frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[RFC] allow cancelation after shutdown in blocking collective_rpc (#23390)

Browse Source 
Signed-off-by: Shiyan Deng <dsy842974287@meta.com>
This commit is contained in:
Shiyan Deng
2025-09-05 14:14:18 -07:00
committed by GitHub
parent eedb2a2a10
commit 9dfbeb41e5
1 changed files with 8 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
vllm/v1/executor/multiproc_executor.py
View File

@ -253,7 +253,8 @@ class MultiprocExecutor(Executor):
if not non_block:
result = result.result()
elif not non_block:
result = get_response(w, dequeue_timeout)
result = get_response(w, dequeue_timeout,
self.shutdown_event)
else:
raise RuntimeError("non_block can only be used when"
" max_concurrent_batches > 1")
@ -295,12 +296,8 @@ class MultiprocExecutor(Executor):
"""Properly shut down the executor and its workers"""
if not getattr(self, 'shutting_down', False):
self.shutting_down = True
self.shutdown_event.set()
if self.io_thread_pool is not None:
self.io_thread_pool.shutdown(wait=False, cancel_futures=True)
self.io_thread_pool = None
# Make sure all the worker processes are terminated first.
if workers := getattr(self, 'workers', None):
for w in workers:
# Close death_writer to signal child processes to exit
@ -310,6 +307,11 @@ class MultiprocExecutor(Executor):
w.worker_response_mq = None
self._ensure_worker_termination([w.proc for w in workers])
self.shutdown_event.set()
if self.io_thread_pool is not None:
self.io_thread_pool.shutdown(wait=False, cancel_futures=True)
del self.io_thread_pool
self.rpc_broadcast_mq = None
def check_health(self) -> None: