Overview

Datadog’s OpenTelemetry Protocol (OTLP) metrics intake API endpoint allows you to send metrics directly to Datadog. With this feature, you don’t need to run the Datadog Agent or OpenTelemetry Collector + Datadog Exporter.

You might prefer this option if you’re looking for a straightforward setup and want to send metrics directly to Datadog without using the Datadog Agent or OpenTelemetry Collector.

This endpoint is particularly useful in the following scenarios:

• OpenTelemetry distributions without Datadog Exporter support: Some OpenTelemetry distributions, such as the AWS Distro for OpenTelemetry (ADOT), have removed vendor-specific exporters in favor of a unified OTLP exporter. The OTLP metrics endpoint enables these distributions to send metrics directly to Datadog seamlessly.

• Technical constraints using the Datadog Exporter or Agent: Ideal for scenarios where installing additional software is impractical or restrictive, such as third-party managed services (for example, Vercel), applications on customer devices, or environments requiring streamlined, Agentless observability pipelines. The OTLP metrics endpoint enables direct OTLP metric ingestion in these scenarios.

Configuration

To export OTLP metrics data to the Datadog OTLP metrics intake endpoint:

1. Ensure only delta metrics are sent.
2. Configure the OTLP HTTP exporter.
   • Set the Datadog OTLP metrics intake endpoint.
   • Configure the required HTTP headers.
3. (Optional) Set the dd-otel-metric-config HTTP header to configure the metric translator behavior.

Configure the exporter

To send OTLP data to the Datadog OTLP metrics intake endpoint, use the OTLP HTTP exporter. For metrics, the exporter supports both HTTP Protobuf and HTTP JSON. HTTP Protobuf is recommended for better performance.

The process differs depending on whether you’re using automatic or manual instrumentation for OpenTelemetry.

Ensure only delta metrics are sent

The Datadog OTLP metrics intake endpoint accepts only delta metrics. If you attempt to send cumulative metrics (the default in most SDKs), you will receive an error. Make sure to configure your OpenTelemetry SDK or Collector to produce delta metrics.

• For supported languages, set the OTEL_EXPORTER_OTLP_METRICS_TEMPORALITY_PREFERENCE environment variable to delta.
• Otherwise, set the temporality preference in your code. For more information, read Configure delta temporality in OpenTelemetry.

Automatic instrumentation

If you are using OpenTelemetry automatic instrumentation, set the following environment variables:

export OTEL_EXPORTER_OTLP_METRICS_PROTOCOL="http/protobuf"
export OTEL_EXPORTER_OTLP_METRICS_ENDPOINT=""
export OTEL_EXPORTER_OTLP_METRICS_HEADERS="dd-api-key=${DD_API_KEY},dd-otlp-source=${YOUR_SITE}"
export OTEL_EXPORTER_OTLP_METRICS_TEMPORALITY_PREFERENCE="delta"

Manual instrumentation

If you are using manual instrumentation with OpenTelemetry SDKs, configure the OTLP HTTP Protobuf exporter programmatically.

const { OTLPMetricExporter } = require('@opentelemetry/exporter-metrics-otlp-proto');

const exporter = new OTLPMetricExporter({
  url: 'https://api.datadoghq.com/api/intake/otlp/v1/metrics',
  temporalityPreference: AggregationTemporalityPreference.DELTA, // Ensure delta temporality
  headers: {
    'dd-api-key': process.env.DD_API_KEY,
    'dd-otel-metric-config': '{resource_attributes_as_tags: true}',
    'dd-otlp-source': '${YOUR_SITE}', // Replace this with the correct site
  },
});

import io.opentelemetry.exporter.otlp.http.metrics.OtlpHttpMetricExporter;

OtlpHttpMetricExporter exporter = OtlpHttpMetricExporter.builder()
    .setEndpoint("https://api.datadoghq.com/api/intake/otlp/v1/metrics")
    .setAggregationTemporalitySelector(
			AggregationTemporalitySelector.deltaPreferred()) // Ensure delta temporality
    .addHeader("dd-api-key", System.getenv("DD_API_KEY"))
    .addHeader("dd-otel-metric-config", "{resource_attributes_as_tags: true}")
    .addHeader("dd-otlp-source", "${YOUR_SITE}") // Replace this with the correct site
    .build();

import "go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetrichttp"

metricExporter, err := otlpmetrichttp.New(
	ctx,
	otlpmetrichttp.WithEndpoint("api.datadoghq.com"),
	otlpmetrichttp.WithURLPath("/api/intake/otlp/v1/metrics"),
      otlpmetrichttp.WithTemporalitySelector(deltaSelector), // Ensure delta temporality
	otlpmetrichttp.WithHeaders(
		map[string]string{
			"dd-api-key": os.Getenv("DD_API_KEY"),
			"dd-otel-metric-config": "{resource_attributes_as_tags: true}",
      "dd-otlp-source": "${YOUR_SITE}", // Replace this with the correct site
		}),
)

from opentelemetry.exporter.otlp.proto.http.metric_exporter import OTLPMetricExporter

exporter = OTLPMetricExporter(
    endpoint="https://api.datadoghq.com/api/intake/otlp/v1/metrics",
    preferred_temporality=deltaTemporality, # Ensure delta temporality
    headers={
        "dd-api-key": os.environ.get("DD_API_KEY"),
        "dd-otel-metric-config": "{resource_attributes_as_tags: true}",
        "dd-otlp-source": "${YOUR_SITE}" # Replace this with the correct site
    },
)

(Optional) Configure the metric translator

Use the dd-otel-metric-config header to configure how metrics are translated and sent to Datadog. The JSON header contains the following fields:

resource_attributes_as_tags

Type: Boolean
Default: false
If set to true, transforms all resource attributes into metric labels, which are then converted into tags.

instrumentation_scope_metadata_as_tags

Type: Boolean
Default: false
If set to true, adds the name and version of the instrumentation scope that created a metric to the metric tags.

histograms.mode

Type: String
Mode for exporting histograms. Valid values are:

• distributions: sends histograms as Datadog distributions (recommended).
• counters: sends histograms as Datadog counts, one metric per bucket.
• nobuckets: sends no bucket histogram metrics.

histograms.send_aggregation_metrics

Type: Boolean
If set to true, writes additional .sum, .count, .min, and .max metrics for histograms.

summaries.mode

Type: String
Mode for exporting OTLP summaries. Valid values are:

• noquantiles: sends no .quantile metrics. .sum and .count metrics are still sent.
• gauges: sends .quantile metrics as gauges tagged by the quantile.

For example:

{
  "resource_attributes_as_tags": true,
  "instrumentation_scope_metadata_as_tags": true,
  "histograms": {
    "mode": "distributions",
    "send_aggregation_metrics": true
  },
  "summaries": {
    "mode": "gauges"
  }
}

OpenTelemetry Collector

If you are using an OpenTelemetry Collector distribution that doesn’t support the Datadog Exporter, you can configure the otlphttpexporter to export metrics to the Datadog OTLP metrics intake endpoint.

For example, your config.yaml file would look like this:

...
exporters:
  otlphttp:
    metrics_endpoint: 
    headers:
      dd-api-key: ${env:DD_API_KEY}
      dd-otel-metric-config: "{resource_attributes_as_tags: true}"
      dd-otlp-source: "${YOUR_SITE}", # Replace this with the correct site
...

service:
  pipelines:
    metrics:
      receivers: [otlp]
      processors: [batch, cumulativetodelta]
      exporters: [otlphttp]

Troubleshooting

Error: 403 Forbidden

If you receive a 403 Forbidden error when sending metrics to the Datadog OTLP metrics intake endpoint, it indicates one of the following issues:

• The API key belongs to an organization that is not allowed to access the Datadog OTLP metrics intake endpoint. Solution: Verify that you are using an API key from an organization that is allowed to access the Datadog OTLP metrics intake endpoint.

• The dd-otlp-source header is missing or has an incorrect value. Solution: Ensure that the dd-otlp-source header is set with the proper value for your site. You should have received an allowlisted value for this header from Datadog if you are a platform partner.

• The endpoint URL is incorrect for your organization. Solution: Use the correct endpoint URL for your organization. Your site is , so you need to use the endpoint.

Error: 413 Request Entity Too Large

If you receive a 413 Request Entity Too Large error when sending metrics to the Datadog OTLP metrics intake endpoint, it indicates that the payload size sent by the OTLP exporter exceeds the Datadog metrics intake endpoint’s limit of 500KB for uncompressed payloads, or 5MB for compressed payloads after decompression.

This error usually occurs when the OpenTelemetry SDK batches too much telemetry data in a single request payload.

Solution: Reduce the export batch size of the SDK’s batch processor. For example, in the OpenTelemetry Java SDK, you can adjust BatchMetricExportProcessor.

Issue: Missing datapoints or lower than expected metric values

If you notice missing datapoints or lower than expected metric values, it may be because you are sending multiple datapoints for a metric that have the same timestamp (in seconds) and same dimensions. In such cases, Datadog only accepts the last datapoint, and previous datapoints are dropped (last-write-wins). Datadog requires the timeseries data of a metric to be unique in the context of {timestamp + dimensions}.

Solution: Ensure that your datapoints of a given metric at one timestamp are uniquely tagged. For example, if you send multiple datapoints for a metric simultaneously from multiple AWS Lambda invocations, make sure to include unique identifiers (such as the Lambda ARN) as resource attributes in your metrics. Use the resource_attributes_as_tags option to add these resource attributes as metric tags.

Further reading

추가 유용한 문서, 링크 및 기사:

General OpenTelemetry SDK ConfigurationEXTERNAL SITE OpenTelemetry Environment Variable SpecEXTERNAL SITE OpenTelemetry Protocol ExporterEXTERNAL SITE OTLP Metrics ExporterEXTERNAL SITE Configure delta temporality in OpenTelemetryDOCUMENTATION OTLP Metrics Mapping in DatadogDOCUMENTATION