Skip to content

Your lightweight open-source personal agent.

test: await busy-session batch completion instead of polling

Xubin Ren4a6ed1a6733a
diff --git a/tests/agent/test_runner_injections.py b/tests/agent/test_runner_injections.py
index 31344270c..fa7404034 100644
--- a/tests/agent/test_runner_injections.py
+++ b/tests/agent/test_runner_injections.py
@@ -1532,7 +1532,7 @@ async def test_session_inbox_is_installed_before_worker_start(tmp_path, monkeypa
@pytest.mark.asyncio
@pytest.mark.parametrize("log_content", [True, False])
async def test_busy_session_burst_reaches_next_model_call_as_one_ordered_batch(
- tmp_path, request, log_content,
+ tmp_path, request, monkeypatch, log_content,
):
"""Messages accumulated during a model call share its next request snapshot."""
from loguru import logger
@@ -1550,7 +1550,7 @@ async def test_busy_session_burst_reaches_next_model_call_as_one_ordered_batch(
provider.get_default_model.return_value = "test-model"
first_request_started = asyncio.Event()
release_first_request = asyncio.Event()
- second_request_started = asyncio.Event()
+ all_enqueued = asyncio.Event()
captured_messages: list[list[dict]] = []
async def chat_stream_with_retry(*, messages, **kwargs):
@@ -1559,7 +1559,6 @@ async def test_busy_session_burst_reaches_next_model_call_as_one_ordered_batch(
first_request_started.set()
await release_first_request.wait()
return LLMResponse(content="first answer", tool_calls=[], usage=None)
- second_request_started.set()
return LLMResponse(content="batch answer", tool_calls=[], usage=None)
provider.chat_stream_with_retry = chat_stream_with_retry
@@ -1573,38 +1572,44 @@ async def test_busy_session_burst_reaches_next_model_call_as_one_ordered_batch(
await loop.sessions.state.get("cli:c")
await loop.sessions.state.set_policy("cli:c", SessionPolicy(log_content=log_content))
followups = [f"follow-up-{index:02}" for index in range(12)]
+ enqueue = loop._enqueue_session_message
+
+ async def capture_enqueue(msg):
+ await enqueue(msg)
+ if msg.content == followups[-1]:
+ all_enqueued.set()
+
+ monkeypatch.setattr(loop, "_enqueue_session_message", capture_enqueue)
run_task = asyncio.create_task(loop.run())
- await loop.bus.publish_inbound(InboundMessage(
- channel="cli",
- sender_id="u",
- chat_id="c",
- content="initial",
- ))
- await asyncio.wait_for(first_request_started.wait(), timeout=2)
- for content in followups:
+ try:
await loop.bus.publish_inbound(InboundMessage(
channel="cli",
sender_id="u",
chat_id="c",
- content=content,
+ content="initial",
))
- for _ in range(200):
- pending = loop._pending_queues.get("cli:c")
- if pending is not None and pending.qsize() == len(followups):
- break
- await asyncio.sleep(0.01)
- assert loop._pending_queues["cli:c"].qsize() == len(followups)
-
- release_first_request.set()
- await asyncio.wait_for(second_request_started.wait(), timeout=2)
- for _ in range(200):
- if "cli:c" not in loop._pending_queues:
- break
- await asyncio.sleep(0.01)
- loop.stop()
- # Shutdown drains accepted writes and closes the worker's WAL connection.
- await asyncio.wait_for(run_task, timeout=30)
+ await asyncio.wait_for(first_request_started.wait(), timeout=30)
+ for content in followups:
+ await loop.bus.publish_inbound(InboundMessage(
+ channel="cli",
+ sender_id="u",
+ chat_id="c",
+ content=content,
+ ))
+ await asyncio.wait_for(all_enqueued.wait(), timeout=30)
+ assert loop._pending_queues["cli:c"].qsize() == len(followups)
+ workers = list(loop._active_tasks["cli:c"])
+
+ release_first_request.set()
+ # Wait for the real worker, including durable writes, not a disk-speed SLA.
+ await asyncio.wait_for(asyncio.gather(*workers), timeout=30)
+ assert "cli:c" not in loop._pending_queues
+ finally:
+ release_first_request.set()
+ loop.stop()
+ # Shutdown drains accepted writes and closes the worker's WAL connection.
+ await asyncio.wait_for(run_task, timeout=30)
assert len(captured_messages) == 2
second_request = "\n".join(