feat(llmobs): sampling decisions, rates, and propagation#9030
Merged
Conversation
Compute a keep/drop decision + rate once on the root LLMObs span and inherit it across spans and services (via x-datadog-tags). Spans are always shipped; the decision is recorded in the event _dd block so the backend can honor it. Mirrors dd-trace-py. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
🎉 All green!🧪 All tests passed 🔗 Commit SHA: ff8b677 | Docs | Datadog PR Page | Give us feedback! |
Contributor
Overall package sizeSelf size: 6.36 MB Dependency sizes| name | version | self size | total size | |------|---------|-----------|------------| | import-in-the-middle | 3.2.0 | 104.26 kB | 843.44 kB | | opentracing | 0.14.7 | 194.81 kB | 194.81 kB | | dc-polyfill | 0.1.11 | 25.74 kB | 25.74 kB |🤖 This report was automatically generated by heaviest-objects-in-the-universe |
BenchmarksBenchmark execution time: 2026-06-26 17:06:40 Comparing candidate commit ff8b677 in PR branch Found 0 performance improvements and 0 performance regressions! Performance is the same for 1951 metrics, 14 unstable metrics.
|
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
- Match dd-trace-py: only the true trace root makes a fresh sampling decision; a span under a propagated LLMObs parent inherits whatever was propagated (possibly none) instead of starting a divergent one. - Reuse the shared formatKnuthRate (hoisted to sampler.js) instead of a bespoke llmobs formatRate. - Rename handleLLMObsParentIdInjection -> handleLLMObsInjection. - Declare llmobs.sampleRate in the public index.d.ts (fixes config-names lint) and tag sample_rate on enablement telemetry. - Drop redundant comments flagged in review. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…kim/llmobs-sampling
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
- Rebuild the sampler lazily from the live config rate so a disable/ re-enable (or any rate change) is reflected, instead of caching one sampler at construction. - Tweak the injection comment wording. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Keep the formatter in priority_sampler.js (export it) instead of hoisting to sampler.js, per review. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
The sample rate is set at init and not mutable at runtime, so the per-span lazy rebuild was unnecessary work on a hot path. Build it once at construction. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Yun-Kim
commented
Jun 24, 2026
Yun-Kim
commented
Jun 24, 2026
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: dbda2e5b15
ℹ️ About Codex in GitHub
Codex has been enabled to automatically 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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
BridgeAR
reviewed
Jun 24, 2026
…onfig-reactive - Mirror `llmobs.sampleRate` in the v5 type declarations (index.d.v5.ts), per the repo type policy requiring non-v6-only public APIs in both declaration surfaces (PR review). - Register DD_LLMOBS_SAMPLE_RATE as implementation "A" in supported-configurations.json (PR review); generated types are unchanged. - Build the LLMObs sampler lazily and rebuild it when config.llmobs.sampleRate changes, instead of fixing it for the tagger's lifetime, so a future remote config update to the sample rate takes effect without re-instantiation. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…1.0) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
sabrenner
reviewed
Jun 26, 2026
sabrenner
left a comment
sabrenner
left a comment
Collaborator
There was a problem hiding this comment.
just a couple of comments but lgtm!
- Drop the redundant #samplerRate field; use sampler.rate() for the rebuild check.
- Move formatKnuthRate to the shared src/util.js so the LLMObs tagger no longer
imports priority_sampler just for a string formatter.
- Reword the sampleRate JSDoc in index.d.ts and index.d.v5.ts ("honored at
ingestion time").
- Test rates that exercise the formatter: 1/3 (6-decimal cap) and 0.25
(trailing-zero stripping).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
sabrenner
approved these changes
Jun 26, 2026
dd-octo-sts Bot
pushed a commit
that referenced
this pull request
Jun 26, 2026
* feat(llmobs): add sampling decisions, rates, and propagation
Compute a keep/drop decision + rate once on the root LLMObs span and
inherit it across spans and services (via x-datadog-tags). Spans are
always shipped; the decision is recorded in the event _dd block so the
backend can honor it. Mirrors dd-trace-py.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(llmobs): address sampling PR review comments
- Match dd-trace-py: only the true trace root makes a fresh sampling
decision; a span under a propagated LLMObs parent inherits whatever
was propagated (possibly none) instead of starting a divergent one.
- Reuse the shared formatKnuthRate (hoisted to sampler.js) instead of a
bespoke llmobs formatRate.
- Rename handleLLMObsParentIdInjection -> handleLLMObsInjection.
- Declare llmobs.sampleRate in the public index.d.ts (fixes config-names
lint) and tag sample_rate on enablement telemetry.
- Drop redundant comments flagged in review.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* chore(llmobs): regenerate config types after merging master
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(llmobs): address follow-up sampling review comments
- Rebuild the sampler lazily from the live config rate so a disable/
re-enable (or any rate change) is reflected, instead of caching one
sampler at construction.
- Tweak the injection comment wording.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(llmobs): export formatKnuthRate from priority_sampler.js
Keep the formatter in priority_sampler.js (export it) instead of
hoisting to sampler.js, per review.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(llmobs): build the sampler once in the tagger constructor
The sample rate is set at init and not mutable at runtime, so the
per-span lazy rebuild was unnecessary work on a hot path. Build it once
at construction.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Update packages/dd-trace/src/llmobs/tagger.js
* refactor(llmobs): address sampling review comments and make sampler config-reactive
- Mirror `llmobs.sampleRate` in the v5 type declarations (index.d.v5.ts), per
the repo type policy requiring non-v6-only public APIs in both declaration
surfaces (PR review).
- Register DD_LLMOBS_SAMPLE_RATE as implementation "A" in
supported-configurations.json (PR review); generated types are unchanged.
