chore: baseline — ORM migration + systemd deploy + MCP server + daily check

- ORM migration: models/ops 重构
- deploy: 标准化 systemd timer/service 部署体系
- MCP server: 5 个只读工具(任务/状态/数据集)
- daily check: 数据一致性巡检
- watch: task_watch 后台监控
- kline_daily: 麦蕊优先,时间门禁15:00
- 删除: task_industry_sector, task_sector_features

app/tasks/__init__.py

@@ -3,15 +3,14 @@
 所有 `SyncTask` 子类在此集中注册，外部通过 `get_task(dataset_id)` 或
 遍历 `TASKS` 字典使用。
 """
-from app.tasks.task_industry_sector import SyncIndustrySector
 from app.tasks.task_kline_5min import SyncKline5Min
 from app.tasks.task_kline_daily import SyncKlineDaily
 from app.tasks.task_kline_index import SyncKlineIndex
 from app.tasks.task_longhubang import SyncLonghubang
 from app.tasks.task_mairui_ma_daily import SyncMairuiMADaily
+from app.tasks.task_mairui_indicators import SyncMairuiIndicators
 from app.tasks.task_market_regime import SyncMarketRegime
 from app.tasks.task_moneyflow import SyncMoneyflow
-from app.tasks.task_sector_features import SyncSectorFeatures
 from app.tasks.task_share_snapshot import SyncShareSnapshot
 from app.tasks.task_stock_node import SyncStockNode
 from app.tasks.task_stocks_basic import SyncStocksBasic
@@ -26,13 +25,12 @@ TASKS: dict[str, type] = {
         SyncKline5Min,
         SyncTickTrade,
         SyncMoneyflow,
-        SyncIndustrySector,
-        SyncSectorFeatures,
         SyncShareSnapshot,
         SyncMarketRegime,
         SyncLonghubang,
         SyncStockNode,
         SyncMairuiMADaily,
+        SyncMairuiIndicators,
     )
 }
 

app/tasks/task_industry_sector.py

@@ -1,121 +0,0 @@
-"""同步任务：股票-行业映射（Baostock）。
-
-数据流：Baostock query_stock_industry() → 写 industry / sectors / stock_sector_map 三张表。"""
-from __future__ import annotations
-
-import time
-from typing import Any
-
-from sqlalchemy import update
-
-from app.core.db import ops as db_ops
-from app.core.db.models import Stock
-from app.core.datasource.base import registry as ds_registry
-from app.core.datasource.registry import is_source_ready
-from app.core.datasource.utils import to_hermes
-from app.core.sync.base import SyncTask
-from app.core.sync.registry import mark_sync_blocked
-from app.core.utils.logging import get_logger
-
-logger = get_logger("sync.industry_sector")
-
-
-def _build_sector_key(taxonomy: str, sector_name: str) -> str:
-    tax = (taxonomy or "").strip() or "default"
-    name = (sector_name or "").strip()
-    return f"{tax}::{name}"
-
-
-class SyncIndustrySector(SyncTask):
-    dataset_id = "industry_sector"
-
-    def _run(self, *, trigger_source: str = "manual", **kwargs) -> dict[str, Any]:
-        bs = ds_registry.get("datasource_baostock")
-        if bs is None:
-            return {"status": "error", "message": "Baostock 数据源未注册"}
-        ok, reason = is_source_ready(bs.key)
-        if not ok:
-            mark_sync_blocked(self.dataset_id, message=f"Baostock 未就绪: {reason}")
-            return {"status": "blocked", "message": f"Baostock 未就绪: {reason}"}
-
-        self._progress(message="拉取股票-行业映射...")
-        t0 = time.time()
-        try:
-            raw_rows = bs.fetch_industry_map()
-        except Exception as e:
-            return {"status": "error", "message": f"拉取失败: {e}"}
-
-        if not raw_rows:
-            return {"status": "warning", "message": "Baostock 未返回行业数据"}
-
-        # 合并：sectors + stock_sector_map + industry
-        sectors_rows: dict[str, dict] = {}
-        stock_sector_rows: list[dict] = []
-        industry_rows: list[dict] = []
-
-        for r in raw_rows:
-            code6 = str(r["code"]).zfill(6)
-            name = (r.get("industry_name") or "").strip()
-            if not name:
-                continue
-            classification = (r.get("industry_classification") or "").strip()
-            sector_key = _build_sector_key(classification, name)
-
-            sectors_rows[sector_key] = {
-                "sector_key": sector_key,
-                "sector_name": name,
-                "taxonomy": classification,
-                "level": "",
-                "source": "baostock_query_stock_industry",
-                "enabled": 1,
-            }
-            stock_sector_rows.append({"stock_code": to_hermes(code6), "sector_key": sector_key})
-            industry_rows.append({
-                "code": to_hermes(code6),
-                "industry_name": name,
-                "industry_classification": classification,
-                "update_date": r.get("update_date", ""),
-            })
-
-        # 写库（全量替换）
-        try:
-            db_ops.replace_all_sectors(list(sectors_rows.values()))
-            db_ops.replace_all_stock_sector_map(stock_sector_rows)
-            db_ops.replace_all_industries(industry_rows)
-        except Exception as e:
-            return {"status": "error", "message": f"写库失败: {e}"}
-
-        # 回填 stocks.industry 字段（让 stocks 表也能直接看到）
-        # 走 ORM update(Stock) + IN (SHxxx, SZxxx, BJxxx, ...)
-        try:
-            # 按 industry_name 分组，对每组用一个 IN 批量更新
-            by_industry: dict[str, list[str]] = {}
-            for r in industry_rows:
-                name = r["industry_name"]
-                code = r["code"]
-                if name not in by_industry:
-                    by_industry[name] = []
-                by_industry[name].append(code)
-            for ind_name, codes in by_industry.items():
-                # 6 位 code × 3 个交易所前缀（SH/SZ/BJ）
-                code_variants = [
-                    f"{prefix}{c}" for c in codes for prefix in ("SH", "SZ", "BJ")
-                ]
-                with db_ops.get_session() as s:
-                    s.execute(
-                        update(Stock)
-                        .where(Stock.code.in_(code_variants))
-                        .values(industry=ind_name)
-                    )
-        except Exception as e:
-            logger.warning(f"回填 stocks.industry 失败（不影响主任务）: {e}")
-
-        elapsed = round(time.time() - t0, 1)
-        msg = f"行业 {len(sectors_rows)}个 映射{len(stock_sector_rows)}条, {elapsed}s"
-        return {
-            "status": "ok",
-            "message": msg,
-            "sectors": len(sectors_rows),
-            "stock_sector_map": len(stock_sector_rows),
-            "elapsed_sec": elapsed,
-        }

app/tasks/task_kline_5min.py

@@ -17,7 +17,7 @@ from __future__ import annotations
 
 import os
 import time
-from concurrent.futures import ThreadPoolExecutor, as_completed
+from concurrent.futures import ThreadPoolExecutor, as_completed, TimeoutError as FuturesTimeoutError
 from datetime import datetime, timedelta, timezone
 from typing import Any, Optional
 
@@ -37,6 +37,10 @@ MAIRUI_5MIN_FLOOR = datetime(2023, 6, 14, tzinfo=timezone.utc)
 WINDOW_DAYS = 365
 # 增量时往前多取的天数（防交易日历边界漏当天）
 INCREMENT_OVERLAP_DAYS = 2
+# 2026-07-08 教训: 数据源层有 20s timeout,task 层再硬兜底,
+# 防 SDK 升级 / 网络层 bug 让单只股票卡住(7月7日 4.4只/秒 后突然 0 持续 24h)
+# 7月9日 mairui/雪球 间歇性 "服务器连接失败" + Broken pipe,拉慢。给 120s 容忍。
+FETCH_HARD_TIMEOUT = 120.0  # 5min K 一只拉多年,单只允许更久
 
 
 class SyncKline5Min(SyncTask):
@@ -99,7 +103,12 @@ class SyncKline5Min(SyncTask):
 
         # ── 2) 内存算计划 ──
         # PG TIMESTAMPTZ 是 timezone-aware，所以 end 也必须 tz-aware 否则比较会炸
-        end = datetime.now(timezone.utc)
+        # 2026-07-12 修复: end 改为上海时区最近交易日 15:00，避免 UTC 与上海时间混用
+        from app.core.datasource.utils import effective_market_date
+        end_date_str = effective_market_date()
+        end = datetime.strptime(f"{end_date_str} 15:00:00", "%Y-%m-%d %H:%M:%S").replace(
+            tzinfo=timezone(timedelta(hours=8))
+        )
         plans = self._plan(stock_codes, end, snapshots)
         if not plans:
             msg = f"5min K线 — 全部 {len(stock_codes)} 只都已最新（无需同步）"
@@ -129,12 +138,17 @@ class SyncKline5Min(SyncTask):
         t0 = time.time()
         ok_cnt = fail_cnt = 0
         rows_total = 0
-        with ThreadPoolExecutor(max_workers=max_workers) as pool:
-            futures = {pool.submit(self._sync_one, primary, p[0], p[1], p[2]): p[0] for p in plans}
-            for i, future in enumerate(as_completed(futures), 1):
+        # 不用 `with ThreadPoolExecutor(...) as pool:` — 它的 __exit__ 默认 wait=True,
+        # 一旦某个 worker 卡在 xueqiu IO,主线程会在 as_completed 触发 60s timeout 后
+        # 仍被 __exit__ 阻塞等 worker 退出 → 进程挂死(2026-07-09 13:21 教训)。
+        # 改成手动管理 + shutdown(wait=False),主线程能立刻退出。
+        pool = ThreadPoolExecutor(max_workers=max_workers)
+        futures = {pool.submit(self._sync_one, primary, p[0], p[1], p[2]): p[0] for p in plans}
+        try:
+            for i, future in enumerate(as_completed(futures, timeout=FETCH_HARD_TIMEOUT), 1):
                 c6 = futures[future]
                 try:
-                    res = future.result()
+                    res = future.result(timeout=0.1)  # 已被 as_completed 释放
                     if res["status"] == "ok":
                         ok_cnt += 1
                         rows_total += res["rows"]
@@ -142,6 +156,9 @@ class SyncKline5Min(SyncTask):
                         fail_cnt += 1
                         if res.get("error"):
                             logger.warning(f"[5min {c6}] {res['error']}")
+                except FuturesTimeoutError:
+                    fail_cnt += 1
+                    logger.warning(f"[5min {c6}] 内部 race timeout, 记 fail")
                 except Exception as e:
                     fail_cnt += 1
                     logger.warning(f"[5min {c6}] {e}")
@@ -150,6 +167,64 @@ class SyncKline5Min(SyncTask):
                         message=f"5min 进度 {i}/{len(plans)} OK:{ok_cnt} FAIL:{fail_cnt}",
                         current=i, total=len(plans), current_step=c6,
                     )
+        except FuturesTimeoutError:
+            stuck = [c6 for _, c6 in futures.items() if not _.done()]
+            logger.error(
+                "[5min] 所有 fetcher 卡死 (>%ss), %d 只股票未完成",
+                FETCH_HARD_TIMEOUT, len(stuck),
+            )
+            for f, c6 in futures.items():
+                try:
+                    if not f.done() and hasattr(f, "cancel"):
+                        f.cancel()
+                except Exception as e:
+                    logger.warning(f"[5min] cancel future for {c6} 失败: {e}")
+            try:
+                pool.shutdown(wait=False)
+            except Exception as e:
+                logger.warning(f"[5min] pool.shutdown(wait=False) 失败: {e}")
+
+            # ── 补偿：冷却后重入队列 ──
+            # mairui 偶发全 hang（中间件重启/网络抖动），等 30s 再试一次。
+            # 如果还 hang 就放弃，下次调度增量重拉。
+            if stuck:
+                cooldown = 30
+                logger.warning(f"[5min] 冷却 {cooldown}s 后重试 {len(stuck)} 只…")
+                time.sleep(cooldown)
+                retry_ok = retry_fail = retry_rows = 0
+                retry_pool = ThreadPoolExecutor(max_workers=max_workers)
+                plan_dict = {p[0]: (p[1], p[2]) for p in plans}
+                retry_futs = {}
+                for c6 in stuck:
+                    if c6 in plan_dict:
+                        s, e = plan_dict[c6]
+                        # 重试直接用雪球 fallback（Mairui 已全 hang）
+                        retry_futs[retry_pool.submit(self._fallback_xueqiu_5m_and_write, c6, s, e)] = c6
+                if retry_futs:
+                    for fut in as_completed(retry_futs, timeout=FETCH_HARD_TIMEOUT):
+                        c6 = retry_futs[fut]
+                        try:
+                            res = fut.result(timeout=0.1)
+                            if res["status"] == "ok":
+                                retry_ok += 1
+                                retry_rows += res["rows"]
+                            else:
+                                retry_fail += 1
+                        except Exception:
+                            retry_fail += 1
+                    try:
+                        retry_pool.shutdown(wait=False)
+                    except Exception:
+                        pass
+                    ok_cnt += retry_ok
+                    fail_cnt -= retry_ok  # 从 fail 挪到 ok
+                    rows_total += retry_rows
+                    logger.warning(
+                        f"[5min] 重试结果: {retry_ok}成 {retry_fail}败 "
+                        f"共{retry_rows}行, 救回 {retry_ok} 只"
+                    )
+        else:
+            pool.shutdown(wait=True)
         elapsed = round(time.time() - t0, 1)
         msg = (
             f"5min K线 {ok_cnt}成 {fail_cnt}败 共{rows_total}行, "
@@ -192,10 +267,16 @@ class SyncKline5Min(SyncTask):
                         })
                 cur = w_end + timedelta(days=1)
         except Exception as e:
-            return {"status": "fail", "error": str(e)}
+            logger.warning(f"[5min {code6}] mairui 失败, 尝试雪球 5m fallback: {e}")
+            all_rows = self._fallback_xueqiu_5m(code6, start, end)
 
         if not all_rows:
-            return {"status": "fail", "error": "no data"}
+            # Mairui 无数据, 尝试雪球 fallback
+            logger.info(f"[5min {code6}] mairui 无数据, 尝试雪球 fallback")
+            all_rows = self._fallback_xueqiu_5m(code6, start, end)
+
+        if not all_rows:
+            return {"status": "fail", "error": "no data (mairui + xueqiu fallback)"}
         try:
             CHUNK = 5000
             for i in range(0, len(all_rows), CHUNK):
@@ -203,3 +284,77 @@ class SyncKline5Min(SyncTask):
         except Exception as e:
             return {"status": "fail", "error": f"db write: {e}"}
         return {"status": "ok", "rows": len(all_rows)}
+
+    @staticmethod
+    def _fallback_xueqiu_5m_and_write(code6: str, start: datetime, end: datetime) -> dict:
+        """雪球 5m fallback + 直接写库，对齐 _sync_one 返回格式供 retry 逻辑使用。"""
+        rows = SyncKline5Min._fallback_xueqiu_5m(code6, start, end)
+        if not rows:
+            return {"status": "fail", "error": "no data via xueqiu"}
+        try:
+            CHUNK = 5000
+            for i in range(0, len(rows), CHUNK):
+                db_ops.upsert_kline_5min(rows[i: i + CHUNK])
+        except Exception as e:
+            return {"status": "fail", "error": f"db write: {e}"}
+        return {"status": "ok", "rows": len(rows)}
+
+    @staticmethod
+    def _fallback_xueqiu_5m(code6: str, start: datetime, end: datetime) -> list[dict]:
+        """通过雪球 5m K线 fallback 拉取数据，对齐 _sync_one 返回格式。"""
+        import os
+        from datetime import datetime as dt_mod
+        from app.core.datasource.utils import code6_to_xueqiu, to_hermes
+
+        token_raw = os.environ.get("XUEQIU_TOKEN", "").strip()
+        if not token_raw:
+            # 从 .env 读
+            try:
+                with open("/home/gao/Development/quant_home/market_sync/.env") as f:
+                    for line in f:
+                        if line.startswith("XUEQIU_TOKEN="):
+                            token_raw = line.strip().split("=", 1)[1]
+                            break
+            except Exception:
+                pass
+        if not token_raw:
+            logger.warning("[xueqiu_5m] 无 XUEQIU_TOKEN, 跳过 fallback")
+            return []
+
+        import pysnowball as ball
+        ball.set_token(token_raw)
+
+        symbol = code6_to_xueqiu(code6)
+        # 计算需要多少根 5m bar (每天 48 根, 加 buffer)
+        days_needed = max((end - start).days + 2, 1)
+        count = days_needed * 48
+
+        try:
+            data = ball.kline(symbol, "5m", min(count, 5000))
+        except Exception as e:
+            logger.warning(f"[xueqiu_5m {code6}] 请求失败: {e}")
+            return []
+
+        items = data.get("data", {}).get("item", [])
+        if not items:
+            return []
+
+        columns = data["data"]["column"]  # [timestamp, volume, open, high, low, close, ...]
+        idx = {c: i for i, c in enumerate(columns)}
+        hermes = to_hermes(code6)
+        rows = []
+        for item in items:
+            ts = item[idx["timestamp"]] / 1000
+            bar_time = dt_mod.fromtimestamp(ts).strftime("%Y-%m-%d %H:%M:%S")
+            rows.append({
+                "stock_code": hermes,
+                "bar_time": bar_time,
+                "open": float(item[idx["open"]]),
+                "high": float(item[idx["high"]]),
+                "low": float(item[idx["low"]]),
+                "close": float(item[idx["close"]]),
+                "volume": float(item[idx["volume"]]),
+                "amount": 0.0,
+                "turnover_rate": 0.0,
+            })
+        return rows

app/tasks/task_kline_daily.py

@@ -19,6 +19,7 @@ import os
 import queue
 import threading
 import time
+from concurrent.futures import ThreadPoolExecutor, as_completed, TimeoutError as FuturesTimeoutError
 from datetime import datetime, timedelta
 from typing import Any
 
@@ -32,15 +33,19 @@ from app.core.utils.logging import get_logger
 
 logger = get_logger("sync.kline_daily")
 
-# 主源优先级：mairui（5 RPS，免费 + 稳定 + 单接口含5min）→ 雪球 → Baostock → 新浪
-# 经验：sina 在单日窗口 (start=end=某天) 会返回空，雪球更可靠
-PRIMARY_PRIORITY = ["datasource_xueqiu", "datasource_mairui", "datasource_xinlang"]
+# 源优先级：麦蕊 → 雪球 → 新浪
+# 2026-07-21 改: 麦蕊15:10已有数据,雪球要15:40后才齐,麦蕊为主源提速
+PRIMARY_PRIORITY = ["datasource_mairui", "datasource_xueqiu", "datasource_xinlang"]
 DEFAULT_START = "2018-01-01"
 # 写库批量：攒够 BATCH_SIZE 行就 executemany 一次 + commit
 BATCH_SIZE = 2000
 # fetcher 并发：实测单线程最稳（5000 只 × 0.2s = 17 分钟，足够）
 # 雪球 token 限速 10 RPS，单线程 0.1s/只 已经打满。多 worker 会触发风控 hang
 DEFAULT_FETCH_WORKERS = 1
+# 2026-07-08 教训: 数据源层虽然加了 fetch timeout (20s), task 层也加硬上限兜底,
+# 防止数据源 SDK 升级后又被绕过。60s 对应 xueqiu/mairui 的 15-20s 上限 + DB upsert
+# / DataFrame 处理余量 + 7月9日 mairui 间歇慢请求的容忍窗口。
+FETCH_HARD_TIMEOUT = 180.0
 
 
 class SyncKlineDaily(SyncTask):
@@ -57,14 +62,15 @@ class SyncKlineDaily(SyncTask):
         max_workers: int = DEFAULT_FETCH_WORKERS,
         **kwargs,
     ) -> dict[str, Any]:
-        # 1. 选主源
-        primary = self._pick_primary()
-        if primary is None:
+        # 1. 获取所有可用源（按优先级），支持 per-stock fallback
+        sources = self._get_available_sources()
+        if not sources:
             return {"status": "error", "message": "所有 K 线数据源均不可用"}
+        source_names = ", ".join(s.name for s in sources)
 
         self._progress(
-            message=f"使用 {primary.name} 同步日K线 (fetcher={max_workers}, batch={BATCH_SIZE})...",
-            current_step=primary.key,
+            message=f"使用 [{source_names}] 同步日K线 (fetcher={max_workers}, batch={BATCH_SIZE})...",
+            current_step=sources[0].key,
         )
 
         # 2. 拉股票列表
@@ -108,7 +114,6 @@ class SyncKlineDaily(SyncTask):
         )
 
         # 4. 并发 fetch + 攒 batch 写库（线程安全）
-        # mairui 等外部 API 无并发问题；DB 写串行加锁
         from concurrent.futures import ThreadPoolExecutor, as_completed
         t0 = time.time()
         batch_lock = threading.Lock()
@@ -116,6 +121,7 @@ class SyncKlineDaily(SyncTask):
         pending_codes: dict[str, str] = {}
         ok_lock = threading.Lock()
         ok_cnt = [0]
+        fallback_ok_cnt = [0]
         fail_cnt = [0]
         rows_cnt = [0]
 
@@ -138,15 +144,7 @@ class SyncKlineDaily(SyncTask):
                 except Exception as e:
                     logger.warning(f"update synced_at 失败 {c6}: {e}")
 
-        def _fetch_one(code6: str, fetch_start: str) -> tuple[str, list[dict] | None, str | None]:
-            try:
-                df = primary.fetch_kline_daily(code6, fetch_start, _end)
-            except Exception as e:
-                return (code6, None, str(e)[:120])
-            if df is None or df.empty:
-                return (code6, None, "no data")
-            # 写库用 hermes（带 SH/SZ 前缀）—— 2026-07-01 重构：历史上 kline_stock 用
-            # 6位 code 与 stocks.code(SH600519) 格式不一致，跨表 JOIN 失败。现统一。
+        def _rows_from_df(code6: str, df: pd.DataFrame) -> tuple[list[dict], str]:
             hermes = to_hermes(code6)
             rows = []
             for _, r in df.iterrows():
@@ -162,20 +160,56 @@ class SyncKlineDaily(SyncTask):
                 })
             latest = df["trade_date"].max()
             latest = latest.strftime("%Y-%m-%d") if hasattr(latest, "strftime") else str(latest)[:10]
-            return (code6, rows, latest)
+            return (rows, latest)
+
+        def _fetch_one(code6: str, fetch_start: str) -> tuple[str, list[dict] | None, str | None, bool]:
+            errors = []
+            for src in sources:
+                try:
+                    df = src.fetch_kline_daily(code6, fetch_start, _end)
+                except Exception as e:
+                    errors.append(f"{src.key}: {e}")
+                    continue
+                if df is None or df.empty:
+                    errors.append(f"{src.key}: no data")
+                    continue
+                rows, latest = _rows_from_df(code6, df)
+                is_fallback = src is not sources[0]
+                return (code6, rows, latest, is_fallback)
+            return (code6, None, "; ".join(errors)[:200], False)
 
         total = len(jobs)
-        with ThreadPoolExecutor(max_workers=max_workers) as pool:
-            futures = {pool.submit(_fetch_one, c6, fs): c6 for c6, fs in jobs}
-            done_cnt = 0
-            for future in as_completed(futures):
+        # 不用 `with ThreadPoolExecutor(...) as pool:` — 它的 __exit__ 默认 wait=True,
+        # 一旦某个 worker 卡在 xueqiu IO,主线程会在 as_completed 触发 timeout 后
+        # 仍被 __exit__ 阻塞等 worker 退出 → 进程挂死(2026-07-09 13:21 教训)。
+        # 改成手动管理 + shutdown(wait=False),主线程能立刻退出。
+        pool = ThreadPoolExecutor(max_workers=max_workers)
+        futures = {pool.submit(_fetch_one, c6, fs): c6 for c6, fs in jobs}
+        done_cnt = 0
+        try:
+            for future in as_completed(futures, timeout=FETCH_HARD_TIMEOUT):
                 done_cnt += 1
-                code6, rows, latest = future.result()
+                code6 = futures[future]
+                try:
+                    code6_r, rows, latest, is_fallback = future.result(timeout=0.1)
+                except FuturesTimeoutError:
+                    with ok_lock:
+                        fail_cnt[0] += 1
+                    logger.warning(f"[kline {code6}] 内部 race timeout, 记 fail")
+                    continue
+                except Exception as e:
+                    with ok_lock:
+                        fail_cnt[0] += 1
+                    logger.warning(f"[kline {code6}] 异常: {e}")
+                    continue
                 if rows is None:
                     with ok_lock:
                         fail_cnt[0] += 1
                     logger.warning(f"[kline {code6}] {latest}")
                     continue
+                if is_fallback:
+                    with ok_lock:
+                        fallback_ok_cnt[0] += 1
                 with batch_lock:
                     batch_rows.extend(rows)
                     pending_codes[code6] = latest
@@ -194,32 +228,121 @@ class SyncKlineDaily(SyncTask):
                         message=f"进度 {done_cnt}/{total} OK:{ok_cnt[0]} FAIL:{fail_cnt[0]} {rate:.1f}只/秒",
                         current=done_cnt, total=total, current_step=code6,
                     )
+        except FuturesTimeoutError:
+            # 30s 内没 future 完成 → 所有 worker 都被卡死(数据源层 timeout 兜底失败)
+            # 取消 pending futures,标 fail。shutdown(wait=False) 让主线程立刻返回,
+            # 不被卡住的 worker 拖死。
+            # 2026-07-11 修复: 防御性 cancel,避免异常处理链里变量被遮蔽导致
+            # AttributeError: 'str' object has no attribute 'cancel'。
+            stuck = [(f, c6) for f, c6 in futures.items() if not f.done()]
+            logger.error(
+                "所有 fetcher 卡死 (>%ss), %d 只股票未完成, 标 fail 后退出主循环",
+                FETCH_HARD_TIMEOUT, len(stuck),
+            )
+            with ok_lock:
+                fail_cnt[0] += len(stuck)
+            for f, c6 in stuck:
+                try:
+                    if hasattr(f, "cancel"):
+                        f.cancel()
+                except Exception as cancel_err:
+                    logger.warning(f"[kline] cancel future for {c6} 失败: {cancel_err}")
+            try:
+                pool.shutdown(wait=False)
+            except Exception as shutdown_err:
+                logger.warning(f"[kline] pool.shutdown(wait=False) 失败: {shutdown_err}")
+
+            # ── 补偿：冷却后重入队列 ──
+            # 雪球/mairui 偶发全 hang（服务端限流/网络抖动），等 30s 再试一次。
+            # 单线程拉取场景下，卡住通常意味着当前这只股票请求 hang 死；
+            # 冷却后把未完成的股票重新排队重试，救回临时故障。
+            if stuck:
+                cooldown = 30
+                logger.warning(f"[kline] 冷却 {cooldown}s 后重试 {len(stuck)} 只…")
+                time.sleep(cooldown)
+                retry_ok = 0
+                retry_fb = 0
+                retry_rows = 0
+                retry_pool = ThreadPoolExecutor(max_workers=max_workers)
+                jobs_dict = {c6: fs for c6, fs in jobs}
+                retry_futs = {}
+                for _, c6 in stuck:
+                    if c6 in jobs_dict:
+                        retry_futs[retry_pool.submit(_fetch_one, c6, jobs_dict[c6])] = c6
+                if retry_futs:
+                    try:
+                        for fut in as_completed(retry_futs, timeout=FETCH_HARD_TIMEOUT):
+                            c6 = retry_futs[fut]
+                            try:
+                                code6_r, rows, latest, is_fallback = fut.result(timeout=0.1)
+                            except Exception:
+                                logger.warning(f"[kline {c6}] 重试异常")
+                                continue
+                            if rows is None:
+                                logger.warning(f"[kline {c6}] 重试仍失败: {latest}")
+                                continue
+                            with batch_lock:
+                                batch_rows.extend(rows)
+                                pending_codes[c6] = latest
+                                cur_size = len(batch_rows)
+                            with ok_lock:
+                                retry_ok += 1
+                                retry_rows += len(rows)
+                                if is_fallback:
+                                    retry_fb += 1
+                            if cur_size >= BATCH_SIZE:
+                                _flush()
+                    except FuturesTimeoutError:
+                        logger.error(
+                            "[kline] 重试仍卡死 (>%ss), %d 只股票放弃",
+                            FETCH_HARD_TIMEOUT, sum(1 for f in retry_futs if not f.done()),
+                        )
+                    try:
+                        retry_pool.shutdown(wait=False)
+                    except Exception:
+                        pass
+                    with ok_lock:
+                        ok_cnt[0] += retry_ok
+                        fail_cnt[0] -= retry_ok
+                        fallback_ok_cnt[0] += retry_fb
+                        rows_cnt[0] += retry_rows
+                    logger.warning(
+                        f"[kline] 重试结果: {retry_ok}成 "
+                        f"({retry_fb}只备胎) 共{retry_rows}行, 救回 {retry_ok} 只"
+                    )
+        else:
+            pool.shutdown(wait=True)
 
