Skip to content

Latest commit

 

History

History
545 lines (445 loc) · 23.9 KB

README.md

File metadata and controls

545 lines (445 loc) · 23.9 KB
Raw

Customizing OpenTelemetry .NET SDK for Metrics

MeterProvider

As shown in the getting-startedgetting started in 5 minutes - Console Application doc, a valid MeterProvider must be configured and built to collect metrics with OpenTelemetry .NET Sdk. MeterProvider holds all the configuration for metrics like MetricReaders, Views, etc. Naturally, almost all the customizations must be done on the MeterProvider.

Building a MeterProvider

Building a MeterProvider is done using MeterProviderBuilder which must be obtained by calling Sdk.CreateMeterProviderBuilder(). MeterProviderBuilder exposes various methods which configure the provider it is going to build. These include methods like AddMeter, AddView etc, and are explained in subsequent sections of this document. Once configuration is done, calling Build() on the MeterProviderBuilder builds the MeterProvider instance. Once built, changes to its configuration is not allowed. In most cases, a single MeterProvider is created at the application startup, and is disposed when application shuts down.

The snippet below shows how to build a basic MeterProvider. This will create a provider with default configuration, and is not particularly useful. The subsequent sections show how to build a more useful provider.

using OpenTelemetry;
using OpenTelemetry.Metrics;

using var meterProvider = Sdk.CreateMeterProviderBuilder().Build();

In a typical application, a single MeterProvider is created at application startup and disposed at application shutdown. It is important to ensure that the provider is not disposed too early. Actual mechanism depends on the application type. For example, in a typical ASP.NET application, MeterProvider is created in Application_Start, and disposed in Application_End (both methods are a part of the Global.asax.cs file) as shown here. In a typical ASP.NET Core application, MeterProvider lifetime is managed by leveraging the built-in Dependency Injection container as shown here.

MeterProvider configuration

MeterProvider holds the metrics configuration, which includes the following:

  1. The list of Meters from which instruments are created to report measurements.
  2. The list of instrumentations enabled via Instrumentation Library.
  3. The list of MetricReaders, including exporting readers which exports metrics to Exporters
  4. The Resource associated with the metrics.
  5. The list of Views to be used.

Meter

Meter is used for creating Instruments, which are then used to report Measurements. The SDK follows an explicit opt-in model for listening to meters. i.e, by default, it listens to no meters. Every meter which is used to create instruments must be explicitly added to the meter provider.

AddMeter method on MeterProviderBuilder can be used to add a Meter to the provider. The name of the Meter (case-insensitive) must be provided as an argument to this method. AddMeter can be called multiple times to add more than one meters. It also supports wildcard subscription model. It is important to note that all the instruments from the meter will be enabled, when a Meter is added. To selectively drop some instruments from a Meter, use the View feature, as shown here.

It is not possible to add meters once the provider is built by the Build() method on the MeterProviderBuilder.

The snippet below shows how to add meters to the provider.

using OpenTelemetry;
using OpenTelemetry.Metrics;

using var meterProvider = Sdk.CreateMeterProviderBuilder()
    // The following enables instruments from Meter
    // named "MyCompany.MyProduct.MyLibrary" only.
    .AddMeter("MyCompany.MyProduct.MyLibrary")
    // The following enables instruments from all Meters
    // whose name starts with  "AbcCompany.XyzProduct.".
    .AddMeter("AbcCompany.XyzProduct.*")
    .Build();

See Program.cs for complete example.

Note

A common mistake while configuring MeterProvider is forgetting to add the required Meters to the provider. It is recommended to leverage the wildcard subscription model where it makes sense. For example, if your application is expecting to enable instruments from a number of libraries from a company "Abc", the you can use AddMeter("Abc.*") to enable all meters whose name starts with "Abc.".

View

A View provides the ability to customize the metrics that are output by the SDK. Following sections explains how to use this feature. Each section has two code snippets. The first one uses an overload of AddView method that takes in the name of the instrument as the first parameter. The View configuration is then applied to the matching instrument name. The second code snippet shows how to use an advanced selection criteria to achieve the same results. This requires the user to provide a Func<Instrument, MetricStreamConfiguration> which offers more flexibility in filtering the instruments to which the View should be applied.

Rename an instrument

When SDK produces Metrics, the name of Metric is by default the name of the instrument. View may be used to rename a metric to a different name. This is particularly useful if there are conflicting instrument names, and you do not own the instrument to create it with a different name.

    // Rename an instrument to new name.
    .AddView(instrumentName: "MyCounter", name: "MyCounterRenamed")
    // Advanced selection criteria and config via Func<Instrument, MetricStreamConfiguration>
    .AddView((instrument) =>
    {
        if (instrument.Meter.Name == "CompanyA.ProductB.LibraryC" &&
            instrument.Name == "MyCounter")
        {
            return new MetricStreamConfiguration() { Name = "MyCounterRenamed" };
        }

        return null;
    })

Drop an instrument

When using AddMeter to add a Meter to the provider, all the instruments from that Meter gets subscribed. Views can be used to selectively drop an instrument from a Meter. If the goal is to drop every instrument from a Meter, then it is recommended to simply not add that Meter using AddMeter.

    // Drop the instrument "MyCounterDrop".
    .AddView(instrumentName: "MyCounterDrop", MetricStreamConfiguration.Drop)
    // Advanced selection criteria and config via Func<Instrument, MetricStreamConfiguration>
    .AddView((instrument) =>
    {
        if (instrument.Meter.Name == "CompanyA.ProductB.LibraryC" &&
            instrument.Name == "MyCounterDrop")
        {
            return MetricStreamConfiguration.Drop;
        }

        return null;
    })

Select specific tags

When recording a measurement from an instrument, all the tags that were provided are reported as dimensions for the given metric. Views can be used to selectively choose a subset of dimensions to report for a given metric. This is useful when you have a metric for which only a few of the dimensions associated with the metric are of interest to you.

    // Only choose "name" as the dimension for the metric "MyFruitCounter"
    .AddView(
        instrumentName: "MyFruitCounter",
        metricStreamConfiguration: new MetricStreamConfiguration
        {
            TagKeys = new string[] { "name" },
        })

    ...
    // Only the dimension "name" is selected, "color" is dropped
    MyFruitCounter.Add(1, new("name", "apple"), new("color", "red"));
    MyFruitCounter.Add(2, new("name", "lemon"), new("color", "yellow"));
    MyFruitCounter.Add(2, new("name", "apple"), new("color", "green"));
    // Because "color" is dropped the resulting metric values are - name:apple LongSum Value:3 and name:lemon LongSum Value:2
    ...

    // If you provide an empty `string` array as `TagKeys` to the `MetricStreamConfiguration`
    // the SDK will drop all the dimensions associated with the metric
    .AddView(
        instrumentName: "MyFruitCounter",
        metricStreamConfiguration: new MetricStreamConfiguration
        {
            TagKeys = Array.Empty<string>(),
        })

    ...
    // both "name" and "color" are dropped
    MyFruitCounter.Add(1, new("name", "apple"), new("color", "red"));
    MyFruitCounter.Add(2, new("name", "lemon"), new("color", "yellow"));
    MyFruitCounter.Add(2, new("name", "apple"), new("color", "green"));
    // Because both "name" and "color" are dropped the resulting metric value is - LongSum Value:5
    ...
    // Advanced selection criteria and config via Func<Instrument, MetricStreamConfiguration>
    .AddView((instrument) =>
    {
        if (instrument.Meter.Name == "CompanyA.ProductB.LibraryC" &&
            instrument.Name == "MyFruitCounter")
        {
            return new MetricStreamConfiguration
            {
                TagKeys = new string[] { "name" },
            };
        }

        return null;
    })

Configuring the aggregation of a Histogram

There are two types of Histogram aggregations: the Explicit bucket histogram aggregation and the Base2 exponential bucket histogram aggregation. Views can be used to select which aggregation is used and to configure the parameters of the aggregation. By default, the explicit bucket aggregation is used.

Explicit bucket histogram aggregation

By default, the boundaries used for a Histogram are { 0, 5, 10, 25, 50, 75, 100, 250, 500, 750, 1000, 2500, 5000, 7500, 10000}. Views can be used to provide custom boundaries for a Histogram. The measurements are then aggregated using the custom boundaries provided instead of the the default boundaries. This requires the use of ExplicitBucketHistogramConfiguration.

    // Change Histogram boundaries to count measurements under the following buckets:
    // (-inf, 10]
    // (10, 20]
    // (20, +inf)
    .AddView(
        instrumentName: "MyHistogram",
        new ExplicitBucketHistogramConfiguration { Boundaries = new double[] { 10, 20 } })

    // If you provide an empty `double` array as `Boundaries` to the `ExplicitBucketHistogramConfiguration`,
    // the SDK will only export the sum, count, min and max for the measurements.
    // There are no buckets exported in this case.
    .AddView(
        instrumentName: "MyHistogram",
        new ExplicitBucketHistogramConfiguration { Boundaries = Array.Empty<double>() })
    // Advanced selection criteria and config via Func<Instrument, MetricStreamConfiguration>
    .AddView((instrument) =>
    {
        if (instrument.Meter.Name == "CompanyA.ProductB.LibraryC" &&
            instrument.Name == "MyHistogram")
        {
            // `ExplicitBucketHistogramConfiguration` is a child class of `MetricStreamConfiguration`
            return new ExplicitBucketHistogramConfiguration
            {
                Boundaries = new double[] { 10, 20 },
            };
        }

        return null;
    })
Base2 exponential bucket histogram aggregation

By default, a Histogram is configured to use the ExplicitBucketHistogramConfiguration. Views are used to switch a Histogram to use the Base2ExponentialBucketHistogramConfiguration.

The bucket boundaries for a Base2 Exponential Bucket Histogram Aggregation are determined dynamically based on the configured MaxSize and MaxScale parameters. The parameters are used to adjust the resolution of the Histogram buckets. Larger values of MaxScale enables higher resolution, however the scale may be adjusted down such that the full range of recorded values fit within the maximum number of buckets defined by MaxSize. The default MaxSize is 160 buckets and the default MaxScale is 20.

    // Change the maximum number of buckets
    .AddView(
        instrumentName: "MyHistogram",
        new Base2ExponentialBucketHistogramConfiguration { MaxSize = 40 })
    // Configure all histogram instruments to use the Base2 Exponential Histogram aggregation
    .AddView((instrument) =>
    {
        return instrument.GetType().GetGenericTypeDefinition() == typeof(Histogram<>)
            ? new Base2ExponentialBucketHistogramConfiguration()
            : null;
    })

Note

The SDK currently does not support any changes to Aggregation type by using Views.

See Program.cs for a complete example.

Changing maximum Metric Streams

Every instrument results in the creation of a single Metric stream. With Views, it is possible to produce more than one Metric stream from a single instrument. To protect the SDK from unbounded memory usage, SDK limits the maximum number of metric streams. All the measurements from the instruments created after reaching this limit will be dropped. The default is 1000, and SetMaxMetricStreams can be used to override the default.

Consider the below example. Here we set the maximum number of MetricStreams allowed to be 1. This means that the SDK would export measurements from only one MetricStream. The very first instrument that is published (MyFruitCounter in this case) will create a MetricStream and the SDK will thereby reach the maximum MetricStream limit of 1. The measurements from any subsequent instruments added will be dropped.

using System.Diagnostics.Metrics;
using OpenTelemetry;
using OpenTelemetry.Metrics;

Counter<long> MyFruitCounter = MyMeter.CreateCounter<long>("MyFruitCounter");
Counter<long> AnotherFruitCounter = MyMeter.CreateCounter<long>("AnotherFruitCounter");

using var meterProvider = Sdk.CreateMeterProviderBuilder()
    .AddMeter("*")
    .AddConsoleExporter()
    .SetMaxMetricStreams(1) // The default value is 1000
    .Build();

// SDK only exports measurements from `MyFruitCounter`.
MyFruitCounter.Add(1, new("name", "apple"), new("color", "red"));

// The measurements from `AnotherFruitCounter` are dropped as the maximum
// `MetricStream`s allowed is `1`.
AnotherFruitCounter.Add(1, new("name", "apple"), new("color", "red"));

Changing the cardinality limit for a Metric

To set the cardinality limit for an individual metric, use MetricStreamConfiguration.CardinalityLimit setting on the View API:

Note

MetricStreamConfiguration.CardinalityLimit is an experimental API only available in pre-release builds. For details see: OTEL1003.

var meterProvider = Sdk.CreateMeterProviderBuilder()
    .AddMeter("MyCompany.MyProduct.MyLibrary")
    // Set a custom CardinalityLimit (10) for "MyFruitCounter"
    .AddView(
        instrumentName: "MyFruitCounter",
        new MetricStreamConfiguration { CardinalityLimit = 10 })
    .AddConsoleExporter()
    .Build();

Exemplars

Exemplars are example data points for aggregated data. They provide access to the raw measurement value, time stamp when measurement was made, and trace context, if any. It also provides "Filtered Tags", which are attributes (Tags) that are dropped by a view. Exemplars are an opt-in feature, and allow customization via ExemplarFilter and ExemplarReservoir.

Exemplar collection in OpenTelemetry .NET is done automatically (once Exemplar feature itself is enabled on MeterProvider). There is no separate API to report exemplar data. If an app is already using existing Metrics API (manually or via instrumentation libraries), exemplars can be configured/enabled without requiring instrumentation changes.

While the SDK is capable of producing exemplars automatically, the exporters (and the backends) must also support them in order to be useful. OTLP Metric Exporter has support for this today, and this end-to-end tutorial demonstrates how to use exemplars to achieve correlation from metrics to traces, which is one of the primary use cases for exemplars.

ExemplarFilter

ExemplarFilter determines which measurements are offered to the configured ExemplarReservoir, which makes the final decision about whether or not the offered measurement gets recorded as an Exemplar. Generally ExemplarFilter is a mechanism to control the overhead associated with Exemplar offering.

OpenTelemetry SDK comes with the following ExemplarFilters (defined on ExemplarFilterType):

  • AlwaysOff: Makes no measurements eligible for becoming an Exemplar. Using this is as good as turning off the Exemplar feature and is the current default.
  • AlwaysOn: Makes all measurements eligible for becoming an Exemplar.
  • TraceBased: Makes those measurements eligible for becoming an Exemplar which are recorded in the context of a sampled Activity (span).

The SetExemplarFilter extension method on MeterProviderBuilder can be used to set the desired ExemplarFilterType and enable Exemplar collection:

using OpenTelemetry;
using OpenTelemetry.Metrics;

using var meterProvider = Sdk.CreateMeterProviderBuilder()
    // rest of config not shown
    .SetExemplarFilter(ExemplarFilterType.TraceBased)
    .Build();

ExemplarReservoir

ExemplarReservoir receives the measurements sampled by the ExemplarFilter and is responsible for recording Exemplars. The following are the default reservoirs:

  • AlignedHistogramBucketExemplarReservoir is the default reservoir used for Histograms with buckets, and it stores at most one exemplar per histogram bucket. The exemplar stored is the last measurement recorded - i.e. any new measurement overwrites the previous one in that bucket.

  • SimpleFixedSizeExemplarReservoir is the default reservoir used for all metrics except Histograms with buckets. It has a fixed reservoir pool, and implements the equivalent of naive reservoir. The reservoir pool size (currently defaulting to 1) determines the maximum number of exemplars stored.

Note

Currently there is no ability to change or configure ExemplarReservoir.

Instrumentation

// TODO

MetricReader

MetricReader allows collecting the pre-aggregated metrics from the SDK. They are typically paired with a MetricExporter which does the actual export of metrics.

Though MetricReader can be added by using the AddReader method on MeterProviderBuilder, most users use the extension methods on MeterProviderBuilder offered by exporter libraries, which adds the correct MetricReader, that is configured to export metrics to the exporter.

Refer to the individual exporter docs to learn how to use them:

Resource

Resource is the immutable representation of the entity producing the telemetry. If no Resource is explicitly configured, the default is to use a resource indicating this Service and Telemetry SDK. The ConfigureResource method on MeterProviderBuilder can be used to configure the resource on the provider. ConfigureResource accepts an Action to configure the ResourceBuilder. Multiple calls to ConfigureResource can be made. When the provider is built, it builds the final Resource combining all the ConfigureResource calls. There can only be a single Resource associated with a provider. It is not possible to change the resource builder after the provider is built, by calling the Build() method on the MeterProviderBuilder.

ResourceBuilder offers various methods to construct resource comprising of attributes from various sources. For example, AddService() adds Service resource. AddAttributes can be used to add any additional attributes to the Resource. It also allows adding ResourceDetectors.

It is recommended to model attributes that are static throughout the lifetime of the process as Resources, instead of adding them as attributes(tags) on each measurement.

Follow this document to learn about writing custom resource detectors.

The snippet below shows configuring the Resource associated with the provider.

using OpenTelemetry;
using OpenTelemetry.Metrics;
using OpenTelemetry.Resources;

using var meterProvider = Sdk.CreateMeterProviderBuilder()
    .ConfigureResource(r => r.AddAttributes(new List<KeyValuePair<string, object>>
                {
                    new KeyValuePair<string, object>("static-attribute1", "v1"),
                    new KeyValuePair<string, object>("static-attribute2", "v2"),
                }))
    .ConfigureResource(resourceBuilder => resourceBuilder.AddService("service-name"))
    .Build();

It is also possible to configure the Resource by using following environmental variables:

Environment variable Description
OTEL_RESOURCE_ATTRIBUTES Key-value pairs to be used as resource attributes. See the Resource SDK specification for more details.
OTEL_SERVICE_NAME Sets the value of the service.name resource attribute. If service.name is also provided in OTEL_RESOURCE_ATTRIBUTES, then OTEL_SERVICE_NAME takes precedence.