Repository navigation
Runner.run_async runs an LlmAgent with neither new_message nor invocation_id instead of raising ValueError #7934
Description
🔴 Required Information
Describe the Bug:
Runner.run_async documents "Raises: ValueError: ... If both invocation_id and new_message are None.", and _run_with_trace has that check ("Running an agent requires either a new_message or an invocation_id ..."). But when the root agent is an LlmAgent, run_async takes the _run_node_async branch first and returns before reaching the check. So the run starts with no user message and calls the model with empty contents.
Steps to Reproduce:
pip install google-adk==2.13.0- Run the script below. It uses a recording model, so no API key is needed.
Expected Behavior:
ValueError, as documented, before the model is called.
Observed Behavior:
No error; the model is called once with no contents.
Environment Details:
- ADK Library Version (pip show google-adk): 2.13.0, and main at 95d8610
- Desktop OS: Linux
- Python Version (python -V): 3.12
Model Information:
- Are you using LiteLLM: No
- Which model is being used: a
BaseLlmsubclass that records its calls (any model shows it)
🟡 Optional Information
Minimal Reproduction Code:
import asyncio
from typing import AsyncGenerator
from google.adk.agents import LlmAgent
from google.adk.models.base_llm import BaseLlm
from google.adk.models.llm_request import LlmRequest
from google.adk.models.llm_response import LlmResponse
from google.adk.runners import Runner
from google.adk.sessions import InMemorySessionService
from google.genai import types
CALLS = []
class Recorder(BaseLlm):
model: str = "recorder"
async def generate_content_async(self, llm_request: LlmRequest, stream: bool = False) -> AsyncGenerator[LlmResponse, None]:
CALLS.append([c.role for c in llm_request.contents])
yield LlmResponse(content=types.Content(role="model", parts=[types.Part(text="hello")]))
async def main():
runner = Runner(app_name="demo", agent=LlmAgent(name="greeter", model=Recorder()), session_service=InMemorySessionService())
await runner.session_service.create_session(app_name="demo", user_id="u", session_id="s")
try:
events = [event async for event in runner.run_async(user_id="u", session_id="s", new_message=None)]
print("no error;", len(events), "events; model calls:", CALLS)
except ValueError as exc:
print("ValueError:", exc)
asyncio.run(main())Logs:
no error; 1 events; model calls: [[]]
Additional Context:
In runners.py the if isinstance(self.agent, LlmAgent): branch (and the BaseNode one after it) return before _run_with_trace, where the check is. Moving the check before those branches covers them all.
How often has this issue occurred?:
Always (100%)
Activity
mohitdixit02 commentedon Nov 2, 2026
Hi! I was able to reproduce this locally on ADK 2.13.0. In
Runner.run_async, theLlmAgentbranch routes into_run_node_asyncand returns early before reaching thenew_message/invocation_idcheck in_run_with_trace, resulting in calling the model with empty contents instead of raisingValueError.I'd like to work on this and submit a PR with unit tests. Could you please assign this to me?
surajksharma07 commentedon Nov 2, 2026
@yifanxiong272. Reproduced this on 2.13.0 and on current main for both an LlmAgent and a Workflow root while a BaseAgent root still raises. In 1.x every root raised so this looks like it came in with the 2.x node path.
@mohitdixit02 one thing on #7936: a few existing tests run with new_message=None on purpose to re-run over the session history (test_llm_agent_include_contents.py and two in test_compaction_runner_e2e.py) and they fail with the check moved to the top as-is.
As a workaround, raising only when there’s no new_message, no invocation_id and the session has no events kept those tests green and still raised for the empty case. Could you both check if that works on your side?
Before #7936 goes for review please give it a full run of the runner, include_contents and compaction tests and add a case for a session that already has history.
fix(runner): permit run_async without new_message when session has hi…
mohitdixit02 commentedon Nov 3, 2026
@surajksharma07
Checked on my end and that works cleanly.
I've updated the check in #7936 to inspect
not session.eventsbefore raising whennew_messageandinvocation_idare absent.Full runs of
test_runners.py,test_compaction_runner_e2e.py, andtest_llm_agent_include_contents.pyare all fine, and I've added a unit test coveringrun_asyncexecution over existing session history withnew_message=None.Pushed the update to the PR branch!