Building a Trace Receiver

If you are reading this tutorial, you probably already have an idea of the OpenTelemetry concepts behind distributed tracing, but if you don’t you can quickly read through it here.

Here is the definition of those concepts according to OpenTelemetry:

Traces track the progression of a single request, called a trace, as it is handled by services that make up an application. The request may be initiated by a user or an application. Distributed tracing is a form of tracing that traverses process, network and security boundaries.

Although the definition seems very application centric, you can leverage the OpenTelemetry trace model as a way to represent a request and quickly understand its duration and the details about every step involved in completing it.

Assuming you already have a system generating some kind of tracing telemetry, the OpenTelemetry Collector is the doorway to help you make it available into the OTel world.

Within the Collector, a trace receiver has the role to receive and convert your request telemetry from its original format into the OTel trace model, so the information can be properly processed through the Collector’s pipelines.

In order to implement a traces receiver you will need the following:

  • A Config implementation to enable the trace receiver to gather and validate its configurations within the Collector’s config.yaml.

  • A receiver.Factory implementation so the Collector can properly instantiate the trace receiver component.

  • A TracesReceiver implementation that is responsible to collect the telemetry, convert it to the internal trace representation, and hand the information to the next consumer in the pipeline.

In this tutorial we will create a sample trace receiver called tailtracer that simulates a pull operation and generates traces as an outcome of that operation. The next sections will guide you through the process of implementing the steps above in order to create the receiver, so let’s get started.

Setting up your receiver development and testing environment

First use the Building a Custom Collector tutorial to create a Collector instance named otelcol-dev; all you need is to copy the builder-config.yaml described on Step 2 and run the builder. As an outcome you should now have a otelcol-dev folder with your Collector’s development instance ready to go.

In order to properly test your trace receiver, you will need a distributed tracing backend so the Collector can send the telemetry to it. We will be using Jaeger, if you don’t have a Jaeger instance running, you can easily start one using Docker with the following command:

docker run -d --name jaeger \
  -e COLLECTOR_OTLP_ENABLED=true \
  -p 16686:16686 \
  -p 14317:4317 \
  -p 14318:4318 \
  jaegertracing/all-in-one:1.41

Now, create a config.yaml file so you can set up your Collector’s components.

cd otelcol-dev
touch config.yaml

For now, you just need a basic traces pipeline with the otlp receiver, the otlp and debug1 exporters, here is what your config.yaml file should look like:

config.yaml

receivers:
  otlp:
    protocols:
      grpc:

processors:

exporters:
  # NOTE: Prior to v0.86.0 use `logging` instead of `debug`.
  debug:
    verbosity: detailed
  otlp/jaeger:
    endpoint: localhost:14317
    tls:
      insecure: true

service:
  pipelines:
    traces:
      receivers: [otlp]
      processors: []
      exporters: [otlp/jaeger, debug]
  telemetry:
    logs:
      level: debug

Notice that I am only using the insecure flag in my otlp exporter config to make my local development setup easier; you should not use this flag when running your collector in production.

In order to verify that your initial pipeline is properly set up, you should have the following output after running your otelcol-dev command:

$ ./otelcol-dev --config config.yaml
2023-09-12T15:22:18.652-0700    info    service/telemetry.go:84 Setting up own telemetry...
2023-09-12T15:22:18.652-0700    info    service/telemetry.go:201        Serving Prometheus metrics      {"address": ":8888", "level": "Basic"}
2023-09-12T15:22:18.652-0700    debug   exporter@v0.85.0/exporter.go:273        Stable component.       {"kind": "exporter", "data_type": "traces", "name": "otlp/jaeger"}
2023-09-12T15:22:18.652-0700    info    exporter@v0.85.0/exporter.go:275        Development component. May change in the future.        {"kind": "exporter", "data_type": "traces", "name": "debug"}
2023-09-12T15:22:18.652-0700    debug   receiver@v0.85.0/receiver.go:294        Stable component.       {"kind": "receiver", "name": "otlp", "data_type": "traces"}
2023-09-12T15:22:18.652-0700    info    service/service.go:138  Starting otelcontribcol...      {"Version": "0.85.0", "NumCPU": 10}
2023-09-12T15:22:18.652-0700    info    extensions/extensions.go:31     Starting extensions...
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1] Channel created     {"grpc_log": true}
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1] original dial target is: "localhost:14317"  {"grpc_log": true}
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1] parsed dial target is: {URL:{Scheme:localhost Opaque:14317 User: Host: Path: RawPath: OmitHost:false ForceQuery:false RawQuery: Fragment: RawFragment:}}   {"grpc_log": true}
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1] fallback to scheme "passthrough"    {"grpc_log": true}
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1] parsed dial target is: {URL:{Scheme:passthrough Opaque: User: Host: Path:/localhost:14317 RawPath: OmitHost:false ForceQuery:false RawQuery: Fragment: RawFragment:}}      {"grpc_log": true}
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1] Channel authority set to "localhost:14317"  {"grpc_log": true}
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1] Channel switches to new LB policy "pick_first"      {"grpc_log": true}
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1 SubChannel #2] Subchannel created    {"grpc_log": true}
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1] Channel Connectivity change to CONNECTING   {"grpc_log": true}
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1 SubChannel #2] Subchannel Connectivity change to CONNECTING  {"grpc_log": true}
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [Server #3] Server created       {"grpc_log": true}
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [pick-first-lb 0x140027d9b30] Received SubConn state update: 0x140027d9ce0, {ConnectivityState:CONNECTING ConnectionError:<nil>}        {"grpc_log": true}
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1 SubChannel #2] Subchannel picks a new address "localhost:14317" to connect   {"grpc_log": true}
2023-09-12T15:22:18.652-0700    info    otlpreceiver@v0.85.0/otlp.go:83 Starting GRPC server    {"kind": "receiver", "name": "otlp", "data_type": "traces", "endpoint": "0.0.0.0:4317"}
2023-09-12T15:22:18.652-0700    info    service/service.go:161  Everything is ready. Begin running and processing data.
2023-09-12T15:22:18.652-0700    info    zapgrpc/zapgrpc.go:178  [core] [Server #3 ListenSocket #4] ListenSocket created {"grpc_log": true}
2023-09-12T15:22:18.662-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1 SubChannel #2] Subchannel Connectivity change to READY       {"grpc_log": true}

Make sure you see the last line, that will confirm that the OTLP exporter has successfully established a gRPC connection to your local Jaeger instance. Now that we have our environment ready, let’s start writing your receiver’s code.

Now, create another folder called tailtracer so we can have a place to host all of our receiver code.

mkdir tailtracer

Every Collector’s component should be created as a Go module, so you will need to properly initialize the tailtracer module. In my case here is what the command looked like:

cd tailtracer
go mod init github.com/rquedas/otel4devs/collector/receiver/trace-receiver/tailtracer

Reading and Validating your Receiver Settings

In order to be instantiated and participate in pipelines the Collector needs to identify your receiver and properly load its settings from within its configuration file.

The tailtracer receiver will have the following settings:

  • interval: a string representing the time interval (in minutes) between telemetry pull operations
  • number_of_traces: the number of mock traces generated for each interval

Here is what the tailtracer receiver settings will look like:

receivers:
  tailtracer: # this line represents the ID of your receiver
    interval: 1m
    number_of_traces: 1

Under the tailtracer folder, create a file named config.go where you will write all the code to support your receiver settings.

cd tailtracer
touch config.go

To implement the configuration aspects of a receiver you need create a Config struct. Add the following code to your config.go file:

package tailtracer

type Config struct{

}

In order to be able to give your receiver access to its settings the Config struct must have a field for each of the receiver’s settings.

Here is what your config.go file should look like after you implemented the requirements above.

config.go

package tailtracer

// Config represents the receiver config settings within the collector's config.yaml
type Config struct {
   Interval    string `mapstructure:"interval"`
   NumberOfTraces int `mapstructure:"number_of_traces"`
}

Now that you have access to the settings, you can provide any kind of validation needed for those values by implementing the Validate method according to the optional ConfigValidator interface.

In this case, the interval value will be optional (we will look at generating default values later) but when defined should be at least 1 minute (1m) and the number_of_traces will be a required value. Here is what the config.go looks like after implementing the Validate method.

config.go

package tailtracer

import (
	"fmt"
	"time"
)

// Config represents the receiver config settings within the collector's config.yaml
type Config struct {
	Interval       string `mapstructure:"interval"`
	NumberOfTraces int    `mapstructure:"number_of_traces"`
}

// Validate checks if the receiver configuration is valid
func (cfg *Config) Validate() error {
	interval, _ := time.ParseDuration(cfg.Interval)
	if interval.Minutes() < 1 {
		return fmt.Errorf("when defined, the interval has to be set to at least 1 minute (1m)")
	}

	if cfg.NumberOfTraces < 1 {
		return fmt.Errorf("number_of_traces must be greater or equal to 1")
	}
	return nil
}

If you want to take a closer look at the structs and interfaces involved in the configuration aspects of a component, take a look at the component/config.go file inside the Collector’s GitHub project.

Enabling the Collector to instantiate your receiver

At the beginning of this tutorial, you created your otelcol-dev instance, which is bootstrapped with the following components:

  • Receivers: OTLP Receiver
  • Processors: Batch Processor
  • Exporters: Debug1 and OTLP Exporters

Go ahead and open the components.go file under the otelcol-dev folder, and let’s take a look at the components() function.

func components() (otelcol.Factories, error) {
	var err error
	factories := otelcol.Factories{}

	factories.Extensions, err = extension.MakeFactoryMap(
	)
	if err != nil {
		return otelcol.Factories{}, err
	}

	factories.Receivers, err = receiver.MakeFactoryMap(
		otlpreceiver.NewFactory(),
	)
	if err != nil {
		return otelcol.Factories{}, err
	}

	factories.Exporters, err = exporter.MakeFactoryMap(
		debugexporter.NewFactory(),
		otlpexporter.NewFactory(),
	)
	if err != nil {
		return otelcol.Factories{}, err
	}

	factories.Processors, err = processor.MakeFactoryMap(
		batchprocessor.NewFactory(),
	)
	if err != nil {
		return otelcol.Factories{}, err
	}

	return factories, nil
}

As you can see, the components() function is responsible to provide the Collector the factories for all its components which is represented by a variable called factories of type otelcol.Factories (here is the declaration of the otelcol.Factories struct), which will then be used to instantiate the components that are configured and consumed by the Collector’s pipelines.

Notice that factories.Receivers is the field holding a map to all the receiver factories (instances of receiver.Factory), and it currently has the otlpreceiver factory only which is instantiated through the otlpreceiver.NewFactory() function call.

The tailtracer receiver has to provide a receiver.Factory implementation, and although you will find a receiver.Factory interface (you can find its definition in the receiver/receiver.go file within the Collector’s project ), the right way to provide the implementation is by using the functions available within the go.opentelemetry.io/collector/receiver package.

Implementing your receiver.Factory

Start by creating a file named factory.go within the tailtracer folder.

cd tailtracer
touch factory.go

Now let’s follow the convention and add a function named NewFactory() that will be responsible to instantiate the tailtracer factory. Go ahead the add the following code to your factory.go file:

package tailtracer

import (
	"go.opentelemetry.io/collector/receiver"
)

// NewFactory creates a factory for tailtracer receiver.
func NewFactory() receiver.Factory {

}

In order to instantiate your tailtracer receiver factory, you will use the following function from the receiver package:

func NewFactory(cfgType component.Type, createDefaultConfig component.CreateDefaultConfigFunc, options ...FactoryOption) Factory

The receiver.NewFactory() instantiates and returns a receiver.Factory and it requires the following parameters:

  • component.Type: a unique string identifier for your receiver across all Collector’s components.

  • component.CreateDefaultConfigFunc: a reference to a function that returns the component.Config instance for your receiver.

  • ...FactoryOption: the slice of receiver.FactoryOptions that will determine what type of signal your receiver is capable of processing.

Let’s now implement the code to support all the parameters required by receiver.NewFactory().

Identifying and Providing default settings for the receiver

Previously, we said that the interval setting for our tailtracer receiver would be optional, in that case you will need to provide a default value for it so it can be used as part of the default settings.

Go ahead and add the following code to your factory.go file:

const (
	typeStr = "tailtracer"
	defaultInterval = 1 * time.Minute
)

As for default settings, you just need to add a function that returns a component.Config holding the default configurations for the tailtracer receiver.

To accomplish that, go ahead and add the following code to your factory.go file:

func createDefaultConfig() component.Config {
	return &Config{
		Interval: string(defaultInterval),
	}
}

After these two changes you will notice a few imports are missing, so here is what your factory.go file should look like with the proper imports:

factory.go

package tailtracer

import (
	"time"

	"go.opentelemetry.io/collector/component"
	"go.opentelemetry.io/collector/receiver"
)

const (
	typeStr = "tailtracer"
	defaultInterval = 1 * time.Minute
)

func createDefaultConfig() component.Config {
	return &Config{
		Interval: string(defaultInterval),
	}
}

// NewFactory creates a factory for tailtracer receiver.
func NewFactory() receiver.Factory {
	return nil
}

Enabling the factory to describe the receiver as capable of processing traces

The same receiver component can process traces, metrics, and logs. The receiver’s factory is responsible for describing those capabilities.

Given that traces are the subject of the tutorial, that’s the only signal we will enable the tailtracer receiver to work with. The receiver package provides the following function and type to help the factory describe the trace processing capabilities:

func WithTraces(createTracesReceiver CreateTracesFunc, sl component.StabilityLevel) FactoryOption

The receiver.WithTraces() instantiates and returns a receiver.FactoryOption and it requires the following parameters:

  • createTracesReceiver: A reference to a function that matches the receiver.CreateTracesFunc type

The receiver.CreateTracesFunc type is a pointer to a function that is responsible to instantiate and return a receiver.Traces instance and it requires the following parameters:

  • context.Context: the reference to the Collector’s context.Context so your trace receiver can properly manage its execution context.
  • receiver.CreateSettings: the reference to some of the Collector’s settings under which your receiver is created.
  • component.Config: the reference for the receiver config settings passed by the Collector to the factory so it can properly read its settings from the Collector config.
  • consumer.Traces: the reference to the next consumer.Traces in the pipeline, which is where received traces will go. This is either a processor or an exporter.

Start by adding the bootstrap code to properly implement the receiver.CreateTracesFunc function pointer. Go ahead and add the following code to your factory.go file:

func createTracesReceiver(_ context.Context, params receiver.CreateSettings, baseCfg component.Config, consumer consumer.Traces) (receiver.Traces, error) {
	return nil, nil
}

You now have all the necessary components to successfully instantiate your receiver factory using the receiver.NewFactory function. Go ahead and and update your NewFactory() function in your factory.go file as follow:

// NewFactory creates a factory for tailtracer receiver.
func NewFactory() receiver.Factory {
	return receiver.NewFactory(
		typeStr,
		createDefaultConfig,
		receiver.WithTraces(createTracesReceiver, component.StabilityLevelAlpha))
}

After these two changes you will notice a few imports are missing, so here is what your factory.go file should look like with the proper imports:

factory.go

package tailtracer

import (
	"context"
	"time"

	"go.opentelemetry.io/collector/component"
	"go.opentelemetry.io/collector/consumer"
	"go.opentelemetry.io/collector/receiver"
)

const (
	typeStr = "tailtracer"
	defaultInterval = 1 * time.Minute
)

func createDefaultConfig() component.Config {
	return &Config{
		Interval: string(defaultInterval),
	}
}

func createTracesReceiver(_ context.Context, params receiver.CreateSettings, baseCfg component.Config, consumer consumer.Traces) (receiver.Traces, error) {
	return nil, nil
}

// NewFactory creates a factory for tailtracer receiver.
func NewFactory() receiver.Factory {
	return receiver.NewFactory(
		typeStr,
		createDefaultConfig,
		receiver.WithTraces(createTracesReceiver, component.StabilityLevelAlpha))
}

At this point, you have the tailtracer factory and config code needed for the Collector to validate the tailtracer receiver settings if they are defined within the config.yaml. You just need to add it to the Collector’s initialization process.

Adding the receiver factory to the Collector’s initialization

As explained before, all the Collector components are instantiated by the components() function within the components.go file.

The tailtracer receiver factory instance has to be added to the factories map so the Collector can load it properly as part of its initialization process.

Here is what the components.go file looks like after making the changes to support that:

components.go

// Code generated by "go.opentelemetry.io/collector/cmd/builder". DO NOT EDIT.

package main

import (
	"go.opentelemetry.io/collector/exporter"
	"go.opentelemetry.io/collector/extension"
	"go.opentelemetry.io/collector/otelcol"
	"go.opentelemetry.io/collector/processor"
	"go.opentelemetry.io/collector/receiver"
	debugexporter "go.opentelemetry.io/collector/exporter/debugexporter"
	otlpexporter "go.opentelemetry.io/collector/exporter/otlpexporter"
	batchprocessor "go.opentelemetry.io/collector/processor/batchprocessor"
	otlpreceiver "go.opentelemetry.io/collector/receiver/otlpreceiver"
	tailtracer "github.com/rquedas/otel4devs/collector/receiver/trace-receiver/tailtracer"
)

func components() (otelcol.Factories, error) {
	var err error
	factories := otelcol.Factories{}

	factories.Extensions, err = extension.MakeFactoryMap(
	)
	if err != nil {
		return otelcol.Factories{}, err
	}

	factories.Receivers, err = receiver.MakeFactoryMap(
		otlpreceiver.NewFactory(),
		tailtracer.NewFactory(),
	)
	if err != nil {
		return otelcol.Factories{}, err
	}

	factories.Exporters, err = exporter.MakeFactoryMap(
		debugexporter.NewFactory(),
		otlpexporter.NewFactory(),
	)
	if err != nil {
		return otelcol.Factories{}, err
	}

	factories.Processors, err = processor.MakeFactoryMap(
		batchprocessor.NewFactory(),
	)
	if err != nil {
		return otelcol.Factories{}, err
	}

	return factories, nil
}

We added the tailtracer receiver settings to the config.yaml previously, so here is what the beginning of the output for running your Collector with otelcol-dev command should look like after building it with the current codebase:

$ ./otelcol-dev --config config.yaml
2023-09-28T08:56:53.027-0700    info    service@v0.86.0/telemetry.go:84 Setting up own telemetry...
2023-09-28T08:56:53.027-0700    info    service@v0.86.0/telemetry.go:201        Serving Prometheus metrics      {"address": ":8888", "level": "Basic"}
2023-09-28T08:56:53.027-0700    debug   exporter@v0.86.0/exporter.go:273        Stable component.       {"kind": "exporter", "data_type": "traces", "name": "otlp/jaeger"}
2023-09-28T08:56:53.027-0700    info    exporter@v0.86.0/exporter.go:275        Development component. May change in the future.        {"kind": "exporter", "data_type": "traces", "name": "debug"}
2023-09-28T08:56:53.027-0700    debug   receiver@v0.86.0/receiver.go:294        Alpha component. May change in the future.      {"kind": "receiver", "name": "tailtracer", "data_type": "traces"}
2023-09-28T08:56:53.027-0700    debug   receiver@v0.86.0/receiver.go:294        Stable component.       {"kind": "receiver", "name": "otlp", "data_type": "traces"}
2023-09-28T08:56:53.027-0700    info    service@v0.86.0/service.go:138  Starting otelcol-dev... {"Version": "1.0.0", "NumCPU": 10}
2023-09-28T08:56:53.027-0700    info    extensions/extensions.go:31     Starting extensions...

Look for the log line for “builder/receivers_builder.go:111” (it’s the 4th line from the bottom at the snippet showed here), you can see that the Collector found the settings for the tailtracer receiver, validated them (the current settings are all correct), but ignores the receiver given that it’s not used in any pipeline.

Let’s check if the tailtracer factory is validating the receiver settings correctly, the interval setting isn’t required, so if you remove it from the config.yaml and run the command again you should get the same output.

Now, let’s test one of the tailtracer settings validation rules. Remove the number_of_traces setting from the config.yaml, and here is what the output for running the Collector will look like:

$ ./otelcol-dev --config config.yaml
Error: invalid configuration: receiver "tailtracer" has invalid configuration: number_of_traces must be at least 1
2022/02/24 13:00:20 collector server run finished with error: invalid configuration: receiver "tailtracer" has invalid configuration: number_of_traces must be at least 1

The tailtracer receiver factory and config requirements are done and the Collector is properly loading your component. You can now move to the core of your receiver, the implementation of the component itself.

Implementing the trace receiver component

In the previous section, I mentioned the fact that a receiver can process any of the OpenTelemetry signals, and the Collector’s API is designed to help you accomplish that.

All the receiver APIs responsible to enable the signals are currently declared in the receiver/receiver.go file within the OTel Collector’s project in GitHub, open the file and take a minute to browse through all the interfaces declared in it.

Notice that receiver.Traces (and its siblings receiver.Metrics and receiver.Logs) at this point in time, doesn’t describe any specific methods other than the ones it “inherits” from component.Component.

It might feel weird, but remember, the Collector’s API was meant to be extensible, and the components and their signals might evolve in different ways, so the role of those interfaces exist to help support that.

So, to create a receiver.Traces, you just need to implement the following methods described by component.Component interface:

Start(ctx context.Context, host Host) error
Shutdown(ctx context.Context) error

Both methods actually act as event handlers used by the Collector to communicate with its components as part of their lifecycle.

The Start() represents a signal of the Collector telling the component to start its processing. As part of the event, the Collector will pass the following information:

  • context.Context: Most of the time, a receiver will be processing a long-running operation, so the recommendation is to ignore this context and actually create a new one from context.Background().
  • Host: The host is meant to enable the receiver to communicate with the Collector’s host once it’s up and running.

The Shutdown() represents a signal of the Collector telling the component that the service is getting shutdown and as such the component should stop its processing and make all the necessary cleanup work required:

  • context.Context: the context passed by the Collector as part of the shutdown operation.

You will start the implementation by creating a new file called trace-receiver.go within your project’s tailtracer folder and add the declaration to a type type called tailtracerReceiver as follow:

type tailtracerReceiver struct{

}

Now that you have the tailtracerReceiver type you can implement the Start() and Shutdown() methods so the receiver type can be compliant with the receiver.Traces interface.

Here is what the tailtracer/trace-receiver.go file should look like with the methods implementation:

trace-receiver.go

package tailtracer

import (
	"context"
	"go.opentelemetry.io/collector/component"
)

type tailtracerReceiver struct {
}

func (tailtracerRcvr *tailtracerReceiver) Start(ctx context.Context, host component.Host) error {
	return nil
}

func (tailtracerRcvr *tailtracerReceiver) Shutdown(ctx context.Context) error {
	return nil
}

The Start() method is passing 2 references (context.Context and component.Host) that your receiver might need to keep so they can be used as part of its processing operations.

The context.Context reference should be used for creating a new context to support you receiver processing operations, and in that case you will need to decide the best way to handle context cancellation so you can finalize it properly as part of the component’s shutdown within the Shutdown() method.

The component.Host can be useful during the whole lifecycle of the receiver so you should keep that reference within your tailtracerReceiver type.

Here is what the tailtracerReceiver type declaration will look like after you include the fields for keeping the references suggested above:

type tailtracerReceiver struct {
  host component.Host
  cancel context.CancelFunc
}

Now you need to update the Start() methods so the receiver can properly initialize its own processing context and have the cancellation function kept in the cancel field and also initialize its host field value. You will also update the Stop() method in order to finalize the context by calling the cancel function.

Here is what the trace-receiver.go file look like after making the changes above:

trace-receiver.go

package tailtracer

import (
	"context"
	"go.opentelemetry.io/collector/component"
)

type tailtracerReceiver struct {
    host component.Host
	cancel context.CancelFunc
}

func (tailtracerRcvr *tailtracerReceiver) Start(ctx context.Context, host component.Host) error {
    tailtracerRcvr.host = host
    ctx = context.Background()
	ctx, tailtracerRcvr.cancel = context.WithCancel(ctx)

	return nil
}

func (tailtracerRcvr *tailtracerReceiver) Shutdown(ctx context.Context) error {
	tailtracerRcvr.cancel()
	return nil
}

Keeping information passed by the receiver’s factory

Now that you have implemented the receiver.Traces interface methods, your tailtracer receiver component is ready to be instantiated and returned by its factory.

Open the tailtracer/factory.go file and navigate to the createTracesReceiver() function. Notice that the factory will pass references as part of the createTracesReceiver() function parameters that your receiver actually requires to work properly like its configuration settings (component.Config), the next Consumer in the pipeline that will consume the generated traces (consumer.Traces) and the Collector’s logger so the tailtracer receiver can add meaningful events to it (receiver.CreateSettings).

Given that all this information will be only be made available to the receiver at the moment its instantiated by the factory, The tailtracerReceiver type will need fields to keep that information and use it within other stages of its lifecycle.

