Repository navigation
fix(langchain): don't mark spans errored when a stream is closed early - #1078
Merged
Abhijeet Prasad (AbhiPrasad) merged 2 commits intoNov 6, 2026
Merged
Abhijeet Prasad (AbhiPrasad) merged 2 commits into
Abhijeet Prasad (AbhiPrasad) merged 2 commits into
Conversation
> 🤖 **This text was AI-generated.**
When code stops reading a LangGraph stream partway through, Python closes
the stream's generator with `GeneratorExit`. LangGraph passes that to
`on_chain_error`, and `BraintrustCallbackHandler` logged it as
`error=str(error)`. `str(GeneratorExit())` is `""`, so the root "LangGraph"
span showed up as failed with an empty Error panel:
```python
async for mode, chunk in graph.astream(..., config={"callbacks": [handler]}):
if done(chunk):
return # -> GeneratorExit -> on_chain_error -> span error: ""
```
The handler now treats `GeneratorExit` as an intentional stop, not a
failure. A new `_error_for_span()` helper returns `None` for it and
`str(error)` for everything else. All four error callbacks use it
(`on_llm_error`, `on_chain_error`, `on_tool_error`,
`on_retriever_error`), because LLM streams can be stopped early the same
way:
```
GeneratorExit -> span ends, no error
other errors -> span ends, error=str(error) (unchanged)
```
`asyncio.CancelledError` is still recorded as an error. Unlike the ADK
integration, this handler does not treat it as cleanup, because it can
also come from timeouts.
Fixes SDK-797
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
> 🤖 **This text was AI-generated.** The previous commit moved the `langgraph.graph` import to the top of `test_callbacks.py`. The `pylint` session in `static_checks` doesn't install langgraph, so it failed with: ``` test_callbacks.py:20:0: E0401: Unable to import 'langgraph.graph' (import-error) ``` pylint skips `import-error` when the import is inside a `try/except ImportError` block. This puts the guarded imports back inside the two langgraph tests. Refs SDK-797
Abhijeet Prasad (AbhiPrasad)
enabled auto-merge (squash)
November 6, 2026 16:52
Andrew Kent (realark)
approved these changes
Nov 6, 2026
Abhijeet Prasad (AbhiPrasad)
deleted the
abhi/sdk-797-python-langchain-integration-logs-an
branch
November 6, 2026 18:25
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
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.
When code stops reading a LangGraph stream partway through, Python closes the stream's generator with
GeneratorExit. LangGraph passes that toon_chain_error, andBraintrustCallbackHandlerlogged it aserror=str(error).str(GeneratorExit())is"", so the root "LangGraph" span showed up as failed with an empty Error panel:The handler now treats
GeneratorExitas an intentional stop, not a failure. A new_error_for_span()helper returnsNonefor it andstr(error)for everything else. All four error callbacks use it (on_llm_error,on_chain_error,on_tool_error,on_retriever_error), because LLM streams can be stopped early the same way:asyncio.CancelledErroris still recorded as an error. Unlike the ADK integration, this handler does not treat it as cleanup, because it can also come from timeouts.Fixes https://linear.app/braintrustdata/issue/SDK-797