Quick links:
- Building your own exporter
- Building your own instrumentation library
- Building your own processor
- Building your own sampler
- Building your own resource detector
- Registration extension method guidance for library authors
- References
OpenTelemetry .NET SDK has provided the following built-in trace exporters:
Custom exporters can be implemented to send telemetry data to places which are not covered by the built-in exporters:
- Exporters should derive from
OpenTelemetry.BaseExporter<Activity>
(which belongs to the OpenTelemetry package) and implement theExport
method. - Exporters can optionally implement the
OnForceFlush
andOnShutdown
method. - Depending on user's choice and load on the application,
Export
may get called with one or more activities. - Exporters will only receive sampled-in and ended activities.
- Exporters should not throw exceptions from
Export
,OnForceFlush
andOnShutdown
. - Exporters should not modify activities they receive (the same activity may be exported again by different exporter).
- Exporters are responsible for any retry logic needed by the scenario. The SDK does not implement any retry logic.
- Exporters should avoid generating telemetry and causing live-loop, this can be
done via
OpenTelemetry.SuppressInstrumentationScope
. - Exporters should use
Activity.TagObjects
collection instead ofActivity.Tags
to obtain the full set of attributes (tags).Activity.Tags
only returns tags whose value are of typestring
(source). For improved performance, use Activity.EnumerateTagObjects if planning to enumerate over all TagObjects. - Exporters should use
ParentProvider.GetResource()
to get theResource
associated with the provider.
class MyExporter : BaseExporter<Activity>
{
public override ExportResult Export(in Batch<Activity> batch)
{
using var scope = SuppressInstrumentationScope.Begin();
foreach (var activity in batch)
{
Console.WriteLine($"Export: {activity.DisplayName}");
}
return ExportResult.Success;
}
}
A demo exporter which simply writes activity name to the console is shown here.
Apart from the exporter itself, you should also provide extension methods as
shown here. This allows users to add the Exporter
to the TracerProvider
as shown in the example here. See
here for more
detailed extension method guidance.
DiagnosticSource
package did not originally have a dedicated field for storing
Status,
and hence, users were encouraged to follow the convention of storing status
using tags "otel.status_code" and "otel.status_description".
DiagnosticSource
version 6.0.0 added Status
and StatusDescription
to Activity
class.
Exporters which support reading status from Activity
directly should fall back
to retrieving status from the tags described above, to maintain backward
compatibility.
ConsoleActivityExporter
may be used as a reference.
The inspiration of the OpenTelemetry project is to make every library observable out of the box by having them call OpenTelemetry API directly. However, many libraries will not have such integration, and as such there is a need for a separate library which would inject such calls, using mechanisms such as wrapping interfaces, subscribing to library-specific callbacks, or translating existing telemetry into OpenTelemetry model.
A library which enables instrumentation for another library is called Instrumentation Library and the library it instruments is called the Instrumented Library. If a given library has built-in instrumentation with OpenTelemetry, then instrumented library and instrumentation library will be the same.
The OpenTelemetry .NET Github repo ships the following instrumentation libraries. The individual docs for them describes the library they instrument, and steps for enabling them.
More community contributed instrumentations are available in OpenTelemetry .NET Contrib. If you are writing an instrumentation library yourself, use the following guidelines.
This section describes the steps required to write a custom instrumentation library.
Note
If you are writing a new library or modifying an existing library the
recommendation is to use the ActivitySource API/OpenTelemetry
API
to emit activity/span instances directly. If a library is instrumented using the
ActivitySource
API then there isn't a need for a separate instrumentation
library to exist. Users simply need to configure the OpenTelemetry SDK to listen
to the ActivitySource
used by the library by calling AddSource
on the
TracerProviderBuilder
being configured. The following section is applicable
only if you are writing an instrumentation library for something you cannot
modify to emit activity/span instances directly.
Writing an instrumentation library typically involves 3 steps.
-
The first step involves attaching to the target library. The exact attachment mechanism will depend on the implementation details of the target library itself. For example, System.Data.SqlClient when running on .NET Framework happens to publish events using an
EventSource
which the SqlClient instrumentation library listens to in order to trigger code as Sql commands are executed. The .NET Framework HttpWebRequest instrumentation patches the runtime code (using reflection) and swaps a static reference that gets invoked as requests are processed for custom code. Every library will be different. -
The second step is to emit activity instances using the ActivitySource API on behalf of the target library. Irrespective of the actual mechanism used in first step, this should be uniform across all instrumentation libraries. The
ActivitySource
must be created using the name and version of the instrumentation library (eg: "OpenTelemetry.Instrumentation.Http") and NOT the instrumented library (eg: "System.Net.Http")- Context
Propagation:
If your library initiates out of process requests or accepts them, the
library needs to inject the
PropagationContext
to outgoing requests and extract the context and hydrate the Activity/Baggage upon receiving incoming requests. This is only required if you're using your own protocol to communicate over the wire. (i.e. If you're using an already instrumented HttpClient or GrpcClient, this is already provided to you and do not require injecting/extractingPropagationContext
explicitly again.)
