Role

The OpenTelemetry Collector is a vendor-neutral service for receiving, processing, and exporting telemetry. Applications can send one protocol and the Collector can batch, filter, enrich, sample, and forward the data to one or more backends.

A Collector does not store telemetry as its primary job. It is a pipeline component, so design for backpressure, retries, queue limits, and what happens when the destination is unavailable.

Pipeline model

Each signal pipeline is assembled from three component types:

• receivers accept telemetry, for example OTLP;
• processors transform or buffer it, for example batch or memory_limiter;
• exporters send it to another Collector or backend.

receivers:
  otlp:
    protocols:
      grpc: {}
      http: {}

processors:
  memory_limiter:
    check_interval: 1s
    limit_mib: 512
  batch: {}

exporters:
  debug: {}

service:
  pipelines:
    traces:
      receivers: [otlp]
      processors: [memory_limiter, batch]
      exporters: [debug]

Defining a component is not enough; it must also be referenced from a pipeline under service.

Deployment patterns

• Agent: one Collector near each workload or node; useful for host-local data.
• Gateway: a shared tier for central policy, sampling, and export.
• Agent plus gateway: local collection with centralized processing.

Start with the fewest moving parts that meet the reliability requirement. A gateway needs horizontal scaling and load balancing for the protocol and any stateful processors it uses.

Further reading: Collector documentation and configuration.