From 8355358fb3d0709c2be2c48a5f805cd15f6304cd Mon Sep 17 00:00:00 2001 From: clark Date: Tue, 11 Mar 2025 16:17:36 +0800 Subject: [PATCH] add unlimited HWM Signed-off-by: clark --- vllm/entrypoints/disagg_connector.py | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-) diff --git a/vllm/entrypoints/disagg_connector.py b/vllm/entrypoints/disagg_connector.py index ba0b0d748ad89..e7e7e55627b99 100644 --- a/vllm/entrypoints/disagg_connector.py +++ b/vllm/entrypoints/disagg_connector.py @@ -42,12 +42,16 @@ async def log_stats(request_queues: dict[str, asyncio.Queue]): # create async socket use ZMQ_DEALER async def create_socket(url: str, zmqctx: zmq.asyncio.Context) -> zmq.asyncio.Socket: - sock = zmqctx.socket(zmq.DEALER) + socket = zmqctx.socket(zmq.DEALER) identity = f"connector-{uuid.uuid4()}" - sock.setsockopt(zmq.IDENTITY, identity.encode()) - sock.connect(url) + # unlimited HWM + hwm_limit = 0 + socket.setsockopt(zmq.IDENTITY, identity.encode()) + socket.setsockopt(zmq.SNDHWM, hwm_limit) + socket.setsockopt(zmq.RCVHWM, hwm_limit) + socket.connect(url) logger.info("%s started at %s", identity, url) - return sock + return socket @asynccontextmanager