add cli flags

flags/flags.go

@@ -113,6 +113,9 @@ type Flags struct {
 	RubyUnwindingDisable   bool                `default:"false" help:"[deprecated] Disable Ruby unwinder."`
 	JavaUnwindingDisable   bool                `default:"true"  help:"[deprecated] Disable Java unwinder."`
 
+	// which metrics producers (e.g. nvidia) to enable
+	MetricsProducer FlagsMetricProducer `embed:"" prefix:"metrics-producer-"`
+
 	CollectCustomLabels bool `default:"false" help:"Attempt to collect custom labels (e.g. trace ID) from the process."`
 
 	AnalyticsOptOut bool `default:"false" help:"Opt out of sending anonymous usage statistics."`
@@ -311,6 +314,12 @@ type FlagsDWARFUnwinding struct {
 	Mixed   bool `default:"true"                                    help:"[deprecated] Unwind using .eh_frame information and frame pointers."`
 }
 
+// FlagsMetricProducer contains flags that configure arrow metrics production.
+type FlagsMetricProducer struct {
+	NvidiaGpu     bool `default:"false" help:"Collect metrics related to Nvidia GPUs."`
+	NvidiaGpuMock bool `default:"false" help:"Generate fake Nvidia GPU metrics." hidden:""`
+}
+
 type FlagsTelemetry struct {
 	DisablePanicReporting bool  `default:"false"`
 	StderrBufferSizeKb    int64 `default:"4096"`

main.go

@@ -293,17 +293,30 @@ func mainWithExitCode() flags.ExitCode {
 
 	arrowClient := arrowpb.NewArrowMetricsServiceClient(grpcConn)
 	arrowMetricsExporter := arrowmetrics.NewExporter(arrowClient, time.Second*10, map[string]any{"foo": "bar"})
-	// nvidia, err := arrowmetrics.NewNvidiaProducer()
-	nvidia := arrowmetrics.NewNvidiaMockProducer(3, time.Now())
-	err = nil
-	if err != nil {
-		// XXX
-		panic(err)
+	const nvidiaMetricsScopeName = "parca.nvidia_gpu_metrics"
+	if f.MetricsProducer.NvidiaGpu {
+		nvidia, err := arrowmetrics.NewNvidiaProducer()
+		if err != nil {
+			return flags.Failure("Failed to instantiate nvidia metrics producer: %v. Are the Nvidia drivers installed?", err)
+		}
+		arrowMetricsExporter.AddProducer(arrowmetrics.ProducerConfig{
+			Producer:  nvidia,
+			ScopeName: nvidiaMetricsScopeName,
+		})
+	}
+	if f.MetricsProducer.NvidiaGpuMock {
+		mock := arrowmetrics.NewNvidiaMockProducer(3, time.Now())
+		scopeName := nvidiaMetricsScopeName
+		if f.MetricsProducer.NvidiaGpu {
+			// don't conflict with the real producer
+			scopeName = scopeName + "_mock"
+		}
+		arrowMetricsExporter.AddProducer(arrowmetrics.ProducerConfig{
+			Producer:  mock,
+			ScopeName: scopeName,
+		})
+
 	}
-	arrowMetricsExporter.AddProducer(arrowmetrics.ProducerConfig{
-		Producer: nvidia,
-		ScopeName: "parca.nvidia_gpu_metrics",
-	})
 	arrowMetricsExporter.Start(ctx)
 
 	// Network operations to CA start here