- Context
Propagation:
If your library initiates out of process requests or accepts them, the
library needs to inject the
-
The third step is an optional step, and involves providing extension methods on
TracerProviderBuilder
and/orIServiceCollection
to enable the instrumentation. For help in choosing see: Registration extension method guidance for library authors. This is optional, and the below guidance should be followed:-
If the instrumentation library requires state management tied to that of
TracerProvider
then it should:-
Implement
IDisposable
. -
Provide an extension method which calls
AddSource
(to enable itsActivitySource
) andAddInstrumentation
(to enable state management) on theTracerProviderBuilder
being configured.An example instrumentation using this approach is SqlClient instrumentation.
[!WARNING] The instrumentation libraries requiring state management are usually hard to auto-instrument. Therefore, they take the risk of not being supported by OpenTelemetry .NET Automatic Instrumentation.
-
-
If the instrumentation library does not require any state management, then providing an extension method is optional.
-
If an extension is provided it should call
AddSource
on theTracerProviderBuilder
being configured to enable itsActivitySource
. -
If an extension is not provided, then the name of the
ActivitySource
used by the instrumented library must be documented so that end users can enable it by callingAddSource
on theTracerProviderBuilder
being configured.[!NOTE] Changing the name of the source should be considered a breaking change.
-
-
There is a special case for libraries which are already instrumented to produce
Activity,
but using the
DiagnosticSource
method. These are referred to as "legacy Activity" in this repo. These libraries
already create activities but they do so by using the Activity
constructor
directly, rather than using ActivitySource.StartActivity
method. These
activities does not by default runs through the sampler, and will have their
Kind
set to internal and they'll have empty ActivitySource name associated
with it.
Some common examples of such libraries include ASP.NET Core, HTTP client .NET Core . Instrumentation libraries for these are already provided in this repo. The OpenTelemetry .NET Contrib repository also has instrumentations for libraries like ElasticSearchClient etc. which fall in this category.
If you are writing instrumentation for such library, it is recommended to refer to one of the above as a reference.
OpenTelemetry .NET SDK has provided the following built-in processors:
Custom processors can be implemented to cover more scenarios:
- Processors should inherit from
OpenTelemetry.BaseProcessor<Activity>
(which belongs to the OpenTelemetry package), and implement theOnStart
andOnEnd
methods. - Processors can optionally implement the
OnForceFlush
andOnShutdown
methods.OnForceFlush
should be thread safe. - Processors should not throw exceptions from
OnStart
,OnEnd
,OnForceFlush
andOnShutdown
. OnStart
andOnEnd
should be thread safe, and should not block or take long time, since they will be called on critical code path.- Processors should use
Activity.TagObjects
collection instead ofActivity.Tags
to obtain the full set of attributes (tags).Activity.Tags
only returns tags whose value are of typestring
(source). For improved performance, use Activity.EnumerateTagObjects if planning to enumerate over all TagObjects.
class MyProcessor : BaseProcessor<Activity>
{
public override void OnStart(Activity activity)
{
Console.WriteLine($"OnStart: {activity.DisplayName}");
}
public override void OnEnd(Activity activity)
{
Console.WriteLine($"OnEnd: {activity.DisplayName}");
}
}
A demo processor is shown here.
A common use case of writing custom processor is to enrich activities with additional tags. An example of such an "EnrichingProcessor" is shown here. Such processors must be added before the exporters.
This processor also shows how to enrich Activity
with additional tags from the
Baggage
.
Many instrumentation libraries shipped from this
repo provides a built-in Enrich
option, which may also be used to enrich
activities. Instrumentation library provided approach may offer additional
capabilities such as offering easy access to more context (library specific).
Another common use case of writing custom processor is to filter Activities from
being exported. Such a "FilteringProcessor" can be written to toggle the
Activity.Recorded
flag. An example "FilteringProcessor" is shown
here.
When using such a filtering processor it should be registered BEFORE the processor containing the exporter which should be bypassed:
using var tracerProvider = Sdk.CreateTracerProviderBuilder()
.SetSampler(new MySampler())
.AddSource("OTel.Demo")
.AddProcessor(new MyFilteringProcessor(activity => true))
.AddProcessor(new SimpleActivityExportProcessor(new MyExporter("ExporterX")))
.Build();
Most instrumentation libraries shipped from this
repo provides a built-in Filter
option to achieve the same effect. In such
cases, it is recommended to use that option as it offers higher performance.
