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 14, 05:30 UTC
[Misc] Remove unnecessary `_send_reconfig_message()` in `core_client.py` (#23127)
Signed-off-by: Nick Hill <nhill@redhat.com>
Nick Hill committed
Aug 27, 2025 at 05:50 UTC
a403d0fa41cc68e3b6da4e1097dc896fde2f1a6a
1 file changed
+9
-22
vllm/v1/engine/core_client.py
+9
-22
@@ -1190,21 +1190,6 @@ class DPLBAsyncMPClient(DPAsyncMPClient):
1190
await self._send_input(EngineCoreRequestType.ABORT, request_ids,
1191
engine)
1192
1193
- async def _send_reconfig_message(
1194
- self, reconfig_request: ReconfigureDistributedRequest,
1195
- engine: EngineIdentity) -> asyncio.Future:
1196
- """Send reconfiguration message and return the result future without
1197
- waiting for completion."""
1198
- call_id = uuid.uuid1().int >> 64
1199
- future = asyncio.get_running_loop().create_future()
1200
- self.utility_results[call_id] = future
1201
- message = (EngineCoreRequestType.UTILITY.value, *self.encoder.encode(
1202
- (self.client_index, call_id, "reinitialize_distributed",
1203
- (reconfig_request, ))))
1204
- await self._send_input_message(message, engine, reconfig_request)
1205
- self._ensure_output_queue_task()
1206
- return future
1207
-
1193
async def scale_elastic_ep(self, new_data_parallel_size: int) -> None:
1194
"""Scale elastic EP data parallel size"""
1195
cur_data_parallel_size = len(self.core_engines)
@@ -1214,7 +1199,7 @@ class DPLBAsyncMPClient(DPAsyncMPClient):
1199
f"different from cur_data_parallel_size {cur_data_parallel_size}")
1200
1201
assert self.vllm_config.parallel_config.data_parallel_backend == \
1217
- "ray", ("Only ray DP backend supports scaling elastic EP")
1202
+ "ray", "Only ray DP backend supports scaling elastic EP"
1203
1204
scale_up = new_data_parallel_size > cur_data_parallel_size
1205
@@ -1246,9 +1231,10 @@ class DPLBAsyncMPClient(DPAsyncMPClient):
1231
data_parallel_master_ip,
1232
new_data_parallel_master_port=self.vllm_config.parallel_config.
1233
data_parallel_master_port)
1249
- future = await self._send_reconfig_message(reconfig_request,
1250
- engine)
1251
- reconfig_futures.append(future)
1234
+ coro = self._call_utility_async("reinitialize_distributed",
1235
+ reconfig_request,
1236
+ engine=engine)
1237
+ reconfig_futures.append(asyncio.create_task(coro))
1238
1239
logger.info("All reconfigure messages sent, starting engine creation")
1240
@@ -1318,9 +1304,10 @@ class DPLBAsyncMPClient(DPAsyncMPClient):
1304
if cur_dp_rank >= new_data_parallel_size:
1305
reconfig_request.new_data_parallel_rank = \
1306
ReconfigureRankType.SHUTDOWN_CURRENT_RANK
1321
- future = await self._send_reconfig_message(reconfig_request,
1322
- engine)
1323
- reconfig_futures.append(future)
1307
+ coro = self._call_utility_async("reinitialize_distributed",
1308
+ reconfig_request,
1309
+ engine=engine)
1310
+ reconfig_futures.append(asyncio.create_task(coro))
1311
1312
for _ in range(new_data_parallel_size, cur_data_parallel_size):
1313
self.core_engines.pop()