Here is what the trace-receiver.go file looks like with the updated tailtracerReceiver type declaration:

trace-receiver.go

package tailtracer

import (
	"context"
	"time"
	"go.opentelemetry.io/collector/component"
	"go.opentelemetry.io/collector/consumer"
	"go.uber.org/zap"
)

type tailtracerReceiver struct {
	host         component.Host
	cancel       context.CancelFunc
	logger       *zap.Logger
	nextConsumer consumer.Traces
	config       *Config
}

func (tailtracerRcvr *tailtracerReceiver) Start(ctx context.Context, host component.Host) error {
	tailtracerRcvr.host = host
	ctx = context.Background()
	ctx, tailtracerRcvr.cancel = context.WithCancel(ctx)

	interval, _ := time.ParseDuration(tailtracerRcvr.config.Interval)
	go func() {
		ticker := time.NewTicker(interval)
		defer ticker.Stop()

		for {
			select {
				case <-ticker.C:
					tailtracerRcvr.logger.Info("I should start processing traces now!")
				case <-ctx.Done():
					return
			}
		}
	}()

	return nil
}

func (tailtracerRcvr *tailtracerReceiver) Shutdown(ctx context.Context) error {
	tailtracerRcvr.cancel()
	return nil
}

The tailtracerReceiver type is now ready to be instantiated and keep all meaningful information passed by its factory.

Open the tailtracer/factory.go file and navigate to the createTracesReceiver() function.

The receiver is only instantiated if it’s declared as a component within a pipeline and the factory is responsible to make sure the next consumer (either a processor or exporter) in the pipeline is valid otherwise it should generate an error.

The Collector’s API provides some standard error types to help the factory handle pipeline configurations. Your receiver factory should throw a component.ErrNilNextConsumer in case the next consumer has an issue and is passed as nil.

The createTracesReceiver() function will need a guard clause to make that validation.

You will also need variables to properly initialize the config and the logger fields of the tailtracerReceiver instance.

Here is what the factory.go file looks like with the updated createTracesReceiver() function:

factory.go

package tailtracer

import (
	"context"
	"time"

	"go.opentelemetry.io/collector/component"
	"go.opentelemetry.io/collector/consumer"
	"go.opentelemetry.io/collector/receiver"
)

const (
	typeStr = "tailtracer"
	defaultInterval = 1 * time.Minute
)

func createDefaultConfig() component.Config {
	return &Config{
		Interval: string(defaultInterval),
	}
}

func createTracesReceiver(_ context.Context, params receiver.CreateSettings, baseCfg component.Config, consumer consumer.Traces) (receiver.Traces, error) {
	if consumer == nil {
		return nil, component.ErrNilNextConsumer
	}

	logger := params.Logger
	tailtracerCfg := baseCfg.(*Config)

	traceRcvr := &tailtracerReceiver{
		logger:       logger,
		nextConsumer: consumer,
		config:       tailtracerCfg,
	}

	return traceRcvr, nil
}

// NewFactory creates a factory for tailtracer receiver.
func NewFactory() receiver.Factory {
	return receiver.NewFactory(
		typeStr,
		createDefaultConfig,
		receiver.WithTraces(createTracesReceiver, component.StabilityLevelAlpha))
}

With the factory fully implemented and instantiating the trace receiver component you are ready to test the receiver as part of a pipeline. Go ahead and add the tailtracer receiver to your traces pipeline in the config.yaml as follow:

service:
  pipelines:
    traces:
      receivers: [otlp, tailtracer]
      processors: []
      exporters: [otlp/jaeger, debug]

Here is what the output for running your Collector with otelcol-dev command should look like after you updated the traces pipeline:

$ ./otelcol-dev --config config.yaml
2023-09-28T08:59:52.111-0700    info    service@v0.86.0/telemetry.go:84 Setting up own telemetry...
2023-09-28T08:59:52.111-0700    info    service@v0.86.0/telemetry.go:201        Serving Prometheus metrics      {"address": ":8888", "level": "Basic"}
2023-09-28T08:59:52.111-0700    debug   exporter@v0.86.0/exporter.go:273        Stable component.       {"kind": "exporter", "data_type": "traces", "name": "otlp/jaeger"}
2023-09-28T08:59:52.112-0700    info    exporter@v0.86.0/exporter.go:275        Development component. May change in the future.        {"kind": "exporter", "data_type": "traces", "name": "debug"}
2023-09-28T08:59:52.112-0700    debug   receiver@v0.86.0/receiver.go:294        Stable component.       {"kind": "receiver", "name": "otlp", "data_type": "traces"}
2023-09-28T08:59:52.112-0700    debug   receiver@v0.86.0/receiver.go:294        Alpha component. May change in the future.      {"kind": "receiver", "name": "tailtracer", "data_type": "traces"}
2023-09-28T08:59:52.112-0700    info    service@v0.86.0/service.go:138  Starting otelcol-dev... {"Version": "1.0.0", "NumCPU": 10}
2023-09-28T08:59:52.112-0700    info    extensions/extensions.go:31     Starting extensions...
2023-09-28T08:59:52.113-0700    info    otlpreceiver@v0.86.0/otlp.go:83 Starting GRPC server    {"kind": "receiver", "name": "otlp", "data_type": "traces", "endpoint": "0.0.0.0:4317"}
2023-09-28T08:59:52.113-0700    info    service@v0.86.0/service.go:161  Everything is ready. Begin running and processing data.
2023-09-28T09:00:52.113-0700    info    tailtracer/receiver.go:33       I should start processing traces now!   {"kind": "receiver", "name": "tailtracer", "data_type": "traces"}

Look for the log line for “builder/receivers_builder.go:68 Receiver is starting… {“kind”: “receiver”, “name”: “tailtracer”}”, you can see that the Collector found the settings for the tailtracer receiver within the traces pipeline and is now instantiating it and starting it given that 1 minute after the Collector has started, you can see the info line we added to the ticker function within the Start() method.

Now, go ahead and press Ctrl + C in your Collector’s terminal so you want watch the shutdown process happening. Here is what the output should look like:

^C2023-09-28T09:01:18.784-0700  info    otelcol@v0.86.0/collector.go:250        Received signal from OS {"signal": "interrupt"}
2023-09-28T09:01:18.784-0700    info    service@v0.86.0/service.go:170  Starting shutdown...
2023-09-28T09:01:18.784-0700    info    zapgrpc/zapgrpc.go:178  [core] [Server #3 ListenSocket #4] ListenSocket deleted {"grpc_log": true}
2023-09-28T09:01:18.784-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1] Channel Connectivity change to SHUTDOWN     {"grpc_log": true}
2023-09-28T09:01:18.784-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1] ccBalancerWrapper: closing  {"grpc_log": true}
2023-09-28T09:01:18.785-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1] Closing the name resolver   {"grpc_log": true}
2023-09-28T09:01:18.785-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1 SubChannel #2] Subchannel Connectivity change to SHUTDOWN    {"grpc_log": true}
2023-09-28T09:01:18.785-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1 SubChannel #2] Subchannel deleted    {"grpc_log": true}
2023-09-28T09:01:18.785-0700    info    zapgrpc/zapgrpc.go:178  [transport] [client-transport 0x140002c8000] Closing: rpc error: code = Canceled desc = grpc: the client connection is closing   {"grpc_log": true}
2023-09-28T09:01:18.785-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1] Channel deleted     {"grpc_log": true}
2023-09-28T09:01:18.785-0700    info    extensions/extensions.go:45     Stopping extensions...
2023-09-28T09:01:18.785-0700    info    service@v0.86.0/service.go:184  Shutdown complete.

As you can see there is an info log line for the tailtracer receiver which means the component is responding correctly to the Shutdown() event. In the next section you will learn more about the OpenTelemetry Trace data model so the tailtracer receiver can finally generate traces!

The Collector’s Trace Data Model

You might be familiar with OpenTelemetry traces by using the SDKs and instrumenting an application so you can see and evaluate your traces within a distributed tracing backend like Jaeger.

Here is what a trace looks like in Jaeger:

Jaeger trace

Granted, this is a Jaeger trace, but it was generated by a trace pipeline within the Collector, therefore you can use it to learn a few things about the OTel trace data model :

  • A trace is made of one or multiple spans structured within a hierarchy to represent dependencies.
  • The spans can represent operations within a service and/or across services.

Creating a trace within the trace receiver will be slightly different than the way you would do it with the SDKs, so let’s start reviewing the high level concepts.

Working with Resources

In the OTel world, all telemetry is generated by a Resource, here is the definition according to the OTel spec:

A Resource is an immutable representation of the entity producing telemetry as Attributes. For example, a process producing telemetry that is running in a container on Kubernetes has a Pod name, it is in a namespace and possibly is part of a Deployment which also has a name. All three of these attributes can be included in the Resource.

Traces are most commonly used to represent a service request (the Services entity described by Jaeger’s model), which are normally implemented as processes running in a compute unit, but OTel’s API approach to describe a Resource through attributes is flexible enough to represent any entity that you may require like ATMs, IoT sensors, the sky is the limit.

So it’s safe to say that for a trace to exist, a Resource will have to start it.

In this tutorial we will simulate a system that has telemetry that demonstrate ATMs located in 2 different states (eg: Illinois and California) accessing the Account’s backend system to execute balance, deposit and withdraw operations, therefore we will have to implement code to create the Resource types representing the ATM and the backend system.

Go ahead and create a file named model.go inside the tailtracer folder

cd tailtracer
touch model.go

Now, within the model.go file, add the definition for the Atm and the BackendSystem types as follow:

model.go

package tailtracer

type Atm struct{
    ID           int64
	Version      string
	Name         string
	StateID      string
	SerialNumber string
	ISPNetwork   string
}

type BackendSystem struct{
	Version       string
	ProcessName   string
	OSType        string
    OSVersion     string
	CloudProvider string
	CloudRegion   string
	ServiceName   string
	Endpoint      string
}

These types are meant to represent the entities as they are within the system being observed and they contain information that would be quite meaningful to be added to the traces as part of the Resource definition. You will add some helper functions to generate the instances of those types.

Here is what the model.go file will look with the helper functions:

model.go

package tailtracer

import (
	"math/rand"
	"time"
)

type Atm struct{
    ID           int64
	Version      string
	Name         string
	StateID      string
	SerialNumber string
	ISPNetwork   string
}

type BackendSystem struct{
	Version       string
	ProcessName   string
	OSType        string
    OSVersion     string
	CloudProvider string
	CloudRegion   string
	Endpoint      string
}

func generateAtm() Atm{
	i := getRandomNumber(1, 2)
    var newAtm Atm

	switch i {
		case 1:
			newAtm = Atm{
				ID: 111,
				Name: "ATM-111-IL",
				SerialNumber: "atmxph-2022-111",
				Version: "v1.0",
				ISPNetwork: "comcast-chicago",
				StateID: "IL",

			}

		case 2:
			newAtm = Atm{
				ID: 222,
				Name: "ATM-222-CA",
				SerialNumber: "atmxph-2022-222",
				Version: "v1.0",
				ISPNetwork: "comcast-sanfrancisco",
				StateID: "CA",
			}
	}

	return newAtm
}

func generateBackendSystem() BackendSystem{
    i := getRandomNumber(1, 3)

	newBackend := BackendSystem{
    	ProcessName: "accounts",
		Version: "v2.5",
		OSType: "lnx",
		OSVersion: "4.16.10-300.fc28.x86_64",
		CloudProvider: "amzn",
		CloudRegion: "us-east-2",
	}

	switch i {
		case 1:
		 	newBackend.Endpoint = "api/v2.5/balance"
		case 2:
		  	newBackend.Endpoint = "api/v2.5/deposit"
		case 3:
			newBackend.Endpoint = "api/v2.5/withdrawn"

	}

	return newBackend
}

func getRandomNumber(min int, max int) int {
	rand.Seed(time.Now().UnixNano())
	i := (rand.Intn(max - min + 1) + min)
    return i
}

Now that you have the functions to generate object instances representing the entities generating telemetry, you are ready to represent those entities in the OTel Collector world.

The Collector’s API provides a package named ptrace (nested under the pdata package) with all the types, interfaces and helper functions required to work with traces within the Collector’s pipeline components.

Open the tailtracer/model.go file and add go.opentelemetry.io/collector/pdata/ptrace to the import clause so you can have access to the ptrace package capabilities.

Before you can define a Resource, you need to create a ptrace.Traces that will be responsible to propagate the traces through the Collector’s pipeline and you can use the helper function ptrace.NewTraces() to instantiate it. You will also need to create instances of the Atm and BackendSystem types so you can have data to represent the telemetry sources involved in your trace.

Open the tailtracer/model.go file and add the following function to it:

func generateTraces(numberOfTraces int) ptrace.Traces{
	traces := ptrace.NewTraces()

	for i := 0; i <= numberOfTraces; i++{
		newAtm := generateAtm()
		newBackendSystem := generateBackendSystem()
	}

	return traces
}

By now you have heard and read enough about how traces are made up of Spans. You have probably also written some instrumentation code using the SDK’s functions and types available to create them, but what you probably didn’t know, is that within the Collector’s API, that there are other types of “spans” involved in creating a trace.

You will start with a type called ptrace.ResourceSpans which represents the resource and all the operations that it either originated or received while participating in a trace. You can find its definition within the /pdata/internal/data/protogen/trace/v1/trace.pb.go.

ptrace.Traces has a method named ResourceSpans() which returns an instance of a helper type called ptrace.ResourceSpansSlice. The ptrace.ResourceSpansSlice type has methods to help you handle the array of ptrace.ResourceSpans that will contain as many items as the number of Resource entities participating in the request represented by the trace.

ptrace.ResourceSpansSlice has a method named AppendEmpty() that adds a new ptrace.ResourceSpan to the array and return its reference.

Once you have an instance of a ptrace.ResourceSpan you will use a method named Resource() which will return the instance of the pcommon.Resource associated with the ResourceSpan.

Update the generateTrace() function with the following changes:

  • add a variable named resourceSpan to represent the ResourceSpan
  • add a variable named atmResource to represent the pcommon.Resource associated with the ResourceSpan.
  • Use the methods mentioned above to initialize both variables respectively.

Here is what the function should look like after you implemented these changes:

func generateTraces(numberOfTraces int) ptrace.Traces{
	traces := ptrace.NewTraces()

	for i := 0; i <= numberOfTraces; i++{
		newAtm := generateAtm()
		newBackendSystem := generateBackendSystem()

		resourceSpan := traces.ResourceSpans().AppendEmpty()
		atmResource := resourceSpan.Resource()
	}

	return traces
}

Describing Resources through attributes

The Collector’s API provides a package named pcommon (nested under the pdata package) with all the types and helper functions required to describe a Resource.

In the Collector’s world, a Resource is described by attributes in a key/value pair format represented by the pcommon.Map type.

You can check the definition of the pcommon.Map type and the related helper functions to create attribute values using the supported formats in the /pdata/pcommon/common.go file within the OTel Collector’s GitHub project.

Key/value pairs provide a lot of flexibility to help model your Resource data, so the OTel specification has some guidelines in place to help organize and minimize the conflicts across all the different types of telemetry generation entities that it may need to represent.

These guidelines are known as Resource Semantic Conventions and are documented in the OTel specification.

When creating your own attributes to represent your own telemetry generation entities, you should follow the guideline provided by the specification:

Attributes are grouped logically by the type of the concept that they described. Attributes in the same group have a common prefix that ends with a dot. For example all attributes that describe Kubernetes properties start with k8s.

Let’s start by opening the tailtracer/model.go and adding go.opentelemetry.io/collector/pdata/pcommon to the import clause so you can have access to the pcommon package capabilities.

Now go ahead and add a function to read the field values from an Atm instance and write them as attributes (grouped by the prefix “atm.”) into a pcommon.Resource instance. Here is what the function looks like:

func fillResourceWithAtm(resource *pcommon.Resource, atm Atm){
   atmAttrs := resource.Attributes()
   atmAttrs.PutInt("atm.id", atm.ID)
   atmAttrs.PutStr("atm.stateid", atm.StateID)
   atmAttrs.PutStr("atm.ispnetwork", atm.ISPNetwork)
   atmAttrs.PutStr("atm.serialnumber", atm.SerialNumber)
}

The resource semantic conventions also have prescriptive attribute names and well-known values to represent telemetry generation entities that are common and applicable across different domains like compute unit, environment and others.

So, when you look at the BackendSystem entity, it has fields representing OS related information and Cloud related information, and we will use the attribute names and values prescribed by the resource semantic convention to represent that information on its Resource.

All the resource semantic convention attribute names and well known-values are kept within the /semconv/v1.9.0/generated_resource.go file within the Collector’s GitHub project.

Let’s create a function to read the field values from an BackendSystem instance and write them as attributes into a pcommon.Resource instance. Open the tailtracer/model.go file and add the following function:

func fillResourceWithBackendSystem(resource *pcommon.Resource, backend BackendSystem){
	backendAttrs := resource.Attributes()
	var osType, cloudProvider string

	switch {
		case backend.CloudProvider == "amzn":
			cloudProvider = conventions.AttributeCloudProviderAWS
		case backend.OSType == "mcrsft":
			cloudProvider = conventions.AttributeCloudProviderAzure
		case backend.OSType == "gogl":
			cloudProvider = conventions.AttributeCloudProviderGCP
	}

	backendAttrs.PutStr(conventions.AttributeCloudProvider, cloudProvider)
	backendAttrs.PutStr(conventions.AttributeCloudRegion, backend.CloudRegion)

	switch {
		case backend.OSType == "lnx":
			osType = conventions.AttributeOSTypeLinux
		case backend.OSType == "wndws":
			osType = conventions.AttributeOSTypeWindows
		case backend.OSType == "slrs":
			osType = conventions.AttributeOSTypeSolaris
	}

	backendAttrs.PutStr(conventions.AttributeOSType, osType)
	backendAttrs.PutStr(conventions.AttributeOSVersion, backend.OSVersion)
 }

Notice that I didn’t add an attribute named “atm.name” or “backendsystem.name” to the pcommon.Resource representing the Atm and BackendSystem entity names, that’s because most (not to say all) distributed tracing backend systems that are compatible with the OTel trace specification, interpret the pcommon.Resource described in a trace as a Service, therefore they expect the pcommon.Resource to carry a required attribute named service.name as prescribed by the resource semantic convention.

