330. )
  331. // 解析并缓存环境变量中的前缀
  332. func initAllowedOps() {
  333. allowedOps = make(map[string]struct{})
  334. env := os.Getenv("ALLOWED_SQL_PREFIXES")
  335. for _, s := range strings.Split(env, ",") {
  336. if trimmed := strings.ToUpper(strings.TrimSpace(s)); trimmed != "" {
  337. allowedOps[trimmed] = struct{}{}
  338. }
  339. }
  340. }
  341. func isCURDOperation(q string) bool {
  342. if len(q) < 6 {
  343. return false
  344. }
  345. // 提取前缀(考虑最长10个字符)
  346. prefix := strings.ToUpper(strings.TrimSpace(q))
  347. if len(prefix) > 10 {
  348. prefix = prefix[:10]
  349. }
  350. // 提取第一个单词作为操作关键字
  351. spaceIdx := strings.IndexAny(prefix, " \t\n")
  352. if spaceIdx > 0 {
  353. prefix = prefix[:spaceIdx]
  354. }
  355. // 先查固定允许的操作
  356. if _, ok := builtinOps[prefix]; ok {
  357. return true
  358. }
  359. // 加载环境变量控制的操作(只执行一次)
  360. initOnce.Do(initAllowedOps)
  361. // 查额外允许操作
  362. _, ok := allowedOps[prefix]
  363. return ok
  364. }
  365. func (t *Trace) SQLTraceQueryEvent(l7Type l7.Protocol, semconvVal attribute.KeyValue, query string, r *l7.RequestData, destination netaddr.IPPort) {
  366. //fmt.Println("query", query)
  367. if t == nil {
  368. return
  369. }
  370. t.addEvent()
  371. // 只保留增删改查基本操作
  372. if !isCURDOperation(query) {
  373. return
  374. }
  375. var attr []attribute.KeyValue
  376. attr = append(attr,
  377. semconvVal,
  378. semconv.DBStatement(query),
  379. semconv.NetPeerName(destination.IP().String()),
  380. semconv.NetPeerPort(int(destination.Port())),
  381. attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  382. attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  383. )
  384. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  385. attr = append(attr,
  386. attribute.Bool("sql.exception", r.Status.Error()),
  387. )
  388. t.createTraceEvent(l7Type.String(), ebpftracer.EventTypeL7Request.Int(), l7Type.Int(), attr...)
  389. }
  390. //func (t *Trace) MysqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  391. // fmt.Println("query", query)
  392. // if t == nil {
  393. // return
  394. // }
  395. //
  396. // var attr []attribute.KeyValue
  397. // attr = append(attr,
  398. // semconv.DBSystemMySQL,
  399. // semconv.DBStatement(query),
  400. // semconv.NetPeerName(destination.IP().String()),
  401. // semconv.NetPeerPort(int(destination.Port())),
  402. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  403. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  404. // )
  405. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  406. //
  407. // attr = append(attr,
  408. // attribute.Bool("sql.exception", r.Status.Error()),
  409. // )
  410. // t.createTraceEvent(l7.ProtocolMysql.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolMysql), attr...)
  411. //}
  412. //func (t *Trace) PostGreSqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
  413. // if t == nil || query == "" {
  414. // return
  415. // }
  416. // t.createTraceSpan(l7.ProtocolPostgres.String(), duration, error,
  417. // semconv.DBSystemPostgreSQL,
  418. // semconv.DBStatement(query),
  419. // semconv.NetPeerName(destination.IP().String()),
  420. // semconv.NetPeerPort(int(destination.Port())),
  421. // )
  422. //}
  423. //func (t *Trace) PostGreSqlTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  424. // fmt.Println("query", query)
  425. // if t == nil {
  426. // return
  427. // }
  428. //
  429. // var attr []attribute.KeyValue
  430. // attr = append(attr,
  431. // semconv.DBSystemPostgreSQL,
  432. // semconv.DBStatement(query),
  433. // semconv.NetPeerName(destination.IP().String()),
  434. // semconv.NetPeerPort(int(destination.Port())),
  435. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  436. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  437. // )
  438. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  439. // attr = append(attr,
  440. // attribute.Bool("sql.exception", r.Status.Error()),
  441. // )
  442. // t.createTraceEvent(l7.ProtocolPostgres.String(), int(ebpftracer.EventTypeL7Request), int(l7.ProtocolPostgres), attr...)
