diff --git a/litellm/litellm_core_utils/logging_worker.py b/litellm/litellm_core_utils/logging_worker.py
index cb3d8bf4fe5..2f8e7bdccea 100644
--- a/litellm/litellm_core_utils/logging_worker.py
+++ b/litellm/litellm_core_utils/logging_worker.py
@@ -485,10 +485,8 @@ class LoggingWorker:
         callback hasn't finished yet — ``queue.empty()`` would return True in
         that window and cause us to skip the wait.
 
-        ``start()`` runs first so that, after an event loop change, the tasks
-        still on the previous loop's queue move onto this loop and a worker
-        here drains them; joining the old queue directly would wait on a
-        counter nothing on this loop ever decrements.
+        ``start()`` runs first so a queue left behind by a previous event loop
+        is carried onto this one and drained here instead of joined forever.
         """
         if self._queue is None:
             return
diff --git a/tests/test_litellm/litellm_core_utils/test_logging_worker.py b/tests/test_litellm/litellm_core_utils/test_logging_worker.py
index 891bd0686b2..336067976fa 100644
--- a/tests/test_litellm/litellm_core_utils/test_logging_worker.py
+++ b/tests/test_litellm/litellm_core_utils/test_logging_worker.py
@@ -207,10 +207,29 @@ class TestLoggingWorker:
         assert (worker._queue.qsize(), len(worker._unstarted_dequeued_tasks())) == expected_shape
         assert fired == [], "precondition: the callback never ran before the first loop closed"
 
-        async def flush_on_second_loop():
+        async def flush_twice_on_second_loop():
+            await asyncio.wait_for(worker.flush(), timeout=5)
             await asyncio.wait_for(worker.flush(), timeout=5)
 
-        asyncio.run(flush_on_second_loop())
+        asyncio.run(flush_twice_on_second_loop())
+
+        assert fired == [True]
+
+    def test_flush_starts_a_worker_when_the_queue_has_none(self):
+        """``flush()`` must drain a queue that exists on the current loop without a running worker."""
+        worker = LoggingWorker(timeout=1.0, max_queue_size=10)
+        fired = []
+
+        async def marker():
+            fired.append(True)
+
+        async def enqueue_then_flush():
+            worker._ensure_queue()
+            worker.enqueue(marker())
+            assert worker._worker_task is None, "precondition: nothing is draining the queue yet"
+            await asyncio.wait_for(worker.flush(), timeout=3)
+
+        asyncio.run(enqueue_then_flush())
 
         assert fired == [True]