Add custom traces and metrics to your app

This document describes how to add observability code to your application by using OpenTelemetry. OpenTelemetry provides instrumentation libraries that generate telemetry for popular frameworks. You can augment the library-generated telemetry by adding custom instrumentation that measures your application-specific behavior.

The principles and concepts described in this document can be applied to apps written in all languages supported by OpenTelemetry.

The sample code, which is the same Go app that is described in Getting Started guide for Go, is available in GitHub. To view the full sample, click More, and then select View on GitHub.

Create custom traces

To generate custom traces from your application, you add instrumentation code that creates OpenTelemetry spans. In OpenTelemetry, spans are the building blocks for traces.

To create a span, do the following:

  1. Modify your app to acquire an OpenTelemetry Tracer. In OpenTelemetry, a tracer is a creator of spans. You can acquire a tracer as demonstrated in the following code:

    const scopeName = "github.com/GoogleCloudPlatform/golang-samples/opentelemetry/instrumentation/app/work"

var (
	meter                = otel.Meter(scopeName)
	tracer               = otel.Tracer(scopeName)
	sleepHistogram       metric.Float64Histogram
	subRequestsHistogram metric.Int64Histogram
)

    The tracer name, which is represented by scopeName, identifies the instrumentation scope of the generated traces.

  2. Use the tracer instance to create spans. In the following code sample, the computeSubrequests function generates a span whenever it is called:

    func computeSubrequests(r *http.Request, subRequests int) error {
	// Add custom span representing the work done for the subrequests
	ctx, span := tracer.Start(r.Context(), "subrequests")
	defer span.End()

	// Make specified number of http requests to the /single endpoint.
	for i := 0; i < subRequests; i++ {
		if err := callSingle(ctx); err != nil {
			return err
		}
	}
	// record number of sub-requests made
	subRequestsHistogram.Record(ctx, int64(subRequests))
	return nil
}

    In the previous code sample, the span generated from the computeSubrequests function represents the work done by the entire function. This is because the first step of the function is to start a new span using tracer.Start and the defer keyword before the span.End() ensures that the span is ended right before the function exits.

Create custom metrics

To generate metrics from your application, you add instrumentation code that records measurements taken during your app's execution.

To create metrics, do the following:

  1. Modify your app to acquire an OpenTelemetry Meter. In OpenTelemetry, a meter provides access to metric instruments for recording metrics. You can acquire a meter as demonstrated in the following code:

    const scopeName = "github.com/GoogleCloudPlatform/golang-samples/opentelemetry/instrumentation/app/work"

var (
	meter                = otel.Meter(scopeName)
	tracer               = otel.Tracer(scopeName)
	sleepHistogram       metric.Float64Histogram
	subRequestsHistogram metric.Int64Histogram
)

    The meter name, which is represented by scopeName, identifies the instrumentation scope of the generated metrics.

  2. Use the meter instance to create instruments which can record metrics. For example, in the following code, we use the meter to create an OpenTelemetry Histogram:

    sleepHistogram, err = meter.Float64Histogram("example.sleep.duration",
	metric.WithDescription("Sample histogram to measure time spent in sleeping"),
	metric.WithExplicitBucketBoundaries(0.05, 0.075, 0.1, 0.125, 0.150, 0.2),
	metric.WithUnit("s"))
if err != nil {
	panic(err)
}

    This previous code generates a histogram named sleepHistogram.

  3. Use the sleepHistogram instance to record the sleep time, which is determined when the function randomSleep is invoked:

    func randomSleep(r *http.Request) time.Duration {
	// simulate the work by sleeping 100 to 200 ms
	sleepTime := time.Duration(100+rand.Intn(100)) * time.Millisecond
	time.Sleep(sleepTime)

	hostValue := attribute.String("host.value", r.Host)
	// custom histogram metric to record time slept in seconds
	sleepHistogram.Record(r.Context(), sleepTime.Seconds(), metric.WithAttributes(hostValue))
	return sleepTime
}

    The recorded metrics from these instruments are exported from your application based on your OpenTelemetry exporter configuration.

Using OpenTelemetry Exemplars

Exemplars in OpenTelemetry provide examples of measurements recorded with a metric instrument, and provide a link to the current Span. Cloud Monitoring and Google Cloud Managed Service for Prometheus only support exemplars on histograms.

A histogram takes many measurements and groups them together into buckets. Histogram data is typically displayed by using a heatmap. You can also calculate percentiles such as the 95th percentile, based on the bucket distribution. Histograms discards information about individual measurements to efficiently store large numbers of them.

Exemplars provide "example" measurements which link to spans. By selecting an exemplar and navigating to the trace, you can see more detailed information about the example measurement, and also see where the measurement occurred within your broader application.

Exemplars appear as dots in a histogram heatmap:

metric-trace-correlation

OpenTelemetry instrumentation libraries automatically include exemplars when you have configured both the OpenTelemetry metric and trace SDKs. To include exemplars on your custom instrumentation, you need to ensure the metric is recorded within a span, and that you pass the context from the span with the metric measurement. In the computeSubrequests function previously shown, these actions are performed by invoking subRequestsHistogram.Record between tracer.Start and span.End (which is deferred), and by passing the context, ctx, from tracer.Start to the Record function.

For more information on exemplars, see OpenTelemetry documentation on Exemplars.

What's next