A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 17, 00:30 UTC
Remove Ray health check (#4693)
Antoni Baum committed
Jun 7, 2024 at 03:01 UTC
18a277b52dd2a64ee4c0111fc8cda126031e5889
1 file changed
-17
vllm/executor/ray_gpu_executor.py
-17
@@ -293,23 +293,6 @@ class RayGPUExecutor(DistributedGPUExecutor):
293
])
294
return forward_dag.experimental_compile()
295
296
- def check_health(self) -> None:
297
- """Raises an error if engine is unhealthy."""
298
- self._check_if_any_actor_is_dead()
299
-
300
- def _check_if_any_actor_is_dead(self):
301
- if not self.workers:
302
- return
303
-
304
- dead_actors = []
305
- for actor in self.workers:
306
- actor_state = ray.state.actors(actor._ray_actor_id.hex()) # pylint: disable=protected-access
307
- if actor_state["State"] == "DEAD":
308
- dead_actors.append(actor)
309
- if dead_actors:
310
- raise RuntimeError("At least one Worker is dead. "
311
- f"Dead Workers: {dead_actors}. ")
312
-
296
297
class RayGPUExecutorAsync(RayGPUExecutor, DistributedGPUExecutorAsync):
298