- Build the LLMObs sampler lazily and rebuild it when config.llmobs.sampleRate
changes, instead of fixing it for the tagger's lifetime, so a future remote
config update to the sample rate takes effect without re-instantiation.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(llmobs): match DD_LLMOBS_SAMPLE_RATE default to config registry (1.0)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(llmobs): address remaining sampling review comments
- Drop the redundant #samplerRate field; use sampler.rate() for the rebuild check.
- Move formatKnuthRate to the shared src/util.js so the LLMObs tagger no longer
imports priority_sampler just for a string formatter.
- Reword the sampleRate JSDoc in index.d.ts and index.d.v5.ts ("honored at
ingestion time").
- Test rates that exercise the formatter: 1/3 (6-decimal cap) and 0.25
(trailing-zero stripping).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Merged
pabloerhard
pushed a commit
that referenced
this pull request
Jun 26, 2026
* feat(llmobs): add sampling decisions, rates, and propagation
Compute a keep/drop decision + rate once on the root LLMObs span and
inherit it across spans and services (via x-datadog-tags). Spans are
always shipped; the decision is recorded in the event _dd block so the
backend can honor it. Mirrors dd-trace-py.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(llmobs): address sampling PR review comments
- Match dd-trace-py: only the true trace root makes a fresh sampling
decision; a span under a propagated LLMObs parent inherits whatever
was propagated (possibly none) instead of starting a divergent one.
- Reuse the shared formatKnuthRate (hoisted to sampler.js) instead of a
bespoke llmobs formatRate.
- Rename handleLLMObsParentIdInjection -> handleLLMObsInjection.
- Declare llmobs.sampleRate in the public index.d.ts (fixes config-names
lint) and tag sample_rate on enablement telemetry.
- Drop redundant comments flagged in review.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* chore(llmobs): regenerate config types after merging master
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(llmobs): address follow-up sampling review comments
- Rebuild the sampler lazily from the live config rate so a disable/
re-enable (or any rate change) is reflected, instead of caching one
sampler at construction.
- Tweak the injection comment wording.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(llmobs): export formatKnuthRate from priority_sampler.js
Keep the formatter in priority_sampler.js (export it) instead of
hoisting to sampler.js, per review.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(llmobs): build the sampler once in the tagger constructor
The sample rate is set at init and not mutable at runtime, so the
per-span lazy rebuild was unnecessary work on a hot path. Build it once
at construction.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Update packages/dd-trace/src/llmobs/tagger.js
* refactor(llmobs): address sampling review comments and make sampler config-reactive
- Mirror `llmobs.sampleRate` in the v5 type declarations (index.d.v5.ts), per
the repo type policy requiring non-v6-only public APIs in both declaration
surfaces (PR review).
- Register DD_LLMOBS_SAMPLE_RATE as implementation "A" in
supported-configurations.json (PR review); generated types are unchanged.
- Build the LLMObs sampler lazily and rebuild it when config.llmobs.sampleRate
changes, instead of fixing it for the tagger's lifetime, so a future remote
config update to the sample rate takes effect without re-instantiation.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(llmobs): match DD_LLMOBS_SAMPLE_RATE default to config registry (1.0)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(llmobs): address remaining sampling review comments
- Drop the redundant #samplerRate field; use sampler.rate() for the rebuild check.
- Move formatKnuthRate to the shared src/util.js so the LLMObs tagger no longer
imports priority_sampler just for a string formatter.
- Reword the sampleRate JSDoc in index.d.ts and index.d.v5.ts ("honored at
ingestion time").
- Test rates that exercise the formatter: 1/3 (6-decimal cap) and 0.25
(trailing-zero stripping).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
pabloerhard
pushed a commit
that referenced
this pull request
Jun 26, 2026
The sampling tests in #9030 build their own taggers with `{ llmobs: { enabled: true } }`, and #8943 renamed that config key to `DD_LLMOBS_ENABLED` everywhere it could see. The two landed in parallel, so #8943 normalized the rest of the file but never saw these four fixtures. On master the tagger now reads `DD_LLMOBS_ENABLED`, finds it undefined, and returns before registering the span; `Tagger.tagMap.get` then yields undefined and the "DROPPED at sampleRate 0" test throws synchronously, aborting the whole `test:llmobs:sdk:ci` run with exit 7. Fixes: https://github.com/DataDog/dd-trace-js/actions/runs/28265509637/job/83751636644
dd-octo-sts Bot
pushed a commit
that referenced
this pull request
Jun 29, 2026
The sampling tests in #9030 build their own taggers with `{ llmobs: { enabled: true } }`, and #8943 renamed that config key to `DD_LLMOBS_ENABLED` everywhere it could see. The two landed in parallel, so #8943 normalized the rest of the file but never saw these four fixtures. On master the tagger now reads `DD_LLMOBS_ENABLED`, finds it undefined, and returns before registering the span; `Tagger.tagMap.get` then yields undefined and the "DROPPED at sampleRate 0" test throws synchronously, aborting the whole `test:llmobs:sdk:ci` run with exit 7. Fixes: https://github.com/DataDog/dd-trace-js/actions/runs/28265509637/job/83751636644
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Make and record LLMO sampling decisions on LLMO spans (always ship, we drop in the backend). Allow users to set via DD_LLMOBS_SAMPLE_RATE or via
tracer.init({ llmobs: {sampleRate: ...} })x-datadog-tags(_dd.p.llmobs_sr/_dd.p.llmobs_sd).Claude session:
1688bbe5-066d-47dc-ae28-8678aef20ee2Resume:
claude --resume 1688bbe5-066d-47dc-ae28-8678aef20ee2🤖 Generated with Claude Code