[Misc] allow pulling vllm in Ray runtime environment

vllm/config.py

@@ -56,6 +56,7 @@
 
 if TYPE_CHECKING:
     from _typeshed import DataclassInstance
+    from ray.runtime_env import RuntimeEnv
     from ray.util.placement_group import PlacementGroup
     from transformers.configuration_utils import PretrainedConfig
 
@@ -73,6 +74,7 @@
 else:
     DataclassInstance = Any
     PlacementGroup = Any
+    RuntimeEnv = Any
     PretrainedConfig = Any
     ExecutorBase = Any
     QuantizationConfig = Any
@@ -1902,6 +1904,9 @@ class ParallelConfig:
     placement_group: Optional["PlacementGroup"] = None
     """ray distributed model workers placement group."""
 
+    runtime_env: Optional["RuntimeEnv"] = None
+    """ray runtime environment for distributed workers"""
+
     distributed_executor_backend: Optional[Union[DistributedExecutorBackend,
                                                  type["ExecutorBase"]]] = None
     """Backend to use for distributed model

vllm/engine/arg_utils.py

@@ -1088,12 +1088,14 @@ def create_engine_config(
         # we are in a Ray actor. If so, then the placement group will be
         # passed to spawned processes.
         placement_group = None
+        runtime_env = None
         if is_in_ray_actor():
             import ray
 
             # This call initializes Ray automatically if it is not initialized,
             # but we should not do this here.
             placement_group = ray.util.get_current_placement_group()
+            runtime_env = ray.get_runtime_context().runtime_env
 
         data_parallel_external_lb = self.data_parallel_rank is not None
         if data_parallel_external_lb:
@@ -1170,6 +1172,7 @@ def create_engine_config(
             disable_custom_all_reduce=self.disable_custom_all_reduce,
             ray_workers_use_nsight=self.ray_workers_use_nsight,
             placement_group=placement_group,
+            runtime_env=runtime_env,
             distributed_executor_backend=self.distributed_executor_backend,
             worker_cls=self.worker_cls,
             worker_extension_cls=self.worker_extension_cls,

vllm/executor/ray_utils.py

@@ -288,14 +288,16 @@ def initialize_ray_cluster(
     elif current_platform.is_rocm() or current_platform.is_xpu():
         # Try to connect existing ray instance and create a new one if not found
         try:
-            ray.init("auto")
+            ray.init("auto", runtime_env=parallel_config.runtime_env)
         except ConnectionError:
             logger.warning(
                 "No existing RAY instance detected. "
                 "A new instance will be launched with current node resources.")
-            ray.init(address=ray_address, num_gpus=parallel_config.world_size)
+            ray.init(address=ray_address,
+                     num_gpus=parallel_config.world_size,
+                     runtime_env=parallel_config.runtime_env)
     else:
-        ray.init(address=ray_address)
+        ray.init(address=ray_address, runtime_env=parallel_config.runtime_env)
 
     device_str = current_platform.ray_device_key
     if not device_str: