From c7acd449f1fa35d4421ad7905256df2355fbfeb4 Mon Sep 17 00:00:00 2001
From: yumoqing <yumoqing@gmail.com>
Date: Sun, 24 May 2026 17:21:22 +0800
Subject: [PATCH] =?UTF-8?q?feat(llmage):=20=E6=97=A5=E6=9C=9F=E5=8F=98?=
 =?UTF-8?q?=E6=9B=B4=E8=A7=A6=E5=8F=91=E5=A4=87=E4=BB=BD(use=5Fdate<?=
 =?UTF-8?q?=E6=98=A8=E5=A4=A9)=20+=20llmusage=E6=B7=BB=E5=8A=A0(accounting?=
 =?UTF-8?q?=5Fstatus,use=5Fdate)=E7=BB=84=E5=90=88=E7=B4=A2=E5=BC=95?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 dat/qwen3.7-max.txt                           |  53 ++++++++++++++++++
 llmage/__pycache__/accounting.cpython-310.pyc | Bin 0 -> 9604 bytes
 llmage/accounting.py                          |  31 +++++-----
 models/llmusage.json                          |   8 +++
 scripts/migrate_llmusage_history.sql          |   3 +
 5 files changed, 79 insertions(+), 16 deletions(-)
 create mode 100644 dat/qwen3.7-max.txt
 create mode 100644 llmage/__pycache__/accounting.cpython-310.pyc

diff --git a/dat/qwen3.7-max.txt b/dat/qwen3.7-max.txt
new file mode 100644
index 0000000..f07f135
--- /dev/null
+++ b/dat/qwen3.7-max.txt
@@ -0,0 +1,53 @@
+    已生成完整的 qwen3.7-max 配置SQL。以下是配置方案：
+    
+    模型摘要
+    - 模型名称: 千问3.7-Max
+    - API model: qwen3.7-max (严格按API文档填写)
+    - 分类: text2text (文生文)
+    - 供应商: 阿里百炼 (ali-qwen)
+    - 接口: OpenAI兼容 /chat/completions，同步流式
+    
+    复用vs新建
+    - upapp: 复用 ali-qwen (阿里百炼)
+    - upappkey: 复用 gCMl-BnrTrfoqWAEPPoH8 (API Key已配置)
+    - uapi: 复用 t2t (BY2cA4RD_axNxTG4m9Coa)
+    - uapiio: 复用 Is8l4TGkcZcqFSjbbeIK2 (文本会话)
+    - llm: 新建
+    - llm_api_map: 新建
+    - pricing_program: 新建
+    - pricing_program_timing: 新建
+    
+    生成的SQL (4条)
+    
+    sql
+    -- 1. llm表
+    INSERT INTO llm VALUES ('Jt26VKlUFsJxABuVLauBc','千问3.7-Max','qwen3.7-max','通义千问3.7系列旗舰模型，相较3.6版本在推理能力、代码能力、多模态理解等方面全面升级。适用于复杂任务处理、智能体开发、代码生成等高级应用场景。','text2text','qwen','ali-qwen','t2t','6fadgewjraOyvxC_EkHou','0','2026-05-23','9999-12-31','',NULL,10.00);
+    
+    -- 2. llm_api_map表
+    INSERT INTO llm_api_map VALUES ('rFS0Ag-epBUxE5LCF6glW','Jt26VKlUFsJxABuVLauBc','text2text','t2t',NULL,NULL,'0B-JNSUEelsY5TQ_Sawa7','1');
+    
+    -- 3. pricing_program表
+    INSERT INTO pricing_program VALUES ('0B-JNSUEelsY5TQ_Sawa7','qwen3.7-max定价','fields:\r\n  formula:\r\n    label: 计算公式\r\n    type: str\r\n  model:\r\n    label: 模型\r\n    type: str\r\n  prompt_tokens:\r\n    label: 输入词元数\r\n    type: int\r\n    value_mode: between\r\nformula: null\r\npricings:\r\n- formula: 2.5  prompt_tokens / 1000000 + completion_tokens  10 /1000000\r\n  model: qwen3.7-max\r\n  prompt_tokens: -0.1 ~= 32000\r\n- formula: 4  prompt_tokens / 1000000 + completion_tokens  16 /1000000\r\n  model: qwen3.7-max\r\n  prompt_tokens: 32000 ~= 128000\r\n- formula: 7  prompt_tokens / 1000000 + completion_tokens  28 /1000000\r\n  model: qwen3.7-max\r\n  prompt_tokens: 128000 ~= 256000\r\n- formula: 10  prompt_tokens / 1000000 + completion_tokens  40 /1000000\r\n  model: qwen3.7-max\r\n  prompt_tokens: 256000 ~= 1000000','2026-05-23','9999-12-31');
+    
+    -- 4. pricing_program_timing表
+    INSERT INTO pricing_program_timing VALUES ('C-IIEG_9_O9beN8S2vdUI','0B-JNSUEelsY5TQ_Sawa7','qwen3.7-max计费','fields:\r\n  formula:\r\n    label: 计算公式\r\n    type: str\r\n  model:\r\n    label: 模型\r\n    type: str\r\n  prompt_tokens:\r\n    label: 输入词元数\r\n    type: int\r\n    value_mode: between\r\nformula: null\r\npricings:\r\n- formula: 2.5  prompt_tokens / 1000000 + completion_tokens  10 /1000000\r\n  model: qwen3.7-max\r\n  prompt_tokens: -0.1 ~= 32000\r\n- formula: 4  prompt_tokens / 1000000 + completion_tokens  16 /1000000\r\n  model: qwen3.7-max\r\n  prompt_tokens: 32000 ~= 128000\r\n- formula: 7  prompt_tokens / 1000000 + completion_tokens  28 /1000000\r\n  model: qwen3.7-max\r\n  prompt_tokens: 128000 ~= 256000\r\n- formula: 10  prompt_tokens / 1000000 + completion_tokens  40 /1000000\r\n  model: qwen3.7-max\r\n  prompt_tokens: 256000 ~= 1000000','2026-05-23','9999-12-31');
+    
+    
+    验证查询
+    sql
+    SELECT id, name, model, upappid, apiname FROM llm WHERE model = 'qwen3.7-max';
+    SELECT id, llmid, llmcatelogid, apiname, ppid FROM llm_api_map WHERE llmid = 'Jt26VKlUFsJxABuVLauBc';
