diff --git a/tests/async_engine/test_api_server.py b/tests/async_engine/test_api_server.py
index 248bfbc8ab5c..7f57d5cf9b18 100644
--- a/tests/async_engine/test_api_server.py
+++ b/tests/async_engine/test_api_server.py
@@ -25,21 +25,30 @@ def _query_server_long(prompt: str) -> dict:
 
 
 @pytest.fixture
-def api_server(tokenizer_pool_size: int):
+def api_server(tokenizer_pool_size: int, engine_use_ray: bool,
+               worker_use_ray: bool):
     script_path = Path(__file__).parent.joinpath(
         "api_server_async_engine.py").absolute()
-    uvicorn_process = subprocess.Popen([
+    commands = [
         sys.executable, "-u",
         str(script_path), "--model", "facebook/opt-125m", "--host",
         "127.0.0.1", "--tokenizer-pool-size",
         str(tokenizer_pool_size)
-    ])
+    ]
+    if engine_use_ray:
+        commands.append("--engine-use-ray")
+    if worker_use_ray:
+        commands.append("--worker-use-ray")
+    uvicorn_process = subprocess.Popen(commands)
     yield
     uvicorn_process.terminate()
 
 
 @pytest.mark.parametrize("tokenizer_pool_size", [0, 2])
-def test_api_server(api_server, tokenizer_pool_size: int):
+@pytest.mark.parametrize("worker_use_ray", [False, True])
+@pytest.mark.parametrize("engine_use_ray", [False, True])
+def test_api_server(api_server, tokenizer_pool_size: int, worker_use_ray: bool,
+                    engine_use_ray: bool):
     """
     Run the API server and test it.
 
diff --git a/vllm/engine/async_llm_engine.py b/vllm/engine/async_llm_engine.py
index f61049513512..1dbf58904541 100644
--- a/vllm/engine/async_llm_engine.py
+++ b/vllm/engine/async_llm_engine.py
@@ -333,8 +333,7 @@ class AsyncLLMEngine:
         if engine_config.device_config.device_type == "neuron":
             raise NotImplementedError("Neuron is not supported for "
                                       "async engine yet.")
-        elif (engine_config.parallel_config.worker_use_ray
-              or engine_args.engine_use_ray):
+        elif engine_config.parallel_config.worker_use_ray:
             initialize_ray_cluster(engine_config.parallel_config)
             from vllm.executor.ray_gpu_executor import RayGPUExecutorAsync
             executor_class = RayGPUExecutorAsync
@@ -410,8 +409,8 @@ class AsyncLLMEngine:
         else:
             # FIXME(woosuk): This is a bit hacky. Be careful when changing the
             # order of the arguments.
-            cache_config = args[1]
-            parallel_config = args[2]
+            cache_config = kwargs["cache_config"]
+            parallel_config = kwargs["parallel_config"]
             if parallel_config.tensor_parallel_size == 1:
                 num_gpus = cache_config.gpu_memory_utilization
             else: