perf(langchain): send trace updates on root span

[hassiebp]

Important

Enhance trace updates in on_chain_start by parsing and applying specific metadata attributes in CallbackHandler.py.

• Behavior:
  • In on_chain_start, update trace with parsed attributes from metadata using _parse_langfuse_trace_attributes_from_metadata().
  • _parse_langfuse_trace_attributes_from_metadata() extracts session_id, user_id, and tags from metadata.
• Utility Functions:
  • Update _strip_langfuse_keys_from_dict() to remove langfuse_session_id, langfuse_user_id, and langfuse_tags from metadata.

^{This description was created by for 15f26fc. You can customize this summary. It will automatically update as commits are pushed.}

Disclaimer: Experimental PR review

Greptile Summary

This PR adds performance optimizations to the Langchain callback handler by enabling trace-level attributes to be set through Langchain's metadata parameter. The key changes include:

New Functionality:

• Added _parse_langfuse_trace_attributes_from_metadata() method that extracts and validates session_id, user_id, and tags from metadata
• Modified on_chain_start() to detect root chain spans (where parent_run_id is None) and apply trace-level attributes via update_trace()
• Extended the metadata key stripping logic to remove langfuse-specific keys after processing

Integration with Existing Codebase:
This change fits naturally into the existing Langchain callback handler architecture. The handler already processes metadata and manages trace/span relationships. By leveraging the root span detection logic (parent_run_id is None), the implementation correctly identifies when to apply trace-level attributes versus span-level attributes.

The metadata stripping mechanism already existed for other langfuse-specific keys, so extending it to include the new trace attribute keys (langfuse_session_id, langfuse_user_id, langfuse_tags) maintains consistency with the existing pattern of preventing langfuse-internal metadata from polluting the actual span metadata.

User Experience Improvement:
This allows users to pass trace attributes directly through Langchain's standard metadata parameter when invoking chains, rather than requiring separate Langfuse API calls. For example, users can now do:

chain.run(input_text, metadata={"langfuse_session_id": "session123", "langfuse_user_id": "user456"})

The implementation includes proper type validation to ensure attributes are correctly typed before being passed to the trace update.

Confidence score: 4/5

• This PR is safe to merge with minimal risk as it only adds new functionality without breaking existing behavior
• Score reflects well-structured implementation that follows existing patterns, with proper validation and error handling
• Pay close attention to the metadata parsing logic in _parse_langfuse_trace_attributes_from_metadata method

[greptile-apps]

_{1 file reviewed, no comments}

_{Edit Code Review Bot Settings | Greptile}