+    SELECT id, name FROM pricing_program WHERE id = '0B-JNSUEelsY5TQ_Sawa7';
+    
+    
+    ppid三处一致
+    - pricing_program.id = 0B-JNSUEelsY5TQ_Sawa7
+    - pricing_program_timing.ppid = 0B-JNSUEelsY5TQ_Sawa7
+    - llm_api_map.ppid = 0B-JNSUEelsY5TQ_Sawa7
+    
+    注意事项
+    1. 定价参考qwen3.6-plus的阶梯式计费模式，具体价格需根据百炼官方定价调整
+    2. qwen3.7-max使用已有的t2t uapi，OpenAI兼容接口
+    3. 如果qwen3.7-max有特殊的推理模式（如思考/非思考切换），可能需要额外的配置
+    4. 之前提到的 startReasoning is not defined 前端报错问题需要单独排查，可能与harnessed_reasoning的bricks前端代码有关
+
diff --git a/llmage/__pycache__/accounting.cpython-310.pyc b/llmage/__pycache__/accounting.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..efd3830c5996b9a0313b7c0a0ea40ec21f93d9a1
GIT binary patch
literal 9604
zcmai4Ym6LMR<2i9S5-f!r{`rmPLYf?ZevfJO|p@UCu9?QiC63-!pTM%6q=sCHFmZ8
zF}_u8$Cc{Hn+arBC@8BH4>=AyUM<lEvHw;eK!P9e3nYXDzv>5Q6_kzn0qid9Bb&s0
z=iaLB_A5-Q?(^JJ_ue}9obQ}FgHp*+@OS;MYn?Y5it-*+rvG$Q9>){>6B4E{wX1B%
zT-{JbUE9!*YhAskZ|ExZ>0P6j+sO6I4O7&OZoX%2SfZTk+C68(5oNPm=oL4LqMYxR
zdgYC>C|li1uewncWxG4mt8LU!c37c1`>eu>tn`|)QD^tCGON6%Zp^VNn?cJwT54<-
zEej~s*&IrXY@RKkZ;9<^OKcBX_OQKdA6oV@^|aEs_c~UktT)ueVy(~%{hrs*65D&J
z?QMttpr7QJw>dn6{@fWaJaqyE=Y-!5pWf_v?GUvJnw!CZH`{}L=)FWu_Gyn_@c7C8
z1ypKi4~9Lj*=n~3!+z-ZiCGP!o*V|@py%-?THRK^?Oj(f5p=1C+8WI%{iBa#;rUxU
z!QUbIvKlHo%1DdVjylpqE!IPQN8>Z0v73waYiNtLT{Bit)=uGLJf)0`*kH;~;SYy-
zrpC%!8p9~0I(S&fjkohV>L?f5aqcZ;M^%+L_o~WrBNI<PHclunRz_BA##YC9R2flT
z;HTm|<*4V+AHy4q;8ns?9>3KND=hCT7-z<IU5WF|ysCdo{axC2Ydy+sZ~JU5vJbn%
zpmoLz)}ms!+Y{wg>Nw!KUjM@C=5TAv;~s0a&$jp(?B2nK%e`>O`|kRn?{TqfdQus1
zAHA(^v(=+*PRukn+Jhi`S7}r^?P6lemnDuL`28THjZgF-<cZM<=xcWeEfyp*x4kv7
zPmV=G^BJd8!ON6Xi0vFnK|B4RP38yUTPOL!MNH9Wi9HNFKEV7LH$mTU5`&hXSlN;j
z9gF2N#LL{~e%lKY)th>Ox(xwM+43sAmy!?b7>~_+U?N<r+m{mB$nk#BB!kZ&(Jq^+
zsXF-8R7>5D+)xe8K-m%P+GPtZ&J9!3P}fxFW4?@GZ39X6S3tqWBc$;l_WsL|Cm~yf
z5UPP>hiRvkP=$mVkWENaDucN=*MVf7)3^&tT~HdQ6_T>hh%qvy%g1?s9291phh*m8
z7L;KQ(x&cY@@)1W7*kj&g$WtAceIBU$>+-K$O<8$u{Ig?JZ5U(x1&oMRp}QVOzDHI
zi<mq8UceKZVom3n(^q42gmclcM9a$z>u6f2+k;oBHMY+wBL}l7P#b1kq*mmm7`3-e
zahe=Re0e=O&_3(6&)stD4>pdu_*_17{Xj_c=c7vCb+=q0EjSw<fA&na{ZTI5Znf8<
z+F=(0=TgPpl<Fgzm2KTYyVcDa6K5hA;-o}2-nA-E9+*(2=l7c_kG04=OheZi4yUg!
zk4&Cq8<czxnvxRLXjV;nds3w;l`<YYPXkG7C50*0d>>jG1spJ1D;+~R7kn?#*==av
z?tl)*eyZ!BN%HBO5<|R4hl)*fv?V%D|KtSl`)Ne|PCL7Tp5Q1F5^NG}Q!PW(sbrvJ
zs8t+<s#ZsO@J3amLr}%BFjW2!2DM4CWPkH0kmw7skD#7YB@E$*64oH-DujP5d~d2(
z_K$R?LGkLlIs{WEI$s^D#vDXGx1%xx$KeF}ax4ess6n;ryC%!WW=5$hP}vaTw-jcR
zppWvQg@YlLbpBN}&M{+2>DWRz{r{JfBRCcA$VurqQ1VKrK;kaA7DrB6Dvb*0DUXUs
zD<i1+-71-y^vIS`pSh-tDq#)Bz8Y6??3}nvb~-M`cC4}TReLmpwav0hJTtX6(1TiI
z)wsxJuI5HHK~sxs6Ph#1XqNb$)A{$pI{4SYwceaVTfI4t_tcvUaRF&5E+buxD@d1O
z4e1`dz24jl-Uge!N-g{1+1-2M`ZXMN!E27y<2k`=PVkx&yygV2Il*fVyb7Qxf#V!F
zmXPjcb8*R6QIh=T1i$^Uj<w8Alna}WmFB(To6O(s>)x)Cm~vXll+v4O@Y!j#M175g
z^~A&hf!2>MLQw#PxUJPgIJe?hyIXwFqjT-H#PM#nw*6+Wwe4=Y&c(AZq#%ZMU4B`d
z!_l(q*sVSjjqVzFh~~6$bJY)+x78YU!}NrLWaYsXhabXO5c(uBw_1MJWAAEw5pBuL
zMBg-FUBh9(??HQ1rWy4Ip}Pg+%np92HX_ml?!oAuhRX)jbG8+@EthN`jC^yO56<vb
zFHs*%^k+WzLc^B)oy^8UhdA8B`n$ruIs?w2k_f^jMOl`wPYRnuSb8rAnuBd=Pk)gn
zYU2fL*bd1yZ~9SFjNx>~8#(!^yl8?>dYtGBDSfjIY*7q(s~==YqcR@Pyb!z5(1t|}
z`S-v4*86|>_3N*H?fN(09ui294ex*TcfS8u-}&IXe-Rm2*C9cA{q@&B`0jV!|H_;1
zU;Wd>PPfow06K(O<ZhE%P0ZkY_vA}JEM2cJXQ@8N7l=lugmkP>shrj*^|gbEU+}Rx
zSjhyh%#hk1?~L3w!oJwF-ZtP6&w)XrJ}J-&uoO0G(C5EIT#r+aMQawr?C~(UQI^=7
zRxeDf3oY)qHoG3iZ*O~j#tAv_N0Bt<g;)r6FPBGp8bIuPH!--^4ic>&aF@u%JH>M7
z#rzADoIsK|@*@zu<4;rTGn70*T{-dTl7h(OM@jUSA5e0Xk_T@!AS9nt2>o$z3BHX)
zVy1Z*g~ePEXbH&ZlbTT#*a;@&rh101D6hiI07Iew=AZeKT7gE%pZW7zp70lqxcNsT
ztxp`|=2lrVH1md|*Dyl9ziu3Thrfu)-ej`qTay?HljTmCEGSehRCiS<Y<LeaPvkSe
zL)bN#dQFWD7$a>D5ZYVDf&!nwglRJ-rtQkVO0NL;Ea$6aF!&=t_IdIH;B#2aIu7)-
z3ltYynG%OVvEf7f8O({3;;2sKz}s*-&KUO<o>%%cLkSC)u#6F0!pM9cSd5|kO7Z;T
zGQX%^RN}&Jk=dB3100s|+X+ida0l`d<_!wIjNS?vp*K?8s49X_DcTEa>aBvhLOhUH
zi8@1U;SBZ0&KNtEFuPSTyMIh)cVAe$jmJz}d`pp-FMWGWv;YlG=5Uu)l{&Ln(QI7m
z)RE)q(9F=*sfTkN7`n~^n<c0A4V}$jHOL>^U1STBT?T5^c9&v@jM4eWZ#OJ4aQT#B
zv8N3S^)(jPquPNNjvVM6Il$ZlPaQk(#bXCfulgTAwIuq{hZuht;uQVcw9RQto8x3g
zM>toh<#9J}HbCvM{gF0?tPN0aW`DBAwlqPqxqUYalo~Cm<H;(We(uR9Po6w+^2AEU
z5D2qLy2(B6F25WG&=RBNtnW4}wYJjc9*hTDNpgZVs=1k6Bx5=B*=RX2C{J=?m66R?
