Instrumentation

Manual instrumentation for OpenTelemetry Go

Instrumentation is the act of adding observability code to an app yourself.

If you’re instrumenting an app, you need to use the OpenTelemetry SDK for your language. You’ll then use the SDK to initialize OpenTelemetry and the API to instrument your code. This will emit telemetry from your app, and any library you installed that also comes with instrumentation.

If you’re instrumenting a library, only install the OpenTelemetry API package for your language. Your library will not emit telemetry on its own. It will only emit telemetry when it is part of an app that uses the OpenTelemetry SDK. For more on instrumenting libraries, see Libraries.

For more information about the OpenTelemetry API and SDK, see the specification.

Setup

Traces

Getting a Tracer

To create spans, you’ll need to acquire or initialize a tracer first.

Ensure you have the right packages installed:

go get go.opentelemetry.io/otel \
  go.opentelemetry.io/otel/trace \
  go.opentelemetry.io/otel/sdk \

Then initialize an exporter, resources, tracer provider, and finally a tracer.

package app

import (
	"context"
	"fmt"
	"log"

	"go.opentelemetry.io/otel"
	"go.opentelemetry.io/otel/sdk/resource"
	sdktrace "go.opentelemetry.io/otel/sdk/trace"
	semconv "go.opentelemetry.io/otel/semconv/v1.26.0"
	"go.opentelemetry.io/otel/trace"
)

var tracer trace.Tracer

func newExporter(ctx context.Context)  /* (someExporter.Exporter, error) */ {
	// Your preferred exporter: console, jaeger, zipkin, OTLP, etc.
}

func newTraceProvider(exp sdktrace.SpanExporter) *sdktrace.TracerProvider {
	// Ensure default SDK resources and the required service name are set.
	r, err := resource.Merge(
		resource.Default(),
		resource.NewWithAttributes(
			semconv.SchemaURL,
			semconv.ServiceName("ExampleService"),
		),
	)

	if err != nil {
		panic(err)
	}

	return sdktrace.NewTracerProvider(
		sdktrace.WithBatcher(exp),
		sdktrace.WithResource(r),
	)
}

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

	exp, err := newExporter(ctx)
	if err != nil {
		log.Fatalf("failed to initialize exporter: %v", err)
	}

	// Create a new tracer provider with a batch span processor and the given exporter.
	tp := newTraceProvider(exp)

	// Handle shutdown properly so nothing leaks.
	defer func() { _ = tp.Shutdown(ctx) }()

	otel.SetTracerProvider(tp)

	// Finally, set the tracer that can be used for this package.
	tracer = tp.Tracer("example.io/package/name")
}

You can now access tracer to manually instrument your code.

Creating Spans

Spans are created by tracers. If you don’t have one initialized, you’ll need to do that.

To create a span with a tracer, you’ll also need a handle on a context.Context instance. These will typically come from things like a request object and may already contain a parent span from an instrumentation library.

func httpHandler(w http.ResponseWriter, r *http.Request) {
	ctx, span := tracer.Start(r.Context(), "hello-span")
	defer span.End()

	// do some work to track with hello-span
}

In Go, the context package is used to store the active span. When you start a span, you’ll get a handle on not only the span that’s created, but the modified context that contains it.

Once a span has completed, it is immutable and can no longer be modified.

Get the current span

To get the current span, you’ll need to pull it out of a context.Context you have a handle on:

// This context needs contain the active span you plan to extract.
ctx := context.TODO()
span := trace.SpanFromContext(ctx)

// Do something with the current span, optionally calling `span.End()` if you want it to end

This can be helpful if you’d like to add information to the current span at a point in time.

Create nested spans

You can create a nested span to track work in a nested operation.

If the current context.Context you have a handle on already contains a span inside of it, creating a new span makes it a nested span. For example:

func parentFunction(ctx context.Context) {
	ctx, parentSpan := tracer.Start(ctx, "parent")
	defer parentSpan.End()

	// call the child function and start a nested span in there
	childFunction(ctx)

	// do more work - when this function ends, parentSpan will complete.
}

func childFunction(ctx context.Context) {
	// Create a span to track `childFunction()` - this is a nested span whose parent is `parentSpan`
	ctx, childSpan := tracer.Start(ctx, "child")
	defer childSpan.End()

	// do work here, when this function returns, childSpan will complete.
}

Once a span has completed, it is immutable and can no longer be modified.

Span Attributes

Attributes are keys and values that are applied as metadata to your spans and are useful for aggregating, filtering, and grouping traces. Attributes can be added at span creation, or at any other time during the lifecycle of a span before it has completed.

// setting attributes at creation...
ctx, span = tracer.Start(ctx, "attributesAtCreation", trace.WithAttributes(attribute.String("hello", "world")))
// ... and after creation
span.SetAttributes(attribute.Bool("isTrue", true), attribute.String("stringAttr", "hi!"))

Attribute keys can be precomputed, as well:

var myKey = attribute.Key("myCoolAttribute")
span.SetAttributes(myKey.String("a value"))

Semantic Attributes

Semantic Attributes are attributes that are defined by the OpenTelemetry Specification in order to provide a shared set of attribute keys across multiple languages, frameworks, and runtimes for common concepts like HTTP methods, status codes, user agents, and more. These attributes are available in the go.opentelemetry.io/otel/semconv/v1.26.0 package.

For details, see Trace semantic conventions.

Events

An event is a human-readable message on a span that represents “something happening” during it’s lifetime. For example, imagine a function that requires exclusive access to a resource that is under a mutex. An event could be created at two points - once, when we try to gain access to the resource, and another when we acquire the mutex.

span.AddEvent("Acquiring lock")
mutex.Lock()
span.AddEvent("Got lock, doing work...")
// do stuff
span.AddEvent("Unlocking")
mutex.Unlock()

A useful characteristic of events is that their timestamps are displayed as offsets from the beginning of the span, allowing you to easily see how much time elapsed between them.

Events can also have attributes of their own -

span.AddEvent("Cancelled wait due to external signal", trace.WithAttributes(attribute.Int("pid", 4328), attribute.String("signal", "SIGHUP")))

Set span status

A Status can be set on a Span, typically used to specify that a Span has not completed successfully - Error. By default, all spans are Unset, which means a span completed without error. The Ok status is reserved for when you need to explicitly mark a span as successful rather than stick with the default of Unset (i.e., “without error”).

The status can be set at any time before the span is finished.

import (
	// ...
	"go.opentelemetry.io/otel/codes"
	// ...
)

// ...

result, err := operationThatCouldFail()
if err != nil {
	span.SetStatus(codes.Error, "operationThatCouldFail failed")
}

Record errors

If you have an operation that failed and you wish to capture the error it produced, you can record that error.

import (
	// ...
	"go.opentelemetry.io/otel/codes"
	// ...
)

// ...

result, err := operationThatCouldFail()
if err != nil {
	span.SetStatus(codes.Error, "operationThatCouldFail failed")
	span.RecordError(err)
}

It is highly recommended that you also set a span’s status to Error when using RecordError, unless you do not wish to consider the span tracking a failed operation as an error span. The RecordError function does not automatically set a span status when called.

Propagators and Context

Traces can extend beyond a single process. This requires context propagation, a mechanism where identifiers for a trace are sent to remote processes.

In order to propagate trace context over the wire, a propagator must be registered with the OpenTelemetry API.

import (
  "go.opentelemetry.io/otel"
  "go.opentelemetry.io/otel/propagation"
)
...
otel.SetTextMapPropagator(propagation.TraceContext{})

OpenTelemetry also supports the B3 header format, for compatibility with existing tracing systems (go.opentelemetry.io/contrib/propagators/b3) that do not support the W3C TraceContext standard.

After configuring context propagation, you’ll most likely want to use automatic instrumentation to handle the behind-the-scenes work of actually managing serializing the context.

