[jbachorik]

What does this PR do?:

• Adds ThreadContextBenchmark for the all-native context write API (setTraceContext, setContextValue, clearTraceContext). The class replaces the benchmark that Remove legacy DirectByteBuffer-based OTEP context storage (Phase 3) #696 deleted together with ThreadContext.
• Updates doc/performance/reports/ThreadContextBenchmark.md with a three-way comparison on one Linux x86_64 host: JNI (pre-2026-04-08), ByteBuffer (Add OTEP #4947 OTEL-compatible context storage #347 to Remove legacy DirectByteBuffer-based OTEP context storage (Phase 3) #696), and all-native (current).

Benchmark (ns/op)	JNI (pre-2026-04-08)	ByteBuffer	All-native
setContextFull	37.5	17.1	31.3
setAttrCacheHit	172.4	28.1	60.9
spanLifecycle	226.4	43.8	91.2

Motivation:

The report measured only the ByteBuffer implementation, on a laptop. #696 removed that implementation and its benchmark, so the current write path had no benchmark and no current numbers.

The all-native implementation is about 2x slower than the ByteBuffer implementation. It is 1.2x to 2.8x faster than the JNI implementation, but the JNI put() did different work per call (it kept the custom attributes). None of the three implementations shows false sharing from 1 to 4 threads.

Additional Notes:

• The report contains the source of the historical benchmark, with a link to the original file at 86367c704. ThreadContext no longer exists, so that source cannot be in the tree.
• The ported benchmark wraps its route counter. The historical benchmark used counter++ % 5, which throws after 2^31 calls on one thread. The benchmark also stops the profiler and deletes its JFR file in a trial @TearDown.
• The report documents two problems in the benchmark setup:
  • The Gradle jmh {} block overrides the class annotations.
  • stresstests.jar contains both the debug and the release libjavaProfiler under the same entry name.
    The procedure runs the jar directly, with the release library only.
• The report does not explain the 2x cost against the ByteBuffer implementation. It lists the candidate costs and recommends a profile.
• Review: two passes of the Claude Code /code-review (high). Sphinx was unavailable. The first pass found 10 issues, and this PR fixes all of them. The second pass ran over the fixed diff and found 7 issues. This PR fixes 4 of them: the javadoc and the procedure now name the jar Main defaults, the report notes that the JNI comparison covers different work, and the cost model is marked as an estimate. 3 are declined:
  • The two findings against the implementation-history and historical-results sections. Comparing implementations is the purpose of the report.
  • The finding about a shared spanLifecycle helper. The committed benchmark stays identical to the code that produced the numbers.

How to test the change?:

• :ddprof-stresstest:compileJmhJava and :ddprof-lib:qualityChecks pass.
• Follow "Reproducing the Numbers" in the report: build :ddprof-stresstest:jmhJar, keep only the release library in the jar, and run java -jar bench.jar 'ThreadContextBenchmark' -f 2 -wi 3 -w 1s -i 5 -r 2s -tu ns -bm avgt.
• The numbers in the report come from that procedure on an AWS Xeon Platinum 8175M host (Ubuntu 22.04, Temurin 25.0.3).

For Datadog employees:

• If this PR touches code that signs or publishes builds or packages, or handles
  credentials of any kind, I've requested a security review (run the dd:platform-security-review
  skill, or file a request via the PSEC review form).
  bewaire also runs automatically on every PR.
• This PR doesn't touch any of that.
• JIRA: [JIRA-XXXX]

Unsure? Have a question? Request a review!

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-26T11:59:03.366903Z	a92490c	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[datadog-datadog-prod-us1-2]

More details

The benchmark calls match the current native context API, and the reproduction procedure accounts for runner defaults and duplicate debug/release library entries. The reported timings were not independently rerun.

Was this helpful? React 👍 or 👎

_{Open Bits AI session}

_{🤖 Bits Code Review · Commit a92490c}

[dd-octo-sts]

CI Test Results

Run: #40316770477 | Commit: 8040fdc | Duration: 18m 11s (longest job)

✅ All 32 test jobs passed

Status Overview

JDK	glibc-aarch64/debug	glibc-amd64/debug	musl-aarch64/debug	musl-amd64/debug
8	-	✅	-	-
8-ibm	-	✅	-	-
8-j9	✅	✅	-	-
8-librca	-	-	✅	✅
8-orcl	-	✅	-	-
11	-	✅	-	-
11-j9	✅	✅	-	-
11-librca	-	-	✅	✅
17	✅	✅	-	-
17-graal	✅	✅	-	-
17-j9	✅	✅	-	-
17-librca	-	-	✅	✅
21	✅	✅	-	-
21-graal	✅	✅	-	-
21-librca	-	-	✅	✅
25	✅	✅	-	-
25-graal	✅	✅	-	-
25-librca	-	-	✅	✅

Legend: ✅ passed | ❌ failed | ⚪ skipped | 🚫 cancelled

Summary: Total: 32 | Passed: 32 | Failed: 0

---

Updated: 2026-10-26 12:01:10 UTC