zr7yV_?aQ$HVV5}j(#?0JX^hH}ETef>O-#HC-e#2NDHJapb|>6Sl*s{JATvI8<KXs6
zeutyE>GrhsnEUzb1lbY1D0&Dua0DEJw(lpV-)CMwY#72`OiTc8fm&$c#vY0D4caYE
zv4^BYA9SqP6APQSMOA|IZ9WJ*1T~mM`oBVKgx^{genJaDjbXSwga_^q{QlNJ*r!BC
zr|>C~8qHpQfHO_GBsn!<y_SicFlJ+8#p^Vs92>$i$yuKwqSHtkW&WE~PThtfoaQjm
zSjd~yN;ZW*OUaigc@9aU^2C($6XqpR*{!Bvk(MzfAegTFfZxFQfFU8HAl>FgwN53A
zY8JWG@FFy{mDQqZUDj?CQLmw{fl5<dbzUuN`_(<N_hZX|F9HXgJQ4Gwnpsq}_o_cI
zjx7GOafDyMi*5=85fmU8Ar$(5ppXTELKO&l%KNzT>j(kqyBdN#P~*(Fs%458Dgyop
z)8ND^Q{D)aKYtNkin&Az-vGv@kPU)B9dg>C=sS6GYVnXG44<G=h}G@S!B0Wh=Nepf
zW-<F}2?jvlB{iI%yzh23L>TtYt(kHP`qQdOVBW&IP(BhgJc70J&r(8=Q@XWNxHCNv
ziF&kAk{HR!J|Q5GK&}E#2-njhEg^7>D2Wp=7CamS0dAY}xF&_PDo+Brp_dT#qhM41
zER+RcaIz<R)wBf3VUSgU=7fJt5}c^NCScYrbK;k1RS_lQnebnw>KBnTEGhAAY9O(f
zQckkX10;!^PLcrfc>3*Igq-(5`Y(8bV@ODz>wpskO3I(i=OiQ{2tu+s`4Rje8uLf>
zO8xs<e)VU94^3tKM-Y!5A*EMEfJlTZ5fa+ft|^c{fQ%9R=3N5-1TqQ`0%=4*NP5r4
zQ_7WZj0}W@a)2a|Jphf&b4DLYX0QgI7~vq8T8+XhHp!DQ4(xy`4MM?lhPR9mOy)gP
zU<W|N8zC%7a@w)Kt|D-g?>NYd$P36z$cxC!$V<p8$jiv9$ScTakXMn{kk25WMP5T*
zM?Q;u4tX8<Jn}i@3&`h@FCt$+zJz=c`5xp;$oIyYul_MXDPwmJ+m{-2u?n>8MI>9y
zG(E4fU1$3Nxb*X<ZeMvC(gI!ZH`C_`dv-5^z6x_?z%15b@_!+E`iWNi+;H2y1nBQ^
z*1EI;FHA<NJ%&JQ8yGrp!-0F&hhpcK+>j8xu#Md74K8?Nhf^rY)yT>qggBVu*se!k
zN<qV+o2kbx={Pi=I*n^9<2kIP2q8PD(wq-4QyAo2NQLN^H^m~dGlyVtb_|cf>6~7L
zB!m~rDY7O`J{U%OKZOC1z|=NKLaBL#E}5KGp;gEd@-C7jFZ`4KAen)mx{XT!OiLQh
zASrM8eLpxmuA8Z|YKE=g+_a-6y*UcJ3BL*6tX!#3LDg)P&Gh3){a65Xbl_7~L|yo}
z`0A0Bt*x<7YA0d-_5R=@_vxiV?+A@0>^vb1{uCub^?vFG@Yjg$w<!5-N?xbrE0j#s
z8tGu+ri#@G1(&EN;Oq;{wu9b83j>M;hKTU7DT_W)c!hNF(9*Z6xn$|CyYNHs4nibQ
zep(sE+aOb}&Fd5z)YSQ()@_SaovKkh{1a2xJ~j>QBQvM{+%&Zh&Aj$A)6#xw+S*S{
zNBgl^(EisfYX4)Fv>%ye?T2PX`+-^2{@a|<{>!MD|5>Xq5;XYW;E${KBf$73UVNQm
z@@*YS_E$n-tUKdWcTy;QY~3|z$9G0LTsXsrt%mc(ph<C&b4}&<L0iDu6Z|KAx&-^p
zbU0(wpF{s=M8BE#=R|)l?ayCRnKo1rizFUaN|P5f`IH7O7k^gt3l{_LutmR}_7~94
zM1L_IUl9F;;|eQ5ZyN}>;J$9xh>dG1TuT!g)g&D*a03PSa{@PDjnIV_e+|y$3}TyR
zA37H<BsA_KT%0<>?a;Xx35`o}(Ye?mrwb8E9T#8-!;iU8>=D=q@jp0B&np_PCdO6*
zZ&9;I8_L0-PwPs$!e}hC7Vrs9doA8Rn_`MgGa^8hog4R}AD(pw;;s;J_btB*H5#ls
