ebpf/tracing/apm_tracing.go

package tracing

import (
	"context"
	"fmt"
	"sort"
	"time"

	"strconv"

	"github.com/coroot/coroot-node-agent/ebpftracer"
	"github.com/coroot/coroot-node-agent/ebpftracer/l7"
	"go.opentelemetry.io/otel/attribute"
	"go.opentelemetry.io/otel/codes"
	semconv "go.opentelemetry.io/otel/semconv/v1.18.0"
	"go.opentelemetry.io/otel/trace"
	"inet.af/netaddr"
)

/**
 * Trace
 */

func (t *Trace) setContext(ctx context.Context) {
	t.lock.Lock()
	defer t.lock.Unlock()
	t.ctx = ctx
}

func (t *Trace) setSpan(span trace.Span) {
	t.lock.Lock()
	defer t.lock.Unlock()
	t.span = span
}

type TimeMap struct {
	Time uint64
	Type int
	Map  *ebpftracer.StackFunEvent
}

func (t *Trace) buildFun() {
	mapSlice := []TimeMap{}
	for i, v := range t.stack {
		timeStartMap := TimeMap{}
		if v.StackEvent.Location == 0 {
			timeStartMap = TimeMap{
				Time: v.StackEvent.TimeNsStart,
				Type: 0,
				Map:  &t.stack[i],
			}
		} else {
			timeStartMap = TimeMap{
				Time: v.StackEvent.TimeNsEnd,
				Type: 1,
				Map:  &t.stack[i],
			}
		}
		mapSlice = append(mapSlice, timeStartMap)
	}
	sort.Slice(mapSlice, func(i, j int) bool {
		return mapSlice[i].Time < mapSlice[j].Time
	})

	funStack := []TimeMap{}

	currentfunNum := 1

	// for k, v := range mapSlice {
	// 	fmt.Println("---SliceSliceindex", k, "value", v.Time, v.Type, v.Map.Uprobe.Funcname, v.Map.StackEvent.Nid)
	// }

	mapSliceLen := len(mapSlice)
	for k, v := range mapSlice {
		// fmt.Println("SliceSliceindex", k, "value", v.Time, v.Type, v.Map.Uprobe.Funcname, v.Map.StackEvent.Nid)
		if v.Type == 0 {
			// 函数入口
			funStack = append(funStack, v)
		} else if v.Type == 1 {
			// 函数出口
			len := len(funStack)
			if len < 1 {
				fmt.Printf("buildFun ErrorError return before enter: %v\n", v)
				continue
			}
			currnt := funStack[len-1]
			if currnt.Map.StackEvent.Location != 0 {
				fmt.Printf("currnt StackEvent ErrorError is not enter: %v\n", v)
				continue
			}
			if k < mapSliceLen-1 && len >= 2 {
				nextfun := mapSlice[k+1]
				preCurrnt := funStack[len-2]
				// // 处理 .netcore 多次 returun
				// 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
				if nextfun.Map.StackEvent.Location == 1 && nextfun.Map.Uprobe.Funcname == currnt.Map.Uprobe.Funcname && preCurrnt.Map.Uprobe.Funcname != currnt.Map.Uprobe.Funcname {
					currentfunNum++
					continue
				}
			}
			funStack = funStack[:len-1]
			duration := v.Map.StackEvent.TimeNsEnd - currnt.Map.StackEvent.TimeNsStart
			t.FuncTraceQuery(currnt.Map.Uprobe.Funcname, time.Duration(duration), currnt.Map.StackEvent.TimeNsStart, v.Map.StackEvent.TimeNsEnd, currentfunNum)
			currentfunNum = 1
		}
	}
}

func (t *Trace) TraceStart(method, path string, status l7.Status, duration time.Duration) {
	if t == nil || method == "" {
		return
	}
	t.createParentSpan("APPLICATION", duration, status >= 400,
		semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
		semconv.HTTPMethod(method),
		//semconv.HTTPStatusCode(int(status)),
		attribute.String("http.uri", path),
	)
}

func (t *Trace) TraceEnd(r *l7.RequestData) {
	if t == nil {
		return
	}
	t.span.SetAttributes(
		semconv.HTTPStatusCode(int(r.Status)),
		attribute.String("server.trace_id_from", r.ParentSpanContext.TraceIdFrom),
	)

	CalledId, err := strconv.ParseInt(r.ParentSpanContext.CalledId, 10, 64)
	if err == nil && CalledId != 0 {
		t.span.SetAttributes(attribute.Int64("server.called_id", CalledId))
	}

	InstanceIdFrom, err := strconv.ParseInt(r.ParentSpanContext.InstanceIdFrom, 10, 64)
	if err == nil && InstanceIdFrom != 0 {
		t.span.SetAttributes(attribute.Int64("server.instance_id_from", InstanceIdFrom))
	}

	AppIdFrom, err := strconv.ParseInt(r.ParentSpanContext.AppIdFrom, 10, 64)
	if err == nil && AppIdFrom != 0 {
		t.span.SetAttributes(attribute.Int64("server.app_id_from", AppIdFrom))
	}
	if r.ParentSpanContext.SpanIdFrom != "0000000000000000" {
		t.span.SetAttributes(attribute.String("server.span_id_from", r.ParentSpanContext.SpanIdFrom))
	}
	// for _, v := range t.stack {
	// 	fmt.Printf("TraceEndTraceEndTraceEnd%s\n", v)
	// }
	t.buildFun()
	t.span.End(trace.WithTimestamp(time.Now()))
}

func (t *Trace) createParentSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
	end := time.Now()
	start := end.Add(-duration)
	ctx, span := tracer(t.containerId).Start(context.Background(), name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
	span.SetAttributes(attrs...)
	span.SetAttributes(t.commonAttrs...)
	if error {
		span.SetStatus(codes.Error, "")
	}
	t.setContext(ctx)
	t.setSpan(span)
}

func (t *Trace) createTraceSpan(name string, duration time.Duration, error bool, attrs ...attribute.KeyValue) {
	end := time.Now()
	start := end.Add(-duration)
	//fmt.Println("createTraceSpan:", t.ctx)
	_, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(start), trace.WithSpanKind(trace.SpanKindClient))
	span.SetAttributes(t.commonAttrs...)
	span.SetAttributes(attrs...)
	if error {
		span.SetStatus(codes.Error, "")
	}
	span.End(trace.WithTimestamp(end))
}

func (t *Trace) MysqlTraceQuery(query string, error bool, duration time.Duration, destination netaddr.IPPort) {
	if t == nil || query == "" {
		return
	}
	t.createTraceSpan(l7.ProtocolMysql.String(), duration, error,
		semconv.DBSystemMySQL,
		semconv.DBStatement(query),
		semconv.NetPeerName(destination.IP().String()),
		semconv.NetPeerPort(int(destination.Port())),
	)
}

func (t *Trace) RedisTraceQuery(cmd, args string, error bool, duration time.Duration) {
	if t == nil || cmd == "" {
		return
	}
	statement := cmd
	if args != "" {
		statement += " " + args
	}
	t.createTraceSpan(l7.ProtocolRedis.String(), duration, error,
		semconv.DBSystemRedis,
		semconv.DBOperation(cmd),
		semconv.DBStatement(statement),
	)
}

func (t *Trace) HttpTraceRequest(method, path, ip string, port uint16, r *l7.RequestData) {
	if t == nil || method == "" {
		return
	}
	assumedAppID, err := strconv.ParseInt(r.AssumedAppId, 10, 64)
	if err != nil {
		assumedAppID = 0
	}
	status := r.Status
	duration := r.Duration
	t.createTraceSpan(l7.ProtocolHTTP.String(), duration, status >= 400,
		semconv.HTTPURL(fmt.Sprintf("http://%s%s", t.destination.String(), path)),
		semconv.HTTPMethod(method),
		semconv.HTTPStatusCode(int(status)),
		attribute.String("http.uri", path),
		attribute.String("http.ip", ip),
		attribute.Int64("http.assumed_app_id", assumedAppID),
		attribute.String("http.span_id", r.SpanId),
		attribute.Int("http.port", int(port)),
	)
}

func (t *Trace) FuncTraceQuery(funcname string, duration time.Duration, start uint64, end uint64, num int) {
	if t == nil || funcname == "" {
		return
	}
	t.createTraceSpanNoTime(funcname, duration, false, start, end, attribute.Int("num", num))
}

func (t *Trace) createTraceSpanNoTime(name string, duration time.Duration, error bool, start uint64, end uint64, attrs ...attribute.KeyValue) {
	// end := time.Now()
	// start := end.Add(-duration)
	startTime := time.Unix(0, int64(start))
	endTime := time.Unix(0, int64(end))
	//fmt.Println("createTraceSpan:", t.ctx)
	_, span := tracer(t.containerId).Start(t.ctx, name, trace.WithTimestamp(startTime), trace.WithSpanKind(trace.SpanKindClient))
	span.SetAttributes(t.commonAttrs...)
	span.SetAttributes(attrs...)
	if error {
		span.SetStatus(codes.Error, "")
	}
	span.End(trace.WithTimestamp(endTime))
}