fix comment

llumnix/arg_utils.py

@@ -25,6 +25,8 @@
 from llumnix.entrypoints.utils import LaunchMode
 
 
+# All the default values of llumnix arguments are set in default.py. So all the arguments here are set to None for default.
+
 class LlumnixArgumentParser(argparse.ArgumentParser):
     def __init__(self, *args, **kwargs):
         self.cur_namespace = "llumnix"

llumnix/launcher.py

@@ -152,22 +152,22 @@ def _get_next_instance_args(self, instance_args) -> InstanceArgs:
         assert not self.enablde_engine_pd_disagg, \
             "Currently not support engine based pd-disaggregation in global launch mode."
 
-        config: InstanceArgs = copy.deepcopy(instance_args)
+        next_instance_args: InstanceArgs = copy.deepcopy(instance_args)
         cur_num_prefill = len(self.global_scheduler.dispatch_scheduler.available_dispatch_instance_set)
         cur_num_decode = len(self.global_scheduler.instance_id_set -
                                 self.global_scheduler.dispatch_scheduler.available_dispatch_instance_set)
-        config.instance_type = self._get_next_instance_type(cur_num_prefill, cur_num_decode, self.pd_ratio)
-        return config
+        next_instance_args.instance_type = self._get_next_instance_type(cur_num_prefill, cur_num_decode, self.pd_ratio)
+        return next_instance_args
 
     def _get_next_entrypoints_args(self, entrypoints_args: EntrypointsArgs) -> EntrypointsArgs:
-        config = copy.deepcopy(entrypoints_args)
+        next_entrypoints_args = copy.deepcopy(entrypoints_args)
         if self.enable_port_increment:
-            config.port += self.port_offset
-            config.request_output_queue_port += self.port_offset
+            next_entrypoints_args.port += self.port_offset
+            next_entrypoints_args.request_output_queue_port += self.port_offset
             self.port_offset += 1
             if self.enable_port_offset_store:
                 put_actor_data_to_ray_internal_kv("manager", "port_offset", self.port_offset)
-        return config
+        return next_entrypoints_args
 
     def init_server_and_instance(self, instance_id: str, entrypoints_args: EntrypointsArgs,
                                  instance_args: InstanceArgs, engine_args, backend_type: BackendType,

tests/unit_test/global_scheduler/test_manager.py

@@ -397,7 +397,7 @@ async def test_init_server_and_get_instance_deployment_states_and_instance_and_c
 @pytest.mark.parametrize("request_output_queue_type", ['rayqueue', 'zmq'])
 async def test_auto_scale_up_loop_and_get_cluster_deployment(ray_env, request_output_queue_type):
     manager, _, _, _, _ = init_manager_with_launch_mode(LaunchMode.GLOBAL, request_output_queue_type)
-    await asyncio.sleep(30.0)
+    await asyncio.sleep(60.0)
 
     num_instances = manager.scale_up([], [], [])
     assert num_instances == 4
@@ -410,7 +410,7 @@ async def test_auto_scale_up_loop_and_get_cluster_deployment(ray_env, request_ou
     assert len(instance_ids) == 4
     await manager.clear_instance_ray_resources(instance_ids[0])
     await manager.clear_instance_ray_resources(instance_ids[1])
-    await asyncio.sleep(30.0)
+    await asyncio.sleep(60.0)
 
     num_instances = manager.scale_up([], [], [])
     assert num_instances == 4
@@ -421,7 +421,7 @@ async def test_auto_scale_up_loop_and_get_cluster_deployment(ray_env, request_ou
 @pytest.mark.parametrize("request_output_queue_type", ['rayqueue', 'zmq'])
 async def test_check_deployment_states_loop_and_auto_scale_up_loop(ray_env, request_output_queue_type):
     manager, _, _, _, _ = init_manager_with_launch_mode(LaunchMode.GLOBAL, request_output_queue_type)
-    await asyncio.sleep(30.0)
+    await asyncio.sleep(60.0)
 
     num_instances = manager.scale_up([], [], [])
     assert num_instances == 4
@@ -466,7 +466,7 @@ def test_pd_disagg_gloal_launch_instance_type():
 async def test_pd_disagg_gloal_launch_deployment_and_auto_scale_up_loop(ray_env, request_output_queue_type):
     manager, _, _, _, _ = init_manager_with_launch_mode(LaunchMode.GLOBAL, request_output_queue_type,
                                                         enable_pd_disagg=True, pd_ratio="1:1")
-    await asyncio.sleep(30.0)
+    await asyncio.sleep(60.0)
 
     num_instances = manager.scale_up([], [], [])
     assert num_instances == 4