API Server: Fix stack overflow panic when tracing is disabled (#90075)

pkg/cmd/grafana/apiserver/cmd.go

@@ -3,6 +3,7 @@ package apiserver
 import (
 	"context"
 	"os"
+	"sync"
 
 	"github.com/spf13/cobra"
 	"go.opentelemetry.io/otel"
@@ -116,16 +117,17 @@ type lateInitializedTracingProvider struct {
 }
 
 func (tp lateInitializedTracingProvider) Tracer(name string, options ...trace.TracerOption) trace.Tracer {
-	return tp.tracer
+	return tp.tracer.getTracer()
 }
 
 type lateInitializedTracingService struct {
 	tracing.Tracer
+	mutex sync.RWMutex
 }
 
 func newLateInitializedTracingService() *lateInitializedTracingService {
 	ts := &lateInitializedTracingService{
-		Tracer: tracing.InitializeTracerForTest(),
+		Tracer: tracing.NewNoopTracerService(),
 	}
 
 	tp := &lateInitializedTracingProvider{
@@ -137,8 +139,17 @@ func newLateInitializedTracingService() *lateInitializedTracingService {
 	return ts
 }
 
-func (s *lateInitializedTracingService) InitTracer(tracer tracing.Tracer) {
+func (s *lateInitializedTracingService) getTracer() tracing.Tracer {
+	s.mutex.RLock()
+	t := s.Tracer
+	s.mutex.RUnlock()
+	return t
+}
+
+func (s *lateInitializedTracingService) InitTracer(tracer *tracing.TracingService) {
+	s.mutex.Lock()
 	s.Tracer = tracer
+	s.mutex.Unlock()
 }
 
 var _ tracing.Tracer = &lateInitializedTracingService{}

pkg/infra/tracing/tracing.go

@@ -99,38 +99,20 @@ func ProvideService(tracingCfg *TracingConfig) (*TracingService, error) {
 	return ots, nil
 }
 
+func NewNoopTracerService() *TracingService {
+	tp := &noopTracerProvider{TracerProvider: noop.NewTracerProvider()}
+	otel.SetTracerProvider(tp)
+
+	cfg := NewEmptyTracingConfig()
+	ots := &TracingService{cfg: cfg, tracerProvider: tp}
+	_ = ots.initOpentelemetryTracer()
+	return ots
+}
+
 func (ots *TracingService) GetTracerProvider() tracerProvider {
 	return ots.tracerProvider
 }
 
-func TraceIDFromContext(ctx context.Context, requireSampled bool) string {
-	spanCtx := trace.SpanContextFromContext(ctx)
-	if !spanCtx.HasTraceID() || !spanCtx.IsValid() || (requireSampled && !spanCtx.IsSampled()) {
-		return ""
-	}
-
-	return spanCtx.TraceID().String()
-}
-
-// Error sets the status to error and record the error as an exception in the provided span.
-func Error(span trace.Span, err error) error {
-	attr := []attribute.KeyValue{}
-	grafanaErr := errutil.Error{}
-	if errors.As(err, &grafanaErr) {
-		attr = append(attr, attribute.String("message_id", grafanaErr.MessageID))
-	}
-
-	span.SetStatus(codes.Error, err.Error())
-	span.RecordError(err, trace.WithAttributes(attr...))
-	return err
-}
-
-// Errorf wraps fmt.Errorf and also sets the status to error and record the error as an exception in the provided span.
-func Errorf(span trace.Span, format string, args ...any) error {
-	err := fmt.Errorf(format, args...)
-	return Error(span, err)
-}
-
 type noopTracerProvider struct {
 	trace.TracerProvider
 }
@@ -389,3 +371,31 @@ func (rl *rateLimiter) ShouldSample(p tracesdk.SamplingParameters) tracesdk.Samp
 }
 
 func (rl *rateLimiter) Description() string { return rl.description }
+
+func TraceIDFromContext(ctx context.Context, requireSampled bool) string {
+	spanCtx := trace.SpanContextFromContext(ctx)
+	if !spanCtx.HasTraceID() || !spanCtx.IsValid() || (requireSampled && !spanCtx.IsSampled()) {
+		return ""
+	}
+
+	return spanCtx.TraceID().String()
+}
+
+// Error sets the status to error and record the error as an exception in the provided span.
+func Error(span trace.Span, err error) error {
+	attr := []attribute.KeyValue{}
+	grafanaErr := errutil.Error{}
+	if errors.As(err, &grafanaErr) {
+		attr = append(attr, attribute.String("message_id", grafanaErr.MessageID))
+	}
+
+	span.SetStatus(codes.Error, err.Error())
+	span.RecordError(err, trace.WithAttributes(attr...))
+	return err
+}
+
+// Errorf wraps fmt.Errorf and also sets the status to error and record the error as an exception in the provided span.
+func Errorf(span trace.Span, format string, args ...any) error {
+	err := fmt.Errorf(format, args...)
+	return Error(span, err)
+}