IndexError in ParallelAgent with LlmAgent #3523
Description
Summary
IndexError: list index out of range occurs in llm_agent.py line 449 when using LlmAgent instances within a ParallelAgent. The code attempts to access events[-2] without checking if the events list has at least 2 items.
Environment
- ADK Version: google-adk - 1.18.0
Summary: Agent Development Kit - Python Version: 3.12
- OS: macOS
Bug Location
File: google/adk/agents/llm_agent.py
Line: 449
Current Code (Buggy)
if ctx.is_resumable:
events = ctx._get_events(current_invocation=True, current_branch=True)
if events and (
ctx.should_pause_invocation(events[-1])
or ctx.should_pause_invocation(events[-2]) # ← BUG: No length check
):
returnThe Problem
The code checks if events (ensuring the list is not empty), but then attempts to access events[-2] without verifying the list has at least 2 elements. This causes an IndexError when an agent generates only 1 event.
Proposed Fix
if ctx.is_resumable:
events = ctx._get_events(current_invocation=True, current_branch=True)
if events and (
ctx.should_pause_invocation(events[-1])
or (len(events) >= 2 and ctx.should_pause_invocation(events[-2])) # ← Fixed
):
returnHow to Reproduce
Minimal Example
from google.adk.agents import Agent, SequentialAgent, ParallelAgent
from google.adk.models.google_llm import Gemini
from google.adk.runners import InMemoryRunner
from google.adk.tools.tool_context import ToolContext
from typing import Dict, Any
import asyncio
# Create a simple tool
def example_tool(tool_context: ToolContext, data: str) -> Dict[str, Any]:
"""A simple tool that saves to session state."""
tool_context.state["result"] = data
return {"status": "success"}
# Create agents with tools
agent1 = Agent(
name="Agent1",
model=Gemini(model="gemini-2.5-flash"),
instruction="Analyze the input and call example_tool with your analysis.",
tools=[example_tool],
output_key="analysis1"
)
agent2 = Agent(
name="Agent2",
model=Gemini(model="gemini-2.5-flash"),
instruction="Analyze the input and call example_tool with your analysis.",
tools=[example_tool],
output_key="analysis2"
)
# Create parallel agent
parallel = ParallelAgent(
name="ParallelAnalysis",
sub_agents=[agent1, agent2]
)
# Wrap in sequential agent
pipeline = SequentialAgent(
name="Pipeline",
sub_agents=[parallel]
)
# Run
async def test():
runner = InMemoryRunner(agent=pipeline)
response = await runner.run_debug("Analyze this data")
print(response)
asyncio.run(test())Expected Behavior
The parallel agents should execute successfully and return their results.
Actual Behavior
IndexError: list index out of range
File ".../google/adk/agents/llm_agent.py", line 449, in _run_async_impl
or ctx.should_pause_invocation(events[-2])
~~~~~~^^^^
Stack Trace
ExceptionGroup: unhandled errors in a TaskGroup (1 sub-exception)
+-+---------------- 1 ----------------
| Traceback (most recent call last):
| File ".../google/adk/agents/parallel_agent.py", line 136, in process_an_agent
| async for event in events_for_one_agent:
| File ".../google/adk/agents/base_agent.py", line 294, in run_async
| async for event in agen:
| File ".../google/adk/agents/llm_agent.py", line 449, in _run_async_impl
| or ctx.should_pause_invocation(events[-2])
| ~~~~~~^^^^
| IndexError: list index out of range
Impact
- Severity: High
- Affected: Any use of
LlmAgentwith tools insideParallelAgent(at least in my testing) - Workaround: Use
SequentialAgentinstead ofParallelAgent(loses parallel execution benefits)
Tested Fix
Tested the proposed fix.
Related
This may be related to the resumability feature (ctx.is_resumable) and how events are tracked during parallel execution?