test: use a more robust mechanism for detect loop blocking

tests/unit/providers/inference/test_remote_vllm.py

@@ -5,6 +5,7 @@
 # the root directory of this source tree.
 
 import asyncio
+import contextlib
 import json
 import logging
 import threading
@@ -182,13 +183,41 @@ async def test_process_vllm_chat_completion_stream_response_no_choices():
     assert len(chunks) == 0
 
 
-def test_chat_completion_doesnt_block_event_loop(caplog):
-    loop = asyncio.new_event_loop()
-    loop.set_debug(True)
-    caplog.set_level(logging.WARNING)
+@contextlib.asynccontextmanager
+async def detect_blocking(threshold=0.1):
+    """Context manager to detect blocking in an event loop."""
+    block_detected = False
+    watchdog_active = True
 
-    # Log when event loop is blocked for more than 200ms
-    loop.slow_callback_duration = 0.2
+    last_check = time.monotonic()
+
+    async def watchdog():
+        nonlocal block_detected, last_check
+        while watchdog_active:
+            now = time.monotonic()
+            # If our check is significantly delayed, we might be blocked
+            if now - last_check > threshold:
+                block_detected = True
+            last_check = now
+            await asyncio.sleep(threshold / 3)
+
+    watchdog_task = asyncio.create_task(watchdog())
+
+    try:
+        yield
+    finally:
+        watchdog_active = False
+        watchdog_task.cancel()
+        try:
+            await watchdog_task
+        except asyncio.CancelledError:
+            pass
+
+        assert not block_detected, f"Event loop was blocked for more than {threshold}s"
+
+
+@pytest.mark.asyncio
+async def test_chat_completion_doesnt_block_event_loop():
     # Sleep for 500ms in our delayed http response
     sleep_time = 0.5
 
@@ -220,15 +249,7 @@ def test_chat_completion_doesnt_block_event_loop(caplog):
     with MockInferenceAdapterWithSleep(sleep_time, mock_response) as inference_adapter:
         inference_adapter.model_store = AsyncMock()
         inference_adapter.model_store.get_model.return_value = mock_model
-        loop.run_until_complete(inference_adapter.initialize())
+        await inference_adapter.initialize()
 
-        # Clear the logs so far and run the actual chat completion we care about
-        caplog.clear()
-        loop.run_until_complete(do_chat_completion())
-
-    # Ensure we don't have any asyncio warnings in the captured log
-    # records from our chat completion call. A message gets logged
-    # here any time we exceed the slow_callback_duration configured
-    # above.
-    asyncio_warnings = [record.message for record in caplog.records if record.name == "asyncio"]
-    assert not asyncio_warnings
+        async with detect_blocking(0.1):
+            await do_chat_completion()