[Bug]: List index out of range when using output schema with ADK, trace fails to upload.

[bugthedebugger]

What component(s) are affected?

• [x] Python SDK
• [ ] Opik UI
• [ ] Opik Server
• [ ] Documentation

Opik version

• Opik version: 1.7.16

Describe the problem

When using the output schema in Google ADK, we get this error stack after which the full trace fails to upload.

Reproduction steps and code snippets

agent = LlmAgent(
        model=selected_model,
        name=agent_name,
        instruction=system_prompt,
        tools=tools,
        before_agent_callback=opik_tracer.before_agent_callback,
        after_agent_callback=opik_tracer.after_agent_callback,
        before_model_callback=opik_tracer.before_model_callback,
        after_model_callback=opik_tracer.after_model_callback,
        before_tool_callback=opik_tracer.before_tool_callback,
        after_tool_callback=opik_tracer.after_tool_callback,
        output_schema=output_schema,
        output_key=output_key,
    )

Error logs or stack trace

Error message: Unable to serialize unknown type: <class 'pydantic._internal._model_construction.ModelMetaclass'>
Traceback (most recent call last):
  File "/Users/<username>/Documents/Projects/<project name>/<project name>-ai-agents/.venv/lib/python3.13/site-packages/opik/decorator/base_track_decorator.py", line 385, in _before_call
    start_span_arguments = self._start_span_inputs_preprocessor(
        func=func,
    ...<2 lines>...
        kwargs=kwargs,
    )
  File "/Users/<username>/Documents/Projects/<project name>/<project name>-ai-agents/.venv/lib/python3.13/site-packages/opik/integrations/adk/decorators.py", line 39, in _start_span_inputs_preprocessor
    input=convert_adk_base_models(args[0]),
          ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^
  File "/Users/<username>/Documents/Projects/<project name>/<project name>-ai-agents/.venv/lib/python3.13/site-packages/opik/integrations/adk/decorators.py", line 19, in convert_adk_base_models
    return arg.model_dump(mode="json", exclude_unset=True)
           ~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/<username>/Documents/Projects/<project name>/<project name>-ai-agents/.venv/lib/python3.13/site-packages/pydantic/main.py", line 463, in model_dump
    return self.__pydantic_serializer__.to_python(
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        self,
        ^^^^^
    ...<11 lines>...
        serialize_as_any=serialize_as_any,
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
pydantic_core._pydantic_core.PydanticSerializationError: Unable to serialize unknown type: <class 'pydantic._internal._model_construction.ModelMetaclass'>
OPIK: Unexpected exception happened when tried to finalize span.
Output: content=Content(parts=[Part(video_metadata=None, thought=None, code_execution_result=None, executable_code=None, file_data=None, function_call=None, function_response=None, inline_data=None, text='{"session_required": true}')], role='model') grounding_metadata=None partial=None turn_complete=None error_code=None error_message=None interrupted=None custom_metadata=None
Error message: list index out of range
Traceback (most recent call last):
  File "/Users/<username>/Documents/Projects/<project name>/<project name>-ai-agents/.venv/lib/python3.13/site-packages/opik/decorator/base_track_decorator.py", line 427, in _after_call
    span_data_to_end, trace_data_to_end = pop_end_candidates()
                                          ~~~~~~~~~~~~~~~~~~^^
  File "/Users/<username>/Documents/Projects/<project name>/<project name>-ai-agents/.venv/lib/python3.13/site-packages/opik/decorator/base_track_decorator.py", line 529, in pop_end_candidates
    span_data_to_end = context_storage.pop_span_data()
  File "/Users/<username>/Documents/Projects/<project name>/<project name>-ai-agents/.venv/lib/python3.13/site-packages/opik/context_storage.py", line 50, in pop_span_data
    return stack[-1]
           ~~~~~^^^^
IndexError: list index out of range

Healthcheck results

No response

[Lothiraldan]

Hi, @bugthedebugger thank your for reporting this issue. I will investigate the issue shortly and will keep you posted.

[bugthedebugger]

Awesome thank you!

[Lothiraldan]

@bugthedebugger pydantic._internal._model_construction.ModelMetaclass seems to be a private class in pydantic according to https://github.com/pydantic/pydantic/issues/6381. I didn't find any usage of it in https://github.com/google/adk-python. Are you importing and using it in your own code? Can you also share the version of adk and pydantic that you are using in your environment?

[bugthedebugger]

google-adk = "^0.3.0" pydantic = ">=2.0.0,<3.0.0" <- this is part of adk dependency not added manually by as atm.

Here's a snippet of outputschema that we passed.

from pydantic import BaseModel, Field

class SessionSelectorResponse(BaseModel):
    session_required: bool = Field(description="Whether previous conversations are needed to answer the user query.")


agent = LlmAgent(
        model=selected_model,
        name=agent_name,
        instruction=system_prompt,
        before_agent_callback=opik_tracer.before_agent_callback,
        after_agent_callback=opik_tracer.after_agent_callback,
        before_model_callback=opik_tracer.before_model_callback,
        after_model_callback=opik_tracer.after_model_callback,
        before_tool_callback=opik_tracer.before_tool_callback,
        after_tool_callback=opik_tracer.after_tool_callback,
        output_schema=SessionSelectorResponse,
        output_key='data',
    )

[jverre]

@bugthedebugger We have a few issues with the ADK library, @alexkuzmik is working on a number of fixes for the ADK integration that will include support for multip-agent support

[jverre]

First PR with improvements is open here, after this significant revamp of the integration we'll release some additional improvements. More coming soon !

[Lothiraldan]

Hi @bugthedebugger, the PR has been released, did you had a chance to try it out?

[bugthedebugger]

@Lothiraldan updated to latest version and looks good 👍 You can close out the issue if all good from your end too. Thank you!