zwqvK|V=lve;VpFMyi36`$F`4%0KW+JA4><jo0mi*T>!NDe$=ASiy*~u_t`XAZ@|2+
zd+LPfO(DoJcjf2`#)?24y#T6(heFehhmKX>V{230f(yblccDGLhWj|#>DW@Q3+)sw
zfStVCf~PbtkCp|li7H~&lYk?C1yUZ(rn5zRT7ourCR2?TvpGHnl1V$LtZWw1d`5E|
zG+8Ss5IT^d&L}4gRrF|Tegs%Mc&Onb)QC9V&BiFR9u$3)z{!;+Oz6|boh{fM21WO|
zG8#!AGDTpdu3hE@ZMW<9{E+nSU~4PD6;8t-BNpOb!`Mbh7ja3X&A^YmB%hj{MqYYa
z0zLMyd&2d>g(ci8PHH*Pv6x0F$)^&>f1il{fD#c_bI4osNsI#F2_-qPJ-Bxv@Y<xH
z8%U^#HwOIoh@~m#m)Pfr9$+qLg#UC4j!f<$XDMCel+TpT{9YWOU>)qV%kz%Jk#zl7
z#6v0F#$;vLLNsXAYVewh8Yy7o29)rP8o;OwnHe7){T}}&Sl)y>=}whClR%x{z6o_^
zmy#pZRw66ZcG3&Kk%o}Y5<)uqrNap6;Qtn)5^iOHkO)&!kSq<<=&$Jg9Kuh?Mfgbr
zf*g#D=p=Bf(E++lZ$aQI@!tp$EJG_@U5Z}hz_3X3bj6OK6uc-Jg~!I@;3?qA@0jB&
z$%VV}uu>ky*vQ_h0{fnSK&;}A(pBt}t6;<*^GL5pOH@Np3U^4%xh)ulwVi)pYK=g}
zz?Uj?Zt+}>{{v9EioL63dq;0g_wMf}dk5r<-IJL7&U+VDcV}2RyEj6Zs?eycNBadH
zp-YG{43t_47-e;JRUE=+BeOdcQIW!76iCZZh$Mel+<}V!A~-V;<;cN3=)e`%ppk_T
zlo-Z0Asr&Tf1RL3bl+Wp!8o3B-2q{vCSj2PKID>5McY@XDnOsO`lyIo(8(I4TPLuN
z0EHBx75);Rr4m7aB;N`y_1pd+$pu}{+isXLl2J})Ck~~@`8Ex3X@J%p@&~9^9Vhow
zdoA6=DSh-b9lkH1liWMPH*|oi;?N;FeOdd&GN^7KEU~P*z(B_3GBD7x?90g!M88Lb
zGOD`v9{&@tYpClDeO>G&{~Fcz@pq7kC=EYDC4nJJi1|M=-0b=Y=;11da84TknRE>M
zKGe&p6Y{p9KItHfDgVPpUO<$fe+GP&ZEg<+-9V_rq}DnMDZu~Ptjdi3e}YsbI>vIe
zn*Q^RxOqtCraGXv(we-h&J9B!oR?{u4^bjMm-yo3D~r!1^p4nrsqGgZLVN@3IPCGT
l>pg~zSAv5`EQP{CMeK5M>W4PXS+bU#rt^rk-#Jn|`hPtPl;;2d

literal 0
HcmV?d00001

diff --git a/llmage/accounting.py b/llmage/accounting.py
index e076826..ef9087a 100644
--- a/llmage/accounting.py
+++ b/llmage/accounting.py
@@ -1,7 +1,7 @@
 import asyncio
 import json
 import time
-from datetime import datetime
+from datetime import datetime, timedelta
 from appPublic.log import exception, debug
 from appPublic.uniqueID import getID
 from appPublic.dictObject import DictObject
@@ -239,20 +239,18 @@ async def llm_accoung_failed(luid, reason=None):
 			await sor.C('llmusage_accounting_failed', failed_rec)
 
 
-async def backup_accounted_llmusage():
-	"""Backup accounted records with use_date before today to history table."""