  443. //}
  444. //func (t *Trace) DmTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  445. // if t == nil || query == "" {
  446. // return
  447. // }
  448. // l7Type := int(l7.ProtocolDM)
  449. // var attr []attribute.KeyValue
  450. // attr = append(attr,
  451. // semconv.DBSystemDaMengDB,
  452. // semconv.DBStatement(query),
  453. // semconv.NetPeerName(destination.IP().String()),
  454. // semconv.NetPeerPort(int(destination.Port())),
  455. // attribute.String("sql.src_addr", r.ComponentSAddr.String()),
  456. // attribute.String("sql.destination_addr", r.ComponentDAddr.String()),
  457. // )
  458. // t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  459. //
  460. // attr = append(attr,
  461. // attribute.Bool("sql.exception", r.Status.Error()),
  462. // )
  463. // t.createTraceEvent(l7.ProtocolHTTP.String(), int(ebpftracer.EventTypeL7Request), l7Type, attr...)
  464. //}
  465. //func (t *Trace) RedisTraceQuery(cmd, args string, error bool, duration time.Duration) {
  466. // if t == nil || cmd == "" {
  467. // return
  468. // }
  469. // statement := cmd
  470. // if args != "" {
  471. // statement += " " + args
  472. // }
  473. // t.createTraceSpan(l7.ProtocolRedis.String(), duration, error,
  474. // semconv.DBSystemRedis,
  475. // semconv.DBOperation(cmd),
  476. // semconv.DBStatement(statement),
  477. // )
  478. //}
  479. func (t *Trace) RedisTraceQueryEvent(cmd, args string, r *l7.RequestData, destination netaddr.IPPort) {
  480. if t == nil {
  481. return
  482. }
  483. t.addEvent()
  484. if cmd == "" {
  485. return
  486. }
  487. statement := cmd
  488. if args != "" {
  489. statement += " " + args
  490. }
  491. var attr []attribute.KeyValue
  492. attr = append(attr,
  493. semconv.DBSystemRedis,
  494. semconv.DBOperation(cmd),
  495. semconv.DBStatement(statement),
  496. semconv.NetPeerName(destination.IP().String()),
  497. semconv.NetPeerPort(int(destination.Port())),
  498. attribute.String("nosql.src_addr", r.ComponentSAddr.String()),
  499. attribute.String("nosql.destination_addr", r.ComponentDAddr.String()),
  500. )
  501. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  502. t.createTraceEvent(l7.ProtocolRedis.String(), ebpftracer.EventTypeL7Request.Int(), l7.ProtocolRedis.Int(), attr...)
  503. }
  504. func (t *Trace) GrpcServerTraceQueryEvent(r *l7.RequestData, appInfo AppInfo) {
  505. if t == nil {
  506. return
  507. }
  508. t.addEvent()
  509. // if method == "" {
  510. // return
  511. // }
  512. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  513. if err != nil {
  514. assumedAppID = 0
  515. }
  516. var attr []attribute.KeyValue
  517. attr = append(attr,
  518. semconv.RPCSystemGRPC,
  519. attribute.String("rpc.uri", string(r.Payload)),
  520. attribute.String("rpc.schema", "grpc"),
  521. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  522. attribute.String("rpc.span_id", r.SpanId),
  523. attribute.Int("rpc.port", appInfo.Sport),
  524. attribute.String("rpc.ip", appInfo.Sn),
  525. attribute.String("rpc.oper_type", "PROVIDER"),
  526. attribute.String("rpc.method", "test_method"),
  527. )
  528. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  529. t.createTraceEvent(l7.ProtocolGrpc.String(), ebpftracer.EventTypeL7Request.Int(), l7.ProtocolGrpc.Int(), attr...)