OpenTelemetry .NET SDK has provided the following built-in samplers:
Custom samplers can be implemented to cover more scenarios:
- Samplers should inherit from
OpenTelemetry.Trace.Sampler
(which belongs to the OpenTelemetry package), and implement theShouldSample
method. ShouldSample
should be thread safe, and should not block or take long time, since it will be called on critical code path.
class MySampler : Sampler
{
public override SamplingResult ShouldSample(in SamplingParameters samplingParameters)
{
return new SamplingResult(SamplingDecision.RecordAndSampled);
}
}
A demo sampler is shown here.
Note
This information applies to the OpenTelemetry SDK version 1.4.0 and newer only.
Library authors are encouraged to provide extension methods users may call to
register custom OpenTelemetry components into their TracerProvider
s. These
extension methods can target either the TracerProviderBuilder
or the
IServiceCollection
classes. Both of these patterns are described below.
Note
Libraries providing SDK plugins such as exporters, resource detectors,
and/or samplers should take a dependency on the OpenTelemetry SDK
package. Library authors
providing instrumentation should take a dependency on OpenTelemetry.Api
or
OpenTelemetry.Api.ProviderBuilderExtensions
package.
OpenTelemetry.Api.ProviderBuilderExtensions
exposes interfaces for accessing
the IServiceCollection
which is a requirement for supporting the .NET Options
pattern.
When providing registration extensions:
-
DO support the .NET Options pattern and DO support named options. The Options pattern allows users to bind configuration to options classes and provides extension points for working with instances as they are created. Multiple providers may exist in the same application for a single configuration and multiple components (for example exporters) may exist in the same provider. Named options help users target configuration to specific components.
-
Use the Configure extension to register configuration callbacks for a given name.
-
Use the IOptionsMonitor<T>.Get method to access options class instances by name.
-
-
DO throw exceptions for issues that prevent the component being registered from starting. The OpenTelemetry SDK is allowed to crash if it cannot be started. It MUST NOT crash once running.
Note
The SDK implementation of TracerProviderBuilder
ensures that the
.NET
Configuration
engine is always available by creating a root IConfiguration
from environment
variables if it does not already exist in the IServiceCollection
containing
the TracerProvider
. Library authors can rely on IConfiguration
always being
present in the final IServiceProvider
.
When registering pipeline components (for example samplers, exporters, or
resource detectors) it is recommended to use the TracerProviderBuilder
as the
target type for registration extension methods. These extensions will be highly
discoverable for users interacting with the TracerProviderBuilder
in their IDE
of choice.
The following example shows how to register a custom exporter with named options
support using a TracerProviderBuilder
extension.
using System.Diagnostics;
using Microsoft.Extensions.DependencyInjection.Extensions;
using Microsoft.Extensions.Options;
using MyLibrary;
using OpenTelemetry;
using OpenTelemetry.Metrics;
using OpenTelemetry.Trace;
namespace OpenTelemetry.Trace
{
public static class MyLibraryTracerProviderBuilderRegistrationExtensions
{
public static TracerProviderBuilder AddMyLibraryExporter(
this TracerProviderBuilder builder,
string? name = null,
Action<MyExporterOptions>? configureExporterOptions = null,
Action<BatchExportActivityProcessorOptions>? configureBatchProcessorOptions = null)
{
ArgumentNullException.ThrowIfNull(builder);
// Support named options.
name ??= Options.DefaultName;
builder.ConfigureServices(services =>
{
if (configureExporterOptions != null)
{
// Support configuration through Options API.
services.Configure(name, configureExporterOptions);
}
if (configureBatchProcessorOptions != null)
{
// Support configuration through Options API.
services.Configure(name, configureBatchProcessorOptions);
}
// Register custom service as a singleton.
services.TryAddSingleton<MyCustomService>();
});
builder.AddProcessor(serviceProvider =>
{
// Retrieve MyExporterOptions instance using name.
var exporterOptions = serviceProvider.GetRequiredService<IOptionsMonitor<MyExporterOptions>>().Get(name);
// Retrieve BatchExportActivityProcessorOptions instance using name.
var batchOptions = serviceProvider.GetRequiredService<IOptionsMonitor<BatchExportActivityProcessorOptions>>().Get(name);
// Retrieve MyCustomService singleton.
var myCustomService = serviceProvider.GetRequiredService<MyCustomService>();
// Return a batch export processor using MyCustomExporter.
return new BatchActivityExportProcessor(
new MyCustomExporter(exporterOptions, myCustomService),
batchOptions.MaxQueueSize,
batchOptions.ScheduledDelayMilliseconds,
batchOptions.ExporterTimeoutMilliseconds,
batchOptions.MaxExportBatchSize);
});
// Return builder for call chaining.
return builder;
}
}
}
namespace MyLibrary
{
// Options class can be bound to IConfiguration or configured by code.
public class MyExporterOptions
{
public Uri? IngestionUri { get; set; }
}
internal sealed class MyCustomExporter : BaseExporter<Activity>
{
public MyCustomExporter(
MyExporterOptions options,
MyCustomService myCustomService)
{
// Implementation not shown.
}
public override ExportResult Export(in Batch<Activity> batch)
{
// Implementation not shown.
return ExportResult.Success;
}
}
internal sealed class MyCustomService
{
// Implementation not shown.
}
}
When providing TracerProviderBuilder
registration extensions:
-
DO Use the
OpenTelemetry.Trace
namespace forTracerProviderBuilder
registration extensions to help with discoverability. -
DO Return the
TracerProviderBuilder
passed in to support call chaining of registration methods. -
DO Use the
TracerProviderBuilder.ConfigureServices
extension method to register dependent services. -
DO Use the dependency injection extension methods utilizing factory patterns to perform configuration once the final
IServiceProvider
is available.
When registering instrumentation or listening to telemetry in a library
providing other features it is recommended to use the IServiceCollection
as
the target type for registration extension methods.
The following example shows how a library might enable tracing and metric
support using an IServiceCollection
extension by calling
ConfigureOpenTelemetryTracerProvider
and
ConfigureOpenTelemetryMeterProvider
.
using Microsoft.Extensions.DependencyInjection.Extensions;
using Microsoft.Extensions.Options;
using MyLibrary;
using OpenTelemetry.Metrics;
using OpenTelemetry.Trace;
namespace Microsoft.Extensions.DependencyInjection
{
public static class MyLibraryServiceCollectionRegistrationExtensions
{
public static IServiceCollection AddMyLibrary(
this IServiceCollection services,
string? name = null,
Action<MyLibraryOptions>? configure = null)
{
ArgumentNullException.ThrowIfNull(services);
// Register library services.
services.TryAddSingleton<IMyLibraryService, MyLibraryService>();
// Support named options.
name ??= Options.Options.DefaultName;
if (configure != null)
{
// Support configuration through Options API.
services.Configure(name, configure);
}
// Configure OpenTelemetry tracing.
services.ConfigureOpenTelemetryTracerProvider((sp, builder) =>
{
var options = sp.GetRequiredService<IOptionsMonitor<MyLibraryOptions>>().Get(name);
if (options.EnableTracing)
{
builder.AddSource("MyLibrary");
}
});
// Configure OpenTelemetry metrics.
services.ConfigureOpenTelemetryMeterProvider((sp, builder) =>
{
var options = sp.GetRequiredService<IOptionsMonitor<MyLibraryOptions>>().Get(name);
if (options.EnableMetrics)
{
builder.AddMeter("MyLibrary");
}
});
return services;
}
}
}
namespace MyLibrary
{
// Options class can be bound to IConfiguration or configured by code.
public class MyLibraryOptions
{
public bool EnableTracing { get; set; }
public bool EnableMetrics { get; set; }
}
internal sealed class MyLibraryService : IMyLibraryService
{
// Implementation not shown.
}
public interface IMyLibraryService
{
// Implementation not shown.
}
}
The benefit to using the IServiceCollection
style is users only need to call a
single AddMyLibrary
extension to configure the library itself and optionally
turn on OpenTelemetry integration for multiple signals (tracing & metrics in
this case).
Note
ConfigureOpenTelemetryTracerProvider
and
ConfigureOpenTelemetryMeterProvider
do not automatically start OpenTelemetry.
The host is responsible for either calling AddOpenTelemetry
in the
OpenTelemetry.Extensions.Hosting
package, calling Build
when using the Sdk.CreateTracerProviderBuilder
and
Sdk.CreateMeterProviderBuilder
methods, or by accessing the TracerProvider
and MeterProvider
from the IServiceCollection
where configuration was
performed.
When providing IServiceCollection
registration extensions:
-
DO Use the
Microsoft.Extensions.DependencyInjection
namespace forIServiceCollection
registration extensions to help with discoverability. -
DO Return the
IServiceCollection
passed in to support call chaining of registration methods. -
DO Use the
IServiceCollection
directly to register dependent services. -
DO Use the dependency injection extension methods utilizing factory patterns to perform configuration once the final
IServiceProvider
is available.