v5.118.0 proposal

[dd-octo-sts]

Features

• LLM Observability: Add support for OpenAI Agents (trace-processor) #8044

Fixes

• Feature Flags: Allow custom agentless endpoints #9481
• General: Make peer service finalization idempotent #9364
• LLM Observability: Llm observability traces have custom trace IDs #9460
• standalone: Stamp _dd.apm.enabled on every span #9506
• webpack: Opt out of typescript auto resolve #9521

[dd-octo-sts]

Overall package size

Self size: 7.52 MB
Deduped: 8.19 MB
No deduping: 8.19 MB

Dependency sizes

| name | version | self size | total size | |------|---------|-----------|------------| | import-in-the-middle | 3.3.2 | 124.41 kB | 440.65 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}

[datadog-prod-us1-5]

🎉 All green!

🧪 All tests passed
❄️ No new flaky tests detected

🎯 Code Coverage (details)
• Patch Coverage: 99.46%
• Overall Coverage: 98.48%

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 376fc9b | Docs | Datadog PR Page | Give us feedback!}

[codecov]

Codecov Report

❌ Patch coverage is 99.45652% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 98.48%. Comparing base (6b35e7d) to head (376fc9b).
⚠️ Report is 1922 commits behind head on v5.x.

Files with missing lines	Patch %	Lines
...ages/datadog-instrumentations/src/openai-agents.js	94.49%	6 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##             v5.x    #9519       +/-   ##
===========================================
+ Coverage   83.19%   98.48%   +15.29%     
===========================================
  Files         476      947      +471     
  Lines       20153   128035   +107882     
  Branches        0    10906    +10906     
===========================================
+ Hits        16766   126096   +109330     
+ Misses       3387     1939     -1448     

Flag	Coverage Δ
aiguard	57.37% <48.31%> (?)
aiguard-integration	56.18% <50.87%> (?)
apm-bucket-0	58.60% <52.80%> (?)
apm-bucket-1	63.91% <52.80%> (?)
apm-bucket-2	62.67% <52.80%> (?)
apm-bucket-3	60.12% <52.80%> (?)
apm-capabilities-tracing	64.30% <34.60%> (?)
apm-integrations-aerospike	56.70% <52.80%> (?)
apm-integrations-confluentinc-kafka-javascript	61.52% <52.80%> (?)
apm-integrations-couchbase	57.06% <48.31%> (?)
apm-integrations-http	62.69% <52.80%> (?)
apm-integrations-kafkajs	62.16% <52.80%> (?)
apm-integrations-next	59.15% <48.31%> (?)
apm-integrations-prisma	58.64% <52.80%> (?)
appsec	72.83% <52.80%> (?)
appsec-express_fastify_graphql	70.32% <52.80%> (?)
appsec-integration	51.48% <45.95%> (?)
appsec-kafka_ldapjs_lodash	63.90% <52.80%> (?)
appsec-mongodb-core_mongoose_mysql	67.63% <52.80%> (?)
appsec-next	57.50% <52.80%> (?)
appsec-node-serialize_passport_postgres	67.30% <52.80%> (?)
appsec-sourcing_stripe_template	65.64% <52.80%> (?)
debugger	64.84% <52.80%> (?)
instrumentations-bucket-0	51.71% <47.19%> (?)
instrumentations-bucket-1	60.21% <48.31%> (?)
instrumentations-bucket-10	62.01% <52.80%> (?)
instrumentations-bucket-11	51.62% <47.19%> (?)
instrumentations-bucket-12	52.20% <47.19%> (?)
instrumentations-bucket-13	51.74% <47.19%> (?)
instrumentations-bucket-2	53.63% <47.19%> (?)
instrumentations-bucket-3	59.21% <48.31%> (?)
instrumentations-bucket-4	52.28% <47.19%> (?)
instrumentations-bucket-5	57.54% <52.80%> (?)
instrumentations-bucket-6	60.73% <52.80%> (?)
instrumentations-bucket-7	58.41% <52.80%> (?)
instrumentations-bucket-8	59.55% <52.80%> (?)
instrumentations-bucket-9	61.54% <48.31%> (?)
instrumentations-instrumentation-couchbase	51.02% <47.19%> (?)
instrumentations-integration-esbuild	34.32% <43.58%> (?)
llmobs-ai_anthropic_bedrock	62.78% <84.76%> (?)
llmobs-bucket-1	61.61% <80.00%> (?)
llmobs-openai	62.33% <48.99%> (?)
llmobs-openai-agents_vertex-ai	59.69% <97.53%> (?)
llmobs-sdk	65.48% <98.82%> (?)
openfeature	56.00% <50.51%> (?)
openfeature-unit	53.36% <51.54%> (?)
platform-core_esbuild_instrumentations-misc	40.64% <45.45%> (?)
platform-integration	61.35% <48.31%> (?)
platform-shimmer_unit-guardrails_webpack	39.06% <45.45%> (?)
plugins-bucket-0	57.04% <52.80%> (?)
plugins-bucket-1	54.29% <48.31%> (?)
plugins-bucket-11	62.26% <48.31%> (?)
plugins-bucket-18	61.96% <52.80%> (?)
plugins-bucket-19	60.03% <48.31%> (?)
plugins-bucket-20	62.04% <48.31%> (?)
plugins-bucket-4	58.56% <48.31%> (?)
plugins-bullmq_cassandra_cookie	61.73% <52.80%> (?)
plugins-cookie-parser_crypto_dd-trace-api	56.71% <48.31%> (?)
plugins-fetch_fs_generic-pool	58.76% <52.80%> (?)
plugins-google-cloud-pubsub_grpc_handlebars	64.67% <52.80%> (?)
plugins-hapi_hono_ioredis	60.21% <52.80%> (?)
plugins-jest_knex_langgraph	55.58% <48.31%> (?)
plugins-ldapjs_light-my-request_limitd-client	58.49% <52.80%> (?)
plugins-lodash_mariadb_memcached	58.07% <52.80%> (?)
plugins-moleculer_mongodb_mongodb-core	61.90% <52.80%> (?)
plugins-mongoose_multer_mysql	59.07% <52.80%> (?)
plugins-mysql2_nats_node-serialize	60.70% <52.80%> (?)
plugins-opensearch_passport-http_pino	59.53% <52.80%> (?)
plugins-postgres_process_pug	58.26% <52.80%> (?)
plugins-redis_router_sequelize	62.08% <52.80%> (?)
plugins-test-and-upstream-rhea_undici_url	61.62% <52.80%> (?)
plugins-valkey_vm_winston	58.00% <52.80%> (?)
plugins-ws	59.58% <48.31%> (?)
profiling	62.14% <52.80%> (?)
serverless-aws-sdk-aws-sdk	55.05% <52.80%> (?)
serverless-aws-sdk-bedrockruntime	54.75% <52.80%> (?)
serverless-aws-sdk-client	56.42% <52.80%> (?)
serverless-aws-sdk-dynamodb	55.66% <52.80%> (?)
serverless-aws-sdk-eventbridge	49.38% <47.19%> (?)
serverless-aws-sdk-kinesis	59.33% <52.80%> (?)
serverless-aws-sdk-lambda	57.38% <52.80%> (?)
serverless-aws-sdk-s3	55.75% <52.80%> (?)
serverless-aws-sdk-serverless-peer-service	59.73% <52.80%> (?)
serverless-aws-sdk-sns	60.19% <52.80%> (?)
serverless-aws-sdk-sqs	60.61% <52.80%> (?)
serverless-aws-sdk-stepfunctions	55.58% <52.80%> (?)
serverless-aws-sdk-util	51.47% <47.19%> (?)
serverless-bucket-0	54.15% <48.31%> (?)
serverless-bucket-1	59.24% <48.31%> (?)
test-optimization-cucumber	72.01% <48.31%> (?)
test-optimization-cypress	65.50% <48.31%> (?)
test-optimization-jest	73.40% <48.31%> (?)
test-optimization-mocha	73.43% <48.31%> (?)
test-optimization-playwright-playwright-atr	60.25% <48.31%> (?)
test-optimization-playwright-playwright-efd	60.49% <48.31%> (?)
test-optimization-playwright-playwright-final-status	60.46% <48.31%> (?)
test-optimization-playwright-playwright-impacted-tests	60.19% <48.31%> (?)
test-optimization-playwright-playwright-reporting	61.67% <48.31%> (?)
test-optimization-playwright-playwright-test-management	60.99% <48.31%> (?)
test-optimization-playwright-playwright-test-span	60.38% <48.31%> (?)
test-optimization-selenium	60.12% <48.31%> (?)
test-optimization-testopt	58.55% <48.31%> (?)
test-optimization-vitest	70.33% <48.31%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[datadog-prod-us1-5]

More details

The changed LLMObs propagation, OpenFeature endpoint, peer-service finalization, and OpenAI Agents helper paths behaved correctly across targeted normal, boundary, malformed, and propagation scenarios. The full OpenAI Agents versioned integration suite could not run because its local version fixtures are not materialized, so that path remains the main validation gap.

Was this helpful? React 👍 or 👎

_{📊 Validated against 10 scenarios · Open Bits AI session}

_{🤖 Datadog Autotest · Commit 9b61525 · What is Autotest? · @DataDog review to ask questions · Any feedback? Reach out in #autotest}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9b61525319

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".

[chatgpt-codex-connector]

Preserve active parent for agent workflows

When an agent run is started inside an existing Datadog span, such as an HTTP request handler, this direct startSpan() call creates the workflow as a new root because dd-trace only uses a parent when childOf/references are provided. The child agent and openai.request spans then form a separate trace, so the agent work no longer appears under the request that triggered it; pass the currently active span as the workflow parent.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Trace generation spans from chat-completions agents

When users configure the Agents SDK to use Chat Completions (setOpenAIAPI('chat_completions') or OpenAIChatCompletionsModel), the SDK reports the LLM call as a generation span rather than a response span. Since this mapping only treats response as an LLM span, onSpanStart() drops those model calls, so chat-completions agents lose their LLM/APM span and LLMObs input/output tagging even though the integration advertises @openai/agents support.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Preserve non-text agent input parts

When an Agents run sends Responses API message content with images or files, this filter keeps only text parts and drops every non-text part before tagging the LLM span and workflow input. A multimodal prompt can therefore be recorded as incomplete or even as an empty user message, so LLMObs no longer reflects the actual model input; include placeholders or structured handling for image/file parts instead of filtering them out.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Create LLMObs tagger for late enablement

When @openai/agents is loaded before tracer.llmobs.enable(), this one-time check leaves #tagger undefined because DD_LLMOBS_ENABLED is still false at plugin construction. A later programmatic enable flips the shared config and starts the writers, but every openai-agents LLMObs path is guarded by this.#tagger, so this integration keeps emitting only APM spans; create the tagger unconditionally or lazily when LLMObs becomes enabled.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Activate tool spans while user tools run

When a function tool does instrumented work such as HTTP or database calls, this startSpan() only creates the tool span with childOf; it never makes that span active for the duration of the agents-core span. Since the plugin only binds scope around model response calls, downstream spans emitted from user tool execution inherit the prior active span instead of the function tool span, breaking the trace hierarchy for tool internals.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9b61525319

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".

[chatgpt-codex-connector]

Preserve active parent for agent workflows

When an agent run is started inside an existing Datadog span, such as an HTTP request handler, this direct startSpan() call creates the workflow as a new root because dd-trace only uses a parent when childOf/references are provided. The child agent and openai.request spans then form a separate trace, so the agent work no longer appears under the request that triggered it; pass the currently active span as the workflow parent.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Trace generation spans from chat-completions agents

When users configure the Agents SDK to use Chat Completions (setOpenAIAPI('chat_completions') or OpenAIChatCompletionsModel), the SDK reports the LLM call as a generation span rather than a response span. Since this mapping only treats response as an LLM span, onSpanStart() drops those model calls, so chat-completions agents lose their LLM/APM span and LLMObs input/output tagging even though the integration advertises @openai/agents support.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Preserve non-text agent input parts

When an Agents run sends Responses API message content with images or files, this filter keeps only text parts and drops every non-text part before tagging the LLM span and workflow input. A multimodal prompt can therefore be recorded as incomplete or even as an empty user message, so LLMObs no longer reflects the actual model input; include placeholders or structured handling for image/file parts instead of filtering them out.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Create LLMObs tagger for late enablement

When @openai/agents is loaded before tracer.llmobs.enable(), this one-time check leaves #tagger undefined because DD_LLMOBS_ENABLED is still false at plugin construction. A later programmatic enable flips the shared config and starts the writers, but every openai-agents LLMObs path is guarded by this.#tagger, so this integration keeps emitting only APM spans; create the tagger unconditionally or lazily when LLMObs becomes enabled.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Activate tool spans while user tools run

When a function tool does instrumented work such as HTTP or database calls, this startSpan() only creates the tool span with childOf; it never makes that span active for the duration of the agents-core span. Since the plugin only binds scope around model response calls, downstream spans emitted from user tool execution inherit the prior active span instead of the function tool span, breaking the trace hierarchy for tool internals.

Useful? React with 👍 / 👎.

[pr-commenter]

Benchmarks

Benchmark execution time: 2026-07-24 16:46:37

Comparing candidate commit 376fc9b in PR branch v5.118.0-proposal with baseline commit 9c6a17b in branch v5.x.

📊 Benchmarking dashboard

Found 3 performance improvements and 0 performance regressions! Performance is the same for 2313 metrics, 42 unstable metrics.

Explanation

This is an A/B test comparing a candidate commit's performance against that of a baseline commit. Performance changes are noted in the tables below as:

• 🟩 = significantly better candidate vs. baseline
• 🟥 = significantly worse candidate vs. baseline

We compute a confidence interval (CI) over the relative difference of means between metrics from the candidate and baseline commits, considering the baseline as the reference.

If the CI is entirely outside the configured SIGNIFICANT_IMPACT_THRESHOLD (or the deprecated UNCONFIDENCE_THRESHOLD), the change is considered significant.

Feel free to reach out to #apm-benchmarking-platform on Slack if you have any questions.

More details about the CI and significant changes

You can imagine this CI as a range of values that is likely to contain the true difference of means between the candidate and baseline commits.

CIs of the difference of means are often centered around 0%, because often changes are not that big:

---------------------------------(------|---^--------)-------------------------------->
                              -0.6%    0%  0.3%     +1.2%
                                 |          |        |
         lower bound of the CI --'          |        |
sample mean (center of the CI) -------------'        |
         upper bound of the CI ----------------------'

As described above, a change is considered significant if the CI is entirely outside the configured SIGNIFICANT_IMPACT_THRESHOLD (or the deprecated UNCONFIDENCE_THRESHOLD).

For instance, for an execution time metric, this confidence interval indicates a significantly worse performance:

----------------------------------------|---------|---(---------^---------)---------->
                                       0%        1%  1.3%      2.2%      3.1%
                                                  |   |         |         |
       significant impact threshold --------------'   |         |         |
                      lower bound of CI --------------'         |         |
       sample mean (center of the CI) --------------------------'         |
                      upper bound of CI ----------------------------------'

scenario:llmobs-span-processor-agent-26

• 🟩 cpu_user_time [-486.952ms; -195.344ms] or [-12.101%; -4.855%]
• 🟩 execution_time [-513.701ms; -206.450ms] or [-12.308%; -4.946%]
• 🟩 throughput [+11257.485op/s; +28320.853op/s] or [+4.893%; +12.311%]

Unstable benchmarks

These benchmarks have a confidence interval too wide to call a change; treat them as noise rather than signal.

scenario:appsec-appsec-enabled-20

• unstable execution_time [-221.084ms; +228.545ms] or [-5.912%; +6.112%]

scenario:appsec-appsec-enabled-24

• unstable execution_time [-211.860ms; +209.102ms] or [-7.928%; +7.825%]

scenario:appsec-appsec-enabled-26

• unstable execution_time [-233.585ms; +238.223ms] or [-9.092%; +9.272%]

scenario:appsec-appsec-enabled-with-attacks-24

• unstable execution_time [-165.124ms; +155.399ms] or [-5.343%; +5.028%]

scenario:appsec-appsec-enabled-with-attacks-26

• unstable execution_time [-186.822ms; +193.929ms] or [-6.418%; +6.662%]

scenario:appsec-control-20

• unstable execution_time [-179.590ms; +186.078ms] or [-10.041%; +10.404%]

scenario:appsec-control-24

• unstable execution_time [-114942.254µs; +115637.021µs] or [-9.246%; +9.302%]