Metrics

To start producing metrics, you’ll need to have an initialized MeterProvider that lets you create a Meter. Meters let you create instruments that you can use to create different kinds of metrics. OpenTelemetry Go currently supports the following instruments:

  • Counter, a synchronous instrument that supports non-negative increments
  • Asynchronous Counter, an asynchronous instrument which supports non-negative increments
  • Histogram, a synchronous instrument that supports arbitrary values that are statistically meaningful, such as histograms, summaries, or percentile
  • Synchronous Gauge, a synchronous instrument that supports non-additive values, such as room temperature
  • Asynchronous Gauge, an asynchronous instrument that supports non-additive values, such as room temperature
  • UpDownCounter, a synchronous instrument that supports increments and decrements, such as the number of active requests
  • Asynchronous UpDownCounter, an asynchronous instrument that supports increments and decrements

For more on synchronous and asynchronous instruments, and which kind is best suited for your use case, see Supplementary Guidelines.

If a MeterProvider is not created either by an instrumentation library or manually, the OpenTelemetry Metrics API will use a no-op implementation and fail to generate data.

Here you can find more detailed package documentation for:

Initialize Metrics

To enable metrics in your app, you’ll need to have an initialized MeterProvider that will let you create a Meter.

If a MeterProvider is not created, the OpenTelemetry APIs for metrics will use a no-op implementation and fail to generate data. Therefore, you have to modify the source code to include the SDK initialization code using the following packages:

Ensure you have the right Go modules installed:

go get go.opentelemetry.io/otel \
  go.opentelemetry.io/otel/exporters/stdout/stdoutmetric \
  go.opentelemetry.io/otel/sdk \
  go.opentelemetry.io/otel/sdk/metric

Then initialize a resources, metrics exporter, and metrics provider:

package main

import (
	"context"
	"log"
	"time"

	"go.opentelemetry.io/otel"
	"go.opentelemetry.io/otel/exporters/stdout/stdoutmetric"
	"go.opentelemetry.io/otel/sdk/metric"
	"go.opentelemetry.io/otel/sdk/resource"
	semconv "go.opentelemetry.io/otel/semconv/v1.26.0"
)

func main() {
	// Create resource.
	res, err := newResource()
	if err != nil {
		panic(err)
	}

	// Create a meter provider.
	// You can pass this instance directly to your instrumented code if it
	// accepts a MeterProvider instance.
	meterProvider, err := newMeterProvider(res)
	if err != nil {
		panic(err)
	}

	// Handle shutdown properly so nothing leaks.
	defer func() {
		if err := meterProvider.Shutdown(context.Background()); err != nil {
			log.Println(err)
		}
	}()

	// Register as global meter provider so that it can be used via otel.Meter
	// and accessed using otel.GetMeterProvider.
	// Most instrumentation libraries use the global meter provider as default.
	// If the global meter provider is not set then a no-op implementation
	// is used, which fails to generate data.
	otel.SetMeterProvider(meterProvider)
}

func newResource() (*resource.Resource, error) {
	return resource.Merge(resource.Default(),
		resource.NewWithAttributes(semconv.SchemaURL,
			semconv.ServiceName("my-service"),
			semconv.ServiceVersion("0.1.0"),
		))
}

func newMeterProvider(res *resource.Resource) (*metric.MeterProvider, error) {
	metricExporter, err := stdoutmetric.New()
	if err != nil {
		return nil, err
	}

	meterProvider := metric.NewMeterProvider(
		metric.WithResource(res),
		metric.WithReader(metric.NewPeriodicReader(metricExporter,
			// Default is 1m. Set to 3s for demonstrative purposes.
			metric.WithInterval(3*time.Second))),
	)
	return meterProvider, nil
}

Now that a MeterProvider is configured, you can acquire a Meter.

Acquiring a Meter

Anywhere in your application where you have manually instrumented code you can call otel.Meter to acquire a meter. For example:

import "go.opentelemetry.io/otel"

var meter = otel.Meter("example.io/package/name")

Synchronous and asynchronous instruments

OpenTelemetry instruments are either synchronous or asynchronous (observable).

Synchronous instruments take a measurement when they are called. The measurement is done as another call during program execution, just like any other function call. Periodically, the aggregation of these measurements is exported by a configured exporter. Because measurements are decoupled from exporting values, an export cycle may contain zero or multiple aggregated measurements.

Asynchronous instruments, on the other hand, provide a measurement at the request of the SDK. When the SDK exports, a callback that was provided to the instrument on creation is invoked. This callback provides the SDK with a measurement that is immediately exported. All measurements on asynchronous instruments are performed once per export cycle.

Asynchronous instruments are useful in several circumstances, such as:

  • When updating a counter is not computationally cheap, and you don’t want the current executing thread to wait for the measurement
  • Observations need to happen at frequencies unrelated to program execution (i.e., they cannot be accurately measured when tied to a request lifecycle)
  • There is no known timestamp for a measurement value

In cases like these, it’s often better to observe a cumulative value directly, rather than aggregate a series of deltas in post-processing (the synchronous example).

Using Counters

Counters can be used to measure a non-negative, increasing value.

For example, here’s how you report the number of calls for an HTTP handler:

import (
	"net/http"

	"go.opentelemetry.io/otel/metric"
)

func init() {
	apiCounter, err := meter.Int64Counter(
		"api.counter",
		metric.WithDescription("Number of API calls."),
		metric.WithUnit("{call}"),
	)
	if err != nil {
		panic(err)
	}
	http.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
		apiCounter.Add(r.Context(), 1)

		// do some work in an API call
	})
}

Using UpDown Counters

UpDown counters can increment and decrement, allowing you to observe a cumulative value that goes up or down.

For example, here’s how you report the number of items of some collection:

import (
	"context"

	"go.opentelemetry.io/otel/metric"
)

var itemsCounter metric.Int64UpDownCounter

func init() {
	var err error
	itemsCounter, err = meter.Int64UpDownCounter(
		"items.counter",
		metric.WithDescription("Number of items."),
		metric.WithUnit("{item}"),
	)
	if err != nil {
		panic(err)
	}
}

func addItem() {
	// code that adds an item to the collection

	itemsCounter.Add(context.Background(), 1)
}

func removeItem() {
	// code that removes an item from the collection

	itemsCounter.Add(context.Background(), -1)
}

Using Gauges

Gauges are used to measure non-additive values when changes occur.

For example, here’s how you might report the current speed of a CPU fan:

import (
	"net/http"

	"go.opentelemetry.io/otel/metric"
)

var fanSpeedSubscription chan int64

func init() {
	speedGauge, err := meter.Int64Gauge(
		"cpu.fan.speed",
		metric.WithDescription("Speed of CPU fan"),
		metric.WithUnit("RPM"),
	)
	if err != nil {
		panic(err)
	}

	getCPUFanSpeed := func() int64 {
		// Generates a random fan speed for demonstration purpose.
		// In real world applications, replace this to get the actual fan speed.
		return int64(1500 + rand.Intn(1000))
	}

	fanSpeedSubscription = make(chan int64, 1)
	go func() {
		defer close(fanSpeedSubscription)

		for idx := 0; idx < 5; idx++ {
			// Synchronous gauges are used when the measurement cycle is
			// synchronous to an external change.
			time.Sleep(time.Duration(rand.Intn(3)) * time.Second)
			fanSpeed := getCPUFanSpeed()
			fanSpeedSubscription <- fanSpeed
		}
	}()
}

func recordFanSpeed() {
	ctx := context.Background()
	for fanSpeed := range fanSpeedSubscription {
		speedGauge.Record(ctx, fanSpeed)
	}
}

Using Histograms

Histograms are used to measure a distribution of values over time.

For example, here’s how you report a distribution of response times for an HTTP handler:

import (
	"net/http"
	"time"

	"go.opentelemetry.io/otel/metric"
)

func init() {
	histogram, err := meter.Float64Histogram(
		"task.duration",
		metric.WithDescription("The duration of task execution."),
		metric.WithUnit("s"),
	)
	if err != nil {
		panic(err)
	}
	http.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
		start := time.Now()

		// do some work in an API call

		duration := time.Since(start)
		histogram.Record(r.Context(), duration.Seconds())
	})
}

Using Observable (Async) Counters

Observable counters can be used to measure an additive, non-negative, monotonically increasing value.

For example, here’s how you report time since the application started:

import (
	"context"
	"time"

	"go.opentelemetry.io/otel/metric"
)

func init() {
	start := time.Now()
	if _, err := meter.Float64ObservableCounter(
		"uptime",
		metric.WithDescription("The duration since the application started."),
		metric.WithUnit("s"),
		metric.WithFloat64Callback(func(_ context.Context, o metric.Float64Observer) error {
			o.Observe(float64(time.Since(start).Seconds()))
			return nil
		}),
	); err != nil {
		panic(err)
	}
}

Using Observable (Async) UpDown Counters

Observable UpDown counters can increment and decrement, allowing you to measure an additive, non-negative, non-monotonically increasing cumulative value.

For example, here’s how you report some database metrics:

import (
	"context"
	"database/sql"

	"go.opentelemetry.io/otel/metric"
)

// registerDBMetrics registers asynchronous metrics for the provided db.
// Make sure to unregister metric.Registration before closing the provided db.
func registerDBMetrics(db *sql.DB, meter metric.Meter, poolName string) (metric.Registration, error) {
	max, err := meter.Int64ObservableUpDownCounter(
		"db.client.connections.max",
		metric.WithDescription("The maximum number of open connections allowed."),
		metric.WithUnit("{connection}"),
	)
	if err != nil {
		return nil, err
	}

	waitTime, err := meter.Int64ObservableUpDownCounter(
		"db.client.connections.wait_time",
		metric.WithDescription("The time it took to obtain an open connection from the pool."),
		metric.WithUnit("ms"),
	)
	if err != nil {
		return nil, err
	}

	reg, err := meter.RegisterCallback(
		func(_ context.Context, o metric.Observer) error {
			stats := db.Stats()
			o.ObserveInt64(max, int64(stats.MaxOpenConnections))
			o.ObserveInt64(waitTime, int64(stats.WaitDuration))
			return nil
		},
		max,
		waitTime,
	)
	if err != nil {
		return nil, err
	}
	return reg, nil
}

Using Observable (Async) Gauges

Observable Gauges should be used to measure non-additive values.

For example, here’s how you report memory usage of the heap objects used in application:

import (
	"context"
	"runtime"

	"go.opentelemetry.io/otel/metric"
)

func init() {
	if _, err := meter.Int64ObservableGauge(
		"memory.heap",
		metric.WithDescription(
			"Memory usage of the allocated heap objects.",
		),
		metric.WithUnit("By"),
		metric.WithInt64Callback(func(_ context.Context, o metric.Int64Observer) error {
			var m runtime.MemStats
			runtime.ReadMemStats(&m)
			o.Observe(int64(m.HeapAlloc))
			return nil
		}),
	); err != nil {
		panic(err)
	}
}

Adding attributes

You can add Attributes by using the WithAttributeSet or WithAttributes options.

import (
	"net/http"

	"go.opentelemetry.io/otel/metric"
	semconv "go.opentelemetry.io/otel/semconv/v1.26.0"
)

func init() {
	apiCounter, err := meter.Int64UpDownCounter(
		"api.finished.counter",
		metric.WithDescription("Number of finished API calls."),
		metric.WithUnit("{call}"),
	)
	if err != nil {
		panic(err)
	}
	http.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
		// do some work in an API call and set the response HTTP status code

		apiCounter.Add(r.Context(), 1,
			metric.WithAttributes(semconv.HTTPResponseStatusCode(statusCode)))
	})
}

Registering Views

A view provides SDK users with the flexibility to customize the metrics output by the SDK. You can customize which metric instruments are to be processed or ignored. You can also customize aggregation and what attributes you want to report on metrics.

Every instrument has a default view, which retains the original name, description, and attributes, and has a default aggregation that is based on the type of instrument. When a registered view matches an instrument, the default view is replaced by the registered view. Additional registered views that match the instrument are additive, and result in multiple exported metrics for the instrument.

You can use the NewView function to create a view and register it using the WithView option.

For example, here’s how you create a view that renames the latency instrument from the v0.34.0 version of the http instrumentation library to request.latency:

view := metric.NewView(metric.Instrument{
	Name: "latency",
	Scope: instrumentation.Scope{
		Name:    "http",
		Version: "0.34.0",
	},
}, metric.Stream{Name: "request.latency"})

meterProvider := metric.NewMeterProvider(
	metric.WithView(view),
)

For example, here’s how you create a view that makes the latency instrument from the http instrumentation library to be reported as an exponential histogram:

view := metric.NewView(
	metric.Instrument{
		Name:  "latency",
		Scope: instrumentation.Scope{Name: "http"},
	},
	metric.Stream{
		Aggregation: metric.AggregationBase2ExponentialHistogram{
			MaxSize:  160,
			MaxScale: 20,
		},
	},
)

meterProvider := metric.NewMeterProvider(
	metric.WithView(view),
)

The SDK filters metrics and attributes before exporting metrics. For example, you can use views to reduce memory usage of high cardinality metrics or drop attributes that might contain sensitive data.

Here’s how you create a view that drops the latency instrument from the http instrumentation library:

view := metric.NewView(
  metric.Instrument{
    Name:  "latency",
    Scope: instrumentation.Scope{Name: "http"},
  },
  metric.Stream{Aggregation: metric.AggregationDrop{}},
)

meterProvider := metric.NewMeterProvider(
	metric.WithView(view),
)

Here’s how you create a view that removes the http.request.method attribute recorded by the latency instrument from the http instrumentation library:

view := metric.NewView(
  metric.Instrument{
    Name:  "latency",
    Scope: instrumentation.Scope{Name: "http"},
  },
  metric.Stream{AttributeFilter: attribute.NewDenyKeysFilter("http.request.method")},
)

meterProvider := metric.NewMeterProvider(
	metric.WithView(view),
)

The Name field of criteria supports wildcard pattern matching. The * wildcard is recognized as matching zero or more characters, and ? is recognized as matching exactly one character. For example, a pattern of * matches all instrument names.

The following example shows how you create a view that sets unit to milliseconds for any instrument with a name suffix of .ms:

view := metric.NewView(
  metric.Instrument{Name: "*.ms"},
  metric.Stream{Unit: "ms"},
)

meterProvider := metric.NewMeterProvider(
	metric.WithView(view),
)

The NewView function provides a convenient way of creating views. If NewView can’t provide the functionalities you need, you can create a custom View directly.

For example, here’s how you create a view that uses regular expression matching to ensure all data stream names have a suffix of the units it uses:

re := regexp.MustCompile(`[._](ms|byte)$`)
var view metric.View = func(i metric.Instrument) (metric.Stream, bool) {
	// In a custom View function, you need to explicitly copy
	// the name, description, and unit.
	s := metric.Stream{Name: i.Name, Description: i.Description, Unit: i.Unit}
	// Any instrument that does not have a unit suffix defined, but has a
	// dimensional unit defined, update the name with a unit suffix.
	if re.MatchString(i.Name) {
		return s, false
	}
	switch i.Unit {
	case "ms":
		s.Name += ".ms"
	case "By":
		s.Name += ".byte"
	default:
		return s, false
	}
	return s, true
}

meterProvider := metric.NewMeterProvider(
	metric.WithView(view),
)

Logs

Logs are distinct from metrics and traces in that there is no user-facing OpenTelemetry logs API. Instead, there is tooling to bridge logs from existing popular log packages (such as slog, logrus, zap, logr) into the OpenTelemetry ecosystem. For rationale behind this design decision, see Logging specification.

The two typical workflows discussed below each cater to different application requirements.

Direct-to-Collector

Status: Experimental

In the direct-to-Collector workflow, logs are emitted directly from an application to a collector using a network protocol (e.g. OTLP). This workflow is simple to set up as it doesn’t require any additional log forwarding components, and allows an application to easily emit structured logs that conform to the log data model. However, the overhead required for applications to queue and export logs to a network location may not be suitable for all applications.

To use this workflow:

  • Configure the OpenTelemetry Log SDK to export log records to desired target destination (the collector or other).
  • Use an appropriate Log Bridge.

Logs SDK

The logs SDK dictates how logs are processed when using the direct-to-Collector workflow. No log SDK is needed when using the log forwarding workflow.

The typical log SDK configuration installs a batching log record processor with an OTLP exporter.

To enable logs in your app, you’ll need to have an initialized LoggerProvider that will let you use a Log Bridge.

If a LoggerProvider is not created, the OpenTelemetry APIs for logs will use a no-op implementation and fail to generate data. Therefore, you have to modify the source code to include the SDK initialization code using the following packages:

Ensure you have the right Go modules installed:

go get go.opentelemetry.io/otel \
  go.opentelemetry.io/otel/exporters/otlp/otlplog/otlploghttp \
  go.opentelemetry.io/otel/sdk \
  go.opentelemetry.io/otel/sdk/log

Then initialize a logger provider:

package main

import (
	"context"
	"fmt"

	"go.opentelemetry.io/otel/exporters/otlp/otlplog/otlploghttp"
	"go.opentelemetry.io/otel/log/global"
	"go.opentelemetry.io/otel/sdk/log"
	"go.opentelemetry.io/otel/sdk/resource"
	semconv "go.opentelemetry.io/otel/semconv/v1.26.0"
)

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

	// Create resource.
	res, err := newResource()
	if err != nil {
		panic(err)
	}

	// Create a logger provider.
	// You can pass this instance directly when creating bridges.
	loggerProvider, err := newLoggerProvider(ctx, res)
	if err != nil {
		panic(err)
	}

	// Handle shutdown properly so nothing leaks.
	defer func() {
		if err := loggerProvider.Shutdown(ctx); err != nil {
			fmt.Println(err)
		}
	}()

	// Register as global logger provider so that it can be accessed global.LoggerProvider.
	// Most log bridges use the global logger provider as default.
	// If the global logger provider is not set then a no-op implementation
	// is used, which fails to generate data.
	global.SetLoggerProvider(loggerProvider)
}

func newResource() (*resource.Resource, error) {
	return resource.Merge(resource.Default(),
		resource.NewWithAttributes(semconv.SchemaURL,
			semconv.ServiceName("my-service"),
			semconv.ServiceVersion("0.1.0"),
		))
}

func newLoggerProvider(ctx context.Context, res *resource.Resource) (*log.LoggerProvider, error) {
	exporter, err := otlploghttp.New(ctx)
	if err != nil {
		return nil, err
	}
	processor := log.NewBatchProcessor(exporter)
	provider := log.NewLoggerProvider(
		log.WithResource(res),
		log.WithProcessor(processor),
	)
	return provider, nil
}

Now that a LoggerProvider is configured, you can use it to set up a Log Bridge.

Log Bridge

A log bridge is a component that bridges logs from an existing log package into the OpenTelemetry Log SDK using the Logs Bridge API.

A full list of log bridges available can be found in the OpenTelemetry registry.

Each log bridge package documentation should have a usage example.

Via file or stdout

In the file or stdout workflow, logs are written to files or standout output. Another component (e.g. FluentBit) is responsible for reading / tailing the logs, parsing them to more structured format, and forwarding them a target, such as the collector. This workflow may be preferable in situations where application requirements do not permit additional overhead from direct-to-Collector. However, it requires that all log fields required down stream are encoded into the logs, and that the component reading the logs parse the data into the log data model. The installation and configuration of log forwarding components is outside the scope of this document.

Next Steps

You’ll also want to configure an appropriate exporter to export your telemetry data to one or more telemetry backends.