feat: ETF 支持(选股 / 回测 / 监控) (#61)

* feat(screener): 选股引擎支持 ETF

- 12 个内置策略打 asset_types 白名单 + strategy_supports_asset;涨停类
  (连板/断板反包)仅股票,其余 10 个技术类对 ETF 开放
- ScreenerService(repo, asset_type) 分流取数,ETF 复用 kline_etf_enriched,
  跳过股票专用历史缓存与涨停信号;进程级 _history_cache key 含 asset_type
- API /run、/run_preset 透传 asset_type;/strategies 按资产过滤;
  股票专有策略在 ETF 下返回空
- 新增 enriched_dirname(asset_type) 共享 helper;get_enriched_latest_asset
  增 refresh 参数(供轮询线程避免冷缓存同步重算)
- 前端「策略」页加 股票/ETF 切换,ETF 走实时单跑(空日期→用 ETF 自身最新日);
  QK.screenerStrategies 按 asset_type keyed
- 测试:test_screener_etf.py

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(backtest): 回测支持 ETF(个股/因子/策略组合)

- 三条回测路径 + 共用 BacktestEngine 面板加载按 asset_type 路由到
  kline_etf_enriched(复用 enriched_dirname);PanelCache key 隔离资产;
  ETF 跳过股票专用 get_enriched_range 缓存
- 面板 compute_all/名称 JOIN 按 asset_type 取维表(get_instruments_asset),
  修复 ETF 策略回测用错股票维表致名称为空/涨停信号算错
- BacktestConfig/FactorConfig/StrategyBacktestConfig 增 asset_type
- 三个回测 API + SSE stream 透传 asset_type;_make_job_key 纳入 asset_type
  (修复 stream 与 cancel job_key 不对齐致取消失效的回归)
- 前端策略组合页/因子页加 股票/ETF 切换,标的搜索与策略列表跟随资产;
  assetType 持久化
- 测试:test_backtest_etf.py(含 job_key 一致性回归);既有回测测试替身同步

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(monitor): 监控规则支持 ETF

- engine.evaluate(df, asset_type) 按规则 asset_type 分轮评估;quote_service
  增开 ETF 评估轮(用 ETF enriched 快照),股票轮不受影响、不重置其策略结果
- ETF 评估轮独立 try(异常不丢弃已算出的股票告警)+ refresh=False(不在轮询
  线程触发 ETF 冷缓存同步重算)
- ETF 版历史加载器(main.py 注入)+ 按规则 asset_type 选加载器
- _strategy_pools 按 (sid, asset_type) 键,避免同策略股票/ETF 规则互相覆盖
- name_map 仅在有 ETF 规则时补 ETF 维表, setdefault 保股票名优先
- RuleModel/normalize 增 asset_type(默认 stock,持久化往返)
- 前端 RuleEditor 加 股票/ETF 选择,策略列表与标的搜索跟随资产
- 测试:test_monitor_etf.py

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(etf): 前端 API 绑定透传 asset_type + 文档

- api.ts: screener/backtest 绑定加 assetType 参数,MonitorRule 类型加 asset_type
- docs/features.md: 标注选股/回测/监控的 ETF 支持范围与前提

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(reliability): 管道并发/原子写/能力探测/监控告警多处加固

后端可靠性专项修复(均带回归测试, backend 全套 64 passed):

并发与数据完整性:
- 盘后管道单飞: JobStore.create() 去重纳入 pending∨running, 关闭"两次快速点击"
  并发双跑窗口; 新增 _heavy_run_lock 执行槽挡住 reap 后僵尸线程并发写 parquet
- adj_factor/minute 全部改走原子写(tmp+replace), 消除 kill/断电致 all.parquet 损坏
- 分块拉取失败聚合 WARNING 可见化(不再静默当成功); 复权失败标的会保持旧价已提示

能力探测:
- 周期重探(60min)热更新 app.state.capabilities, 付费 Key 过期/续费无需重启即可见
- 瞬时探测失败(超时/连接/5xx, 按 _is_transient 判定)不降级、保留旧付费档;
  真 401/无权限仍正常降级回落 free-api

监控告警:
- 评估仅在连续竞价(9:30-11:30/13:00-15:00)+ 快照当日新鲜度下进行, 避开集合竞价/
  收盘后陈旧价与节假日误告警
- scope=sector fail-closed(validate 拒绝新建 + _apply_scope 返回空), 修复板块规则
  对全市场刷屏
- 飞书 webhook 加退避重试并移到独立线程池 fire-and-forget, 不再阻塞行情轮询线程

单标的新鲜度: 新增 repo.symbols_lagging() 检测掉队标的并 WARNING + 计入 job 结果

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Jinfeng Sun
2026-07-08 12:12:29 +08:00
committed by GitHub
co-authored by Claude Opus 4.8
parent 9c730e4b69
commit e5a94c42d5
35 changed files with 1135 additions and 213 deletions
+16 -10
View File
@@ -134,8 +134,9 @@ class PanelCache:
end: date,
columns: list[str] | None,
compute_fn,
asset_type: str = "stock",
) -> pl.DataFrame:
key = self._make_key(symbols, start, end, columns)
key = self._make_key(symbols, start, end, columns, asset_type)
now = time.monotonic()
if key in self._cache:
@@ -145,7 +146,7 @@ class PanelCache:
return entry.df
del self._cache[key]
df = compute_fn(symbols, start, end, columns)
df = compute_fn(symbols, start, end, columns, asset_type)
self._cache[key] = _CacheEntry(df=df, ts=now)
if len(self._cache) > self._max_size:
self._cache.popitem(last=False)
@@ -155,13 +156,13 @@ class PanelCache:
self._cache.clear()
@staticmethod
def _make_key(symbols: list[str] | None, start: date, end: date, columns: list[str] | None) -> str:
def _make_key(symbols: list[str] | None, start: date, end: date, columns: list[str] | None, asset_type: str = "stock") -> str:
if symbols is None:
h = "all"
else:
h = hashlib.md5(",".join(sorted(symbols)).encode()).hexdigest()[:12]
cols = "all" if columns is None else hashlib.md5(",".join(sorted(columns)).encode()).hexdigest()[:8]
return f"{h}:{start}:{end}:{cols}"
return f"{asset_type}:{h}:{start}:{end}:{cols}"
# ================================================================
@@ -183,9 +184,10 @@ class BacktestEngine:
start: date,
end: date,
columns: list[str] | None = None,
asset_type: str = "stock",
) -> pl.DataFrame:
"""加载 enriched 数据面板,带缓存。"""
return self._cache.get_or_compute(symbols, start, end, columns, self._load_panel_inner)
"""加载 enriched 数据面板,带缓存。asset_type='etf' 时读 ETF enriched。"""
return self._cache.get_or_compute(symbols, start, end, columns, self._load_panel_inner, asset_type=asset_type)
def _load_panel_inner(
self,
@@ -193,12 +195,13 @@ class BacktestEngine:
start: date,
end: date,
columns: list[str] | None = None,
asset_type: str = "stock",
) -> pl.DataFrame:
t0 = time.perf_counter()
# 近期区间优先复用 repository 的预计算 enriched 历史缓存,避免重复 scan_parquet + compute_all
# 近期区间优先复用 repository 的预计算 enriched 历史缓存 (仅 stock: 该缓存为股票专用)
try:
if self.repo is not None and hasattr(self.repo, "get_enriched_range"):
if asset_type == "stock" and self.repo is not None and hasattr(self.repo, "get_enriched_range"):
cached = self.repo.get_enriched_range(start, end, symbols=symbols, columns=columns)
if cached is not None and not cached.is_empty():
elapsed = (time.perf_counter() - t0) * 1000
@@ -207,7 +210,8 @@ class BacktestEngine:
except Exception as e: # noqa: BLE001
logger.debug("backtest load panel cache miss: %s", e)
enriched_glob = str(self.repo.store.data_dir / "kline_daily_enriched" / "**" / "*.parquet")
from app.tickflow.repository import enriched_dirname
enriched_glob = str(self.repo.store.data_dir / enriched_dirname(asset_type) / "**" / "*.parquet")
try:
lf = pl.scan_parquet(enriched_glob)
@@ -242,7 +246,9 @@ class BacktestEngine:
return df
from app.indicators.pipeline import compute_all
instruments = self.repo.get_instruments()
# 按 asset_type 取维表: ETF 回测须用 ETF 维表, 否则名称 JOIN 失败(全 null)、
# 涨停信号算在错误的 instruments 上。
instruments = self.repo.get_instruments_asset(asset_type)
df = compute_all(df, instruments=instruments)
if not instruments.is_empty() and "name" not in df.columns:
inst_cols = [c for c in ["symbol", "name"] if c in instruments.columns]