fix(livekit_agents): skip empty LiveKit function tool spans (#476)

resolves
https://linear.app/braintrustdata/issue/BT-5477/python-sdk-livekit-agents-auto-instrumentation-emits-an-empty-function

Author: AbhiPrasad

py/src/braintrust/integrations/livekit_agents/test_livekit_agents.py

@@ -409,6 +409,7 @@ def _assert_openai_voice_turn_spans(logs, *, speech_text, outer_parent_name):
     _assert_eou_spans(logs)
     _assert_stt_spans(logs, speech_text)
     _assert_session_spans(logs)
+    _assert_no_function_tool_spans_without_tools(logs)
     _assert_custom_metrics_stay_in_metadata(logs)
 
     if outer_parent_name:
@@ -478,13 +479,17 @@ def _assert_session_spans(logs):
     _assert_no_metrics(session_logs, "prompt_tokens", "completion_tokens", "tokens")
 
 
+def _assert_no_function_tool_spans_without_tools(logs):
+    assert not _spans_named(logs, "function_tool")
+
+
 def _assert_voice_turn_parenting(logs, outer_parent_name):
     outer_parent_log = _single_span(logs, outer_parent_name)
     session_log = _single_span(logs, "livekit_agent_session")
     assert session_log.get("span_parents") == [outer_parent_log.get("span_id")]
     assert "end" in session_log.get("metrics", {})
 
-    for child_name in ("function_tool", "llm_request_run", "user_speaking", "vad_endpointing"):
+    for child_name in ("llm_request_run", "user_speaking", "vad_endpointing"):
         child_log = _single_span(logs, child_name)
         assert child_log.get("span_parents") == [session_log.get("span_id")], child_name
     assert _single_span(logs, "llm_request_run")["span_attributes"]["type"].value == "task"

py/src/braintrust/integrations/livekit_agents/tracing.py

@@ -218,18 +218,43 @@ async def traced_llm_stream_run(wrapped: Any, instance: Any, args: tuple[Any, ..
 async def traced_execute_tools_task(wrapped: Any, instance: Any, args: tuple[Any, ...], kwargs: dict[str, Any]) -> Any:
     session = kwargs.get("session")
     parent = getattr(session, _SESSION_PARENT_ATTR, None)
-    result = await _traced_async_span(
-        "function_tool",
-        SpanTypeAttribute.TOOL,
-        wrapped,
-        args,
-        kwargs,
-        parent=parent if isinstance(parent, str) else None,
-        event_builder=lambda: _tool_execution_event(kwargs.get("tool_output")),
-    )
+    parent_arg = parent if isinstance(parent, str) else None
+    start_time = time.time()
+    try:
+        with _without_current_span(parent_arg):
+            result = await wrapped(*args, **kwargs)
+    except Exception as exc:
+        end_time = time.time()
+        event = _tool_execution_event(kwargs.get("tool_output"))
+        if event:
+            _log_function_tool_span(event, parent=parent_arg, start_time=start_time, end_time=end_time, error=exc)
+        raise
+    end_time = time.time()
+    event = _tool_execution_event(kwargs.get("tool_output"))
+    if event:
+        _log_function_tool_span(event, parent=parent_arg, start_time=start_time, end_time=end_time)
     return result
 
 
+def _log_function_tool_span(
+    event: dict[str, Any],
+    *,
+    parent: str | None,
+    start_time: float,
+    end_time: float,
+    error: Exception | None = None,
+) -> None:
+    span = start_span(
+        name="function_tool",
+        type=SpanTypeAttribute.TOOL,
+        parent=parent,
+        start_time=start_time,
+        set_current=False,
+    )
+    span.log(error=error, **event)
+    span.end(end_time=end_time)
+
+
 async def _traced_async_span(
     name: str,
     span_type: str,