We will also use non-required attribute named service.version to represent the version information for both Atm and BackendSystem entities.

Here is what the tailtracer/model.go file looks like after adding the code for properly assign the “service.” group attributes:

model.go

package tailtracer

import (
	"math/rand"
	"time"
	"go.opentelemetry.io/collector/pdata/pcommon"
	"go.opentelemetry.io/collector/pdata/ptrace"
	conventions "go.opentelemetry.io/collector/semconv/v1.9.0"
)

type Atm struct {
	ID           int64
	Version      string
	Name         string
	StateID      string
	SerialNumber string
	ISPNetwork   string
}

type BackendSystem struct {
	Version       string
	ProcessName   string
	OSType        string
	OSVersion     string
	CloudProvider string
	CloudRegion   string
	Endpoint      string
}

func generateAtm() Atm {
	i := getRandomNumber(1, 2)
	var newAtm Atm

	switch i {
		case 1:
			newAtm = Atm{
				ID:           111,
				Name:         "ATM-111-IL",
				SerialNumber: "atmxph-2022-111",
				Version:      "v1.0",
				ISPNetwork:   "comcast-chicago",
				StateID:      "IL",
			}

		case 2:
			newAtm = Atm{
				ID:           222,
				Name:         "ATM-222-CA",
				SerialNumber: "atmxph-2022-222",
				Version:      "v1.0",
				ISPNetwork:   "comcast-sanfrancisco",
				StateID:      "CA",
			}
	}

	return newAtm
}

func generateBackendSystem() BackendSystem {
	i := getRandomNumber(1, 3)

	newBackend := BackendSystem{
		ProcessName:   "accounts",
		Version:       "v2.5",
		OSType:        "lnx",
		OSVersion:     "4.16.10-300.fc28.x86_64",
		CloudProvider: "amzn",
		CloudRegion:   "us-east-2",
	}

	switch i {
		case 1:
			newBackend.Endpoint = "api/v2.5/balance"
		case 2:
			newBackend.Endpoint = "api/v2.5/deposit"
		case 3:
			newBackend.Endpoint = "api/v2.5/withdrawn"
	}

	return newBackend
}

func getRandomNumber(min int, max int) int {
	rand.Seed(time.Now().UnixNano())
	i := (rand.Intn(max-min+1) + min)
	return i
}

func generateTraces(numberOfTraces int) ptrace.Traces {
	traces := ptrace.NewTraces()

	for i := 0; i <= numberOfTraces; i++ {
		newAtm := generateAtm()
		newBackendSystem := generateBackendSystem()

		resourceSpan := traces.ResourceSpans().AppendEmpty()
		atmResource := resourceSpan.Resource()
		fillResourceWithAtm(&atmResource, newAtm)

		resourceSpan = traces.ResourceSpans().AppendEmpty()
		backendResource := resourceSpan.Resource()
		fillResourceWithBackendSystem(&backendResource, newBackendSystem)
	}

	return traces
}

func fillResourceWithAtm(resource *pcommon.Resource, atm Atm) {
	atmAttrs := resource.Attributes()
	atmAttrs.PutInt("atm.id", atm.ID)
	atmAttrs.PutStr("atm.stateid", atm.StateID)
	atmAttrs.PutStr("atm.ispnetwork", atm.ISPNetwork)
	atmAttrs.PutStr("atm.serialnumber", atm.SerialNumber)
	atmAttrs.PutStr(conventions.AttributeServiceName, atm.Name)
	atmAttrs.PutStr(conventions.AttributeServiceVersion, atm.Version)

}

func fillResourceWithBackendSystem(resource *pcommon.Resource, backend BackendSystem) {
	backendAttrs := resource.Attributes()
	var osType, cloudProvider string

	switch {
		case backend.CloudProvider == "amzn":
			cloudProvider = conventions.AttributeCloudProviderAWS
		case backend.OSType == "mcrsft":
			cloudProvider = conventions.AttributeCloudProviderAzure
		case backend.OSType == "gogl":
			cloudProvider = conventions.AttributeCloudProviderGCP
	}

	backendAttrs.PutStr(conventions.AttributeCloudProvider, cloudProvider)
	backendAttrs.PutStr(conventions.AttributeCloudRegion, backend.CloudRegion)

	switch {
		case backend.OSType == "lnx":
			osType = conventions.AttributeOSTypeLinux
		case backend.OSType == "wndws":
			osType = conventions.AttributeOSTypeWindows
		case backend.OSType == "slrs":
			osType = conventions.AttributeOSTypeSolaris
	}

	backendAttrs.PutStr(conventions.AttributeOSType, osType)
	backendAttrs.PutStr(conventions.AttributeOSVersion, backend.OSVersion)

	backendAttrs.PutStr(conventions.AttributeServiceName, backend.ProcessName)
	backendAttrs.PutStr(conventions.AttributeServiceVersion, backend.Version)
}

Representing operations with spans

You now have a ResourceSpan instance with their respective Resource properly filled with attributes to represent the Atm and BackendSystem entities, you are ready to represent the operations that each Resource execute as part of a trace within the ResourceSpan.

In the OTel world, in order for a system to generate telemetry, it needs to be instrumented either manually or automatically through an instrumentation library.

The instrumentation libraries are responsible to set the scope (also known as the instrumentation scope) in which the operations participating on a trace happened and then describe those operations as spans within the context of the trace.

pdata.ResourceSpans has a method named ScopeSpans() which returns an instance of a helper type called ptrace.ScopeSpansSlice. The ptrace.ScopeSpansSlice type has methods to help you handle the array of ptrace.ScopeSpans that will contain as many items as the number of ptrace.ScopeSpan representing the different instrumentation scopes and the spans it generated within the context of a trace.

ptrace.ScopeSpansSlice has a method named AppendEmpty() that adds a new ptrace.ScopeSpans to the array and return its reference.

Let’s create a function to instantiate a ptrace.ScopeSpans representing for the ATM system’s instrumentation scope and its spans. Open the tailtracer/model.go file and add the following function:

 func appendAtmSystemInstrScopeSpans(resourceSpans *ptrace.ResourceSpans) (ptrace.ScopeSpans){
	scopeSpans := resourceSpans.ScopeSpans().AppendEmpty()

    return scopeSpans
}

The ptrace.ScopeSpans has a method named Scope() that returns a reference for the pcommon.InstrumentationScope instance representing the instrumentation scope that generated the spans.

pcommon.InstrumentationScope has the following methods to describe an instrumentation scope:

  • SetName(v string) sets the name for the instrumentation library

  • SetVersion(v string) sets the version for the instrumentation library

  • Name() string returns the name associated with the instrumentation library

  • Version() string returns the version associated with the instrumentation library

Let’s update the appendAtmSystemInstrScopeSpans function so we can set the name and version of the instrumentation scope for the new ptrace.ScopeSpans. Here is what appendAtmSystemInstrScopeSpans looks like after the update:

 func appendAtmSystemInstrScopeSpans(resourceSpans *ptrace.ResourceSpans) (ptrace.ScopeSpans){
	scopeSpans := resourceSpans.ScopeSpans().AppendEmpty()
	scopeSpans.Scope().SetName("atm-system")
	scopeSpans.Scope().SetVersion("v1.0")
	return scopeSpans
}

You can now update the generateTraces function and add variables to represent the instrumentation scope used by both Atm and BackendSystem entities by initializing them with the appendAtmSystemInstrScopeSpans(). Here is what generateTraces() looks like after the update:

func generateTraces(numberOfTraces int) ptrace.Traces{
	traces := ptrace.NewTraces()

	for i := 0; i <= numberOfTraces; i++{
		newAtm := generateAtm()
		newBackendSystem := generateBackendSystem()

		resourceSpan := traces.ResourceSpans().AppendEmpty()
		atmResource := resourceSpan.Resource()
		fillResourceWithAtm(&atmResource, newAtm)

		atmInstScope := appendAtmSystemInstrScopeSpans(&resourceSpan)

		resourceSpan = traces.ResourceSpans().AppendEmpty()
		backendResource := resourceSpan.Resource()
		fillResourceWithBackendSystem(&backendResource, newBackendSystem)

		backendInstScope := appendAtmSystemInstrScopeSpans(&resourceSpan)
	}

	return traces
}

At this point, you have everything needed to represent the telemetry generation entities in your system and the instrumentation scope that is responsible to identify operations and generate the traces for the system. The next step is to finally create the spans representing the operations that the given instrumentation scope generated as part of a trace.

