refactor: fix P0/P1/P2 security and performance issues

P0 fixes:
- CORS: replace wildcard methods/headers with configurable lists
- deps.py: remove unsafe global _warned_unset variable

P1 fixes:
- http_client: read default timeout from Settings
- bzzoiro: replace sync urllib with async httpx
- bzzoiro: normalize validation failures use warning level only
- db pool: read pool config from Settings (default 5+10)
- backtest: add asyncio.Semaphore(8) for concurrent execution
- predict/context_builder: add backtest parameter for cutoff buffer

P2 improvements:
- injuries: enforce int conversion for player_id/fixture_id
- injuries: use system temp dir for cache
- utils.py: extract shared actual_1x2/is_correct_1x2
- validation: downgrade 1x2 mismatch log to debug
- docker-compose: use env vars for all credentials
- .env.example: add POSTGRES_USER/PASSWORD/PORT, API_PORT
This commit is contained in:
shangfangjian
2026-09-16 02:38:25 +08:00
parent fa69795d69
commit 983b620659
14 changed files with 154 additions and 109 deletions
+35 -43
View File
@@ -3,9 +3,11 @@
核心机制:
- build_context 已内置 before=match_date,天然防未来信息泄漏
- 对历史比赛跑预测 → 用实际比分 settle → 统计准确率
- 并发控制: asyncio.Semaphore 限制同时 LLM 调用数
"""
from __future__ import annotations
import asyncio
import logging
from dataclasses import dataclass, field
from datetime import datetime
@@ -17,6 +19,7 @@ from src.db.models import Match
from src.db.unit_of_work import get_uow
from src.llm.eval import settle_prediction
from src.llm.predict import predict_match
from src.llm.utils import actual_1x2
logger = logging.getLogger(__name__)
@@ -69,15 +72,6 @@ class BacktestSummary:
results: list[BacktestMatchResult] = field(default_factory=list)
def _actual_1x2(home: int, away: int) -> str:
"""实际比分 → 胜平负。"""
if home > away:
return "1"
if home == away:
return "X"
return "2"
async def _get_historical_matches(
db,
*,
@@ -156,44 +150,42 @@ async def run_backtest(
summary = BacktestSummary(total=len(candidates), scored=0)
for c in candidates:
try:
# 预测 (build_context 内部已用 before=match_date 防泄漏,
# injuries_slice 也使用 as_of=match_date 过滤 retrieved_at)
# 回测必须禁用结果缓存: 否则命中缓存会复用同一 prediction_id,
# 导致 settle 反复覆盖同一条记录(见 P1-3)。
result = await predict_match(c.match_id, mode=mode, model=model, use_cache=False)
# P1-6: 并发控制,同时最多 8 场预测(避免 LLM API 限流)
sem = asyncio.Semaphore(8)
# 用实际比分 settle
await settle_prediction(result.prediction_id, c.home_goals, c.away_goals)
async def _one(c: BacktestCandidate) -> BacktestMatchResult | None:
async with sem:
try:
result = await predict_match(c.match_id, mode=mode, model=model, use_cache=False, backtest=True)
await settle_prediction(result.prediction_id, c.home_goals, c.away_goals)
actual = actual_1x2(c.home_goals, c.away_goals)
return BacktestMatchResult(
match_id=c.match_id,
league_code=c.league_code,
home_team=c.home_team,
away_team=c.away_team,
match_date=c.match_date.strftime("%Y-%m-%d") if c.match_date else "?",
actual_home=c.home_goals,
actual_away=c.away_goals,
actual_1x2=actual,
pred_home=result.pred_home_goals,
pred_away=result.pred_away_goals,
pred_1x2=result.pred_1x2,
subjective_confidence=result.subjective_confidence,
correct_1x2=result.pred_1x2 == actual,
prediction_id=result.prediction_id,
)
except Exception:
logger.exception("backtest match %s failed", c.match_id)
return None
actual = _actual_1x2(c.home_goals, c.away_goals)
correct = result.pred_1x2 == actual
bt = BacktestMatchResult(
match_id=c.match_id,
league_code=c.league_code,
home_team=c.home_team,
away_team=c.away_team,
match_date=c.match_date.strftime("%Y-%m-%d") if c.match_date else "?",
actual_home=c.home_goals,
actual_away=c.away_goals,
actual_1x2=actual,
pred_home=result.pred_home_goals,
pred_away=result.pred_away_goals,
pred_1x2=result.pred_1x2,
subjective_confidence=result.subjective_confidence,
correct_1x2=correct,
prediction_id=result.prediction_id,
)
summary.results.append(bt)
# 并行执行,保持结果顺序
results = await asyncio.gather(*[_one(c) for c in candidates])
for r in results:
if r is not None:
summary.results.append(r)
summary.scored += 1
except Exception:
# 用 exception 而非 warning:保留堆栈,否则集成层缺陷(如惰性加载
# 在 session 外触发)会只剩一行无堆栈的 warning,极难定位。
logger.exception("backtest match %s failed", c.match_id)
# 汇总统计
if summary.scored > 0:
correct_count = sum(1 for r in summary.results if r.correct_1x2)