How to Exclude Specific Trace Names (e.g., /api/services) from Tail-Based Sampling in Jaeger v2.8.0?

[jjhwan-h]

Hi Jaeger team

I'm currently using jaegertracing/jaeger:2.8.0 and exploring tail-based sampling to filter out unnecessary traces, especially since remote sampling is not an option in my setup.

I’ve reviewed the config-tail-sampling-service-name-policy.yaml example, and it clearly shows how to apply sampling rules based on service names.

However, I would like to go one step further:
👉 Is it possible to exclude traces based on the operation (such as /api/services) in the sampling configuration?

In my case, I’d like to configure the tail sampler to not sample any traces where the root span operation name is /api/services, for example.

Is this currently supported, or is there a workaround you'd recommend for filtering based on the span name?

Thanks in advance!

[dosubot[bot]]

Hi @jjhwan-h! I'm Dosu and I’m helping the jaeger team.

Jaeger v2.8.0's tail-based sampling does not support excluding traces based on the root span operation name (like /api/services) in its YAML configuration. The built-in tail sampler only supports filtering by service name out of the box, and there is no field or policy for operation or span name filtering in the standard config or code path example config.

Operation-level sampling (including setting a probability of 0 for specific operations) is only available in static JSON configs for head-based sampling, not in the tail-based sampling pipeline source.

If you need to filter traces by operation name in tail-based sampling, the main options are to implement a custom policy (which requires code changes) or use the OpenTelemetry Collector's tail sampling processor, which supports more flexible attribute-based policies (including operation name) discussion.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[yurishkuro]

tail sampling processor is provided from OTEL repository, please refer to its documentation there https://github.com/open-telemetry/opentelemetry-collector-contrib/tree/main/processor/tailsamplingprocessor

[jjhwan-h]

Thank you for the help.
I wanted to leave a note for anyone facing a similar issue.

In my case, I resolved the problem by using the filterprocessor in the OpenTelemetry Collector to drop unnecessary spans based on their span.name, even while using Jaeger v2.
This approach worked well to filter out spans like persistence.sql.NextMessages.

processors:
  filter:
    traces:
        span:
          - 'name == "exampleexampleexample"'
          - 'IsMatch(name, "sql-*")'
  batch:

Hope this helps someone else!