[Core] add clean up method for diffusion engine (#219)

Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
This commit is contained in:
Jiangyun Zhu authored and GitHub committed 2025-12-08 03:52:07 +00:00
1 parent 02b1b36c6d
commit aa689e6af0
13 files changed
+309 -10

No files matched your search

+25 -1
View File
@@ -15,7 +15,11 @@ from vllm.distributed.parallel_state import (
from vllm.logger import init_logger
from vllm.utils import DeviceMemoryProfiler, GiB_bytes
from vllm_omni.diffusion.data import DiffusionOutput, OmniDiffusionConfig
from vllm_omni.diffusion.data import (
SHUTDOWN_MESSAGE,
DiffusionOutput,
OmniDiffusionConfig,
)
from vllm_omni.diffusion.model_loader.diffusers_loader import DiffusersPipelineLoader
from vllm_omni.diffusion.request import OmniDiffusionRequest
@@ -91,6 +95,14 @@ class GPUWorker:
output = self.pipeline.forward(req)
return output
def shutdown(self) -> None:
if torch.distributed.is_initialized():
try:
torch.distributed.destroy_process_group()
logger.info("Worker %s: Destroyed process group", self.rank)
except Exception as exc: # pragma: no cover - best effort cleanup
logger.warning("Worker %s: Failed to destroy process group: %s", self.rank, exc)
class WorkerProc:
"""Wrapper that runs one Worker in a separate process."""
@@ -162,6 +174,14 @@ class WorkerProc:
)
continue
if reqs == SHUTDOWN_MESSAGE:
logger.info("Worker %s: Received shutdown message", self.gpu_id)
self._running = False
continue
if reqs is None:
logger.warning("Worker %s: Received empty payload, ignoring", self.gpu_id)
continue
# 2: execute, make sure a reply is always sent
try:
output = self.worker.execute_model(reqs, self.od_config)
@@ -180,6 +200,10 @@ class WorkerProc:
continue
logger.info("event loop terminated.")
try:
self.worker.shutdown()
except Exception as exc: # pragma: no cover - best effort cleanup
logger.warning("Worker %s: Shutdown encountered an error: %s", self.gpu_id, exc)
# if self.result_sender is not None:
# self.result_sender.close()
self.context.term()