From beb65c7433d6a5b8f72e5498200ee119d35476bf Mon Sep 17 00:00:00 2001 From: Liangsheng Yin Date: Sun, 27 Apr 2025 23:03:30 +0800 Subject: [PATCH] [PD]Reduce kv transfer threads (#5791) --- python/sglang/srt/disaggregation/mooncake/conn.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python/sglang/srt/disaggregation/mooncake/conn.py b/python/sglang/srt/disaggregation/mooncake/conn.py index c2a516c46..330be4ab8 100644 --- a/python/sglang/srt/disaggregation/mooncake/conn.py +++ b/python/sglang/srt/disaggregation/mooncake/conn.py @@ -149,7 +149,7 @@ class MooncakeKVManager(BaseKVManager): # Determine the number of threads to use for kv sender cpu_count = os.cpu_count() self.executor = concurrent.futures.ThreadPoolExecutor( - max_workers=cpu_count if cpu_count is not None else 64 + min(cpu_count // 4, 16) ) elif self.disaggregation_mode == DisaggregationMode.DECODE: self.start_decode_thread()