feat(telemetry): bridge configurable gateway trace headers

This commit is contained in:
2026-10-06 21:21:35 +08:00
parent d65af03162
commit 9665fb5974
20 changed files with 484 additions and 25 deletions

173
telemetry/headers.go Normal file
View File

@@ -0,0 +1,173 @@
package telemetry
import (
"context"
"errors"
"net/http"
"sort"
"strings"
"unicode"
"git.yuntuo.hk/k2safe/go-omni/logx"
"go.opentelemetry.io/otel/propagation"
"go.opentelemetry.io/otel/trace"
)
// HeaderPropagationConfig maps a legacy or gateway header pair to the
// standard OpenTelemetry trace context. TraceIDHeader and SpanIDHeader must
// carry a 32- and 16-hexadecimal ID respectively. RequestIDHeader is optional
// and is copied to/from the request correlation context.
//
// Legacy headers do not carry a sampling flag. The adapter treats a valid pair
// as sampled so the trace remains visible to the configured exporter. Services
// that need upstream sampling decisions should use the standard traceparent
// header, which always takes precedence.
type HeaderPropagationConfig struct {
TraceIDHeader string `json:"traceIdHeader"`
SpanIDHeader string `json:"spanIdHeader"`
RequestIDHeader string `json:"requestIdHeader,omitempty"`
}
// HeaderPropagator accepts standard W3C trace context and one configured
// legacy header pair. W3C context wins when both formats are present. It also
// propagates the configured request ID without making it part of the span
// parent. Use it only at a boundary that is authorized to supply these IDs.
type HeaderPropagator struct {
config HeaderPropagationConfig
trace propagation.TraceContext
}
// NewHeaderPropagator creates a generic compatibility propagator. Header names
// are canonicalized and compared case-insensitively. It does not trust or
// inspect a Host header; the caller decides which ingress is trusted.
func NewHeaderPropagator(config HeaderPropagationConfig) (*HeaderPropagator, error) {
traceHeader, err := canonicalPropagationHeader(config.TraceIDHeader)
if err != nil {
return nil, errors.Join(errors.New("telemetry: invalid trace ID header"), err)
}
spanHeader, err := canonicalPropagationHeader(config.SpanIDHeader)
if err != nil {
return nil, errors.Join(errors.New("telemetry: invalid span ID header"), err)
}
requestHeader := ""
if strings.TrimSpace(config.RequestIDHeader) != "" {
requestHeader, err = canonicalPropagationHeader(config.RequestIDHeader)
if err != nil {
return nil, errors.Join(errors.New("telemetry: invalid request ID header"), err)
}
}
if strings.EqualFold(traceHeader, spanHeader) ||
(requestHeader != "" && (strings.EqualFold(traceHeader, requestHeader) || strings.EqualFold(spanHeader, requestHeader))) {
return nil, errors.New("telemetry: propagation headers must be distinct")
}
for _, name := range []string{traceHeader, spanHeader, requestHeader} {
if strings.EqualFold(name, "Traceparent") || strings.EqualFold(name, "Tracestate") {
return nil, errors.New("telemetry: custom headers cannot replace W3C trace context")
}
}
config.TraceIDHeader = traceHeader
config.SpanIDHeader = spanHeader
config.RequestIDHeader = requestHeader
return &HeaderPropagator{config: config}, nil
}
