biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[V1][BugFix] Exit properly if engine core fails during startup (#16137)

Browse Source 
Signed-off-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
Nick Hill
2025-04-07 15:30:15 -07:00
committed by GitHub
parent 3147586ebd
commit 7f6d47c1a2
5 changed files with 67 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
vllm/v1/engine/core_client.py
View File

@@ -411,10 +411,21 @@ class MPClient(EngineCoreClient):
# Wait for engine core process(es) to send ready messages.
identities = set(eng.index for eng in self.resources.core_engines)
poller = zmq.Poller()
poller.register(sync_input_socket, zmq.POLLIN)
for eng in self.resources.core_engines:
poller.register(eng.proc_handle, zmq.POLLIN)
while identities:
while not sync_input_socket.poll(timeout=STARTUP_POLL_PERIOD_MS):
logger.info("Waiting for %d core engine proc(s) to start: %s",
len(identities), identities)
events = poller.poll(STARTUP_POLL_PERIOD_MS)
if not events:
logger.debug("Waiting for %d core engine proc(s) to start: %s",
len(identities), identities)
continue
if len(events) > 1 or events[0][0] != sync_input_socket:
# One of the core processes exited.
raise RuntimeError("Engine core initialization failed. "
"See root cause above.")
eng_id_bytes, msg = sync_input_socket.recv_multipart()
eng_id = int.from_bytes(eng_id_bytes, byteorder="little")
if eng_id not in identities:
@@ -424,12 +435,6 @@ class MPClient(EngineCoreClient):
logger.info("Core engine process %d ready.", eng_id)
identities.discard(eng_id)
# Double check that the process are running.
for engine in self.resources.core_engines:
proc = engine.proc_handle.proc
if proc.exitcode is not None:
raise RuntimeError(f"Engine proc {proc.name} not running")
def _init_core_engines(
self,
vllm_config: VllmConfig,