[misc][distributed] use 127.0.0.1 for single-node (vllm-project#5619)

Author: youkaichao

vllm/executor/multiproc_gpu_executor.py

@@ -10,7 +10,7 @@
 from vllm.logger import init_logger
 from vllm.sequence import ExecuteModelRequest, SamplerOutput
 from vllm.utils import (cuda_device_count_stateless,
-                        get_distributed_init_method, get_ip, get_open_port,
+                        get_distributed_init_method, get_open_port,
                         get_vllm_instance_id, make_async)
 
 logger = init_logger(__name__)
@@ -37,8 +37,11 @@ def _init_executor(self) -> None:
         assert world_size <= cuda_device_count_stateless(), (
             "please set tensor_parallel_size to less than max local gpu count")
 
+        # Multiprocessing-based executor does not support multi-node setting.
+        # Since it only works for single node, we can use the loopback address
+        # 127.0.0.1 for communication.
         distributed_init_method = get_distributed_init_method(
-            get_ip(), get_open_port())
+            "127.0.0.1", get_open_port())
 
         if world_size == 1:
             self.workers = []

vllm/executor/ray_gpu_executor.py

@@ -161,6 +161,16 @@ def _init_workers_ray(self, placement_group: "PlacementGroup",
         self._run_workers("update_environment_variables",
                           all_args=all_args_to_update_environment_variables)
 
+        if len(node_gpus) == 1:
+            # in single node case, we don't need to get the IP address.
+            # the loopback address is sufficient
+            # NOTE: a node may have several IP addresses, one for each
+            # network interface. `get_ip()` might return any of them,
+            # while they might not work for communication inside the node
+            # if the network setup is complicated. Using the loopback address
+            # solves this issue, as it always works for communication inside
+            # the node.
+            driver_ip = "127.0.0.1"
         distributed_init_method = get_distributed_init_method(
             driver_ip, get_open_port())