fix(langchain): clean up streaming run memo on error

[dhionson2]

Summary

Fixes #1977.

This PR fixes stale run ID retention in LangchainCallbackHandler when a streamed LLM execution terminates with an error.

Root cause

on_llm_new_token() adds the run ID to _updated_completion_start_time_memo after recording the first-token timestamp.

While on_llm_end() removes the run ID in its finally block, on_llm_error() did not perform the same cleanup.

This could cause incremental UUID retention when long-lived callback handlers are reused across failed streaming executions.

Changes

• Add per-run memo cleanup to the finally block of on_llm_error().
• Add a regression test covering a failed streamed run and another active run.
• Verify that cleanup removes only the failed run ID.
• Ensure the remaining active run is properly finalized during test teardown.

The change preserves existing public SDK behavior and LangGraph control-flow handling.

Verification

Regression test

• RED: confirmed the failed run ID remained in the memo.
• GREEN: confirmed the ID is removed after the fix.

Targeted validation

• tests/unit/test_langchain.py: 24 passed.
• Ruff lint and formatting: passed.
• git diff --check: passed.

Full unit suite on Windows

• 883 passed
• 2 skipped
• 3 failed outside the modified test module: two subprocess import failures involving pydantic and one Windows/POSIX path serialization mismatch.

The full suite has not yet been confirmed green on Linux. CI validation is pending.

Scope

This is a minimal, backward-compatible fix limited to the LangChain callback implementation and its regression test.

No dependency, generated API, or public interface changes.

Confidence Score: 5/5

This PR appears safe to merge; no actionable issues were found.

Summary

This PR removes a streamed run’s first-token memo entry when on_llm_error finishes.

• Errored LLM streams clear their own first-token memo.

_{Reviews (1) · Last reviewed commit: "fix(langchain): clean up streaming run m..." · Reviewed by Greptile}

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ dhionson2
❌ hassiebp
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[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 2026-10-20T15:34:50.332649Z	924e830	New commits

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.

[chrikrah]

@dhionson2 approving at 11ac8a0. This fixes #1977, and the new test fails when I put back the old on_llm_error.

$ python -m pytest tests/unit/test_langchain.py -q -p no:cacheprovider    # Python 3.14, uv sync --frozen
# head 11ac8a0
24 passed
# head tests, CallbackHandler.py from base bf11ec1
1 failed, 23 passed
FAILED tests/unit/test_langchain.py::test_streamed_llm_error_cleans_only_its_completion_start_time_memo

@hassiebp could you review this one?

[hassiebp]

Thanks a lot for your contribution, @dhionson2

[dhionson2]

Thanks for reviewing and merging the PR! I appreciate your time and feedback. Glad I could contribute to the project!