This exporter sends metric and trace data to Datadog. For environment specific setup instructions visit the Datadog Documentation.

Please review the Collector's security documentation, which contains recommendations on securing sensitive information such as the API key required by this exporter.

Note: The metrics exporter is not considered stable yet and may suffer breaking changes.

The only required setting is a Datadog API key.

datadog:
  api:
    key: "<API key>"

To send data to the Datadog EU site, set the api.site parameter to datadoghq.eu:

datadog:
  api:
    key: "<API key>"
    site: datadoghq.eu

If you want to use the OpenTelemetry Span Name as the Datadog Resource Name you can set the span_name_as_resource_name configuration option to true (default is false). For more info on the downsides of this option check this issue.

datadog:
 api:
   key: "<API key>"
 traces:
   span_name_as_resource_name: true

The hostname, environment, service and version can be set in the configuration for unified service tagging. The exporter will try to retrieve a hostname following the OpenTelemetry semantic conventions if there is one available.

See the sample configuration files under the example folder for other available options, as well as an example K8s Manifest. This exporter also supports the exporterhelper queuing, retry and timeout settings documented here. Retry settings will only affect metrics.

This exporter assumes a pipeline using the datadog exporter also includes a batch processor configured with the following:

• a timeout setting of 10s(10 seconds).

Please make sure to include this processor in your pipeline. An example pipeline can be found below.

A batch representing 10 seconds of traces is a constraint of Datadog's API Intake for Trace Related Statistics. Without this setting, trace related metrics including .hits .errors and .duration for different services and service resources may be inaccurate over periods of time.

Example:

receivers:
  examplereceiver:

processors:
  batch:
    timeout: 10s

exporters:
  datadog/api:
    hostname: customhostname

    host_metadata:
      tags: [example:tag]

    api:
      key: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
      site: datadoghq.eu

service:
  pipelines:
    traces:
      receivers: [examplereceiver]
      processors: [batch]
      exporters: [datadog/api]

Please Note: Currently Span Events are extracted and added to Spans as Json on the Datadog Span Tag events.

While the OpenTelemetry Specification for Sampling remains undecided and in active development, the Datadog Exporter currently supports two sampling approaches to ensure accuracy in generated Trace Stats payloads for details such as hits, errors, and latency within Datadog.

1. ParentBased(root=AlwaysOn): This is the default SDK sampler, and emits 100% of Spans. Depending on your use case, in high volume environments this approach may result in higher than expected resource consumption.

2. DatadogTraceIdRatioBased, a customized TraceIdRatioBased sampler used as the root of a ParentBased Sampler, ex: ParentBased(root=DatadogTraceIdRatioBased): In certain high volume environments, an AlwaysOn Sampler may not be feasible. In these cases, Datadog supports a global, head based, probability sampling approach for OpenTelemetry by including a custom Datadog TraceIdRatioBased Sampler used as the root in a ParentBased composite Sampler. This sampler ensures that the global probability sampling rate information is appended to Sampled Spans as an attribute _sample_rate. This attribute has semantic significance within Datadog and ensures that the Datadog Exporter can correctly "upscale" generated Trace Stats payloads to ensure accurate hits, errors, and latency statistics that include dropped spans. For an example of this custom sampler, see the example application in example/example_custom_sampler.

The metrics exporter does not assume any specific pipeline setup. There are a number of optional settings for configuring how to send your metrics: