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 15, 09:45 UTC
[CI] Fix race condition in test_kv_cache_events test (#18169)
Signed-off-by: Russell Bryant <rbryant@redhat.com>
Russell Bryant committed
May 14, 2025 at 19:27 UTC
78aa341d124e4e2162defdabde8e8b0a97ffb79d
2 files changed
+1
-2
tests/v1/engine/test_engine_core_client.py
-1
@@ -292,7 +292,6 @@ def test_kv_cache_events(
292
log_stats=False,
293
)
294
endpoint = publisher_config.endpoint.replace("*", "127.0.0.1")
295
- time.sleep(0.1)
295
subscriber = MockSubscriber(endpoint,
296
topic=publisher_config.topic,
297
decode_type=KVEventBatch)
vllm/distributed/kv_events.py
+1
-1
@@ -130,6 +130,7 @@ class ZmqEventPublisher(EventPublisher):
130
self._endpoint = endpoint
131
self._replay_endpoint = replay_endpoint
132
self._hwm = hwm
133
+ self._socket_setup()
134
135
# Payload
136
self._seq_gen = count()
@@ -207,7 +208,6 @@ class ZmqEventPublisher(EventPublisher):
208
def _publisher_thread(self) -> None:
209
"""Background thread that processes the event queue."""
210
self._pack = msgspec.msgpack.Encoder()
210
- self._socket_setup()
211
212
assert self._pub is not None # narrows type for mypy
213