What it does
Honeycomb Refinery is an open-source, self-managed proxy that samples traces after their spans have arrived, so a decision can consider the complete trace rather than a single span. Operators configure dynamic or rules-based policies and place a Refinery cluster in the telemetry path before forwarding retained traces to Honeycomb. This makes it a data-volume control tool for teams that want to keep unusual or important traces while sampling routine traffic.
Filter only after seeing the whole trace
Refinery sits between an OpenTelemetry-compatible source and Honeycomb. A typical setup routes trace data through a load-balanced cluster, then defines sampling behavior in rules.yaml. Because the proxy evaluates complete traces, a team can retain every trace with an error condition while sampling routine successes more aggressively. Dynamic samplers can adjust rates from observed fields, and rules can express conditions for traces that should be kept or dropped.
A sensible first change is to begin with a low-risk rule and run the documented Dry Run mode before allowing the policy to drop telemetry. Compare the expected volume and retained trace patterns in Honeycomb, then move the configuration into production when it behaves as intended. The documentation recommends testing the rules this way before dropping traffic.
Refinery is a component to deploy, configure, and monitor, not a hosted dashboard. Honeycomb's setup guide lists Kubernetes or Linux x86-64/AMD64 and ARM64 deployments; its baseline is 2 GB of RAM and 2 cores per server, and cluster members must communicate to concentrate traces. The project is Apache-2.0 licensed, but the license does not include the cost of any downstream Honeycomb service or infrastructure you run.
Strengths
- Samples at the complete-trace level, allowing rules to retain traces based on attributes or conditions found across spans.
- Supports dynamic and rules-based sampling methods, including a documented Dry Run before rules start dropping traffic.
Watch out for
- Requires a self-managed cluster in the telemetry path; the official setup guide specifies Linux or Kubernetes deployment and per-server CPU and memory requirements.
Best for
- Reducing trace ingestion while preserving requests with errors or other high-value conditions
- Teams already using Honeycomb that need to tune trace sampling based on complete request context
- Operators who can run and maintain a telemetry proxy cluster
Sources
- Honeycomb Refinery - Honeycomb Docs (Oct 26, 2026)
- Set Up Honeycomb Refinery - Honeycomb Docs (Oct 26, 2026)
- Specify Sampling Methods in Honeycomb Refinery - Honeycomb Docs (Oct 26, 2026)
- Refinery - the Honeycomb Sampling Proxy (Oct 26, 2026)
- Refinery LICENSE (Oct 26, 2026)