fix(openai_agents): isolate concurrent workflow state

[yshapiro-57]

Human Summary

This is intended to fix the following errors in the SDK:

Error preparing LLMObs span event for span <Span(id=2273126637415999657,trace_id=141943098994275675798334088410531553281,parent_id=17672040544517427409,name=Agent workflow)>, missing span kind in span context.

I had initially put up a different PR for this, #20846. Both PRs are agent-coded. @Yun-Kim pointed out that this other PR fixes the symptom but not the root cause of the following problem:

I think your executor code starts concurrent Runner.run_streamed() asyncio tasks in on experiment span, which means both agent workflow traces then share the same APM trace ID. The problem is that our integration internally stores per-run state keyed by APM trace ID, and there's likely a race of that unwittingly shared entry that ends up being wiped before the last run finishes.
This is definitely an instrumentation issue where we assumed agent runs should be 1:1 with APM trace IDs. When someone chooses to fork shouldn't break our LLMO tracing. We just need to use the OpenAI trace ID (instead of the APM or LLMO trace IDs) as keys in the internal llmobs_traces map to solve this issue.

AI Summary

This fixes cross-talk between concurrent OpenAI Agents workflows that execute within the same APM trace. Workflow lifecycle bookkeeping now follows each OpenAI Agents trace identity, keeping LLM Observability metadata isolated through completion. Regression coverage exercises two concurrent agent runs under one traced operation.

AI Verification

• Targeted concurrency regression:
  scripts/run-tests --venv 213dcfe -- tests/contrib/openai_agents/test_openai_agents_llmobs.py -k test_llmobs_concurrent_agents_in_same_apm_trace
• Full OpenAI Agents integration suite:
  scripts/run-tests -s --venv 213dcfe
• Targeted type checking:
  scripts/lint typing -- ddtrace/contrib/internal/openai_agents/processor.py ddtrace/llmobs/_integrations/openai_agents.py
• Repository lint suite:
  scripts/lint checks

Human Verification

I have installed dd-trace-py from local and confirmed that the "missing span kind" error still happens for me on trunk, but no longer happens on this branch.

Made with Cursor

[cit-pr-commenter-54b7da]

Codeowners resolved as

Resolved from the full PR diff against main using the target branch CODEOWNERS file.
CODEOWNERS team requests not listed below are not required by the current file set.

ddtrace/contrib/internal/openai_agents/processor.py                     @DataDog/ml-observability
ddtrace/llmobs/_integrations/openai_agents.py                           @DataDog/ml-observability
releasenotes/notes/fix-openai-agents-concurrent-traces-f1f791d2e925dc92.yaml  @DataDog/apm-python
tests/contrib/openai_agents/test_openai_agents_llmobs.py                @DataDog/ml-observability

[cit-pr-commenter-54b7da]

Circular import analysis

⚠️ Existing circular imports

There are 1 circular imports that already exist on the base branch and have not been changed by this PR.

ddtrace.errortracking._handled_exceptions.bytecode_injector -> ddtrace.errortracking._handled_exceptions.callbacks -> ddtrace.errortracking._handled_exceptions.collector -> ddtrace.errortracking._handled_exceptions.bytecode_reporting -> ddtrace.errortracking._handled_exceptions.bytecode_injector

[cit-pr-commenter-54b7da]

Dependency direction analysis

⚠️ Existing dependency direction violations

There are 201 dependency direction violations that already exist on the base branch and have not been changed by this PR.

Show existing violations (showing 5 of 201 highest severity)

ddtrace.internal.tracemethods -×-> ddtrace.trace  (internal-core -> product:tracing, score=132)
ddtrace.internal.ci_visibility.git_client -×-> ddtrace.trace  (product:ci_visibility -> product:tracing, score=130)
ddtrace.llmobs._integrations.litellm -×-> ddtrace.trace  (product:llmobs -> product:tracing, score=130)
ddtrace.appsec._listeners -×-> ddtrace.trace  (product:appsec -> product:tracing, score=130)
ddtrace.internal.ci_visibility.recorder -×-> ddtrace.trace  (product:ci_visibility -> product:tracing, score=130)

To see all violations, download the layers-base.json and layers-pr.json artifacts from this CI job and run:

uv run --script scripts/import-analysis/layers.py compare layers-base.json layers-pr.json

[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 2026-10-09T20:10:07.923140Z	62b2428	Draft marked ready
🔒 Security Review	✅ Completed 2026-10-09T20:11:21.012062Z	62b2428	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-prod-us1-5]

More details

Workflow metadata creation, lookup, and cleanup consistently use the OpenAI trace ID, keeping concurrent workflows independent when they share an APM trace.

Was this helpful? React 👍 or 👎

_{Open Bits AI session}

_{🤖 Bits Code Review · Commit 62b2428}