         # 收尾 flush
         _flush()
 
         elapsed = round(time.time() - t0, 1)
-        msg = f"日K线 {ok_cnt[0]}成 {fail_cnt[0]}败 {skipped}跳 共{rows_cnt[0]}行, {elapsed}s"
+        fb = fallback_ok_cnt[0]
+        fb_info = f" ({fb}只备胎)" if fb else ""
+        msg = f"日K线 {ok_cnt[0]}成 {fail_cnt[0]}败 {skipped}跳 共{rows_cnt[0]}行{fb_info}, {elapsed}s"
         return {
             "status": "ok" if fail_cnt[0] == 0 else "warning",
             "message": msg,
             "ok": ok_cnt[0],
             "fail": fail_cnt[0],
+            "fallback_ok": fb,
             "skip": skipped,
             "rows": rows_cnt[0],
             "elapsed_sec": elapsed,
-            "primary_source": primary.key,
+            "primary_source": sources[0].key if sources else "",
+            "sources": [s.key for s in sources],
         }
 
-    def _pick_primary(self):
+    def _get_available_sources(self):
         from app.core.datasource.registry import is_source_ready, run_health_check, get_health_status
         if not get_health_status():
             run_health_check()
+        available = []
         for key in PRIMARY_PRIORITY:
             ok, _ = is_source_ready(key)
             if ok:
-                return ds_registry.get(key)
-        return None
+                available.append(ds_registry.get(key))
+        return available
 
     def _sync_one(self, primary, code6: str, start: str, end: str) -> dict:
         """保留这个方法以兼容外部调用（已不用，但单只测试可能用到）"""

app/tasks/task_mairui_indicators.py

@@ -0,0 +1,261 @@
+"""同步任务：mairui 日 K 级别技术指标 MACD / KDJ / BOLL。
+
+数据源：mairui /hsstock/history/{indicator}/{symbol}/d/n/{licence}
+  - macd → market_data.kline_stock_macd_daily (diff, dea, macd, ema12, ema26)
+  - kdj  → market_data.kline_stock_kdj_daily  (k, d, j)
+  - boll → market_data.kline_stock_boll_daily (upper, mid, lower ← mairui u/m/d)
+
+设计（对齐 task_kline_daily）：
+  - 全市场 A 股循环，每只票分别拉 3 个指标
+  - 增量：读各指标表 MAX(trade_date)，只补新增区间（回看 5 天防漏）
+  - 并发 fetch + 攒 BATCH 批量 upsert，写库串行加锁
+  - stock_code 落库用 hermes 格式（SH600519），与项目其他表一致
+
+参考实现：pre_limit_seeker/production/sync_mairui_indicators.py（parquet 版）。
+"""
+from __future__ import annotations
+
+import os
+import threading
+import time
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from datetime import datetime, timedelta
+from typing import Any
+
+from app.core.db import ops as db_ops
+from app.core.datasource.base import registry as ds_registry
+from app.core.datasource.utils import is_a_share_code, to_code6, to_hermes
+from app.core.sync.base import SyncTask, _effective_sync_end
+from app.core.utils.logging import get_logger
+
+logger = get_logger("sync.mairui_indicators")
+
+DEFAULT_START = "2018-01-01"
+BATCH_SIZE = 2000
+# mairui 钻石档 100 RPS；5 workers × 10 RPS/worker = 50 RPS，留 buffer
+DEFAULT_FETCH_WORKERS = 5
+
+# 指标 → (mairui 源字段, 落库列名, upsert 函数)
+INDICATORS: dict[str, dict[str, Any]] = {
+    "macd": {
+        "src_fields": ["diff", "dea", "macd", "ema12", "ema26"],
+        "db_cols": ["diff", "dea", "macd", "ema12", "ema26"],
+        "upsert": db_ops.upsert_kline_stock_macd_daily_rows,
+    },
+    "kdj": {
+        "src_fields": ["k", "d", "j"],
+        "db_cols": ["k", "d", "j"],
+        "upsert": db_ops.upsert_kline_stock_kdj_daily_rows,
+    },
+    "boll": {
+        # mairui: u=上轨, m=中轨, d=下轨 → 落库 upper/mid/lower
+        "src_fields": ["u", "m", "d"],
+        "db_cols": ["upper", "mid", "lower"],
+        "upsert": db_ops.upsert_kline_stock_boll_daily_rows,
+    },
+}
+
+
+class SyncMairuiIndicators(SyncTask):
+    dataset_id = "mairui_indicators"
+
+    def _run(
+        self,
+        *,
+        trigger_source: str = "manual",
+        indicators: list[str] | None = None,
+        codes: list[str] | None = None,
+        start: str | None = None,
+        end: str | None = None,
+        incremental: bool = True,
+        max_workers: int = DEFAULT_FETCH_WORKERS,
+        **kwargs,
+    ) -> dict[str, Any]:
+        targets = [i.lower() for i in (indicators or ["macd", "kdj", "boll"])]
+        for i in targets:
+            if i not in INDICATORS:
+                return {"status": "error", "message": f"未知指标 {i!r}，可选 {list(INDICATORS)}"}
+
+        source = self._pick_source()
+        if source is None:
+            return {"status": "error", "message": "mairui 数据源不可用（licence 未配置？）"}
+
+        # 股票列表（hermes 格式）
+        if codes:
+            hermes_codes = [to_hermes(c) for c in codes]
+        else:
+            hermes_codes = [
+                to_hermes(c) for c in db_ops.iter_stock_codes(active_only=True)
+                if is_a_share_code(c)
+            ]
+        limit_raw = os.environ.get("MARKET_DATA_STOCK_LIMIT", "").strip()
+        if limit_raw.isdigit() and int(limit_raw) > 0:
+            hermes_codes = hermes_codes[: int(limit_raw)]
+        if not hermes_codes:
+            return {"status": "error", "message": "无股票代码（stocks 表为空？）"}
+
+        _end = end or _effective_sync_end()
+        end_date_obj = datetime.strptime(_end, "%Y-%m-%d").date()
+
+        t0 = time.time()
+        agg_ok = 0
+        agg_fail = 0
+        agg_rows = 0
+        per_indicator: dict[str, dict[str, int]] = {}
+
+        for ind in targets:
+            self._progress(message=f"[{ind}] 规划增量区间...", current_step=ind)
+            jobs = self._plan_jobs(ind, hermes_codes, start, incremental, end_date_obj)
+            total = len(jobs)
+            skipped = len(hermes_codes) - total
+            logger.info(f"[mairui_indicators] {ind}: to_pull={total}, skip={skipped}")
+            self._progress(
+                message=f"[{ind}] 开始同步 {total} 只 (跳过 {skipped} 只已最新)",
+                current=0, total=total, current_step=ind,
+            )
+
+            ok, fail, rows = self._sync_indicator(
+                source, ind, jobs, _end, max_workers,
+            )
+            per_indicator[ind] = {"ok": ok, "fail": fail, "skip": skipped, "rows": rows}
+            agg_ok += ok
+            agg_fail += fail
+            agg_rows += rows
+
+        elapsed = round(time.time() - t0, 1)
+        parts = ", ".join(
+            f"{k}({v['ok']}成/{v['fail']}败/{v['rows']}行)"
+            for k, v in per_indicator.items()
+        )
+        msg = f"指标 {parts}；共 {agg_rows} 行, {elapsed}s"
+        logger.info(f"[mairui_indicators] {msg}")
+        return {
+            "status": "ok" if agg_fail == 0 else "warning",
+            "message": msg,
+            "ok": agg_ok,
+            "fail": agg_fail,
+            "rows": agg_rows,
+            "elapsed_sec": elapsed,
+            "per_indicator": per_indicator,
+        }
+
+    def _plan_jobs(
+        self, indicator: str, hermes_codes: list[str],
+        start: str | None, incremental: bool, end_date_obj,
+    ) -> list[tuple[str, str]]:
+        """返回 [(hermes_code, fetch_start), ...]。"""
+        jobs: list[tuple[str, str]] = []
+        if incremental and not start:
+            for hc in hermes_codes:
+                last = db_ops.get_indicator_max_date(indicator, hc)
+                if last is None:
+                    fetch_start = DEFAULT_START
+                elif datetime.strptime(last, "%Y-%m-%d").date() >= end_date_obj:
+                    continue
+                else:
+                    fetch_start = (
+                        datetime.strptime(last, "%Y-%m-%d").date() - timedelta(days=5)
+                    ).strftime("%Y-%m-%d")
+                jobs.append((hc, fetch_start))
+        else:
+            fetch_start = start or DEFAULT_START
+            jobs = [(hc, fetch_start) for hc in hermes_codes]
+        return jobs
+
+    def _sync_indicator(
+        self, source, indicator: str, jobs: list[tuple[str, str]],
+        end: str, max_workers: int,
+    ) -> tuple[int, int, int]:
+        cfg = INDICATORS[indicator]
+        upsert_fn = cfg["upsert"]
+        src_fields = cfg["src_fields"]
+        db_cols = cfg["db_cols"]
+
+        batch_lock = threading.Lock()
+        batch_rows: list[dict] = []
+        ok_lock = threading.Lock()
+        ok_cnt = [0]
+        fail_cnt = [0]
+        rows_cnt = [0]
+
+        def _flush():
+            with batch_lock:
+                if not batch_rows:
+                    return
+                to_write = list(batch_rows)
+                batch_rows.clear()
+            try:
+                upsert_fn(to_write)
+            except Exception as e:
+                logger.error(f"[{indicator}] batch upsert 失败 ({len(to_write)} 行): {e}")
+
+        def _fetch_one(hermes_code: str, fetch_start: str):
+            code6 = to_code6(hermes_code)
+            try:
+                df = source.fetch_indicator_daily(indicator, code6, fetch_start, end)
+            except Exception as e:
+                return (hermes_code, None, str(e)[:120])
+            if df is None or df.empty:
+                return (hermes_code, None, "no data")
+            rows = []
+            for _, r in df.iterrows():
+                td = r["trade_date"]
+                # 2026-07-11 修复: mairui 可能返回整行 NaN/NaT(如 688 早期无指标),
+                # float(NaT) 会抛 TypeError。用 pandas.isna 统一跳过无效值。
+                row = {
+                    "stock_code": hermes_code,
+                    "trade_date": td.strftime("%Y-%m-%d") if hasattr(td, "strftime") else str(td)[:10],
+                    "source": "mairui",
+                }
+                for src_f, db_c in zip(src_fields, db_cols):
+                    v = r.get(src_f)
+                    if v is None or (hasattr(v, "isna") and v.isna()) or (isinstance(v, float) and v != v):
+                        row[db_c] = None
+                    else:
+                        try:
+                            row[db_c] = float(v)
+                        except Exception:
+                            row[db_c] = None
+                rows.append(row)
+            return (hermes_code, rows, None)
+
+        total = len(jobs)
+        t0 = time.time()
+        with ThreadPoolExecutor(max_workers=max_workers) as pool:
+            futures = {pool.submit(_fetch_one, hc, fs): hc for hc, fs in jobs}
+            done = 0
+            for fut in as_completed(futures):
+                done += 1
+                hermes_code, rows, err = fut.result()
+                if rows is None:
+                    with ok_lock:
+                        fail_cnt[0] += 1
+                    continue
+                with batch_lock:
+                    batch_rows.extend(rows)
+                    cur = len(batch_rows)
+                with ok_lock:
+                    ok_cnt[0] += 1
+                    rows_cnt[0] += len(rows)
+                if cur >= BATCH_SIZE:
+                    _flush()
+                if done % 100 == 0 or done == total:
+                    el = time.time() - t0
+                    rate = done / el if el > 0 else 0
+                    self._progress(
+                        message=f"[{indicator}] {done}/{total} OK:{ok_cnt[0]} FAIL:{fail_cnt[0]} {rate:.1f}只/秒",
+                        current=done, total=total, current_step=indicator,
+                    )
+        _flush()
+        return ok_cnt[0], fail_cnt[0], rows_cnt[0]
+
+    def _pick_source(self):
+        from app.core.datasource.registry import (
+            get_health_status, is_source_ready, run_health_check,
+        )
+        if not get_health_status():
+            run_health_check()
+        ok, _ = is_source_ready("datasource_mairui")
+        if ok:
+            return ds_registry.get("datasource_mairui")
+        return None

app/tasks/task_market_regime.py

@@ -3,7 +3,6 @@
 基于本地 kline_stock 数据聚合：advancers / decliners / turnover / advance_ratio。"""
 from __future__ import annotations
 
-from datetime import datetime
 from typing import Any
 
 import numpy as np
@@ -12,7 +11,7 @@ from sqlalchemy import text
 
 from app.core.db import ops as db_ops
 from app.core.db.orm import engine as pg_engine
-from app.core.sync.base import SyncTask
+from app.core.sync.base import SyncTask, _effective_sync_end
 from app.core.utils.logging import get_logger
 
 logger = get_logger("sync.market_regime")
@@ -34,7 +33,16 @@ class SyncMarketRegime(SyncTask):
     dataset_id = "market_regime"
 
     def _run(self, *, trigger_source: str = "manual", start: str = DEFAULT_START, **kwargs) -> dict[str, Any]:
-        end = datetime.now().strftime("%Y-%m-%d")
+        # 2026-07-12 修复: end 用 kline_stock 最大 trade_date，而不是 datetime.now()
+        # 避免节假日/周末跑时把非交易日当 end，导致 SQL 条件不精确
+        try:
+            with pg_engine.connect() as conn:
+                max_date = conn.execute(
+                    text("SELECT MAX(trade_date) FROM market_data.kline_stock")
+                ).scalar()
+            end = str(max_date) if max_date else _effective_sync_end()
+        except Exception:
+            end = _effective_sync_end()
         self._progress(message=f"构建 market_regime {start} ~ {end}...")
 
         # 1. 取所有非退市股票代码

app/tasks/task_sector_features.py

@@ -1,183 +0,0 @@
-"""同步任务：行业聚合特征（衍生计算）。
-
-输入：kline_stock（日 K 线）+ industry（股票→行业映射）
-输出：
-  - sector_indices         ：行业日线（trade_date × sector_name × close × sector_amplitude）
-  - sector_features_daily  ：行业特征（+ sector_ret + ema10/20/200 + score）
-
-算法（参考 dashboard/api/services/sync/sector.py 重写，持久层改为 PG）：
-  1) 每只股票日涨跌幅 pct_chg = (close/prev_close - 1) * 100
-  2) 按行业聚合 sector_ret = 该行业所有股票当日 pct_chg 的均值
-  3) 行业指数 close = 100 × ∏(1 + sector_ret/100) （基点 100，复合收益）
-  4) 行业振幅 sector_amplitude = 该行业所有股票当日 (high-low)/close 的均值
-  5) EMA10/20/200  = close 的指数移动平均（adjust=False）
-  6) score: 0/1/2 — close>ema200 + 1，ema10>ema20 + 1
-
-无外部 API 调用，纯本地计算；增量逻辑以"全量重算"实现（数据量小 ~75K 行）。
-"""
-from __future__ import annotations
-
-import time
-from datetime import datetime, timedelta
-from typing import Any
-
-import pandas as pd
-from sqlalchemy import create_engine
-
-from app.core.config import settings
-from app.core.db import ops as db_ops
-from app.core.sync.base import SyncTask
-from app.core.utils.logging import get_logger
-
-logger = get_logger("sync.sector_features")
-
-
-# 起算点：DB 里 kline_stock 最早一天 往前 5 年，或固定一个默认起点
-# （不用写死 2021-04-24，让数据自己决定；下面自动算）
-DEFAULT_START_YEARS_BACK = 5
-
-# 输出基点（行业指数 close 从多少开始）
-INDEX_BASE = 100.0
-
-
-class SyncSectorFeatures(SyncTask):
-    dataset_id = "sector_features"
-
-    def _run(
-        self,
-        *,
-        trigger_source: str = "manual",
-        codes: list[str] | None = None,
-        max_workers: int = 1,  # 本任务纯本地计算，单线程足够
-        **kwargs,
-    ) -> dict[str, Any]:
-        t0 = time.time()
-        logger.info("[sector] 启动行业聚合特征计算")
-
-        # ── 1) 拉 kline_stock（日线）只取需要的列 ──
-        # 走 SQLAlchemy engine + PG URL（pd.read_sql 一次性读 1.1kw 行 → pandas DataFrame）
-        engine = create_engine(settings.pg_sqlalchemy_url())
-        df_kline = pd.read_sql(
-            'SELECT stock_code, trade_date, open, high, low, "close", volume '
-            'FROM market_data.kline_stock ORDER BY stock_code, trade_date',
-            engine,
-        )
-        if df_kline.empty:
-            return {"status": "error", "message": "kline_stock 为空"}
-        df_kline["stock_code"] = df_kline["stock_code"].astype(str).str.zfill(6)
-        df_kline["trade_date"] = pd.to_datetime(df_kline["trade_date"], errors="coerce")
-        logger.info(f"[sector] kline_stock: {len(df_kline):,} 行, {df_kline['stock_code'].nunique()} 只, "
-                    f"{df_kline['trade_date'].min().date()} ~ {df_kline['trade_date'].max().date()}")
-
-        # ── 2) 拉 industry（股票→行业映射）──
-        df_ind = pd.read_sql(
-            "SELECT code, industry_name FROM market_data.industry WHERE industry_name IS NOT NULL",
-            engine,
-        )
-        if df_ind.empty:
-            return {"status": "error", "message": "industry 表为空"}
-        df_ind["code"] = df_ind["code"].astype(str).str.zfill(6)
-        df_ind = df_ind.drop_duplicates(subset=["code"], keep="first")
-        logger.info(f"[sector] industry 映射: {len(df_ind):,} 行, {df_ind['industry_name'].nunique()} 个行业")
-
-        # ── 3) 算每只股票日涨跌幅 + 振幅 ──
-        df = df_kline.merge(df_ind, left_on="stock_code", right_on="code", how="inner")
-        df = df.sort_values(["stock_code", "trade_date"])
-        df["prev_close"] = df.groupby("stock_code")["close"].shift(1)
-        df["stock_amplitude"] = (df["high"] - df["low"]) / df["close"]
-        df = df.dropna(subset=["prev_close", "industry_name"])
-        df["pct_chg"] = (df["close"] / df["prev_close"] - 1.0) * 100.0
-        logger.info(f"[sector] 合并后: {len(df):,} 行, 行业 {df['industry_name'].nunique()} 个")
-
-        # ── 4) 按行业 + 日期聚合 ──
-        sector_daily = (
-            df.groupby(["industry_name", "trade_date"], as_index=False).agg(
-                sector_ret=("pct_chg", "mean"),
-                sector_amplitude=("stock_amplitude", "mean"),
-            )
-            .rename(columns={"industry_name": "sector_name"})
-            .sort_values(["sector_name", "trade_date"])
-            .reset_index(drop=True)
-        )
-        logger.info(f"[sector] 行业日聚合: {len(sector_daily):,} 行")
-
-        # ── 5) 算行业指数 close（基点 100，复合）──
-        sector_index = self._build_index(sector_daily)
-
-        # ── 6) EMA + score ──
-        sector_index = self._calc_ema(sector_index)
-        sector_index["score"] = sector_index.apply(
-            lambda r: self._calc_score(r["close"], r["ema10"], r["ema20"], r["ema200"]),
-            axis=1,
-        )
-
-        # 整理列名
-        out_cols = ["trade_date", "sector_name", "sector_ret", "sector_amplitude",
-                    "close", "ema10", "ema20", "ema200", "score"]
-        sector_index = sector_index[out_cols]
-        sector_index["trade_date"] = sector_index["trade_date"].dt.strftime("%Y-%m-%d")
-        # NaN → None（让 MySQL 接受 NULL）
-        sector_index = sector_index.where(pd.notnull(sector_index), None)
-
-        # ── 7) 写入两张表 ──
-        # 7a. sector_indices (4 列)
-        si_rows = sector_index[["trade_date", "sector_name", "close", "sector_amplitude"]].to_dict("records")
-        db_ops.replace_all_sector_indices(si_rows)
-        # 7b. sector_features_daily (9 列)
-        sf_rows = sector_index.to_dict("records")
-        db_ops.replace_all_sector_features(sf_rows)
-
-        elapsed = round(time.time() - t0, 1)
-        msg = (
-            f"行业聚合 {sector_index['sector_name'].nunique()} 个行业 "
-            f"× {sector_index['trade_date'].nunique()} 天, "
-            f"共 {len(sector_index):,} 行, {elapsed}s"
-        )
-        logger.info(f"[sector] {msg}")
-        return {
-            "status": "ok",
-            "message": msg,
-            "sectors": int(sector_index["sector_name"].nunique()),
-            "days": int(sector_index["trade_date"].nunique()),
-            "rows": int(len(sector_index)),
-            "elapsed_sec": elapsed,
-        }
-
-    @staticmethod
-    def _build_index(sector_daily: pd.DataFrame) -> pd.DataFrame:
-        """行业指数 close = INDEX_BASE × ∏(1 + sector_ret/100)"""
-        out = []
-        for sector_name, group in sector_daily.groupby("sector_name", sort=False):
-            g = group.sort_values("trade_date").copy()
-            close_vals = []
-            cur_base = INDEX_BASE
-            for ret in g["sector_ret"].to_numpy(dtype=float):
-                if pd.isna(ret):
-                    close_vals.append(cur_base)
-                else:
-                    cur_base = cur_base * (1 + float(ret) / 100.0)
-                    close_vals.append(cur_base)
-            g["close"] = close_vals
-            out.append(g)
-        return pd.concat(out, ignore_index=True)
-
-    @staticmethod
-    def _calc_ema(sector_index: pd.DataFrame) -> pd.DataFrame:
-        """每个行业分别算 EMA10/20/200"""
-        out = []
-        for sector_name, group in sector_index.groupby("sector_name", sort=False):
-            g = group.sort_values("trade_date").copy()
-            g["ema10"] = g["close"].ewm(span=10, adjust=False).mean()
-            g["ema20"] = g["close"].ewm(span=20, adjust=False).mean()
-            g["ema200"] = g["close"].ewm(span=200, adjust=False).mean()
-            out.append(g)
-        return pd.concat(out, ignore_index=True)
-
-    @staticmethod
-    def _calc_score(close, ema10, ema20, ema200) -> int:
-        score = 0
-        if pd.notna(close) and pd.notna(ema200) and close > ema200:
-            score += 1
-        if pd.notna(ema10) and pd.notna(ema20) and ema10 > ema20:
-            score += 1
-        return int(score)

app/tasks/task_share_snapshot.py

@@ -91,16 +91,25 @@ class SyncShareSnapshot(SyncTask):
                 # 与 stocks.code 保持一致，方便 JOIN。
                 # 历史 bug：之前 share 表写的是裸 6 位 c6，导致 share JOIN stocks 失败。
                 # 一次性 UPDATE（2026-07-01）已把存量 6.3 万行加上前缀。
+                # trade_date 必须用源数据返回的交易日(数据源层负责"最近已完成交易日"
+                # 兜底),不允许用 today 兜底 — 7月9日 早盘跑任务时 today=7月9日,
+                # 但股本快照实际对应 7月8日 收盘。早期代码用 today 会让 share 表里
+                # 同一股本被重复写多行,统计/回溯出错。
+                trade_date = r.get("trade_date")
+                if not trade_date:
+                    fail_cnt += 1
+                    logger.warning(f"[share {c6}] 源数据未返回 trade_date,跳过(避免用运行日兜底)")
+                    continue
                 hermes = self._to_hermes(c6)
                 db_ops.update_stock_share_snapshot(
                     code=hermes,
                     total_share=r["total_share"],
                     float_share=r["float_share"],
-                    trade_date=r.get("trade_date", today),
+                    trade_date=trade_date,
                 )
                 rows_to_share_table.append({
                     "stock_code": hermes,
-                    "trade_date": r.get("trade_date", today),
+                    "trade_date": trade_date,
                     "total_share": r["total_share"],
                     "float_share": r["float_share"],
                 })

app/tasks/task_stocks_basic.py

@@ -104,7 +104,6 @@ class SyncStocksBasic(SyncTask):
                     exchange=old.get("exchange", ""),
                     list_date=old.get("list_date", ""),
                     listing_status="delisted",
-                    industry=old.get("industry", ""),
                 )
                 delisted += 1
 
@@ -197,11 +196,19 @@ class SyncStocksBasic(SyncTask):
                 if r is None:
                     fail += 1
                 else:
+                    # trade_date 必须是源数据返回的交易日(数据源层用 effective_market_date
+                    # 兜底为"最近已完成交易日"),绝不允许用 today 兜底 — 7月9日 早盘跑
+                    # 时 today=7月9日,但股本快照实际对应 7月8日 收盘。
+                    trade_date = r.get("trade_date")
+                    if not trade_date:
+                        fail += 1
+                        logger.warning(f"[share {c6}] 源数据未返回 trade_date,跳过(避免用运行日兜底)")
+                        continue
                     db_ops.update_stock_share_snapshot(
                         code=self._to_hermes(c6),
                         total_share=r["total_share"],
                         float_share=r["float_share"],
-                        trade_date=r.get("trade_date", today),
+                        trade_date=trade_date,
                     )
                     ok += 1
                 if done % 100 == 0:

app/tasks/task_tick_trade.py

@@ -12,12 +12,15 @@
      trade_date 可能比 wall-clock 早一天；重跑靠 PK + ON CONFLICT 幂等
   4) 单只股票日均 5w-10w tick，CHUNK=2000 upsert 防 driver 撑爆
   5) 数据是"当天 only"，无 start/end 窗口；不存增量概念
+  6) 2026-07-13 加固: 加上外层 FETCH_HARD_TIMEOUT,避免 mairui 挂起时
+     as_completed 无 timeout 导致整个 task 永久卡住(systemd 2h 超时杀)。
+     改用手动 pool 管理 + shutdown(wait=False),与 kline_daily 对齐。
 """
 from __future__ import annotations
 
 import os
 import time
-from concurrent.futures import ThreadPoolExecutor, as_completed
+from concurrent.futures import ThreadPoolExecutor, as_completed, TimeoutError as FuturesTimeoutError
 from datetime import datetime
 from typing import Any
 
@@ -36,6 +39,10 @@ MAIRUI_TICK_TRADE_PUBLISH_HOUR = 21
 MAIRUI_TICK_TRADE_PUBLISH_MIN = 0
 # 21:00 之后再跑（留 5 分钟缓冲，等 mairui 完整入库）
 TICK_TRADE_RUN_GATE = (MAIRUI_TICK_TRADE_PUBLISH_HOUR, MAIRUI_TICK_TRADE_PUBLISH_MIN + 5)
+# 外层超时: mairui _fetch 有 20s timeout,但 as_completed 无 timeout 的话
+# 若所有 worker 同时被 mairui 挂起(罕见),task 会永久卡住等 systemd 2h 杀。
+# 120s 内无任何 future 完成 → 判全部失败退出。
+FETCH_HARD_TIMEOUT = 120.0
 
 
 def _passes_publish_gate(now: Optional[datetime] = None, *, force: bool = False) -> tuple[bool, str]:
@@ -108,15 +115,18 @@ class SyncTickTrade(SyncTask):
         )
 
         # ── 执行 ──
+        # 不用 `with ThreadPoolExecutor` — 与外层 as_completed timeout 配合,
+        # 超时后 `with` 的 shutdown(wait=True) 会阻塞等卡死线程 → 进程挂死(同 kline_daily 教训)。
         t0 = time.time()
         ok_cnt = fail_cnt = 0
         rows_total = 0
-        with ThreadPoolExecutor(max_workers=max_workers) as pool:
-            futures = {pool.submit(self._sync_one, mr, c6): c6 for c6 in stock_codes}
-            for i, future in enumerate(as_completed(futures), 1):
+        pool = ThreadPoolExecutor(max_workers=max_workers)
+        futures = {pool.submit(self._sync_one, mr, c6): c6 for c6 in stock_codes}
+        try:
+            for i, future in enumerate(as_completed(futures, timeout=FETCH_HARD_TIMEOUT), 1):
                 c6 = futures[future]
                 try:
-                    res = future.result()
+                    res = future.result(timeout=0.1)
                     if res["status"] == "ok":
                         ok_cnt += 1
                         rows_total += res["rows"]
@@ -124,6 +134,9 @@ class SyncTickTrade(SyncTask):
                         fail_cnt += 1
                         if res.get("error"):
                             logger.warning(f"[tick_trade {c6}] {res['error']}")
+                except FuturesTimeoutError:
+                    fail_cnt += 1
+                    logger.warning(f"[tick_trade {c6}] 内部 race timeout, 记 fail")
                 except Exception as e:
                     fail_cnt += 1
                     logger.warning(f"[tick_trade {c6}] {e}")
@@ -132,6 +145,25 @@ class SyncTickTrade(SyncTask):
                         message=f"tick_trade 进度 {i}/{total} OK:{ok_cnt} FAIL:{fail_cnt} 行:{rows_total}",
                         current=i, total=total, current_step=c6,
                     )
+        except FuturesTimeoutError:
+            stuck = [(f, c6) for f, c6 in futures.items() if not f.done()]
+            logger.error(
+                "[tick_trade] 所有 fetcher 卡死 (>%ss), %d 只股票未完成, 标 fail 后退出",
+                FETCH_HARD_TIMEOUT, len(stuck),
+            )
+            fail_cnt += len(stuck)
+            for f, c6 in stuck:
+                try:
+                    if hasattr(f, "cancel"):
+                        f.cancel()
+                except Exception as e:
+                    logger.warning(f"[tick_trade] cancel future for {c6} 失败: {e}")
+            try:
+                pool.shutdown(wait=False)
+            except Exception as e:
+                logger.warning(f"[tick_trade] pool.shutdown(wait=False) 失败: {e}")
+        else:
+            pool.shutdown(wait=True)
         elapsed = round(time.time() - t0, 1)
         msg = f"逐笔 {ok_cnt}成 {fail_cnt}败 共{rows_total}行, {elapsed}s"
         return {