pipecat version

v0.0.100 (Unreleased)

Python version

3.12

Operating System

Linux (Docker/Kubernetes on AKS)

Issue description

When using GeminiLiveLLMService (or GeminiLiveVertexLLMService) with a registered function that ends the pipeline (like end_call), a RuntimeWarning is logged because the timeout handler coroutine is never awaited.

Analysis:

The issue appears to be a race condition in LLMService._run_function_call(). When the function call (e.g., end_call) triggers pipeline shutdown via EndTaskFrame/EndFrame, the timeout handler coroutine created for that function call is orphaned before it can be properly cancelled or awaited.

Potential Fix:

The timeout handler task should be properly cancelled in a finally block or when the function call completes, ensuring it doesn't get garbage collected with an unawaited coroutine.

Workaround:

The warning is cosmetic and doesn't affect functionality. Calls complete successfully.

Potential Code Fix:

File: pipecat/services/llm_service.py
Method: _run_function_call() (lines 577-672)

The Bug (Lines 604-612, 669-672)

# Line 612: timeout_task is created
timeout_task = self.create_task(timeout_handler())
# ... handler is called ...
except Exception as e:
    # Line 669-672: No finally block!
    error_message = f"Error executing function call [...]"
    logger.error(...)
    await self.push_error(...)
# Function ends WITHOUT cancelling timeout_task if:
# 1. Handler doesn't call result_callback (rare edge case)
# 2. Exception is raised (timeout_task leaks)
# 3. Handler triggers pipeline end (end_call case)

The Fix

Add a finally block to ensure timeout_task is always cancelled:

# After line 668, add:
        except Exception as e:
            error_message = f"Error executing function call [{runner_item.function_name}]: {e}"
            logger.error(f"{self} {error_message}")
            await self.push_error(error_msg=error_message, exception=e, fatal=False)
        finally:
            # Ensure timeout task is cancelled to prevent "coroutine never awaited" warning
            if timeout_task and not timeout_task.done():
                await self.cancel_task(timeout_task)

Why This Fixes It

When end_call triggers EndTaskFrame/EndFrame:

1. The handler completes (or the pipeline starts shutting down)
2. The _run_function_call method exits
3. Without the fix: timeout_task is orphaned → RuntimeWarning
4. With the fix: finally block cancels timeout_task cleanly

Reproduction steps

1. Register an end_call function that pushes EndTaskFrame and EndFrame to gracefully end the call
2. During a conversation, trigger the function call (e.g., user says goodbye, LLM invokes end_call)
3. Observe the warning in logs

Expected behavior

No warnings - the timeout handler should be properly cancelled or awaited when the pipeline ends.

Actual behavior

/usr/local/lib/python3.12/asyncio/events.py:88: RuntimeWarning: coroutine 'LLMService._run_function_call..timeout_handler' was never awaited
self._context.run(self._callback, *self._args)
RuntimeWarning: Enable tracemalloc to get the object allocation traceback

Logs

2026-01-29 11:17:05.015 | DEBUG | pipecat.observers.loggers.llm_log_observer:on_push_frame:68 - 🧠 GeminiLiveVertexLLMService#5 → LLM GENERATING: ' Bye!' at 114.21s
/usr/local/lib/python3.12/asyncio/events.py:88: RuntimeWarning: coroutine 'LLMService._run_function_call.<locals>.timeout_handler' was never awaited
  self._context.run(self._callback, *self._args)
RuntimeWarning: Enable tracemalloc to get the object allocation traceback
2026-01-29 11:17:05.170 | DEBUG | pipecat.services.llm_service:_run_function_call:585 - GeminiLiveVertexLLMService#5 Calling function [end_call:...] with arguments {'reason': 'conversation concluded.'}
2026-01-29 11:17:05.170 | INFO | bots.base_bot:end_call:313 - end_call triggered for Mohammad Hussam (08767872521). Reason: conversation concluded.