[dougqh]

What Does This Do

The Datadog extractor now keeps the raw x-datadog-tags value and parses it when the context is built. At that point the sampling priority is known, so the extracted PropagationTags starts from the priority the trace will inherit.

Before, the tags were parsed as soon as the header was seen, which isn't necessarily after x-datadog-sampling-priority. They started at UNSET, and the root span's constructor then installed the priority as a second sampling-state snapshot (setSamplingPriority(priority, UNKNOWN)). That install is now a no-op.

With APM tracing disabled, the upstream priority may not be inherited, so the tags still start at UNSET and behave exactly as before.

Motivation

TraceSegment work (APMLP-2049), step 2: one sampling state per extracted trace.

I counted SamplingState constructions with a temporary counter while running PropagationScenariosBenchmark's two upstream header sets through extract, root start, inject and finish. Per extracted trace, the count goes from 4 to 3:

• extract: 3 → 3
• root start: 1 → 0

The remaining extra two are the throwaway W3C PropagationTags from compound extraction and the compound merge's new state. Removing them is the next step.

Allocation (PropagationScenariosBenchmark, JDK 17, @Threads(8), 1 fork × 5 iterations, -prof gc; base = #13009):

scenario	sampler	upstream	base B/op	this PR B/op	Δ
leaf	default	sameVersion	2241.3 ± 5.4	2216.3 ± 9.8	−25.0
leaf	default	withOtelMember	2512.3 ± 29.1	2462.5 ± 2.7	−49.8
leaf	rule	sameVersion	2240.0 ± 6.5	2206.2 ± 8.1	−33.8
leaf	rule	withOtelMember	2373.4 ± 16.0	2325.7 ± 23.6	−47.6
intermediate	default	sameVersion	3848.5 ± 5.6	3809.2 ± 7.5	−39.3
intermediate	default	withOtelMember	4048.0 ± 8.3	3999.1 ± 9.5	−48.9
intermediate	rule	sameVersion	3868.7 ± 5.2	3809.9 ± 19.2	−58.8
intermediate	rule	withOtelMember	4075.3 ± 42.4	3963.0 ± 9.1	−112.3

Throughput is pending. The host was busy during these runs, and all pairs overlap within their error. A quiet run will be added before this is marked ready.

Additional Notes

Output is unchanged. Starting from the priority at extract leads to the same final state as installing it at root start:

• The ot consistency check now runs at the compound merge instead of at the root reconcile. That check is idempotent and order-independent (see Defer parsing the extracted OpenTelemetry tracestate member #12987).
• A reconcile with mechanism UNKNOWN never changes the decision maker.

Tests:

• DatadogHttpExtractorTest: the extracted tags carry the priority whichever header comes first, and don't when APM tracing is disabled.
• DDSpanContextPropagationTagsTest: the root span keeps the extracted SamplingState instance.

The full dd-trace-core suite passes, apart from the 2 TracerConnectionReliabilityTest cases, which need Docker.

Stacked: master ← #13008 ← #13000 ← #13009 ← this PR.

🤖 Generated with Claude Code

[datadog-datadog-prod-us1]

🎯 Code Coverage (details)
• Patch Coverage: 81.25%
• Overall Coverage: 59.86% (-0.01%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 67e1349 | Docs | Give us feedback!}

[dd-octo-sts]

🟢 Java Benchmark SLOs — All performance SLOs passed

Suite	Status
Startup	🟢 pass

SLO thresholds are defined here based on automatically generated metrics. A warning is raised when results are within 5% of the threshold.

PR vs. master results

Scenario	Candidate	master	Δ (95% CI of mean)
startup:insecure-bank:iast:Agent	14.04 s	13.97 s	[-0.4%; +1.4%] (no difference)
startup:insecure-bank:tracing:Agent	12.94 s	12.97 s	[-0.9%; +0.5%] (no difference)
startup:petclinic:appsec:Agent	17.16 s	16.99 s	[+0.0%; +1.9%] (maybe worse)
startup:petclinic:iast:Agent	16.98 s	17.02 s	[-1.0%; +0.5%] (no difference)
startup:petclinic:profiling:Agent	16.69 s	16.73 s	[-1.3%; +0.8%] (no difference)
startup:petclinic:sca:Agent	17.05 s	17.03 s	[-0.6%; +0.9%] (no difference)
startup:petclinic:tracing:Agent	15.71 s	16.15 s	[-7.0%; +1.5%] (no difference)

Commit: 67e1349d · CI Pipeline · Benchmarking Platform UI

---

Load and DaCapo benchmarks can be triggered manually in the GitLab pipeline. Results will appear in the Benchmarking Platform UI after completion.