ptrace.ScopeSpans has a method named Spans() which returns an instance of a helper type called ptrace.SpanSlice. The ptrace.SpanSlice type has methods to help you handle the array of ptrace.Span that will contain as many items as the number of operations the instrumentation scope was able to identify and describe as part of the trace.

ptrace.SpanSlice has a method named AppendEmpty() that adds a new ptrace.Span to the array and return its reference.

ptrace.Span has the following methods to describe an operation:

  • SetTraceID(v pcommon.TraceID) sets the pcommon.TraceID uniquely identifying the trace which this span is associated with

  • SetSpanID(v pcommon.SpanID) sets the pcommon.SpanID uniquely identifying this span within the context of the trace it is associated with

  • SetParentSpanID(v pcommon.SpanID) sets pcommon.SpanID for the parent span/operation in case the operation represented by this span is executed as part of the parent (nested)

  • SetName(v string) sets the name of the operation for the span

  • SetKind(v ptrace.SpanKind) sets ptrace.SpanKind defining what kind of operation the span represents.

  • SetStartTimestamp(v pcommon.Timestamp) sets the pcommon.Timestamp representing the date and time when the operation represented by the span has started

  • SetEndTimestamp(v pcommon.Timestamp) sets the pcommon.Timestamp representing the date and time when the operation represented by the span has ended

As you can see per the methods above, a ptrace.Span is uniquely identified by 2 required IDs; their own unique ID represented by the pcommon.SpanID type and the ID of the trace they are associated with represented by a pcommon.TraceID type.

The pcommon.TraceID has to carry a globally unique ID represented through a 16 byte array and should follow the W3C Trace Context specification while the pcommon.SpanID is a unique ID within the context of the trace they are associated with and it’s represented through a 8 byte array.

The pcommon package provides the following types to generate the span’s IDs:

  • type TraceID [16]byte

  • type SpanID [8]byte

For this tutorial, you will be creating the IDs using functions from github.com/google/uuid package for the pcommon.TraceID and functions from the crypto/rand package to randomly generate the pcommon.SpanID. Open the tailtracer/model.go file and add both packages to the import statement; after that, add the following functions to help generate both IDs:

import (
	crand "crypto/rand"
	"math/rand"
  	...
)

func NewTraceID() pcommon.TraceID {
	return pcommon.TraceID(uuid.New())
}

func NewSpanID() pcommon.SpanID {
	var rngSeed int64
	_ = binary.Read(crand.Reader, binary.LittleEndian, &rngSeed)
	randSource := rand.New(rand.NewSource(rngSeed))

	var sid [8]byte
	randSource.Read(sid[:])
	spanID := pcommon.SpanID(sid)

	return spanID
}

Now that you have a way to properly identify the spans, you can start creating them to represent the operations within and across the entities in your system.

As part of the generateBackendSystem() function, we have randomly assigned the operations that the BackEndSystem entity can provide as services to the system. We will now open the tailtracer/model.go file and a function called appendTraceSpans() that will be responsible to create a trace and append spans representing the BackendSystem operations. Here is what the initial implementation for the appendTraceSpans() function looks like:

func appendTraceSpans(backend *BackendSystem, backendScopeSpans *ptrace.ScopeSpans, atmScopeSpans *ptrace.ScopeSpans) {
	traceId := NewTraceID()
	backendSpanId := NewSpanID()

	backendDuration, _ := time.ParseDuration("1s")
	backendSpanStartTime := time.Now()
	backendSpanFinishTime := backendSpanStartTime.Add(backendDuration)

	backendSpan := backendScopeSpans.Spans().AppendEmpty()
	backendSpan.SetTraceID(traceId)
	backendSpan.SetSpanID(backendSpanId)
	backendSpan.SetName(backend.Endpoint)
	backendSpan.SetKind(ptrace.SpanKindServer)
	backendSpan.SetStartTimestamp(pcommon.NewTimestampFromTime(backendSpanStartTime))
	backendSpan.SetEndTimestamp(pcommon.NewTimestampFromTime(backendSpanFinishTime))
}

You probably noticed that there are 2 references to ptrace.ScopeSpans as parameters in the appendTraceSpans() function, but we only used one of them. Don’t worry about it for now, we will get back to it later.

You will now update the generateTraces() function so it can actually generate the trace by calling the appendTraceSpans() function. Here is what the updated generateTraces() function looks like:

func generateTraces(numberOfTraces int) ptrace.Traces {
	traces := ptrace.NewTraces()

	for i := 0; i <= numberOfTraces; i++ {
		newAtm := generateAtm()
		newBackendSystem := generateBackendSystem()

		resourceSpan := traces.ResourceSpans().AppendEmpty()
		atmResource := resourceSpan.Resource()
		fillResourceWithAtm(&atmResource, newAtm)

		atmInstScope := appendAtmSystemInstrScopeSpans(&resourceSpan)

		resourceSpan = traces.ResourceSpans().AppendEmpty()
		backendResource := resourceSpan.Resource()
		fillResourceWithBackendSystem(&backendResource, newBackendSystem)

		backendInstScope := appendAtmSystemInstrScopeSpans(&resourceSpan)

		appendTraceSpans(&newBackendSystem, &backendInstScope, &atmInstScope)
	}

	return traces
}

You now have the BackendSystem entity and its operations represented in spans within a proper trace context! All you need to do is to push the generated trace through the pipeline so the next consumer (either a processor or an exporter) can receive and process it.

consumer.Traces has a method called ConsumeTraces() which is responsible to push the generated traces to the next consumer in the pipeline. All you need to do now is to update the Start() method within the tailtracerReceiver type and add the code to use it.

Open the tailtracer/trace-receiver.go file and update the Start() method as follow:

func (tailtracerRcvr *tailtracerReceiver) Start(ctx context.Context, host component.Host) error {
    tailtracerRcvr.host = host
    ctx = context.Background()
	ctx, tailtracerRcvr.cancel = context.WithCancel(ctx)

	interval, _ := time.ParseDuration(tailtracerRcvr.config.Interval)
	go func() {
		ticker := time.NewTicker(interval)
		defer ticker.Stop()
		for {
			select {
				case <-ticker.C:
					tailtracerRcvr.logger.Info("I should start processing traces now!")
					tailtracerRcvr.nextConsumer.ConsumeTraces(ctx, generateTraces(tailtracerRcvr.config.NumberOfTraces))
				case <-ctx.Done():
					return
			}
		}
	}()

	return nil
}

If you run your otelcol-dev, here is what the output should look like after 2 minutes running:

Starting: /Users/rquedas/go/bin/dlv dap --check-go-version=false --listen=127.0.0.1:54625 --log-dest=3 from /Users/rquedas/Documents/vscode-workspace/otel4devs/collector/receiver/trace-receiver/otelcol-dev
DAP server listening at: 127.0.0.1:54625
2023-09-28T08:59:52.111-0700    info    service@v0.86.0/telemetry.go:84 Setting up own telemetry...
2023-09-28T08:59:52.111-0700    info    service@v0.86.0/telemetry.go:201        Serving Prometheus metrics      {"address": ":8888", "level": "Basic"}
2023-09-28T08:59:52.111-0700    debug   exporter@v0.86.0/exporter.go:273        Stable component.       {"kind": "exporter", "data_type": "traces", "name": "otlp/jaeger"}
2023-09-28T08:59:52.112-0700    info    exporter@v0.86.0/exporter.go:275        Development component. May change in the future.        {"kind": "exporter", "data_type": "traces", "name": "debug"}
2023-09-28T08:59:52.112-0700    debug   receiver@v0.86.0/receiver.go:294        Stable component.       {"kind": "receiver", "name": "otlp", "data_type": "traces"}
2023-09-28T08:59:52.112-0700    debug   receiver@v0.86.0/receiver.go:294        Alpha component. May change in the future.      {"kind": "receiver", "name": "tailtracer", "data_type": "traces"}
2023-09-28T08:59:52.112-0700    info    service@v0.86.0/service.go:138  Starting otelcol-dev... {"Version": "1.0.0", "NumCPU": 10}
2023-09-28T08:59:52.112-0700    info    extensions/extensions.go:31     Starting extensions...
2023-09-28T08:59:52.113-0700    info    otlpreceiver@v0.86.0/otlp.go:83 Starting GRPC server    {"kind": "receiver", "name": "otlp", "data_type": "traces", "endpoint": "0.0.0.0:4317"}
2023-09-28T08:59:52.113-0700    info    service@v0.86.0/service.go:161  Everything is ready. Begin running and processing data.
2023-09-28T08:59:52.113-0700    info    zapgrpc/zapgrpc.go:178  [core] [Server #3 ListenSocket #4] ListenSocket created {"grpc_log": true}
2023-09-28T08:59:52.124-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1 SubChannel #2] Subchannel Connectivity change to READY       {"grpc_log": true}
2023-09-28T08:59:52.124-0700    info    zapgrpc/zapgrpc.go:178  [core] [pick-first-lb 0x1400054fd10] Received SubConn state update: 0x1400054fec0, {ConnectivityState:READY ConnectionError:<nil>}       {"grpc_log": true}
2023-09-28T08:59:52.124-0700    info    zapgrpc/zapgrpc.go:178  [core] [Channel #1] Channel Connectivity change to READY        {"grpc_log": true}
2023-09-28T09:00:52.113-0700    info    tailtracer/receiver.go:33       I should start processing traces now!   {"kind": "receiver", "name": "tailtracer", "data_type": "traces"}

2023-09-28T09:00:52.743-0700	INFO	debugexporter/debug_exporter.go:40	TracesExporter	{"#spans": 1}
2023-09-28T09:00:52.743-0700	DEBUG	debugexporter/debug_exporter.go:49	ResourceSpans #0
Resource SchemaURL:
Resource labels:
     -> atm.id: INT(222)
     -> atm.stateid: STRING(CA)
     -> atm.ispnetwork: STRING(comcast-sanfrancisco)
     -> atm.serialnumber: STRING(atmxph-2022-222)
     -> service.name: STRING(ATM-222-CA)
     -> service.version: STRING(v1.0)
ScopeSpans #0
ScopeSpans SchemaURL:
InstrumentationScope atm-system v1.0
ResourceSpans #1
Resource SchemaURL:
Resource labels:
     -> cloud.provider: STRING(aws)
     -> cloud.region: STRING(us-east-2)
     -> os.type: STRING(linux)
     -> os.version: STRING(4.16.10-300.fc28.x86_64)
     -> service.name: STRING(accounts)
     -> service.version: STRING(v2.5)
ScopeSpans #0
ScopeSpans SchemaURL:
InstrumentationScope atm-system v1.0
Span #0
    Trace ID       : 5cce8a774d4546c2a5cbdeb607ec74c9
    Parent ID      :
    ID             : bb25c05c7fb13084
    Name           : api/v2.5/balance
    Kind           : SPAN_KIND_SERVER
    Start time     : 2023-09-28 09:00:52.743385 +0000 UTC
    End time       : 2023-09-28 09:00:53.743385 +0000 UTC
    Status code    : STATUS_CODE_OK
    Status message :
2023-09-28T09:00:52.743-0500	info	tailtracer/trace-receiver.go:33	I should start processing traces now!	{"kind": "receiver", "name": "tailtracer"}
2023-09-28T09:00:52.744-0500	INFO	debugexporter/debug_exporter.go:40	TracesExporter	{"#spans": 1}
2023-09-28T09:00:52.744-0500	DEBUG	debugexporter/debug_exporter.go:49	ResourceSpans #0
Resource SchemaURL:
Resource labels:
     -> atm.id: INT(111)
     -> atm.stateid: STRING(IL)
     -> atm.ispnetwork: STRING(comcast-chicago)
     -> atm.serialnumber: STRING(atmxph-2022-111)
     -> service.name: STRING(ATM-111-IL)
     -> service.version: STRING(v1.0)
ScopeSpans #0
ScopeSpans SchemaURL:
InstrumentationScope atm-system v1.0
ResourceSpans #1
Resource SchemaURL:
Resource labels:
     -> cloud.provider: STRING(aws)
     -> cloud.region: STRING(us-east-2)
     -> os.type: STRING(linux)
     -> os.version: STRING(4.16.10-300.fc28.x86_64)
     -> service.name: STRING(accounts)
     -> service.version: STRING(v2.5)
ScopeSpans #0
ScopeSpans SchemaURL:
InstrumentationScope atm-system v1.0
Span #0
    Trace ID       : 8a6ca822db0847f48facfebbb08bbb9e
    Parent ID      :
    ID             : 7cf668c1273ecee5
    Name           : api/v2.5/withdrawn
    Kind           : SPAN_KIND_SERVER
    Start time     : 2023-09-28 09:00:52.74404 +0000 UTC
    End time       : 2023-09-28 09:00:53.74404 +0000 UTC
    Status code    : STATUS_CODE_OK
    Status message :

Here is what the generated trace looks like in Jaeger: Jaeger trace

What you currently see in Jaeger is the representation of a service that is receiving a request from an external entity that isn’t instrumented by an OTel SDK, therefore it can’t be identified as the origin/start of the trace. In order for a ptrace.Span to understand it is representing an operation that was execute as a result of another operation originated either within or outside (nested/child) of the Resource within the same trace context you will need to:

  • Set the same trace context as the caller operation by calling the SetTraceID() method and passing the pcommon.TraceID of the parent/caller ptrace.Span as a parameter.
  • Define who is the caller operation within the context of the trace by calling SetParentId() method and passing the pcommon.SpanID of the parent/caller ptrace.Span as a parameter.

You will now create a ptrace.Span representing the Atm entity operations and set it as the parent for BackendSystem span. Open the tailtracer/model.go file and update the appendTraceSpans() function as follow:

func appendTraceSpans(backend *BackendSystem, backendScopeSpans *ptrace.ScopeSpans, atmScopeSpans *ptrace.ScopeSpans) {
	traceId := NewTraceID()

	var atmOperationName string

	switch {
		case strings.Contains(backend.Endpoint, "balance"):
			atmOperationName = "Check Balance"
		case strings.Contains(backend.Endpoint, "deposit"):
			atmOperationName = "Make Deposit"
		case strings.Contains(backend.Endpoint, "withdraw"):
			atmOperationName = "Fast Cash"
		}

	atmSpanId := NewSpanID()
    atmSpanStartTime := time.Now()
    atmDuration, _ := time.ParseDuration("4s")
    atmSpanFinishTime := atmSpanStartTime.Add(atmDuration)


	atmSpan := atmScopeSpans.Spans().AppendEmpty()
	atmSpan.SetTraceID(traceId)
	atmSpan.SetSpanID(atmSpanId)
	atmSpan.SetName(atmOperationName)
	atmSpan.SetKind(ptrace.SpanKindClient)
	atmSpan.Status().SetCode(ptrace.StatusCodeOk)
	atmSpan.SetStartTimestamp(pcommon.NewTimestampFromTime(atmSpanStartTime))
	atmSpan.SetEndTimestamp(pcommon.NewTimestampFromTime(atmSpanFinishTime))


	backendSpanId := NewSpanID()

	backendDuration, _ := time.ParseDuration("2s")
    backendSpanStartTime := atmSpanStartTime.Add(backendDuration)


	backendSpan := backendScopeSpans.Spans().AppendEmpty()
	backendSpan.SetTraceID(atmSpan.TraceID())
	backendSpan.SetSpanID(backendSpanId)
	backendSpan.SetParentSpanID(atmSpan.SpanID())
	backendSpan.SetName(backend.Endpoint)
	backendSpan.SetKind(ptrace.SpanKindServer)
	backendSpan.Status().SetCode(ptrace.StatusCodeOk)
	backendSpan.SetStartTimestamp(pcommon.NewTimestampFromTime(backendSpanStartTime))
	backendSpan.SetEndTimestamp(atmSpan.EndTimestamp())

}

Go ahead and run your otelcol-dev again and after 2 minutes running, you should start seeing traces in Jaeger like the following: Jaeger trace

We now have services representing both the Atm and the BackendSystem telemetry generation entities in our system and can fully understand how both entities are being used and contributing to the performance of an operation executed by an user.

Here is the detailed view of one of those traces in Jaeger: Jaeger trace

That’s it! You have now reached the end of this tutorial and successfully implemented a trace receiver, congratulations!

  1. Prior to v0.86.0 use the loggingexporter instead of debugexporter. ↩︎ ↩︎

Last modified November 5, 2023: [CI] Terminology fixes from patched terminology textlint rule (#3496) (e97f5220)