  530. }
  531. func (t *Trace) GrpcClientTraceQueryEvent(r *l7.RequestData) {
  532. if t == nil {
  533. return
  534. }
  535. t.addEvent()
  536. // if method == "" {
  537. // return
  538. // }
  539. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  540. if err != nil {
  541. assumedAppID = 0
  542. }
  543. // 解析 RPCTarget 字符串,格式为 "ip:port"
  544. klog.Infof("r.RPCTarget is %s", r.RPCTarget)
  545. var rpcIP, rpcPort string
  546. if r.RPCTarget != "" {
  547. // 清理空字节,只保留有效字符
  548. cleanTarget := strings.TrimRight(r.RPCTarget, "\x00")
  549. cleanTarget = strings.TrimSpace(cleanTarget)
  550. klog.Infof("cleanTarget is %s", cleanTarget)
  551. if colonIndex := strings.LastIndex(cleanTarget, ":"); colonIndex != -1 {
  552. rpcIP = cleanTarget[:colonIndex]
  553. rpcPort = cleanTarget[colonIndex+1:]
  554. klog.Infof("rpcIP is %s, rpcPort is %s", rpcIP, rpcPort)
  555. } else {
  556. // 如果没有找到冒号,可能是只有IP或只有端口
  557. rpcIP = cleanTarget
  558. }
  559. }
  560. // 设置默认值
  561. if rpcIP == "" {
  562. rpcIP = "127.0.0.1"
  563. }
  564. if rpcPort == "" {
  565. rpcPort = "80"
  566. }
  567. // 转换端口为整数
  568. portInt, err := strconv.Atoi(rpcPort)
  569. if err != nil {
  570. klog.Infof("r.RPCTarget convert port err is %s", err)
  571. portInt = 80 // 默认端口
  572. } else {
  573. klog.Infof("r.RPCTarget convert port success, portInt is %d", portInt)
  574. }
  575. var attr []attribute.KeyValue
  576. attr = append(attr,
  577. semconv.RPCSystemGRPC,
  578. attribute.String("rpc.uri", string(r.Payload)),
  579. attribute.String("rpc.schema", "grpc"),
  580. attribute.Int64("rpc.assumed_app_id", assumedAppID),
  581. attribute.String("rpc.span_id", r.SpanId),
  582. attribute.Int("rpc.port", portInt),
  583. attribute.String("rpc.ip", rpcIP),
  584. attribute.String("rpc.oper_type", "CONSUMER"),
  585. attribute.String("rpc.method", "test_method"),
  586. )
  587. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  588. t.createTraceEvent(l7.ProtocolGrpc.String(), ebpftracer.EventTypeL7Request.Int(), l7.ProtocolGrpc.Int(), attr...)
  589. }
  590. func (t *Trace) MongoTraceQueryEvent(query string, r *l7.RequestData, destination netaddr.IPPort) {
  591. if t == nil {
  592. return
  593. }
  594. t.addEvent()
  595. if query == "" {
  596. return
  597. }
  598. var attr []attribute.KeyValue
  599. attr = append(attr,
  600. semconv.DBSystemRedis,
  601. semconv.DBStatement(query),
  602. semconv.NetPeerName(destination.IP().String()),
  603. semconv.NetPeerPort(int(destination.Port())),
  604. attribute.String("nosql.src_addr", r.ComponentSAddr.String()),
  605. attribute.String("nosql.destination_addr", r.ComponentDAddr.String()),
  606. )
  607. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  608. t.createTraceEvent(l7.ProtocolMongo.String(), ebpftracer.EventTypeL7Request.Int(), l7.ProtocolMongo.Int(), attr...)
  609. }
  610. func (t *Trace) DNSTraceQueryEvent(r *l7.RequestData, _type string, fqdn string, ttl uint32, ips []netaddr.IP) {
  611. if t == nil {
  612. return
  613. }
  614. t.addEvent()
  615. var attr []attribute.KeyValue
  616. var ipList []string
  617. for _, ip := range ips {
  618. ipList = append(ipList, ip.String())
  619. }
  620. ip := strings.Join(ipList, ",")
  621. attr = append(attr,
  622. attribute.String("dns.type", _type),
  623. attribute.String("dns.fqdn", fqdn),
  624. attribute.Int("dns.ttl", int(ttl)),
  625. attribute.String("dns.ips", ip),
  626. )
  627. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  628. t.createTraceEvent(l7.ProtocolDNS.String(), ebpftracer.EventTypeL7Request.Int(), l7.ProtocolDNS.Int(), attr...)
  629. }
  630. func (t *Trace) HttpTraceRequest(method, path, ip string, port uint16, r *l7.RequestData) {
  631. if t == nil || method == "" {
  632. return
  633. }
  634. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  635. if err != nil {
  636. assumedAppID = 0
  637. }
  638. status := r.Status
  639. duration := r.Duration
  640. t.createTraceSpan(l7.ProtocolHTTP.String(), duration, status >= 400,
  641. semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
  642. semconv.HTTPMethod(method),
  643. semconv.HTTPStatusCode(int(status)),
  644. attribute.String("http.uri", path),
  645. attribute.String("http.ip", ip),
  646. attribute.Int64("http.assumed_app_id", assumedAppID),
  647. attribute.String("http.span_id", r.SpanId),
  648. attribute.Int("http.port", int(port)),
  649. )
  650. }
  651. // 新增事件处理
  652. func (t *Trace) HttpTraceRequestEvent(method, path, ip string, port uint16, r *l7.RequestData) {
  653. if t == nil {
  654. return
  655. }
  656. t.addEvent()
  657. if method == "" {
  658. return
  659. }
  660. assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
  661. if err != nil {
  662. assumedAppID = 0
  663. }
  664. status := r.Status
  665. var attr []attribute.KeyValue
  666. attr = append(attr,
  667. semconv.HTTPURL(fmt.Sprintf("http://%s:%d%s", ip, port, path)),
  668. semconv.HTTPMethod(method),
  669. semconv.HTTPStatusCode(int(status)),
  670. attribute.Bool("http.status_error", status > 399),
  671. attribute.String("http.uri", path),
  672. attribute.String("http.ip", ip),
  673. attribute.Int64("http.assumed_app_id", assumedAppID),
  674. attribute.String("http.span_id", r.SpanId),
  675. attribute.Int("http.port", int(port)),
  676. attribute.String("http.src_addr", r.ComponentSAddr.String()),
  677. attribute.String("http.destination_addr", r.ComponentDAddr.String()),
  678. )
  679. t.appendTimestamp(&attr, r.StartAt, r.EndAt, r.Duration.Nanoseconds())
  680. t.createTraceEvent(l7.ProtocolHTTP.String(), ebpftracer.EventTypeL7Request.Int(), l7.ProtocolHTTP.Int(), attr...)
  681. }
  682. func (t *Trace) FuncTraceQuery(funcname string, duration time.Duration, start uint64, end uint64) {
  683. if t == nil {
  684. return
  685. }
  686. t.addEvent()
  687. if funcname == "" {
  688. return
  689. }
  690. t.createTraceSpanNoTime2(funcname, duration, false, start, end)
  691. //t.createTraceSpanNoTime2(funcname, duration, false, start, end, attribute.Int("num", num))
  692. }
  693. func (t *Trace) createTraceSpanNoTime(name string, duration time.Duration, error bool, start uint64, end uint64, attrs ...attribute.KeyValue) {
  694. // end := time.Now()
  695. // start := end.Add(-duration)
  696. startTime := time.Unix(0, int64(start))
  697. endTime := time.Unix(0, int64(end))
  698. //fmt.Println("createTraceSpan:", t.ctx)
  699. _, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  700. span.SetAttributes(t.commonAttrs...)
  701. span.SetAttributes(attrs...)
  702. if error {
  703. span.SetStatus(codes.Error, "")
  704. }
  705. span.End(trace.WithTimestamp(endTime))
  706. }
  707. func (t *Trace) createTraceSpanNoTime2(name string, duration time.Duration, error bool, start uint64, end uint64) {
  708. // end := time.Now()
  709. // start := end.Add(-duration)
  710. //startTime := time.Unix(0, int64(start))
  711. //endTime := time.Unix(0, int64(end))
  712. ////fmt.Println("createTraceSpan:", t.ctx)
  713. //_, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
  714. //span.SetAttributes(t.commonAttrs...)
  715. //span.SetAttributes(attrs...)
  716. //if error {
  717. // span.SetStatus(codes.Error, "")
  718. //}
  719. //span.End(trace.WithTimestamp(endTime))
  720. //attrs = append([]attribute.KeyValue{
  721. // attribute.Int64("startAt", int64(start)),
  722. // attribute.Int64("endAt", int64(end)),
  723. //})
  724. var attr []attribute.KeyValue
  725. t.appendTimestamp(&attr, start, end, int64(end-start))
  726. t.createTraceEvent(name, ebpftracer.EventTypeFunEnt.Int(), 0, attr...)
  727. }