Datadog Distributed Tracing: Per-service sampling rule precedence

Direct answer

No. The Datadog Agent does not prioritize per-service rules over global rates based on specificity. Treat sampling_rules as an ordered, first-match list: the first rule that matches a trace wins, and later rules are ignored. A per-service rule only takes effect if it appears before any broader rule that also matches.

Confirmed behavior vs. likely explanation

What is well established is the first-match evaluation model. If your global rate is a separate top-level key such as sampling_rate, it acts as a fallback used only when no rule in sampling_rules matches. If instead the global behavior is implemented as a catch-all rule inside the same array, its position is critical: a catch-all placed above a service-specific rule will shadow that service rule entirely.

The likely explanation for seemingly inconsistent results across versions is not a 7.x-versus-8.x precedence change, but confusion about which layer makes the decision. Tracer-side head sampling can drop a trace before the Agent ever sees it, while Agent-side rules can only filter what they receive. Backend or UI ingestion sampling is a third, separate layer. Mixing these layers makes it look as though rule order is being ignored when in fact the deciding layer never applied the rule.

Steps for this case

1. Determine whether your global rate is a standalone default or a catch-all entry in sampling_rules.
2. Reorder the array so service-specific rules come first and any catch-all or default rule comes last.
3. Verify the Agent actually loaded the keys you expect; unrecognized sampling keys can be silently ignored depending on version.
4. Run a controlled test: set a low default, one high-rate rule for a single service, and compare retained traces. If retention follows the first rule, order controls precedence.

Example ordering:

sampling_rules:
  - service: "my-service"
    sample_rate: 1.0
  - service: "*"
    sample_rate: 0.1

One missing diagnostic detail

To narrow the recommendation further, it would help to know whether your global rate is configured as a top-level sampling_rate or as a catch-all rule inside sampling_rules. That distinction determines whether reordering alone is sufficient or whether you also need to move the default out of the array.

Note: Exact key names and Agent recognition of sampling configuration vary by version. Verify against the Datadog documentation for your deployed Agent release before relying on a specific key.