Configuration

Docs > Observability Pipelines（観測データの制御） > Configuration

このページは日本語には対応しておりません。随時翻訳に取り組んでいます。
翻訳に関してご質問やご意見ございましたら、お気軽にご連絡ください。

Overview

The pipelines page with a source going to two processors groups and two destinations

Observability Pipelines lets you collect and process within your own infrastructure, and then route them to different destinations. A pipeline consists of three core components:

Source: Receives data from a tool like the Datadog Agent.
Processors: Transform, enrich, or filter data.
Destinations: Where data is sent (for example, Datadog, Amazon S3, Splunk, Google Security Operations, and Microsoft Sentinel).

Build and deploy pipelines to collect, transform, and route your data using one of these methods:

Pipeline types

There are two types of pipelines:

Use one of the logs templates to create a log pipeline.

Archive Logs
Dual Ship Logs
Generate Log-based Metrics
Log Enrichment
Log Volume Control
Sensitive Data Redaction
Split Logs

See Set Up Pipelines for more information on setting up a source, processors, and destinations.

Use the Metric Tag Governance template to create a metrics pipeline.

See Set Up Pipelines for more information on setting up a source, processors and destination.

Metrics data

Metrics sent to Observability Pipelines include the following:

name: The metric name.
kind: There are two kinds of metrics:
- absolute metrics: Represents the current value of a measurement at the time it is reported.
- incremental metrics: Represents the change in a measurement since the last reported value, which the system aggregates over time.
value: The metric type:
- counter
- gauge
- distribution
- histogram
timestamp: The date and time the metric is created.
tags: Includes tags such as host.

Whether a received metric is incremental or absolute depends on the source. For example, metrics from OpenTelemetry can either be incremental or absolute based on their temporality. The following table is an example of an OTel counter metric sent with delta versus cumulative temporality.

Metric Type	Incremental	Absolute
Counter	Sent as deltas: `+2`, `+4`, `+6`	Sent as cumulative sum: `2`, `6`, `10`

An example of a metric:

{
  "name":"datadog.agent.retry_queue_duration.bytes_per_sec",
  "tags":{
    "agent":"core",
    "domain":"https://7-72-3-app.agent.datadoghq.com",
    "host":"COMP-YGVQDJG75L",
    "source_type_name":"System",
    "env:prod"
  },
  "timestamp":"2025-11-28T13:03:09Z",
  "kind":"absolute",
  "gauge":{"value":454.1372767857143}
}

Metric types

The available metric types:

Metric type	Description	Example
COUNTER	The total number of event occurrences in one time interval. Can be reset to zero, but cannot be decreased.	You want to count the number of logs with `status:error`.
GAUGE	A snapshot of a value at the time it is reported.	You want to track the latest CPU utilization for each host.
HISTOGRAM	Statistical aggregations (`avg`, `min`, `max`, `count`, `median`, percentiles) computed per host by the Datadog Agent in one time interval, then sent to Datadog.	You want per-host request latency aggregations from each web server.
DISTRIBUTION	Raw values sent to Datadog so percentile aggregations are computed server-side, globally across every host reporting the metric in one time interval.	You want the global p95 latency of an API endpoint, calculated across every host serving it.