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 7, 07:45 UTC
[Bugfix] Improve engine ready timeout error message (#35616)
Signed-off-by: damaozi <1811866786@qq.com>
lailoo committed
Mar 4, 2026 at 13:54 UTC
097eb544e9a22810c9b7a59e586b61627b308362
1 file changed
+13
-4
vllm/v1/engine/core_client.py
+13
-4
@@ -609,8 +609,13 @@ class MPClient(EngineCoreClient):
609
timeout=VLLM_ENGINE_READY_TIMEOUT_S * 1000 # convert to ms
610
):
611
raise TimeoutError(
612
- "Timed out waiting for engines to send "
613
- "initial message on input socket."
612
+ f"Timed out waiting for engine core processes to "
613
+ f"start. This is often caused by slow weight loading "
614
+ f"for large models. Waited "
615
+ f"{VLLM_ENGINE_READY_TIMEOUT_S}s (configured by "
616
+ f"VLLM_ENGINE_READY_TIMEOUT_S). To increase the "
617
+ f"timeout, set the environment variable: "
618
+ f"VLLM_ENGINE_READY_TIMEOUT_S=<seconds>"
619
)
620
identity, _ = sync_input_socket.recv_multipart()
621
identities.remove(identity)
@@ -1586,8 +1591,12 @@ class DPLBAsyncMPClient(DPAsyncMPClient):
1591
timeout=VLLM_ENGINE_READY_TIMEOUT_S * 1000 # convert to ms
1592
):
1593
raise TimeoutError(
1589
- "Timed out waiting for new engines to send initial "
1590
- "message on input socket."
1594
+ f"Timed out waiting for new engine core processes to "
1595
+ f"start. Waited "
1596
+ f"{VLLM_ENGINE_READY_TIMEOUT_S}s (configured by "
1597
+ f"VLLM_ENGINE_READY_TIMEOUT_S). To increase the "
1598
+ f"timeout, set the environment variable: "
1599
+ f"VLLM_ENGINE_READY_TIMEOUT_S=<seconds>"
1600
)
1601
identity, _ = sync_input_socket.recv_multipart()
1602
new_engine_identities.discard(identity)