xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 03:26:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BugFix] Remove default multiproc executor collective_rpc timeout (#17000)

Browse Source 
Signed-off-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
Nick Hill 2025-04-22 16:27:14 -07:00 committed by GitHub
parent 5536b30a4c
commit 5175b884f7
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -38,7 +38,7 @@ logger = init_logger(__name__)
POLLING_TIMEOUT_MS = 5000
POLLING_TIMEOUT_S = POLLING_TIMEOUT_MS // 1000
EXECUTE_MODEL_TIMEOUT_S = 30
EXECUTE_MODEL_TIMEOUT_S = 40
class MultiprocExecutor(Executor):
@ -151,16 +151,16 @@ class MultiprocExecutor(Executor):
def collective_rpc(self,
method: Union[str, Callable],
timeout: Optional[float] = 180.0,
timeout: Optional[float] = None,
args: tuple = (),
kwargs: Optional[dict] = None,
rank0_reply_only: bool = False) -> list[Any]:
start_time = time.monotonic()
kwargs = kwargs or {}
if self.is_failed:
raise RuntimeError("Executor failed.")
deadline = None if timeout is None else time.monotonic() + timeout
kwargs = kwargs or {}
# NOTE: If the args are heterogeneous, then we pack them into a list,
# and unpack them in the method of every worker, because every worker
# knows their own rank.
@ -176,8 +176,8 @@ class MultiprocExecutor(Executor):
workers = (self.workers[0], ) if rank0_reply_only else self.workers
responses = [None] * len(workers)
for w in workers:
dequeue_timeout = timeout - (time.monotonic() - start_time
) if timeout is not None else None
dequeue_timeout = None if deadline is None else (
deadline - time.monotonic())
status, result = w.worker_response_mq.dequeue(
timeout=dequeue_timeout, cancel=self.shutdown_event)