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 2, 14:45 UTC
[UX] Increase DP Coordinator startup timeout from 30s to 120s (#42343)
Signed-off-by: wzhao18 <wzhao18.sz@gmail.com>
Wei Zhao committed
May 28, 2026 at 06:31 UTC
f2caefe226e4c3283d1140aa16125748d1130027
1 file changed
+3
-2
vllm/v1/engine/coordinator.py
+3
-2
@@ -58,13 +58,14 @@ class DPCoordinator:
58
59
def _wait_for_zmq_addrs(self, zmq_addr_pipe) -> tuple[str, str, str]:
60
try:
61
+ timeout = 120
62
ready = multiprocessing.connection.wait(
62
- [zmq_addr_pipe, self.proc.sentinel], timeout=30
63
+ [zmq_addr_pipe, self.proc.sentinel], timeout=timeout
64
)
65
if not ready:
66
raise RuntimeError(
67
"DP Coordinator process failed to report ZMQ addresses "
67
- "during startup."
68
+ f"within timeout={timeout} seconds during startup."
69
)
70
try:
71
return zmq_addr_pipe.recv()