From 18a277b52dd2a64ee4c0111fc8cda126031e5889 Mon Sep 17 00:00:00 2001 From: Antoni Baum Date: Fri, 7 Jun 2024 03:01:56 -0700 Subject: [PATCH] Remove Ray health check (#4693) --- vllm/executor/ray_gpu_executor.py | 17 ----------------- 1 file changed, 17 deletions(-) diff --git a/vllm/executor/ray_gpu_executor.py b/vllm/executor/ray_gpu_executor.py index bed356d1..89d1c4ac 100644 --- a/vllm/executor/ray_gpu_executor.py +++ b/vllm/executor/ray_gpu_executor.py @@ -293,23 +293,6 @@ class RayGPUExecutor(DistributedGPUExecutor): ]) return forward_dag.experimental_compile() - def check_health(self) -> None: - """Raises an error if engine is unhealthy.""" - self._check_if_any_actor_is_dead() - - def _check_if_any_actor_is_dead(self): - if not self.workers: - return - - dead_actors = [] - for actor in self.workers: - actor_state = ray.state.actors(actor._ray_actor_id.hex()) # pylint: disable=protected-access - if actor_state["State"] == "DEAD": - dead_actors.append(actor) - if dead_actors: - raise RuntimeError("At least one Worker is dead. " - f"Dead Workers: {dead_actors}. ") - class RayGPUExecutorAsync(RayGPUExecutor, DistributedGPUExecutorAsync):