Detect OTel instrumentations in applications

.gitignore

@@ -79,4 +79,6 @@ test/integration/components/gokafka/vendor/*
 test/integration/components/gokafka-seg/vendor/*
 test/integration/components/gokafka-seg/gokafka
 test/integration/components/goredis/vendor/*
-test/integration/components/goredis/goredis
+test/integration/components/goredis/goredis
+test/integration/components/go_otel/rolldice
+test/integration/components/go_otel_grpc/rolldice

docs/sources/configure/options.md

@@ -225,13 +225,37 @@ namespace.
 For more details about this section, go to the [discovery services section](#discovery-services-section)
 of this document.
 
+| YAML               | Environment variable | Type            | Default |
+| ------------------ | ------- | --------------- | ------- |
+| `exclude_services` | N/A     | list of objects | (unset) |
+
+This section allows for specifying selection criteria for excluding services from 
+being instrumented. It follows the same definition format as described in the 
+[discovery services section](#discovery-services-section) of this document.
+
+This option is useful for avoiding instrumentation of services which are typically
+found in observability environments. For example, use this option to exclude instrumenting
+Prometheus, the OpenTelemetry collector or Grafana Alloy.
+
 | YAML                       | Environment variable             | Type    | Default |
 | -------------------------- | -------------------------------- | ------- | ------- |
 | `skip_go_specific_tracers` | `BEYLA_SKIP_GO_SPECIFIC_TRACERS` | boolean | false   |
 
-Disables the detection of Go specifics when ebpf tracer inspects executables to be instrumented.
+Disables the detection of Go specifics when the **ebpf** tracer inspects executables to be instrumented.
 The tracer will fallback to using generic instrumentation, which will generally be less efficient.
 
+| YAML                                 | Environment variable                       | Type    | Default |
+| ------------------------------------ | ------------------------------------------ | ------- | ------- |
+| `exclude_otel_instrumented_services` | `BEYLA_EXCLUDE_OTEL_INSTRUMENTED_SERVICES` | boolean | true    |
+
+Disables Beyla instrumentation of services which are already instrumented with OpenTelemetry. Since Beyla
+is often deployed to monitor all services in a Kubernetes cluster, monitoring already instrumented services
+can lead to duplicate telemetry data, unless the instrumentation selection (or exclusion) criteria is 
+carefully crafted. To avoid unnecessary configuration overhead, Beyla monitors for the OpenTelemetry SDK calls
+to publish metrics and traces, and automatically turns off instrumentation of services which publish their own
+telemetry data. Turn this option off if your application generated telemetry data doesn't conflict with the
+Beyla generated metrics and traces.
+
 ### Discovery services section
 
 Example of YAML file allowing the selection of multiple groups of services:

pkg/beyla/config.go

@@ -123,6 +123,9 @@ var DefaultConfig = Config{
 		RunMode:  process.RunModePrivileged,
 		Interval: 5 * time.Second,
 	},
+	Discovery: services.DiscoveryConfig{
+		ExcludeOTelInstrumentedServices: true,
+	},
 }
 
 type Config struct {

pkg/beyla/config_test.go

@@ -25,6 +25,7 @@ import (
 	"github.com/grafana/beyla/pkg/internal/netolly/transform/cidr"
 	"github.com/grafana/beyla/pkg/internal/traces"
 	"github.com/grafana/beyla/pkg/kubeflags"
+	"github.com/grafana/beyla/pkg/services"
 	"github.com/grafana/beyla/pkg/transform"
 )
 
@@ -200,6 +201,9 @@ network:
 			RunMode:  process.RunModePrivileged,
 			Interval: 5 * time.Second,
 		},
+		Discovery: services.DiscoveryConfig{
+			ExcludeOTelInstrumentedServices: true,
+		},
 	}, cfg)
 }
 

pkg/export/alloy/traces.go

@@ -30,6 +30,10 @@ type tracesReceiver struct {
 	hostID     string
 }
 
+func (tr *tracesReceiver) spanDiscarded(span *request.Span) bool {
+	return span.IgnoreTraces() || span.ServiceID.ExportsOTelTraces()
+}
+
 func (tr *tracesReceiver) provideLoop() (pipe.FinalFunc[[]request.Span], error) {
 	if !tr.cfg.Enabled() {
 		return pipe.IgnoreFinal[[]request.Span](), nil
@@ -45,7 +49,7 @@ func (tr *tracesReceiver) provideLoop() (pipe.FinalFunc[[]request.Span], error)
 		for spans := range in {
 			for i := range spans {
 				span := &spans[i]
-				if span.IgnoreTraces() {
+				if tr.spanDiscarded(span) {
 					continue
 				}
 

pkg/export/alloy/traces_test.go

@@ -0,0 +1,81 @@
+package alloy
+
+import (
+	"context"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"go.opentelemetry.io/collector/pdata/ptrace"
+	"go.opentelemetry.io/otel/attribute"
+
+	"github.com/grafana/beyla/pkg/beyla"
+	"github.com/grafana/beyla/pkg/export/attributes"
+	"github.com/grafana/beyla/pkg/export/otel"
+	"github.com/grafana/beyla/pkg/internal/request"
+	"github.com/grafana/beyla/pkg/internal/svc"
+)
+
+func TestTracesSkipsInstrumented(t *testing.T) {
+	svcNoExport := svc.ID{}
+
+	svcNoExportTraces := svc.ID{}
+	svcNoExportTraces.SetExportsOTelMetrics()
+
+	svcExportTraces := svc.ID{}
+	svcExportTraces.SetExportsOTelTraces()
+
+	tests := []struct {
+		name     string
+		spans    []request.Span
+		filtered bool
+	}{
+		{
+			name:     "Foo span is not filtered",
+			spans:    []request.Span{{ServiceID: svcNoExport, Type: request.EventTypeHTTPClient, Method: "GET", Route: "/foo", RequestStart: 100, End: 200}},
+			filtered: false,
+		},
+		{
+			name:     "/v1/metrics span is not filtered",
+			spans:    []request.Span{{ServiceID: svcNoExportTraces, Type: request.EventTypeHTTPClient, Method: "GET", Route: "/v1/metrics", RequestStart: 100, End: 200}},
+			filtered: false,
+		},
+		{
+			name:     "/v1/traces span is filtered",
+			spans:    []request.Span{{ServiceID: svcExportTraces, Type: request.EventTypeHTTPClient, Method: "GET", Route: "/v1/traces", RequestStart: 100, End: 200}},
+			filtered: true,
+		},
+	}
+
+	tr := makeTracesTestReceiver()
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			traces := generateTracesForSpans(t, tr, tt.spans)
+			assert.Equal(t, tt.filtered, len(traces) == 0, tt.name)
+		})
+	}
+}
+
+func makeTracesTestReceiver() *tracesReceiver {
+	return &tracesReceiver{
+		ctx:        context.Background(),
+		cfg:        &beyla.TracesReceiverConfig{},
+		attributes: attributes.Selection{},
+		hostID:     "Alloy",
+	}
+}
+
+func generateTracesForSpans(t *testing.T, tr *tracesReceiver, spans []request.Span) []ptrace.Traces {
+	res := []ptrace.Traces{}
+	traceAttrs, err := otel.GetUserSelectedAttributes(tr.attributes)
+	assert.NoError(t, err)
+	for i := range spans {
+		span := &spans[i]
+		if tr.spanDiscarded(span) {
+			continue
+		}
+		res = append(res, otel.GenerateTraces(span, tr.hostID, traceAttrs, []attribute.KeyValue{}))
+	}
+
+	return res
+}

pkg/export/otel/metrics.go

@@ -756,12 +756,16 @@ func (mr *MetricsReporter) serviceGraphAttributes() []attributes.Field[*request.
 		})
 }
 
+func otelSpanAccepted(span *request.Span, mr *MetricsReporter) bool {
+	return mr.cfg.OTelMetricsEnabled() && !span.ServiceID.ExportsOTelMetrics()
+}
+
 // nolint:cyclop
 func (r *Metrics) record(span *request.Span, mr *MetricsReporter) {
 	t := span.Timings()
 	duration := t.End.Sub(t.RequestStart).Seconds()
 
-	if mr.cfg.OTelMetricsEnabled() {
+	if otelSpanAccepted(span, mr) {
 		switch span.Type {
 		case request.EventTypeHTTP:
 			if mr.is.HTTPEnabled() {

pkg/export/otel/metrics_test.go

@@ -571,6 +571,51 @@ func TestMetricsConfig_Disabled(t *testing.T) {
 	assert.False(t, (&MetricsConfig{Grafana: &GrafanaOTLP{Submit: []string{"traces", "metrics"}, InstanceID: "33221"}}).Enabled())
 }
 
+func TestSpanMetricsDiscarded(t *testing.T) {
+	mc := MetricsConfig{
+		Features: []string{FeatureApplication},
+	}
+	mr := MetricsReporter{
+		cfg: &mc,
+	}
+
+	svcNoExport := svc.ID{}
+
+	svcExportMetrics := svc.ID{}
+	svcExportMetrics.SetExportsOTelMetrics()
+
+	svcExportTraces := svc.ID{}
+	svcExportTraces.SetExportsOTelTraces()
+
+	tests := []struct {
+		name      string
+		span      request.Span
+		discarded bool
+	}{
+		{
+			name:      "Foo span is not filtered",
+			span:      request.Span{ServiceID: svcNoExport, Type: request.EventTypeHTTPClient, Method: "GET", Route: "/foo", RequestStart: 100, End: 200},
+			discarded: false,
+		},
+		{
+			name:      "/v1/metrics span is filtered",
+			span:      request.Span{ServiceID: svcExportMetrics, Type: request.EventTypeHTTPClient, Method: "GET", Route: "/v1/metrics", RequestStart: 100, End: 200},
+			discarded: true,
+		},
+		{
+			name:      "/v1/traces span is not filtered",
+			span:      request.Span{ServiceID: svcExportTraces, Type: request.EventTypeHTTPClient, Method: "GET", Route: "/v1/traces", RequestStart: 100, End: 200},
+			discarded: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			assert.Equal(t, tt.discarded, !otelSpanAccepted(&tt.span, &mr), tt.name)
+		})
+	}
+}
+
 func (f *fakeInternalMetrics) OTELMetricExport(len int) {
 	fakeMux.Lock()
 	defer fakeMux.Unlock()

pkg/export/otel/traces.go

@@ -165,6 +165,10 @@ func GetUserSelectedAttributes(attrs attributes.Selection) (map[attr.Name]struct
 	return traceAttrs, err
 }
 
+func (tr *tracesOTELReceiver) spanDiscarded(span *request.Span) bool {
+	return span.IgnoreTraces() || span.ServiceID.ExportsOTelTraces() || !tr.acceptSpan(span)
+}
+
 func (tr *tracesOTELReceiver) provideLoop() (pipe.FinalFunc[[]request.Span], error) {
 	if !tr.cfg.Enabled() {
 		return pipe.IgnoreFinal[[]request.Span](), nil
@@ -199,7 +203,7 @@ func (tr *tracesOTELReceiver) provideLoop() (pipe.FinalFunc[[]request.Span], err
 		for spans := range in {
 			for i := range spans {
 				span := &spans[i]
-				if span.IgnoreTraces() || !tr.acceptSpan(span) {
+				if tr.spanDiscarded(span) {
 					continue
 				}
 				traces := GenerateTraces(span, tr.ctxInfo.HostID, traceAttrs, envResourceAttrs)

pkg/export/otel/traces_test.go

@@ -990,6 +990,47 @@ func TestTracesAttrReuse(t *testing.T) {
 	}
 }
 
+func TestTracesSkipsInstrumented(t *testing.T) {
+	svcNoExport := svc.ID{}
+
+	svcNoExportTraces := svc.ID{}
+	svcNoExportTraces.SetExportsOTelMetrics()
+
+	svcExportTraces := svc.ID{}
+	svcExportTraces.SetExportsOTelTraces()
+
+	tests := []struct {
+		name     string
+		spans    []request.Span
+		filtered bool
+	}{
+		{
+			name:     "Foo span is not filtered",
+			spans:    []request.Span{{ServiceID: svcNoExport, Type: request.EventTypeHTTPClient, Method: "GET", Route: "/foo", RequestStart: 100, End: 200}},
+			filtered: false,
+		},
+		{
+			name:     "/v1/metrics span is not filtered",
+			spans:    []request.Span{{ServiceID: svcNoExportTraces, Type: request.EventTypeHTTPClient, Method: "GET", Route: "/v1/metrics", RequestStart: 100, End: 200}},
+			filtered: false,
+		},
+		{
+			name:     "/v1/traces span is filtered",
+			spans:    []request.Span{{ServiceID: svcExportTraces, Type: request.EventTypeHTTPClient, Method: "GET", Route: "/v1/traces", RequestStart: 100, End: 200}},
+			filtered: true,
+		},
+	}
+
+	tr := makeTracesTestReceiver([]string{instrumentations.InstrumentationALL})
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			traces := generateTracesForSpans(t, tr, tt.spans)
+			assert.Equal(t, tt.filtered, len(traces) == 0, tt.name)
+		})
+	}
+}
+
 type fakeInternalTraces struct {
 	imetrics.NoopReporter
 	sum  atomic.Int32
@@ -1183,7 +1224,7 @@ func generateTracesForSpans(t *testing.T, tr *tracesOTELReceiver, spans []reques
 	assert.NoError(t, err)
 	for i := range spans {
 		span := &spans[i]
-		if span.IgnoreTraces() || !tr.acceptSpan(span) {
+		if tr.spanDiscarded(span) {
 			continue
 		}
 		res = append(res, GenerateTraces(span, "host-id", traceAttrs, []attribute.KeyValue{}))

pkg/export/prom/prom.go

@@ -575,6 +575,10 @@ func (r *metricsReporter) collectMetrics(input <-chan []request.Span) {
 	}
 }
 
+func (r *metricsReporter) otelSpanObserved(span *request.Span) bool {
+	return r.cfg.OTelMetricsEnabled() && !span.ServiceID.ExportsOTelMetrics()
+}
+
 // nolint:cyclop
 func (r *metricsReporter) observe(span *request.Span) {
 	t := span.Timings()
@@ -584,7 +588,7 @@ func (r *metricsReporter) observe(span *request.Span) {
 	targetInfoLabelValues := r.labelValuesTargetInfo(span.ServiceID)
 	r.targetInfo.WithLabelValues(targetInfoLabelValues...).metric.Set(1)
 
-	if r.cfg.OTelMetricsEnabled() {
+	if r.otelSpanObserved(span) {
 		switch span.Type {
 		case request.EventTypeHTTP:
 			if r.is.HTTPEnabled() {
@@ -637,6 +641,7 @@ func (r *metricsReporter) observe(span *request.Span) {
 			}
 		}
 	}
+
 	if r.cfg.SpanMetricsEnabled() {
 		lv := r.labelValuesSpans(span)
 		r.spanMetricsLatency.WithLabelValues(lv...).metric.Observe(duration)