From d37bfcb26bc6ceaacf2fd497bdf8b051d5bfbdc0 Mon Sep 17 00:00:00 2001 From: jack Date: Fri, 28 Aug 2026 04:24:48 +0800 Subject: [PATCH] =?UTF-8?q?=E5=BF=83=E8=B7=B3=E7=9A=84=E4=BC=98=E5=8C=96?= =?UTF-8?q?=E5=BB=BA=E8=AE=AE=E6=94=B9=E7=9C=8B=E7=BB=9D=E5=AF=B9=E9=87=8F?= =?UTF-8?q?=E7=BA=A7?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit lean + 新引擎后纯计算约 128ms，仍固定提示「需改增量计算」是误导：尾部已由 数据腿主导（lag_data P90 1482ms vs compute P90 237ms），压计算换不到东西。 Co-authored-by: Cursor --- research/live/shadow_hb.py | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) diff --git a/research/live/shadow_hb.py b/research/live/shadow_hb.py index 0a39e28..5a45359 100644 --- a/research/live/shadow_hb.py +++ b/research/live/shadow_hb.py @@ -778,9 +778,14 @@ class Shadow: flush=True) if self.q_hist and self.i_hist: q, i = float(np.median(self.q_hist)), float(np.median(self.i_hist)) - verdict = ("排队为主 → 加 worker/加核直接见效" - if q > i else - "纯计算为主 → 加核帮不上，需改增量计算") + # 建议要看绝对量级：lean + 新引擎后纯计算约 128ms，此时再提 + # 「改增量计算」是误导——尾部已由数据腿主导，压计算换不到东西 + if q > i: + verdict = "排队为主 → 加 worker/加核直接见效" + elif i > 400: + verdict = "纯计算为主且偏高 → 加核帮不上，需改增量计算" + else: + verdict = "纯计算为主但量级已低 → 无需再优化" print(f" [计算] 排队中位 {q:.0f}ms · 纯计算中位 {i:.0f}ms" f" · {verdict}（worker {self.workers} 个 / 币 {len(SYMS)} 个）", flush=True)