From be03d841f6040351fc1b6df63a8866b4ece1568e Mon Sep 17 00:00:00 2001 From: Robert Shaw Date: Sun, 20 Jul 2025 22:33:17 +0000 Subject: [PATCH] stash Signed-off-by: Robert Shaw --- vllm/v1/engine/core_client.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/vllm/v1/engine/core_client.py b/vllm/v1/engine/core_client.py index 643997e34c8ae..93a7eb8b8a9f1 100644 --- a/vllm/v1/engine/core_client.py +++ b/vllm/v1/engine/core_client.py @@ -90,7 +90,8 @@ class EngineCoreClient(ABC): client_args = (vllm_config, executor_class, log_stats, client_addresses, client_index) if parallel_config.data_parallel_size > 1: - if parallel_config.data_parallel_external_lb: + # if parallel_config.data_parallel_external_lb: + if False: # External load balancer - client per DP rank. return DPAsyncMPClient(*client_args) # Internal load balancer - client balances to all DP ranks. @@ -1023,6 +1024,7 @@ class DPLBAsyncMPClient(DPAsyncMPClient): self, request: EngineCoreRequest) -> EngineIdentity: # Engines are in rank order. if (eng_index := request.data_parallel_rank) is None: + # logger.info(f"{self.lb_engines=} | {self.core_engines=}") if not self.lb_engines: return self.core_engine # TODO use P2C alg for larger DP sizes