fix(stats): don't feed inferred proxy spans to client-side stats concentrator

[Leiyks]

Problem

The System Tests: [tracer-release] CI job was failing because client-side stats were reporting 10 hits instead of 5 for 5 HTTP requests. This broke the system test assertion ok_hits == ok_top_hits == 5.

Root Cause

The early-exit stats computation block in ext/serializer.c was feeding both the root span AND the inferred proxy span (API Gateway Tracing) to the stats concentrator. The inferred span has no http.method or http.endpoint in its meta, creating a separate aggregation bucket. With 5 requests:

• Root span bucket: 5 hits (http.method=GET, http.endpoint=/stats-unique)
• Inferred span bucket: 5 hits (empty http.method, empty http.endpoint)

Total: 10 hits instead of 5.

Fix

Remove the inferred span from the early-exit stats path. Inferred spans are proxy-level spans (API Gateway Tracing) that don't represent real service calls and should not be included in client-side stats. The root span already captures the full request statistics.

Analysis

The if (inferred_span) block was added in #3756 (Implement stats computation) with the comment "Inferred span won't be serialized, so feed it to the concentrator here." The intent was to compensate for the fact that the early-exit path returns NULL before the normal inferred span serialization at line 1829. However, inferred spans should not be in the concentrator at all — they are proxy-level spans without meaningful http.method/http.endpoint tags, and feeding them creates a spurious aggregation bucket.

Testing

The system test Test_Client_Stats::test_client_stats in ~/git/system-tests/tests/stats/test_stats.py should now pass with ok_hits == 5.

[datadog-datadog-prod-us1]

🎉 All green!

❄️ No new flaky tests detected
🧪 All tests passed

🎯 Code Coverage (details)
• Patch Coverage: 100.00%
• Overall Coverage: 60.68% (+0.03%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 00acf4c | Docs | Datadog PR Page | Give us feedback!}

[pr-commenter]

Benchmarks [ tracer ]

Benchmark execution time: 2026-04-24 12:12:12

Comparing candidate commit 00acf4c in PR branch leiyks/fix-ci-stats-inferred-span with baseline commit b7eb412 in branch master.

Found 0 performance improvements and 4 performance regressions! Performance is the same for 188 metrics, 2 unstable metrics.

scenario:EmptyFileBench/benchEmptyFileBaseline

• 🟥 execution_time [+94.180µs; +352.360µs] or [+3.073%; +11.498%]

scenario:PDOBench/benchPDOOverhead-opcache

• 🟥 execution_time [+7.630µs; +10.000µs] or [+3.158%; +4.140%]

scenario:PDOBench/benchPDOOverheadWithDBM-opcache

• 🟥 execution_time [+8.964µs; +11.136µs] or [+3.721%; +4.622%]

scenario:PHPRedisBench/benchRedisOverhead-opcache

• 🟥 execution_time [+30.696µs; +39.579µs] or [+3.088%; +3.981%]