func canonicalPropagationHeader(name string) (string, error) {
name = strings.TrimSpace(name)
if name == "" {
return "", errors.New("header name is required")
}
canonical := http.CanonicalHeaderKey(name)
if canonical == "" || strings.IndexFunc(name, unicode.IsSpace) >= 0 || strings.ContainsAny(name, "\r\n:") {
return "", errors.New("header name is malformed")
}
return canonical, nil
}
// Inject writes W3C trace context first and then the configured compatibility
// headers. Existing caller values are replaced so all propagated IDs describe
// the same current span.
func (p *HeaderPropagator) Inject(ctx context.Context, carrier propagation.TextMapCarrier) {
if p == nil || carrier == nil {
return
}
p.trace.Inject(ctx, carrier)
sc := trace.SpanContextFromContext(ctx)
if sc.IsValid() {
carrier.Set(p.config.TraceIDHeader, sc.TraceID().String())
carrier.Set(p.config.SpanIDHeader, sc.SpanID().String())
}
if p.config.RequestIDHeader != "" {
if requestID := validCorrelationID(logx.RequestID(ctx)); requestID != "" {
carrier.Set(p.config.RequestIDHeader, requestID)
}
}
}
// Extract prefers a valid W3C parent. If it is absent or malformed, a valid
// configured legacy pair becomes a remote parent. Invalid legacy IDs are
// ignored, never copied into logs or a new outbound request.
func (p *HeaderPropagator) Extract(ctx context.Context, carrier propagation.TextMapCarrier) context.Context {
if p == nil || carrier == nil {
return ctx
}
ctx = p.trace.Extract(ctx, carrier)
if p.config.RequestIDHeader != "" {
if requestID := validCorrelationID(carrierValue(carrier, p.config.RequestIDHeader)); requestID != "" {
ctx = logx.WithRequestID(ctx, requestID)
}
}
if trace.SpanContextFromContext(ctx).IsValid() {
return ctx
}
traceID, err := trace.TraceIDFromHex(carrierValue(carrier, p.config.TraceIDHeader))
if err != nil || !traceID.IsValid() {
return ctx
}
spanID, err := trace.SpanIDFromHex(carrierValue(carrier, p.config.SpanIDHeader))
if err != nil || !spanID.IsValid() {
return ctx
}
return trace.ContextWithRemoteSpanContext(ctx, trace.NewSpanContext(trace.SpanContextConfig{
TraceID: traceID, SpanID: spanID, TraceFlags: trace.FlagsSampled, Remote: true,
}))
}
// Fields reports both W3C and compatibility fields in stable order.
func (p *HeaderPropagator) Fields() []string {
if p == nil {
return nil
}
fields := []string{"traceparent", "tracestate", p.config.TraceIDHeader, p.config.SpanIDHeader}
if p.config.RequestIDHeader != "" {
fields = append(fields, p.config.RequestIDHeader)
}
sort.Strings(fields)
return fields
}
func carrierValue(carrier propagation.TextMapCarrier, name string) string {
if value := carrier.Get(name); value != "" {
return value
}
for _, key := range carrier.Keys() {
if strings.EqualFold(key, name) {
return carrier.Get(key)
}
}
return ""
}
func validCorrelationID(value string) string {
if value == "" || value != strings.TrimSpace(value) || len(value) > 128 {
return ""
}
for _, r := range value {
if r >= 'a' && r <= 'z' || r >= 'A' && r <= 'Z' || r >= '0' && r <= '9' || r == '-' || r == '_' || r == '.' || r == ':' {
continue
}
return ""
}
return value
}
var _ propagation.TextMapPropagator = (*HeaderPropagator)(nil)

View File

@@ -21,10 +21,11 @@ import (
)
type Config struct {
Service string `json:"service"`
SampleRatio float64 `json:"sampleRatio"`
Exporter sdktrace.SpanExporter `json:"-"`
Reader sdkmetric.Reader `json:"-"`
Service string `json:"service"`
SampleRatio float64 `json:"sampleRatio"`
Exporter sdktrace.SpanExporter `json:"-"`
Reader sdkmetric.Reader `json:"-"`
Propagator propagation.TextMapPropagator `json:"-"`
}
type Runtime struct {
traces *sdktrace.TracerProvider
@@ -48,7 +49,11 @@ func New(cfg Config) (*Runtime, error) {
if cfg.Reader != nil {
metricOptions = append(metricOptions, sdkmetric.WithReader(cfg.Reader))
}
r := &Runtime{traces: sdktrace.NewTracerProvider(traceOptions...), meters: sdkmetric.NewMeterProvider(metricOptions...), propagator: propagation.TraceContext{}}
propagator := cfg.Propagator
if propagator == nil {
propagator = propagation.TraceContext{}
}
r := &Runtime{traces: sdktrace.NewTracerProvider(traceOptions...), meters: sdkmetric.NewMeterProvider(metricOptions...), propagator: propagator}
r.tracer = r.traces.Tracer("git.yuntuo.hk/k2safe/go-omni")
meter := r.meters.Meter("git.yuntuo.hk/k2safe/go-omni")
r.duration, _ = meter.Float64Histogram("http.server.request.duration", metric.WithUnit("s"), metric.WithExplicitBucketBoundaries(.005, .01, .025, .05, .1, .25, .5, 1, 2.5, 5, 10))
@@ -121,8 +126,14 @@ func (r *Runtime) Middleware() httpx.Middleware {
r.duration.Record(ctx, time.Since(started).Seconds(), metric.WithAttributes(attrs...))
span.End()
// Preserve correlation for outer access-log middleware without
// exposing an ended span as its active span.
c.WithContext(logx.WithTraceID(parent, span.SpanContext().TraceID().String()))
// exposing an ended span as its active span. A custom propagator may
// have supplied a request ID that the built-in RequestID middleware
// did not know about, so carry that value back as well.
outer := logx.WithTraceID(parent, span.SpanContext().TraceID().String())
if requestID := logx.RequestID(ctx); requestID != "" {
outer = logx.WithRequestID(outer, requestID)
}
c.WithContext(outer)
if p != nil {
panic(p)
}

View File

@@ -7,6 +7,7 @@ import (
"git.yuntuo.hk/k2safe/go-omni/logx"
"git.yuntuo.hk/k2safe/go-omni/telemetry"
"github.com/redis/go-redis/v9"
"go.opentelemetry.io/otel/propagation"
sdkmetric "go.opentelemetry.io/otel/sdk/metric"
"go.opentelemetry.io/otel/sdk/metric/metricdata"
"go.opentelemetry.io/otel/sdk/trace/tracetest"
@@ -115,3 +116,115 @@ func TestHTTPTracePropagationMetricsAndRedisNil(t *testing.T) {
t.Fatal("request metric", count)
}
}
func TestConfiguredHeaderPropagation(t *testing.T) {
exporter := tracetest.NewInMemoryExporter()
propagator, err := telemetry.NewHeaderPropagator(telemetry.HeaderPropagationConfig{
TraceIDHeader: "X-Trace-Id", SpanIDHeader: "X-Span-Id", RequestIDHeader: "X-Request-Id",
})
if err != nil {
t.Fatal(err)
}
runtime, err := telemetry.New(telemetry.Config{Service: "header-test", SampleRatio: 1, Exporter: exporter, Propagator: propagator})
if err != nil {
t.Fatal(err)
}
defer runtime.Close()
const traceID = "11111111111111111111111111111111"
const parentSpanID = "2222222222222222"
const requestID = "gateway-request-1"
upstream := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if r.Header.Get("X-Trace-Id") != traceID || r.Header.Get("X-Request-Id") != requestID {
t.Errorf("custom propagation = trace %q request %q", r.Header.Get("X-Trace-Id"), r.Header.Get("X-Request-Id"))
}
if r.Header.Get("X-Span-Id") == parentSpanID || len(r.Header.Get("X-Span-Id")) != 16 || !strings.HasPrefix(r.Header.Get("Traceparent"), "00-"+traceID+"-") {
t.Errorf("outbound parent propagation = %q", r.Header)
}
io.WriteString(w, "ok")
}))
defer upstream.Close()
app := omni.New(omni.WithLogger(false), omni.WithRequestIDHeader("X-Request-Id"))
defer app.Close()
app.Use(runtime.Middleware())
app.GET("/", func(c *httpx.Context) error {
if got := logx.TraceID(c.Context()); got != traceID {
t.Errorf("inbound trace correlation = %q", got)
}
if got := logx.RequestID(c.Context()); got != requestID {
t.Errorf("inbound request correlation = %q", got)
}
req, _ := http.NewRequestWithContext(c.Context(), http.MethodGet, upstream.URL, nil)
resp, err := (telemetry.Transport{Runtime: runtime}).RoundTrip(req)
if err != nil {
return err
}
defer resp.Body.Close()
_, _ = io.ReadAll(resp.Body)
return c.OK("ok")
})
req := httptest.NewRequest(http.MethodGet, "/", nil)
req.Header.Set("X-Trace-Id", traceID)
req.Header.Set("X-Span-Id", parentSpanID)
req.Header.Set("X-Request-Id", requestID)
rec := httptest.NewRecorder()
app.ServeHTTP(rec, req)
if rec.Code != http.StatusOK || rec.Header().Get("X-Request-Id") != requestID {
t.Fatalf("response status/request ID = %d/%q", rec.Code, rec.Header().Get("X-Request-Id"))
}
if err := runtime.Flush(context.Background()); err != nil {
t.Fatal(err)
}
spans := exporter.GetSpans()
var server, client *tracetest.SpanStub
for _, span := range spans {
switch span.SpanKind {
case trace.SpanKindServer:
span := span
server = &span
case trace.SpanKindClient:
span := span
client = &span
}
}
if server == nil || client == nil {
t.Fatalf("missing spans: %d", len(spans))
}
if server.SpanContext.TraceID().String() != traceID || server.Parent.TraceID().String() != traceID || server.Parent.SpanID().String() != parentSpanID || !server.Parent.IsRemote() {
t.Fatalf("server parent = %s/%s remote=%v", server.SpanContext.TraceID(), server.Parent.SpanID(), server.Parent.IsRemote())
}
if client.SpanContext.TraceID().String() != traceID || client.Parent.SpanID() != server.SpanContext.SpanID() {
t.Fatalf("client linkage = trace %s parent %s", client.SpanContext.TraceID(), client.Parent.SpanID())
}
}
func TestConfiguredHeaderPropagationPrefersW3CAndRejectsInvalidSetup(t *testing.T) {
propagator, err := telemetry.NewHeaderPropagator(telemetry.HeaderPropagationConfig{TraceIDHeader: "x-trace-id", SpanIDHeader: "x-span-id"})
if err != nil {
t.Fatal(err)
}
carrier := propagation.MapCarrier{
"traceparent": "00-aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa-bbbbbbbbbbbbbbbb-01",
"X-Trace-Id": "11111111111111111111111111111111",
"X-Span-Id": "2222222222222222",
}
ctx := propagator.Extract(context.Background(), carrier)
parent := trace.SpanContextFromContext(ctx)
if parent.TraceID().String() != "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa" || parent.SpanID().String() != "bbbbbbbbbbbbbbbb" {
t.Fatalf("W3C parent was not preferred: %s/%s", parent.TraceID(), parent.SpanID())
}
for _, config := range []telemetry.HeaderPropagationConfig{
{TraceIDHeader: "", SpanIDHeader: "X-Span-Id"},
{TraceIDHeader: "X-Trace-Id", SpanIDHeader: ""},
{TraceIDHeader: "X-Trace-Id", SpanIDHeader: "X-Trace-Id"},
{TraceIDHeader: "traceparent", SpanIDHeader: "X-Span-Id"},
{TraceIDHeader: "X Trace", SpanIDHeader: "X-Span-Id"},
} {
if _, err := telemetry.NewHeaderPropagator(config); err == nil {
t.Fatalf("invalid propagation config accepted: %+v", config)
}
}
invalid := propagation.MapCarrier{"X-Trace-Id": "bad", "X-Span-Id": "bad"}
if got := trace.SpanContextFromContext(propagator.Extract(context.Background(), invalid)); got.IsValid() {
t.Fatal("invalid legacy IDs became a parent")
}
}