Skip to content

README.md

Latest commit

 

History

History
90 lines (70 loc) · 2.61 KB

README.md

File metadata and controls

90 lines (70 loc) · 2.61 KB
Raw

client_golang Tutorial: Export Prometheus instrumentation with OpenTelemetry OTLP

client_golang is built around the Prometheus pull model. For most long-running services, exposing /metrics with promhttp is still the recommended setup.

If you already instrumented your code with client_golang but need to send metrics into an OpenTelemetry pipeline, you can bridge a Prometheus Gatherer into the OpenTelemetry SDK and export it with OTLP.

This is useful for:

  • short-lived batch jobs that may exit before being scraped
  • environments standardized on OTLP and the OpenTelemetry Collector
  • gradual migrations where existing Prometheus instrumentation should stay unchanged

Example

The example below keeps existing Prometheus instrumentation, bridges a custom registry into the OpenTelemetry SDK, and exports the resulting metrics with the OTLP HTTP exporter.

package main

import (
	"context"
	"log"

	bridgeprometheus "go.opentelemetry.io/contrib/bridges/prometheus"
	"go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetrichttp"
	sdkmetric "go.opentelemetry.io/otel/sdk/metric"

	"github.com/prometheus/client_golang/prometheus"
)

func main() {
	ctx := context.Background()

	reg := prometheus.NewRegistry()
	jobsProcessed := prometheus.NewCounter(prometheus.CounterOpts{
		Name: "jobs_processed_total",
		Help: "Total number of processed jobs.",
	})
	reg.MustRegister(jobsProcessed)

	exporter, err := otlpmetrichttp.New(
		ctx,
		otlpmetrichttp.WithEndpoint("localhost:4318"),
		otlpmetrichttp.WithInsecure(),
	)
	if err != nil {
		log.Fatal(err)
	}

	bridge := bridgeprometheus.NewMetricProducer(
		bridgeprometheus.WithGatherer(reg),
	)

	meterProvider := sdkmetric.NewMeterProvider(
		sdkmetric.WithReader(
			sdkmetric.NewPeriodicReader(exporter, sdkmetric.WithProducer(bridge)),
		),
	)
	defer func() {
		if err := meterProvider.Shutdown(ctx); err != nil {
			log.Printf("shutdown metric provider: %v", err)
		}
	}()

	jobsProcessed.Inc()
}

Notes

  • If your application uses the default Prometheus registry, you can omit bridgeprometheus.WithGatherer(reg) and NewMetricProducer will read from prometheus.DefaultGatherer.
  • Shutdown flushes pending metric data. For short-lived jobs, make sure you shut the provider down before exit.
  • If you also use native OpenTelemetry metrics, attach them to the same MeterProvider so both pipelines are exported together.

Related documentation