diff --git a/llumnix/backends/vllm/llm_engine.py b/llumnix/backends/vllm/llm_engine.py
index 46164fa0..0892b4a7 100644
--- a/llumnix/backends/vllm/llm_engine.py
+++ b/llumnix/backends/vllm/llm_engine.py
@@ -249,7 +249,7 @@ def __init__(
         self._thread.start()
 
     def _start_engine_loop(self) -> None:
-        self._stop_event.clear() 
+        self._stop_event.clear()
 
         with self.state_lock:
             self.state = EngineState.RUNNING
diff --git a/llumnix/llumlet/llumlet.py b/llumnix/llumlet/llumlet.py
index 2da4f93e..08bda163 100644
--- a/llumnix/llumlet/llumlet.py
+++ b/llumnix/llumlet/llumlet.py
@@ -109,6 +109,7 @@ def check_state(self):
 
             with self.backend_engine.state_lock:
                 if self.backend_engine.state == EngineState.CRASHED:
+                    # pylint: disable=protected-access
                     self.backend_engine._stop_event.set()
                     if self.backend_engine._thread.is_alive():
                         self.backend_engine._thread.join()
diff --git a/tests/unit_test/llumlet/test_engine_step_exception.py b/tests/unit_test/llumlet/test_engine_step_exception.py
index f395ed89..f96100c1 100644
--- a/tests/unit_test/llumlet/test_engine_step_exception.py
+++ b/tests/unit_test/llumlet/test_engine_step_exception.py
@@ -39,14 +39,14 @@ def set_error_step(self, broken: bool):
             self.backend_engine._thread.join()
 
         def raise_error_step():
-            self.backend_engine.engine.step()
+            self.origin_step()
             raise ValueError("Mock engine step error")
 
         if broken:
             self.backend_engine.engine.step = raise_error_step
         else:
             self.backend_engine.engine.step = self.origin_step
-        
+
         self.backend_engine._thread = threading.Thread(
             target=self.backend_engine._start_engine_loop, args=(), daemon=True, name="engine_loop"
         )
@@ -58,7 +58,7 @@ def test_engine_step_exception(setup_ray_env):
     migration_config = MigrationConfig("LCFS", "rpc", 16, 1, 4, 5, 20)
     node_id = ray.get_runtime_context().get_node_id()
     scheduling_strategy = NodeAffinitySchedulingStrategy(node_id=node_id, soft=False)
-    
+
     origin_free_memory, _ = torch.cuda.mem_get_info()
 
     actor_name = "instance_0"