The OpenTelemetry Collector: One Pipeline for All Telemetry

Apps can send telemetry straight to a vendor, but most production setups put an OpenTelemetry Collector in between. It receives data from many sources, processes it and exports it to one or more backends.

Why use a Collector

• Apps only need to know one endpoint; switch or add backends without redeploying apps.
• Batching, retries and buffering when a backend is slow.
• Drop noisy data, remove sensitive attributes and add metadata (like Kubernetes pod names) centrally.
• Tail sampling: keep all error traces and a sample of the rest.

A configuration

receivers:
  otlp:
    protocols:
      grpc: { endpoint: 0.0.0.0:4317 }
      http: { endpoint: 0.0.0.0:4318 }

processors:
  memory_limiter:
    check_interval: 1s
    limit_percentage: 80
  batch: {}
  attributes/scrub:
    actions:
      - key: user.email
        action: delete

exporters:
  otlphttp/vendor:
    endpoint: https://otlp.example-vendor.com
    headers:
      api-key: ${env:VENDOR_API_KEY}
  debug: {}

service:
  pipelines:
    traces:
      receivers: [otlp]
      processors: [memory_limiter, attributes/scrub, batch]
      exporters: [otlphttp/vendor]
    metrics:
      receivers: [otlp]
      processors: [memory_limiter, batch]
      exporters: [otlphttp/vendor]
    logs:
      receivers: [otlp]
      processors: [memory_limiter, batch]
      exporters: [otlphttp/vendor]

Data flows receivers → processors → exporters, configured separately for each signal. Put memory_limiter first and batch last.

Deployment patterns

• Agent - a Collector on every node (a DaemonSet in Kubernetes), close to the apps. Good for adding host metadata and collecting node metrics and logs.
• Gateway - a central, scalable Collector deployment that agents forward to. Good for tail sampling and holding vendor credentials.
• Many setups use both.

Distributions

The "core" distribution has a small set of components; "contrib" includes many more receivers and exporters (Kafka, Prometheus scraping, cloud services). For production, consider building a custom distribution with only what you need.