Exporting Metrics and Traces with OpenCensus, Zipkin, and Prometheus

This tutorial provides a minimum example to verify that metrics and traces can be exported to OpenCensus from Go tools.

Setting up oragent

  1. Ensure you have docker and docker-compose.
  2. Clone oragent.
  3. In the oragent directory, start the services:
docker-compose up

If everything goes well, you should see output resembling the following:

Starting oragent_zipkin_1 ... done
Starting oragent_oragent_1 ... done
Starting oragent_prometheus_1 ... done
  • You can check the status of the OpenCensus agent using zPages at http://localhost:55679/debug/tracez.
  • You can now access the Prometheus UI at http://localhost:9445.
  • You can now access the Zipkin UI at http://localhost:9444.
  1. To shut down oragent, hit Ctrl+C in the terminal.
  2. You can also start oragent in detached mode by running docker-compose up -d. To stop oragent while detached, run docker-compose down.

Exporting Metrics and Traces

  1. Clone the tools subrepository.
  2. Inside internal, create a file named main.go with the following contents:
package main

import (


type testExporter struct {
	metrics metric.Exporter
	ocagent *ocagent.Exporter

func (e *testExporter) ProcessEvent(ctx context.Context, ev event.Event) (context.Context, event.Event) {
	ctx, ev = export.Tag(ctx, ev)
	ctx, ev = export.ContextSpan(ctx, ev)
	ctx, ev = e.metrics.ProcessEvent(ctx, ev)
	ctx, ev = e.ocagent.ProcessEvent(ctx, ev)
	return ctx, ev

func main() {
	exporter := &testExporter{}

	exporter.ocagent = ocagent.Connect(&ocagent.Config{
		Start:   time.Now(),
		Address: "",
		Service: "go-tools-test",
		Rate:    5 * time.Second,
		Client:  &http.Client{},

	ctx := context.TODO()
	mLatency := event.NewFloat64Key("latency", "the latency in milliseconds")
	distribution := metric.HistogramFloat64Data{
		Info: &metric.HistogramFloat64{
			Name:        "latencyDistribution",
			Description: "the various latencies",
			Buckets:     []float64{0, 10, 50, 100, 200, 400, 800, 1000, 1400, 2000, 5000, 10000, 15000},

	distribution.Info.Record(&exporter.metrics, mLatency)

	for {
		sleep := randomSleep()
		_, end := event.StartSpan(ctx, "main.randomSleep()")
		time.Sleep(time.Duration(sleep) * time.Millisecond)
		event.Record(ctx, mLatency.Of(float64(sleep)))

		fmt.Println("Latency: ", float64(sleep))

func randomSleep() int64 {
	var max int64
	switch modulus := time.Now().Unix() % 5; modulus {
	case 0:
		max = 17001
	case 1:
		max = 8007
	case 2:
		max = 917
	case 3:
		max = 87
	case 4:
		max = 1173
	return rand.Int63n(max)

  1. Run the new file from within the tools repository:
go run internal/main.go
  1. After about 5 seconds, OpenCensus should start receiving your new metrics, which you can see at http://localhost:8844/metrics. This page will look similar to the following:
# HELP promdemo_latencyDistribution the various latencies
# TYPE promdemo_latencyDistribution histogram
promdemo_latencyDistribution_bucket{vendor="otc",le="0"} 0
promdemo_latencyDistribution_bucket{vendor="otc",le="10"} 2
promdemo_latencyDistribution_bucket{vendor="otc",le="50"} 9
promdemo_latencyDistribution_bucket{vendor="otc",le="100"} 22
promdemo_latencyDistribution_bucket{vendor="otc",le="200"} 35
promdemo_latencyDistribution_bucket{vendor="otc",le="400"} 49
promdemo_latencyDistribution_bucket{vendor="otc",le="800"} 63
promdemo_latencyDistribution_bucket{vendor="otc",le="1000"} 78
promdemo_latencyDistribution_bucket{vendor="otc",le="1400"} 93
promdemo_latencyDistribution_bucket{vendor="otc",le="2000"} 108
promdemo_latencyDistribution_bucket{vendor="otc",le="5000"} 123
promdemo_latencyDistribution_bucket{vendor="otc",le="10000"} 138
promdemo_latencyDistribution_bucket{vendor="otc",le="15000"} 153
promdemo_latencyDistribution_bucket{vendor="otc",le="+Inf"} 15
promdemo_latencyDistribution_sum{vendor="otc"} 1641
promdemo_latencyDistribution_count{vendor="otc"} 15
  1. After a few more seconds, Prometheus should start displaying your new metrics. You can view the distribution at http://localhost:9445/graph?g0.range_input=5m&g0.stacked=1&g0.expr=rate(oragent_latencyDistribution_bucket%5B5m%5D)&

  2. Zipkin should also start displaying traces. You can view them at http://localhost:9444/zipkin/?limit=10&lookback=300000&serviceName=go-tools-test.



Package ocagent adds the ability to export all telemetry to an ocagent. This keeps the compile time dependencies to zero and allows the agent to have the exporters needed for telemetry aggregation and viewing systems.



This section is empty.


This section is empty.


This section is empty.


type Config

type Config struct {
	Start   time.Time
	Host    string
	Process uint32
	Client  *http.Client
	Service string
	Address string
	Rate    time.Duration

func Discover

func Discover() *Config

Discover finds the local agent to export to, it will return nil if there is not one running. TODO: Actually implement a discovery protocol rather than a hard coded address

type Exporter

type Exporter struct {
	// contains filtered or unexported fields

func Connect

func Connect(config *Config) *Exporter

Connect creates a process specific exporter with the specified serviceName and the address of the ocagent to which it will upload its telemetry.

func (*Exporter) Flush

func (e *Exporter) Flush()

func (*Exporter) ProcessEvent

func (e *Exporter) ProcessEvent(ctx context.Context, ev core.Event, lm label.Map) context.Context


Path Synopsis