tzt
e5470a3715
- normalizer.canonical_hash 增加可选 route_sig 段：键 = bucket|doc_version|sig|sha256 （旧三段格式向后兼容，旧条目随 TTL 自然淘汰） - ProxyConfig 新增 semcache.route_sig_scope：capabilities（默认，vision/tools 需求 签名）/ model（按模型隔离）/ none（旧行为）；非法值回落 capabilities - semcache：签名升级为条目属性并分区 L2 语义扫描（仅键分桶不够——语义层仍会 跨签名命中）；签名从缓存键第四段解析，重启重建零 schema 变更； 晋升别名键携带签名段；route_sig=None 的旧调用零过滤完全兼容 - routes：lookup/put 共用同一 norm_hash（消除 put 侧重复哈希），签名贯穿两层 - 新增 tests/test_route_sig.py 6 项（键格式/键空间分割/scope 三态/两层隔离/ 重建存活/旧调用兼容）
172 lines
6.9 KiB
Python
172 lines
6.9 KiB
Python
"""ProxyConfig：从 settings 的 proxy 段构建代理层配置（T-P0）。
|
||
|
||
锁定决策落地：
|
||
- D-P1 货币：账本毫元整数。配置里价格是浮点"元/1M tokens"（人类可读），
|
||
**加载期一次性转为整数毫元/1M tokens**，运行期纯整数运算（毫元漂移只可能
|
||
发生在这一次换算，黄金用例锁死）。
|
||
- 售价折扣采用差异化 {in: 0.5, out: 0.8}（设计文档 §2.5 结论：统一 5 折
|
||
盈亏平衡 h0≈32% 结构性危险；执行版 §4 示例与之冲突，按设计文档取差异化，
|
||
纯配置可热改）。
|
||
- D-P7：enabled=False 时不注册任何 /proxy 路由。
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
from dataclasses import dataclass, field
|
||
from typing import Any, Dict, List, Optional, Tuple
|
||
|
||
# 元 -> 毫元换算系数（D-P1：1 元 = 1000 毫元）
|
||
YUAN_TO_MILLI = 1000
|
||
|
||
|
||
@dataclass
|
||
class ModelPrice:
|
||
"""单一模型的整数价格表（毫元 / 1M tokens）。"""
|
||
in_miss: int
|
||
in_hit: int
|
||
out: int
|
||
|
||
|
||
@dataclass
|
||
class BucketCfg:
|
||
"""一个课程桶的配置。"""
|
||
name: str
|
||
system_template: str = "你是校园学习助手。"
|
||
doc_prefix_file: Optional[str] = None
|
||
doc_version: int = 1
|
||
ttl_hours: int = 72
|
||
|
||
|
||
@dataclass
|
||
class ProxyConfig:
|
||
"""代理层运行配置（全部为加载期定型值）。"""
|
||
enabled: bool = False
|
||
admin_key: str = ""
|
||
db_path: str = "data/proxy.sqlite3"
|
||
buckets: Dict[str, BucketCfg] = field(default_factory=dict)
|
||
model_prices: Dict[str, ModelPrice] = field(default_factory=dict)
|
||
peak_start: str = "08:30"
|
||
peak_end: str = "23:59"
|
||
offpeak_factor: float = 0.5
|
||
sale_in: float = 0.5
|
||
sale_out: float = 0.8
|
||
rpm_per_key: int = 10
|
||
day_req_cap: int = 200
|
||
concurrent_per_key: int = 2
|
||
max_body_chars: int = 60000
|
||
semcache_enabled: bool = True
|
||
sim_threshold: float = 0.92
|
||
max_entries: int = 300000
|
||
promote_frequency: int = 5
|
||
# T-X10（采纳 cortiq 路由签名分桶）：语义缓存键的路由签名粒度
|
||
# none | capabilities（默认：vision/tools 需求不同不互串） | model（按模型隔离）
|
||
route_sig_scope: str = "capabilities"
|
||
|
||
def bucket(self, name: str) -> BucketCfg:
|
||
"""取桶配置；未知名回落 default（D-P2 缺省桶）。"""
|
||
if name in self.buckets:
|
||
return self.buckets[name]
|
||
return self.buckets.get("default") or BucketCfg(name="default")
|
||
|
||
def price(self, model: str) -> Optional[ModelPrice]:
|
||
"""取模型价格表；未配置返回 None（调用方应拒绝或用 default 档）。"""
|
||
return self.model_prices.get(model)
|
||
|
||
|
||
def _to_milli_per_m(yuan_per_m: float) -> int:
|
||
"""元/1M -> 毫元/1M（四舍五入取整，加载期唯一换算点）。"""
|
||
return int(round(float(yuan_per_m) * YUAN_TO_MILLI))
|
||
|
||
|
||
def _parse_hhmm(text: str, fallback: str) -> str:
|
||
"""校验 HH:MM 形态，非法回落默认。"""
|
||
t = str(text or fallback).strip()
|
||
parts = t.split(":")
|
||
if len(parts) == 2 and parts[0].isdigit() and parts[1].isdigit():
|
||
h, m = int(parts[0]), int(parts[1])
|
||
if 0 <= h <= 23 and 0 <= m <= 59:
|
||
return f"{h:02d}:{m:02d}"
|
||
return fallback
|
||
|
||
|
||
def build_proxy_config(settings_dict: Dict[str, Any]) -> ProxyConfig:
|
||
"""从 settings.to_dict() 的 proxy 段构建 ProxyConfig（缺省值兜底）。
|
||
|
||
settings_dict 为整个设置 dict（含 worker/architect/... 与 proxy 段）。
|
||
"""
|
||
raw = settings_dict.get("proxy") or {}
|
||
if not isinstance(raw, dict):
|
||
raw = {}
|
||
|
||
# ---- 桶 ----
|
||
buckets: Dict[str, BucketCfg] = {}
|
||
raw_buckets = raw.get("buckets") or {}
|
||
if isinstance(raw_buckets, dict):
|
||
for name, b in raw_buckets.items():
|
||
if not isinstance(b, dict):
|
||
continue
|
||
buckets[str(name)] = BucketCfg(
|
||
name=str(name),
|
||
system_template=str(b.get("system_template") or "你是校园学习助手。"),
|
||
doc_prefix_file=(str(b["doc_prefix_file"])
|
||
if b.get("doc_prefix_file") else None),
|
||
doc_version=int(b.get("doc_version", 1) or 1),
|
||
ttl_hours=int(b.get("ttl_hours", 72) or 72),
|
||
)
|
||
if "default" not in buckets:
|
||
buckets["default"] = BucketCfg(name="default")
|
||
|
||
# ---- 价格表：元/1M 浮点 -> 毫元/1M 整数（D-P1 加载期换算） ----
|
||
model_prices: Dict[str, ModelPrice] = {}
|
||
raw_pricing = raw.get("pricing") or {}
|
||
if isinstance(raw_pricing, dict):
|
||
for model, p in raw_pricing.items():
|
||
if not isinstance(p, dict) or "in_miss" not in p:
|
||
continue # peak_window/offpeak_factor/sale_discount 非模型键
|
||
in_miss = _to_milli_per_m(p.get("in_miss", 0))
|
||
in_hit = _to_milli_per_m(p.get("in_hit", 0))
|
||
out = _to_milli_per_m(p.get("out", 0))
|
||
if in_miss <= 0:
|
||
continue
|
||
if in_hit <= 0:
|
||
in_hit = max(1, in_miss // 30) # 缺省 = in_miss × 1/30（D-P3）
|
||
model_prices[str(model)] = ModelPrice(
|
||
in_miss=in_miss, in_hit=in_hit, out=out)
|
||
|
||
# ---- 峰谷窗口 / 系数 ----
|
||
pw = raw_pricing.get("peak_window") or {} if isinstance(raw_pricing, dict) else {}
|
||
off = float(raw_pricing.get("offpeak_factor", 0.5)) if isinstance(raw_pricing, dict) else 0.5
|
||
sale = raw_pricing.get("sale_discount") or {} if isinstance(raw_pricing, dict) else {}
|
||
|
||
# ---- 限流 / 缓存 ----
|
||
limits = raw.get("limits") or {}
|
||
if not isinstance(limits, dict):
|
||
limits = {}
|
||
sem = raw.get("semcache") or {}
|
||
if not isinstance(sem, dict):
|
||
sem = {}
|
||
|
||
return ProxyConfig(
|
||
enabled=bool(raw.get("enabled", False)),
|
||
admin_key=str(raw.get("admin_key") or ""),
|
||
db_path=str(raw.get("db_path") or "data/proxy.sqlite3"),
|
||
buckets=buckets,
|
||
model_prices=model_prices,
|
||
peak_start=_parse_hhmm(pw.get("start"), "08:30") if isinstance(pw, dict) else "08:30",
|
||
peak_end=_parse_hhmm(pw.get("end"), "23:59") if isinstance(pw, dict) else "23:59",
|
||
offpeak_factor=max(0.0, min(1.0, off)),
|
||
sale_in=max(0.0, min(1.0, float(sale.get("in", 0.5)))),
|
||
sale_out=max(0.0, min(1.0, float(sale.get("out", 0.8)))),
|
||
rpm_per_key=int(limits.get("rpm_per_key", 10) or 10),
|
||
day_req_cap=int(limits.get("day_req_cap", 200) or 200),
|
||
concurrent_per_key=int(limits.get("concurrent_per_key", 2) or 2),
|
||
max_body_chars=int(limits.get("max_body_chars", 60000) or 60000),
|
||
semcache_enabled=bool(sem.get("enabled", True)),
|
||
sim_threshold=float(sem.get("sim_threshold", 0.92) or 0.92),
|
||
max_entries=int(sem.get("max_entries", 300000) or 300000),
|
||
promote_frequency=int(sem.get("promote_frequency", 5) or 5),
|
||
route_sig_scope=(str(sem.get("route_sig_scope"))
|
||
if str(sem.get("route_sig_scope")) in
|
||
("none", "capabilities", "model")
|
||
else "capabilities"),
|
||
)
|