scenario:appsec-control-26

• unstable execution_time [-127488.957µs; +128123.524µs] or [-10.256%; +10.307%]

scenario:child_process-shell-string-24

• unstable execution_time [-18.793ms; +25.616ms] or [-5.753%; +7.842%]
• unstable throughput [-224755.154op/s; +214636.708op/s] or [-6.259%; +5.977%]

scenario:debugger-line-probe-with-snapshot-minimal-24

• unstable cpu_user_time [-1639.495ms; +586.794ms] or [-20.007%; +7.161%]
• unstable execution_time [-1666.770ms; +588.702ms] or [-18.735%; +6.617%]
• unstable instructions [-14.2G instructions; +4.8G instructions] or [-21.248%; +7.261%]
• unstable throughput [-177.895op/s; +465.464op/s] or [-4.809%; +12.583%]

scenario:debugger-line-probe-without-snapshot-24

• unstable cpu_user_time [-1742.999ms; +2750.492ms] or [-21.292%; +33.599%]
• unstable execution_time [-1772.088ms; +2768.681ms] or [-19.918%; +31.120%]
• unstable instructions [-15.0G instructions; +23.6G instructions] or [-22.354%; +35.287%]
• unstable max_rss_usage [-7.440MB; +11.484MB] or [-4.724%; +7.292%]
• unstable throughput [-780.828op/s; +505.952op/s] or [-21.199%; +13.736%]

scenario:debugger-line-probe-without-snapshot-26

• unstable cpu_user_time [-3225.401ms; +2923.067ms] or [-31.767%; +28.789%]
• unstable execution_time [-3256.904ms; +2911.746ms] or [-29.830%; +26.669%]
• unstable instructions [-28.6G instructions; +25.9G instructions] or [-33.599%; +30.426%]
• unstable max_rss_usage [-10055.016KB; +10419.816KB] or [-6.240%; +6.466%]
• unstable throughput [-626.508op/s; +680.294op/s] or [-20.301%; +22.044%]

scenario:dogstatsd-with-tags-20

• unstable cpu_user_time [-430.731ms; +277.423ms] or [-8.515%; +5.484%]
• unstable execution_time [-432.009ms; +272.597ms] or [-8.424%; +5.315%]
• unstable throughput [-95387.392op/s; +145296.563op/s] or [-5.824%; +8.871%]

scenario:plugin-aws-sdk-lambda-inject-with-context-24

• unstable cpu_user_time [-216.449ms; +186.184ms] or [-5.637%; +4.849%]
• unstable execution_time [-217.116ms; +185.454ms] or [-5.618%; +4.798%]

scenario:plugin-claude-agent-sdk-compact-stream-scan-26

• unstable cpu_usage_percentage [-9.004%; +6.055%]
• unstable execution_time [-3346.055µs; +3536.073µs] or [-5.435%; +5.744%]
• unstable throughput [-27965.783op/s; +30572.558op/s] or [-5.073%; +5.546%]

scenario:plugin-graphql-long-with-depth-and-collapse-off-20

• unstable max_rss_usage [-20.624MB; +32.632MB] or [-5.303%; +8.391%]

scenario:plugin-graphql-long-with-depth-off-26

• unstable max_rss_usage [-26.502MB; +15.845MB] or [-11.855%; +7.087%]

scenario:plugin-graphql-long-with-depth-on-max-20

• unstable execution_time [-601.614ms; +589.129ms] or [-5.098%; +4.993%]
• unstable throughput [-3.435op/s; +3.490op/s] or [-5.038%; +5.118%]

scenario:plugin-pg-service-24

• unstable cpu_usage_percentage [-8.342%; +5.119%]
• unstable execution_time [-135.768ms; +202.935ms] or [-7.958%; +11.896%]
• unstable throughput [-301613.336op/s; +206127.463op/s] or [-8.476%; +5.793%]

scenario:sampling-agent-24

• unstable cpu_user_time [-180.046ms; +286.369ms] or [-7.327%; +11.654%]
• unstable execution_time [-190.481ms; +305.062ms] or [-7.735%; +12.389%]
• unstable throughput [-408836.892op/s; +254929.758op/s] or [-8.295%; +5.172%]

[leoromanovsky]

ffe lgtm