feat: JSON 配置、质量分与仪表盘,及设置与爬取流程
- 后端改为 config/app.json;pytest 使用 config/app.test.json 与 set_config_file,不再依赖环境变量;移除 pydantic-settings。 - 前端 API/WebSocket 由 config/webui.json 经 Vite define 注入。 - 代理分数按延迟与随机取用次数计算,新增 use_count 与 proxy_scoring;保存设置时同步调度器启停。 - 仪表盘双饼图(可用/待验证协议);设置页去掉调度器启停按钮并移动立即验证;爬取全部结束后自动提交全量验证。 - 删除 script/settings_maintain.py(此前已标记删除)。 Made-with: Cursor
This commit is contained in:
54
app/services/proxy_scoring.py
Normal file
54
app/services/proxy_scoring.py
Normal file
@@ -0,0 +1,54 @@
|
||||
"""代理质量分:延迟越低越高,被取用次数越多越低。
|
||||
|
||||
设计要点
|
||||
--------
|
||||
1. **延迟项**(0~100):用平滑倒数把毫秒映射到质量,避免线性过于极端。
|
||||
``latency_quality = 100 / (1 + latency_ms / latency_ref_ms)``
|
||||
在 ``latency_ref_ms`` 处约为 50 分;越快越接近 100。
|
||||
|
||||
2. **使用惩罚**:每次通过 API 随机取出代理视为一次「使用」,``use_count`` 递增;
|
||||
惩罚 ``min(max_use_penalty, use_count * use_penalty_per_pick)`` 从延迟项上扣除。
|
||||
|
||||
3. **未知延迟**:尚无 ``response_time_ms`` 时用 ``default_latency_ms`` 代替,避免给满分。
|
||||
|
||||
验证失败仍走 ``update_score`` 扣分;验证成功则用本函数**覆盖**分数(与当前延迟、使用次数一致)。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
from typing import Optional
|
||||
|
||||
from app.core.config import Settings
|
||||
|
||||
|
||||
def compute_proxy_quality_score(
|
||||
latency_ms: Optional[float],
|
||||
use_count: int,
|
||||
settings: Settings,
|
||||
) -> int:
|
||||
"""根据延迟与累计使用次数计算 0~100 的整数分。"""
|
||||
ref = float(settings.score_latency_ref_ms)
|
||||
penalty_per = float(settings.score_use_penalty_per_pick)
|
||||
cap = float(settings.score_max_use_penalty)
|
||||
default_lat = float(settings.score_default_latency_ms)
|
||||
lo = int(settings.score_min)
|
||||
hi = int(settings.score_max)
|
||||
|
||||
if ref <= 0:
|
||||
ref = 500.0
|
||||
if penalty_per < 0:
|
||||
penalty_per = 0.0
|
||||
if cap < 0:
|
||||
cap = 0.0
|
||||
if default_lat <= 0:
|
||||
default_lat = 1500.0
|
||||
|
||||
ms = latency_ms
|
||||
if ms is None or ms <= 0:
|
||||
ms = default_lat
|
||||
|
||||
latency_quality = 100.0 / (1.0 + float(ms) / ref)
|
||||
uses = max(0, int(use_count))
|
||||
usage_penalty = min(cap, uses * penalty_per)
|
||||
raw = latency_quality - usage_penalty
|
||||
score = int(round(raw))
|
||||
return max(lo, min(hi, score))
|
||||
Reference in New Issue
Block a user