+async def backup_accounted_llmusage(cutoff_date):
+	"""Backup accounted records with use_date < cutoff_date to history table."""
 	env = ServerEnv()
-	today = datetime.now().strftime('%Y-%m-%d')
 	ts = env.timestampstr()
 	batched = 0
 	async with get_sor_context(env, 'llmage') as sor:
-		# Select records with use_date < today (i.e. yesterday and earlier)
 		sql = """select * from llmusage 
 where accounting_status='accounted' 
-	and use_date < ${today}$"""
-		recs = await sor.sqlExe(sql, {'today': today})
+	and use_date < ${cutoff_date}$"""
+		recs = await sor.sqlExe(sql, {'cutoff_date': cutoff_date})
 		if not recs:
-			debug(f'backup_accounted_llmusage: no records to backup for use_date < {today}')
+			debug(f'backup_accounted_llmusage: no records to backup for use_date < {cutoff_date}')
 			return 0
 		debug(f'backup_accounted_llmusage: {len(recs)} records to backup')
 		for r in recs:
@@ -280,7 +278,7 @@ where accounting_status='accounted'
 			# Delete from main table
 			await sor.D('llmusage', {'id': r.id})
 			batched += 1
-	debug(f'backup_accounted_llmusage: backed up {batched} records')
+	debug(f'backup_accounted_llmusage: backed up {batched} records for use_date < {cutoff_date}')
 	return batched
 
 
@@ -335,14 +333,13 @@ order by failed_time desc limit {page_size} offset {offset}"""
 async def backend_accounting():
 	env = ServerEnv()
 	debug(f'backend accounting started ...')
-	backup_counter = 0
+	last_backup_date = None
 	while True:
 		try:
 			lus = await get_accounting_llmusages()
 		except Exception as e:
 			exception(f'{e}')
 			lus = []
-		# debug(f'{len(lus)=} need to accounting........')
 		for lu in lus:
 			try:
 				tpac = await get_user_tpac(lu.userid)
@@ -356,12 +353,14 @@ async def backend_accounting():
 				exception(f'{e}, {lu.id=}')
 				await llm_accoung_failed(lu.id, reason=str(e))
 
-		# Run backup every 30 iterations (~5 minutes)
-		backup_counter += 1
-		if backup_counter >= 30:
-			backup_counter = 0
+		# Check if date changed, trigger backup once per day
+		today = datetime.now().strftime('%Y-%m-%d')
+		if today != last_backup_date:
+			yesterday = (datetime.now() - timedelta(days=1)).strftime('%Y-%m-%d')
+			last_backup_date = today
 			try:
-				await backup_accounted_llmusage()
+				debug(f'date changed to {today}, triggering backup for use_date < {yesterday}')
+				await backup_accounted_llmusage(yesterday)
 			except Exception as e:
 				exception(f'backup_accounted_llmusage failed: {e}')
 
diff --git a/models/llmusage.json b/models/llmusage.json
index 7d0f6ca..3371734 100644
--- a/models/llmusage.json
+++ b/models/llmusage.json
@@ -123,6 +123,14 @@
             "idxfields": [
                 "userid"
             ]
+        },
+        {
+            "name": "idx_llmusage_accounting",
+            "idxtype": "index",
+            "idxfields": [
+                "accounting_status",
+                "use_date"
+            ]
         }
     ]
 }
\ No newline at end of file
diff --git a/scripts/migrate_llmusage_history.sql b/scripts/migrate_llmusage_history.sql
index 5c30e0c..9bebf84 100644
--- a/scripts/migrate_llmusage_history.sql
+++ b/scripts/migrate_llmusage_history.sql
@@ -69,6 +69,9 @@ CREATE INDEX idx_laf_llmid ON llmusage_accounting_failed(llmid);
 CREATE INDEX idx_laf_handled ON llmusage_accounting_failed(handled);
 CREATE INDEX idx_laf_failed_time ON llmusage_accounting_failed(failed_time);
 
+-- 3. 为 llmusage 表添加组合索引（优化备份查询: accounting_status + use_date）
+CREATE INDEX idx_llmusage_accounting ON llmusage(accounting_status, use_date);
+
 -- ============================================================
 -- 验证步骤（执行后运行）:
 -- 1. 确认表创建成功: