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

AGENTS.md

@@ -0,0 +1,302 @@
+# Agent 工作记忆
+
+## 活跃监视
+
+task_watch 在后台运行（PID 见 `logs/task_watch.log`），持续跟踪运行中的 task：
+- 查询 `dataset_registry` 中 status='running' 的任务
+- 解析 config 表 lastMessage 中的耗时数据作为历史基准
+- 自适应检查间隔（剩余时间 × 0.15，clamp 15s-300s，无历史默认 60s）
+
+## 待推进的优化（按优先级）
+
+### P0 - 高优
+
+- **task_watch 耗时记录入库**：当前 watch 只读历史不做记录。任务完成后应将 `dataset_registry.message` 中的 elapsed_sec 写入 config 表（或新建 `run_history` 表），供下次估算 ETA 使用。
+- **task_mairui_indicators 外层超时**：与 kline_daily/kline_5min 对齐，给 `as_completed` 加上外层 timeout 兜底（`app/tasks/task_mairui_indicators.py:227`）。
+
+### P1 - 中优
+
+- **节假日自动更新**：`trading_calendar_holidays` 当前硬编码在 config 表。建议接入 https://github.com/shengshiyan/dataconf 或类似日历 API 自动拉取每年交易日历。
+- **`daily_sync_check` 失败告警**：昨日巡检因 `FAILURE_WEBHOOK_URL/SECRET` 未配置而 exit 1（期望的失败）。需确定是否要配 webhook，或改为非致命 warning。
+- **`kline_5min` mairui 偶发全 worker 挂起**：今天 recovery 4734 只失败发生在 mairui 单次 120s 全 hang。`_fetch` 已有 20s timeout + 重试，但仍抗不住 mairui 服务端全挂。建议监控 mairui 健康状态 + 失败股票重入队列。
+
+### P2 - 低优
+
+- **MySQL 残留配置**：`.env` 仍包含 `DB_BACKEND=mysql` 和 `MYSQL_*` 变量，config.py 已只支持 PG。可清理。
+- **`stock_node` 上次 warning**：上次运行 7/10 返回 warning（41/1164 节点失败），需关注是否偶发。
+- **`kline_5min` 大量 fail**：今日 recovery 4734 只失败，需单独排查 mairui 数据源。
+
+## 设计决策记录
+
+- **systemd timer 为主，进程内 scheduler 为辅**：宿主机直接部署用 systemd mode，Docker 部署用 docker mode。二者通过 `RUNTIME_MODE` 隔离。
+- **After= 只依赖 network-online.target**：service 间不设 After= 依赖链，避免一个任务阻塞后续所有任务。
+- **fail-open 策略**：节假日不过滤，让 task 在非交易日返回空数据报 warning 但不阻塞。周末由 systemd OnCalendar=Mon..Fri 保护。
+
+## 数据集来源与依赖关系
+
+> 数据来源与依赖关系是调度正确性的基础。原始源依赖外部 API；衍生源只读本地 PG，依赖上游原始源。
+
+### 数据集清单
+
+| dataset_id | 目标表 | 类型 | 来源 | 说明 |
+|---|---|---|---|---|
+| `stock_basic` | `market_data.stocks` | 原始源 | 麦蕊 `/hslt/list`（主）+ Baostock `query_stock_basic`（备） | 全市场 A 股代码/名称/交易所/上市状态；`stock_basic` 自身可附带雪球股本快照 |
+| `kline_daily` | `market_data.kline_stock` | 原始源 | 雪球 `kline`（主）→ 麦蕊 `hsstock/history/*/d/n` → 新浪 `getKLineData`（备） | 全市场日 K OHLCV；支持 per-stock fallback |
+| `kline_index` | `market_data.kline_index` + `indices` | 原始源 | 麦蕊 `hsindex/history`（主）→ 新浪指数日 K（备） | 上证/深证/创业板/沪深300/中证500/中证1000 |
+| `kline_5min` | `market_data.kline_5min` | 原始源 | 麦蕊 `hsstock/history/*/5/n` | 全市场 5 分钟 K；历史深度 2023-06-14 起 |
+| `tick_trade` | `market_data.tick_trade` | 原始源 | 麦蕊 `hsrl/zbjy` | 当天逐笔成交；每日 21:00 后发布 |
+| `moneyflow` | `market_data.moneyflow` | 原始源 | 麦蕊 `hsstock/history/transaction` | 主力/大/中/小单净额；每日 21:30 后发布 |
+| `longhubang` | `market_data.longhubang_daily` + `longhubang_seat` | 原始源 | akshare `stock_lhb_detail_em` / `stock_lhb_stock_detail_em`（东方财富封装） | 龙虎榜聚合层 + 席位层；每日 22:00 触发 |
+| `stock_node` | `market_data.node_categories` + `nodes` + `stock_node_map` | 原始源 | 麦蕊 `/hszg/{list,gg}` | 股票-指数/行业/概念映射；每周六 11:30 |
+| `mairui_indicators` | `market_data.kline_stock_macd_daily` + `kdj_daily` + `boll_daily` | 原始源 | 麦蕊 `hsstock/history/{macd,kdj,boll}` | 日 K MACD/KDJ/BOLL；增量拉取 |
+| `share_snapshot` | `market_data.stocks` + `share` | 原始源 | 雪球 `quote_detail` | 最新总股本/流通股本；需 `XUEQIU_TOKEN` |
+| `market_regime` | `market_data.market_regime_daily` | 衍生源 | 本地计算（`kline_stock`） | 涨跌家数/ advance_ratio / 恐慌标记 |
+| `mairui_ma_daily` | `market_data.kline_stock_ma_daily` | 衍生源 | 本地计算（`kline_stock.close`） | MA5/10/20/60；未来可切麦蕊 `/d/maN`（需付费 licence） |
+
+### 依赖关系
+
+```
+stock_basic ─┬─► kline_daily ─┬─► market_regime
+             │                └─► mairui_ma_daily
+             ├─► kline_5min
+             ├─► tick_trade
+             ├─► moneyflow
+             ├─► longhubang
+             ├─► stock_node
+             └─► share_snapshot
+
+kline_index （独立，无上游依赖）
+```
+
+- **根任务**：`stock_basic` 是大多数任务的根，提供股票代码列表。`kline_index` 独立运行。
+- **关键链**：`kline_daily` 是 2 个衍生源（`market_regime`、`mairui_ma_daily`）的实际数据基础。
+- **注意**：`mairui_ma_daily` 在 `app/core/sync/registry.py` 中声明依赖 `kline_daily`，但只有**调度顺序**约束；若 `kline_daily` 运行后数据又被回填（如 2026-07-15 场景），`mairui_ma_daily` 不会自动重算，需手动触发或等下次调度。
+
+### 默认调度时间（工作日）
+
+| 时间 | dataset_id | 备注 |
+|---|---|---|
+| 09:00 | `stock_basic` | 开盘前刷新股票列表 |
+| 15:30 | `kline_index` | 指数日 K |
+| 15:40 | `kline_daily` | 个股日 K |
+| 16:00 | `kline_5min` | 5 分钟 K |
+| 15:15 | `market_regime` | 市场情绪 |
+| 16:30 | `mairui_ma_daily` | 日 K MA（本地计算） |
+| 16:40 | `mairui_indicators` | MACD/KDJ/BOLL |
+| 21:35 | `moneyflow` | 资金流（等 21:30 发布） |
+| 22:00 | `longhubang` | 龙虎榜（等 19:00-21:00 出齐） |
+| 周六 11:30 | `stock_node` | 节点映射 |
+| 周六 11:30 | `share_snapshot` | 股本快照（雪球，周度） |
+
+### systemd timer 与数据依赖对齐审核
+
+总体：**基本对齐**，但存在 3 个需要注意的细节。
+
+1. **`mairui_indicators.service` 的 `After=` 多余依赖** ✅ 已修复
+   - 修复前：`After=network-online.target market-sync.service market-sync-mairui-ma-daily.service`
+   - 修复后：`After=network-online.target market-sync.service`
+   - 文件：`deploy/systemd/units/market-sync-mairui-indicators.service`、`bin/systemd/market-sync-mairui-indicators.service`
+   - 待部署：/etc/systemd/system/ 仍保留旧版，需 `sudo bash deploy/systemd/deploy.sh`
+
+2. **`stock_node.service` 未声明对 `stock_basic` 的依赖** ✅ 已修复
+   - 修复前：`After=network-online.target`
+   - 修复后：`After=network-online.target market-sync-morning.service`
+   - 文件：`deploy/systemd/units/market-sync-stock-node.service`、`bin/systemd/market-sync-stock-node.service`
+   - 待部署：/etc/systemd/system/ 仍保留旧版，需 `sudo bash deploy/systemd/deploy.sh`
+
+3. **`After=` 只保证启动顺序，不保证上游成功**
+   - systemd 的 `After=` 不会让下游等待上游**成功完成**，仅控制 timer 触发后的排队顺序。
+   - 如果 `market-sync.service` 在 15:30 启动后因 `kline_daily` 卡住而 3h 超时失败，`mairui_ma_daily` 仍会在 16:30 启动（因为 timer 已触发）。
+   - 结果：`mairui_ma_daily` 可能基于不完整的 `kline_stock` 计算，产生类似 2026-07-15 的 3,404 行缺失。
+
+### 数据回填级联规则
+
+> **核心原则**：上游数据被回填或修复后，下游依赖它的衍生数据集必须手动/自动重跑，否则会出现“上游新、下游旧”的不一致。
+
+| 上游任务/表 | 触发条件 | 必须重跑的下游任务 |
+|---|---|---|
+| `kline_daily` / `kline_stock` | 回填历史 K 线、修复错误日线、补充漏掉的股票/日期 | `market_regime`、`mairui_ma_daily` |
+| `stock_basic` / `stocks` | 新上市/退市股票、代码变更、上市状态修正 | `kline_daily`、`kline_5min`、`tick_trade`、`moneyflow`、`longhubang`、`stock_node`、`share_snapshot` |
+| `kline_stock`（任意修复） | close/volume 等核心字段修正 | `market_regime`、`mairui_ma_daily` |
+
+**操作建议：**
+
+- 单次少量回填（如几只股票、几天）：用 CLI 参数指定 `codes` / `start` / `end`，然后按上表手动触发下游。
+- 大量回填（如全市场、多月/多年历史）：先跑上游，再按依赖链顺序跑下游；必要时禁用当日独立 timer，避免与 runall 并发。
+- 每日巡检（`bin/daily_sync_check.py`）已增加数据一致性检查：对比 `kline_stock` 与 `kline_stock_ma_daily`、`market_regime_daily` 的最新日期与缺失行数，发现缺口即告警。
+
+## 历史问题与解决思路
+
+> 记录真实故障案例、排查路径和修复方法，供后续复察或参考。
+
+### 案例 1：2026-07-15 `kline_stock_ma_daily` 缺失 3,404 行
+
+**现象**
+- `kline_stock` 11,728,408 行，`kline_stock_ma_daily` 只有 11,725,004 行，相差 3,404 行。
+- 差异集中在 2026-07-14（3,400 只股票）+ SH688287 的 4 天（2026-05-29/06-04/06-05/06-08）。
+- `dataset_registry` 中 `mairui_ma_daily` 状态为 `success`，`last_success_at=2026-07-14 17:11:48 UTC`。
+
+**根因**
+- `mairui_ma_daily` 在 2026-07-14 17:11 完成时，`kline_stock` 的 2026-07-14 数据还不完整（少 3,400 只）。
+- `kline_daily` 在 2026-07-15 10:04 又跑了一次，补充了这 3,400 只 2026-07-14 的 K 线。
+- `mairui_ma_daily` 没有自动感知上游变化，导致 MA 表缺了这 3,404 行。
+
+**排查命令**
+```bash
+# 1. 对比行数
+SELECT COUNT(*) FROM market_data.kline_stock;
+SELECT COUNT(*) FROM market_data.kline_stock_ma_daily;
+
+# 2. 查缺失日期分布
+SELECT trade_date, COUNT(*) FROM (
+    SELECT stock_code, trade_date FROM market_data.kline_stock
+    EXCEPT
+    SELECT stock_code, trade_date FROM market_data.kline_stock_ma_daily
+) t GROUP BY trade_date ORDER BY trade_date;
+
+# 3. 查 dataset_registry 时间线
+SELECT dataset_id, last_success_at, finished_at, message
+FROM market_data.dataset_registry
+WHERE dataset_id IN ('kline_daily', 'mairui_ma_daily');
+```
+
+**修复方法**
+```bash
+# 手动重跑 mairui_ma_daily（全量重算 ~22min）
+bin/market_sync_mairui_ma_daily_run.sh
+```
+
+**预防/改进**
+- 在 `bin/daily_sync_check.py` 中增加数据一致性检查：每日 23:00 对比 `kline_stock` 与 `kline_stock_ma_daily` 的覆盖差异。
+- 上游回填历史数据后，按「数据回填级联规则」手动触发下游重跑。
+
+---
+
+### 案例 2：systemd `After=` 依赖设计缺陷
+
+**现象**
+- `market-sync-mairui-indicators.service` 写了 `After=market-sync-mairui-ma-daily.service`。
+- `market-sync-stock-node.service` 没有声明对 `stock_basic` 的依赖。
+
+**根因**
+- 早期设计时对数据依赖关系梳理不够，把无依赖的任务串起来，或遗漏了必要的依赖声明。
+- `systemd` 的 `After=` 只控制**启动顺序**，不检查上游任务是否**成功完成**。
+
+**修复方法**
+1. `mairui_indicators.service`：移除 `market-sync-mairui-ma-daily.service`，改为 `After=network-online.target market-sync.service`。
+2. `stock_node.service`：增加 `After=market-sync-morning.service`，确保 `stock_basic` 已更新。
+3. 同步修改 `deploy/systemd/units/` 和 `bin/systemd/` 下的对应文件，并执行：
+   ```bash
+   sudo bash deploy/systemd/deploy.sh
+   ```
+
+**设计原则**
+- `After=` 只声明最小必要依赖：network-online.target + 直接上游 service。
+- 不要为了让下游等上游成功而过度串 service；成功/失败应通过 `daily_sync_check` 和回填规则兜底。
+
+---
+
+### 案例 3：`dataset_registry` 状态卡在 `running`
+
+**现象**
+- `mairui_ma_daily` 的 `sync_history` 已记录成功，但 `dataset_registry.status='running'`，`finished_at=NULL`。
+- `ps` 中无对应进程。
+
+**根因**
+- 状态更新与 `sync_history` 写入不在同一事务；极端情况下任务进程异常退出，导致 `mark_sync_success` 未执行。
+- 或并发/重入导致后一次 `mark_sync_running` 覆盖了前一次的成功状态。
+
+**排查命令**
+```bash
+SELECT dataset_id, status, started_at, finished_at, last_success_at, message
+FROM market_data.dataset_registry
+WHERE dataset_id = 'mairui_ma_daily';
+
+SELECT * FROM market_data.sync_history
+WHERE dataset_id = 'mairui_ma_daily'
+ORDER BY started_at DESC LIMIT 5;
+```
+
+**修复方法**
+```python
+from app.core.db import ops as db_ops
+from datetime import datetime
+
+db_ops.update_dataset_registry_state(
+    'mairui_ma_daily',
+    status='success',
+    finished_at='2026-07-15 12:02:21',
+    last_success_at='2026-07-15 12:02:21',
+    message='MA [5, 10, 20, 60] 共 11,728,408 行, 5511 只, 1349.2s',
+    last_error=None,
+    needs_resync=0,
+    progress_current=0,
+    progress_total=0,
+    current_step='',
+    updated_at=datetime.now().strftime('%Y-%m-%d %H:%M:%S'),
+)
+```
+
+**改进建议**
+- 启动 `runall_once.py` 时自动调用 `recover_interrupted_dataset_registry()` 清理卡死状态（已实现）。
+- `task_watch` 可升级为：发现任务 `running > 2× 历史平均耗时` 且无进程时，自动标记为 failed。
+
+## Market Supervisor 工作流
+
+> 本项目的 Agent 不仅是 code agent，也是 market data supervisor。修改代码/配置后必须完成部署与验证闭环，不能停留在 repo 层面。
+
+### Agent 职责
+
+1. **主动巡检**：定期/按需运行 `bin/daily_sync_check.py --report-only`，确认 dataset 状态、systemd drift、数据一致性。
+2. **问题闭环**：发现告警后，定位根因 → 修复代码/config → 部署到 /etc → 验证告警消除。
+3. **项目记忆同步**：把已确定的问题、根因、修复方法、验证命令写回 `AGENTS.md`，供后续复察。
+4. **数据依赖守护**：上游数据回填/修复后，按「数据回填级联规则」触发下游重跑，并验证一致性。
+
+### systemd unit 修改后的标准流程
+
+任何修改了 `deploy/systemd/units/` 或 `bin/systemd/` 下 unit 文件的操作，必须：
+
+1. **保持两边同步**：同时修改 `deploy/systemd/units/` 和 `bin/systemd/` 的对应文件（目前两者是镜像关系）。
+2. **执行 deploy**：
+   ```bash
+   cd /home/gao/Development/quant_home/market_sync
+   sudo bash deploy/systemd/deploy.sh
+   # 或跳过确认：sudo bash deploy/systemd/deploy.sh --yes
+   ```
+3. **验证 drift**：
+   ```bash
+   .venv/bin/python bin/daily_sync_check.py --report-only
+   ```
+   确保 `systemd_drift` 中 `has_drift=false`。
+4. **验证 unit 状态**：
+   ```bash
+   systemctl list-timers market-sync-*
+   systemctl status market-sync-mairui-indicators.service
+   systemctl status market-sync-stock-node.service
+   ```
+
+### 部署验证 Checklist
+
+- [ ] repo unit 与 `/etc/systemd/system/` 内容一致（`daily_sync_check` 无 drift 告警）
+- [ ] `systemctl daemon-reload` 已执行
+- [ ] 相关 timer 仍 enabled（`systemctl is-enabled <name>.timer`）
+- [ ] 修改后的 service 无语法错误（`systemd-analyze verify /etc/systemd/system/<name>.service`）
+- [ ] 数据一致性检查通过（`kline_stock` 与下游表日期对齐）
+
+## 已部署的优化（确认完成）
+
+| 项目 | 日期 | 说明 |
+|------|------|------|
+| `RUNTIME_MODE=systemd` 写入 `.env` | 07-13 | 防止双调度器冲突 |
+| `market-sync-worker.service` mask | 07-13 | systemd 模式下禁用进程内 scheduler |
+| mairui_ma_daily / mairui_indicators timer 部署 | 07-13 | 2 个缺失的 timer 补部署并启用 |
+| `/etc` market-sync.service 漂移修复 | 07-13 | deploy.sh 同步 repo→etc，消除 drift |
+| moneyflow/lhb `After=` 依赖链修复 | 07-13 | 去掉 `After=market-sync-tick.service`，任务不再串行阻塞 |
+| `bin/task_watch.py` 创建 | 07-13 | 自适应间隔的任务跟踪监视器（nohup 后台运行） |
+| `runall_once.py` logger f-string 修复 | 07-13 | 超时日志改为 f-string |
+| `sync/base.py` docstring 更新 | 07-13 | `_effective_sync_end` 注释过期问题修复 |
+| `call_with_timeout` 超时后线程阻塞修复 | 07-13 | 改用 daemon 线程替代 `ThreadPoolExecutor`，避免超时后 `shutdown(wait=True)` 阻塞调用方 |
+| `tick_trade` 外层超时加固 | 07-13 | 加上 FETCH_HARD_TIMEOUT=120s + 手动 pool 管理，防止 mairui 挂起时永久卡住 |
+| `kline_daily` per-stock fallback | 07-13 | 雪球失败自动降级到 mairui/新浪，`fallback_ok` 统计备胎救回数 |
+| `kline_5min` 超时后重入队列 | 07-13 | mairui 全 hang 超时后冷却 30s 重试一次失败股票，救回临时故障 |
+| MCP Server SSE 模式 + systemd 服务 | 07-13 | `market-sync-mcp.service` 开机自启，SSE 端点 `http://127.0.0.1:8101/sse` |

Dockerfile

@@ -1,63 +0,0 @@
-# syntax=docker/dockerfile:1.6
-# ─────────────────────────────────────────────────────────────
-# market_sync 数据同步服务
-# 单镜像同时跑：FastAPI dashboard + 进程内 scheduler（worker）
-# 部署：docker compose up -d
-# ─────────────────────────────────────────────────────────────
-
-# --- builder stage: 装依赖到 venv ---
-FROM python:3.11-slim AS builder
-
-ENV PYTHONUNBUFFERED=1 \
-    PYTHONDONTWRITEBYTECODE=1 \
-    PIP_NO_CACHE_DIR=1 \
-    PIP_DISABLE_PIP_VERSION_CHECK=1
-
-# 仅装编译期需要的系统包（很多 wheel 已预编译，但 psycopg2 / cryptography 可能要）
-RUN apt-get update && apt-get install -y --no-install-recommends \
-        build-essential gcc libpq-dev \
-    && rm -rf /var/lib/apt/lists/*
-
-WORKDIR /app
-
-# 先 copy requirements 单独一层（利用 Docker 缓存：依赖不变就不重装）
-COPY requirements.txt .
-RUN python -m venv /app/.venv \
-    && /app/.venv/bin/pip install --upgrade pip \
-    && /app/.venv/bin/pip install -r requirements.txt
-
-
-# --- runtime stage: 极简基础镜像 + 仅复制 venv 与代码 ---
-FROM python:3.11-slim AS runtime
-
-ENV PYTHONUNBUFFERED=1 \
-    PYTHONDONTWRITEBYTECODE=1 \
-    PATH="/app/.venv/bin:${PATH}" \
-    PYTHONPATH=/app \
-    # service 模式默认监听
-    API_HOST=0.0.0.0 \
-    API_PORT=8100
-
-# runtime 只需要 psycopg2 的运行时库 + tzdata（pandas tz aware 需要）
-RUN apt-get update && apt-get install -y --no-install-recommends \
-        libpq5 tzdata curl \
-    && rm -rf /var/lib/apt/lists/*
-
-WORKDIR /app
-
-# 从 builder 复制 venv（已编译好的依赖）
-COPY --from=builder /app/.venv /app/.venv
-
-# 复制应用代码
-COPY app/ ./app/
-COPY bin/ ./bin/
-
-# 默认启动 service 入口（FastAPI + 进程内 scheduler）
-# 也可覆盖为 cli / worker 单跑某个 task：
-#   docker compose run --rm market_sync python -m app.entrypoints.cli sync kline_daily
-EXPOSE 8100
-
-HEALTHCHECK --interval=60s --timeout=5s --start-period=30s --retries=3 \
-    CMD curl -fsS http://localhost:8100/api/health || exit 1
-
-ENTRYPOINT ["/app/bin/service_run.sh"]

Dockerfile

@@ -0,0 +1 @@
+deploy/docker/Dockerfile

README.md

@@ -2,7 +2,7 @@
 
 A 股市场数据**定时同步与管理**框架。
 
-把 `akshare` / `baostock` / 新浪 / 雪球 / 麦蕊智数 等多个数据源接入到 MySQL 库，提供：
+把 `akshare` / `baostock` / 新浪 / 雪球 / 麦蕊智数 等多个数据源接入到 PostgreSQL 库，提供：
 - 后台线程调度器（time-based + interval-based，交易日感知）
 - 多数据源 + 自动降级 + 健康度监控
 - `dataset_registry` 同步状态机（运行中 / 成功 / 失败 / 中断恢复）
@@ -11,15 +11,69 @@ A 股市场数据**定时同步与管理**框架。
 
 > 📖 详细架构说明见 [docs/ARCHITECTURE.md](docs/ARCHITECTURE.md)
 
-## 快速开始
+## 运行模式
+
+项目支持两种互斥的运行模式，由环境变量 `RUNTIME_MODE` 控制：
+
+| 模式 | 值 | 适用场景 | 调度方式 |
+|---|---|---|---|
+| **Docker 模式** | `docker` | `docker compose up` 单容器部署 | 进程内 scheduler（uvicorn + daemon thread） |
+| **systemd 模式** | `systemd` | Linux 宿主机长期运行 | systemd timer/service |
+
+> ⚠️ 两种模式**不能同时启用**，否则同一个 task 会被两边各触发一次，造成状态机抖动和重复 IO。
+
+### Docker 模式
 
 ```bash
 # 1. 准备环境（首次）
-cp .env.example .env                 # 编辑 .env 填写真实 MySQL 凭据 + XUEQIU_TOKEN
+cp .env.example .env                 # 编辑 .env 填写真实 PG 凭据 + XUEQIU_TOKEN + MAIRUI_LICENCE
-chmod +x start.sh
-./start.sh                           # 自动创建 .venv + 装依赖 + 启动 API（http://localhost:8100）
 
-# 2. 常用命令
+# 2. 启动（包含 postgres + market_sync 服务）
+docker compose up -d
+
+# 3. 查看日志
+docker compose logs -f market_sync
+
+# 4. 手动触发一次任务
+docker compose exec market_sync python -m app.entrypoints.cli sync kline_daily
+```
+
+Docker 模式下 `RUNTIME_MODE=docker` 已写在 `deploy/docker/docker-compose.yml` 中，容器内会启动进程内 scheduler 负责所有同步任务。
+
+### systemd 模式
+
+```bash
+# 1. 准备环境（首次）
+cp .env.example .env                 # 编辑 .env 填写真实 PG 凭据
+# 并在 .env 中加上 RUNTIME_MODE=systemd
+
+# 2. 部署 systemd units
+sudo bash deploy/systemd/deploy.sh
+
+# 3. 查看 timer
+systemctl list-timers market-sync*
+
+# 4. 查看 worker / 任务日志
+journalctl -u market-sync -f
+journalctl -u market-sync-moneyflow -f
+```
+
+systemd 模式下：
+- 所有 14 个 task 都由 systemd timer 触发
+- `market-sync-worker.service` 会被部署脚本自动 **stop / disable / mask**，避免进程内 scheduler 与 systemd timers 重叠
+- 每日 23:00 `market-sync-daily-check.timer` 跑巡检，失败时 POST webhook
+
+### 本地开发模式
+
+```bash
+# 1. 准备环境（首次）
+cp .env.example .env                 # 编辑 .env 填写凭据
+chmod +x start.sh
+
+# 2. 启动 API + scheduler（默认 docker 行为，等价于本地单进程）
+./start.sh                           # http://localhost:8100
+
+# 3. 常用命令
 ./start.sh worker                    # 仅启动调度器（无 web）
 ./start.sh list                      # 列出所有同步任务
 ./start.sh sync kline_daily          # 手动触发一次（dataset_id 即可）
@@ -27,39 +81,54 @@ chmod +x start.sh
 ./start.sh datasources               # 查看数据源健康度
 ./start.sh reseed                    # 重新 seed 默认配置到 config 表
 
-# 3. 一键全量同步（首次接入）
+# 4. 一键全量同步（首次接入）
 python bin/runall_once.py
 
-# 4. 跑测试
+# 5. 跑测试
 .venv/bin/pytest tests/ -v
 ```
 
 ## 项目结构
 
 ```
-app/
+app/                          # 应用源码
-├── core/        # 核心抽象（config / db / sync / scheduler / datasource base）
+├── core/                     # 核心抽象（config / db / sync / scheduler / datasource base）
-├── sources/     # 数据源实现（5 个）
+├── sources/                  # 数据源实现
-├── tasks/       # 同步任务实现（8 个）
+├── tasks/                    # 同步任务实现（14 个）
-├── api/         # FastAPI 管理接口
+├── api/                      # FastAPI 管理接口
-└── entrypoints/ # 进程入口（cli + worker）
+└── entrypoints/              # 进程入口（cli + worker）
-bin/             # 一次性脚本
+bin/                          # 通用脚本（运行入口、CLI、一次性任务）
-docs/            # 架构文档
+config/                       # 配置文件、secrets
-tests/           # 单元测试
+deploy/                       # 部署相关
+├── docker/                   # Dockerfile + docker-compose.yml
+└── systemd/                  # systemd units + 部署脚本
+│   ├── units/                # *.service / *.timer
+│   └── deploy.sh             # 一键部署脚本
+docs/                         # 架构文档
+logs/                         # 日志输出
+tests/                        # 单元测试
+Dockerfile -> deploy/docker/Dockerfile
+docker-compose.yml -> deploy/docker/docker-compose.yml
+bin/deploy_systemd.sh -> deploy/systemd/deploy.sh
+bin/systemd_deploy.sh -> deploy/systemd/deploy.sh
 ```
 
 ## 内置同步任务
 
-| dataset_id | 触发时间 | 说明 |
+| dataset_id | systemd 触发时间 | Docker scheduler 时间 | 说明 |
-|---|---|---|
+|---|---|---|---|
-| `stock_basic` | 交易日 09:00 | 全市场股票基础信息 + 股本 |
+| `stock_basic` | 工作日 09:00 | 09:00 | 全市场股票基础信息 + 股本 |
-| `kline_daily` | 交易日 15:40 | 增量日 K 线（OHLCV，多源交叉） |
+| `kline_index` | 工作日 15:30 | 15:30 | 六大指数日线 |
-| `kline_index` | 交易日 15:50 | 六大指数日线 |
+| `kline_daily` | 工作日 15:30（runall）| 15:40 | 增量日 K 线（OHLCV，多源交叉） |
-| `kline_5min` | 交易日 16:00 | 增量 5 分钟 K 线 |
+| `kline_5min` | 工作日 15:30（runall）| 16:00 | 增量 5 分钟 K 线 |
-| `moneyflow` | 交易日 16:30 | 资金流（主力/大/中/小单净额） |
+| `market_regime` | 工作日 15:30（runall）| 16:15 | 市场情绪（衍生源） |
-| `industry_sector` | 周一 09:30 | 股票-行业映射（Baostock） |
+| `share_snapshot` | 工作日 15:30（runall）| 16:20 | 股本快照（雪球） |
-| `share_snapshot` | 交易日 09:30 | 股本快照（雪球） |
+| `mairui_ma_daily` | 工作日 16:30 | 16:30 | 日 K MA5/10/20/60（本地派生） |
-| `market_regime` | 交易日 16:00 | 市场情绪（衍生源） |
+| `mairui_indicators` | 工作日 16:40 | 16:40 | 日 K MACD/KDJ/BOLL（mairui） |
+| `moneyflow` | 工作日 21:35 | 21:35 | 资金流（主力/大/中/小单净额） |
+| `tick_trade` | 工作日 21:05 | 21:05 | 当天逐笔成交（mairui） |
+| `longhubang` | 工作日 22:00 | 22:00 | 龙虎榜（akshare / 东方财富） |
+| `stock_node` | 周六 11:30 | 周六 11:30 | 股票-节点映射（mairui） |
 
 ## 添加新数据源
 
@@ -70,7 +139,9 @@ tests/           # 单元测试
 
 1. 在 `app/tasks/` 新建 `task_xxx.py`，继承 `SyncTask`，设置 `dataset_id`
 2. 在 `app/tasks/__init__.py` 的 `TASKS` 字典注册
-3. （可选）在 `app/core/sync/registry.py` 的 `SYNC_DEFINITIONS` 加条目
+3. 在 `app/core/sync/registry.py` 的 `SYNC_DEFINITIONS` 加条目
+4. 如果使用 systemd 模式，在 `deploy/systemd/units/` 新增对应的 `.service` 和 `.timer`
+5. 如果使用 Docker 模式，在 `app/core/scheduler/scheduler.py` 的 `DEFAULT_SCHEDULES` 新增调度项
 
 ## License
 

app/api/routes/dashboard.py

@@ -84,14 +84,15 @@ def data_stats():
     from sqlalchemy import func, select
     from app.core.db.models import (
         KlineStock, KlineIndex, Kline5Min, Moneyflow, Share, TickTrade,
-        SectorIndices, SectorFeaturesDaily, MarketRegimeDaily,
+        MarketRegimeDaily,
-        LonghubangDaily, LonghubangSeat, Stocks, Industry,
+        LonghubangDaily, LonghubangSeat, Stock,
         NodeCategory, Node, StockNodeMap, KlineStockMADaily,
+        KlineStockMACDDaily, KlineStockKDJDaily, KlineStockBOLLDaily,
     )
     from app.core.db.orm import SessionLocal
 
     targets = [
-        ("stocks", Stocks),
+        ("stocks", Stock),
         ("kline_stock", KlineStock),
         ("kline_index", KlineIndex),
         ("kline_5min", Kline5Min),
@@ -99,15 +100,15 @@ def data_stats():
         ("moneyflow", Moneyflow),
         ("share", Share),
         ("tick_trade", TickTrade),
-        ("sector_indices", SectorIndices),
-        ("sector_features_daily", SectorFeaturesDaily),
         ("market_regime_daily", MarketRegimeDaily),
         ("longhubang_daily", LonghubangDaily),
         ("longhubang_seat", LonghubangSeat),
-        ("industry", Industry),
         ("node_categories", NodeCategory),
         ("nodes", Node),
         ("stock_node_map", StockNodeMap),
+        ("kline_stock_macd_daily", KlineStockMACDDaily),
+        ("kline_stock_kdj_daily", KlineStockKDJDaily),
+        ("kline_stock_boll_daily", KlineStockBOLLDaily),
     ]
     out = []
     with SessionLocal() as s:

app/core/config.py

@@ -30,6 +30,12 @@ if _HAS_DOTENV:
 class Settings(BaseSettings):
     """应用配置。所有字段都可以通过环境变量或 .env 文件覆盖。"""
 
+    # Runtime mode: "docker" | "systemd"
+    #   docker  : 单容器运行，依赖进程内 scheduler 触发所有同步任务
+    #   systemd : Linux 宿主机运行，由 systemd timer/service 触发任务，
+    #             进程内 scheduler 必须关闭，避免双调度器重叠
+    runtime_mode: str = "docker"
+
     # Scheduler
     scheduler_tick_seconds: int = 5
     scheduler_timezone: str = "Asia/Shanghai"

app/core/datasource/utils.py

@@ -1,15 +1,146 @@
 """数据源公共工具：代码转换、K 线标准化、日期过滤。
 
-参考 dashboard/api/services/datasource/fetch_kline.py，但适配新项目。"""
+参考 dashboard/api/services/datasource/fetch_kline.py，但适配新项目。
+"""
 from __future__ import annotations
 
+import logging
 import re
+import threading
+from datetime import datetime, time as dtime, timedelta
+from typing import Any, Callable, TypeVar
 
 import pandas as pd
 
+logger = logging.getLogger("datasource.utils")
+
 KLINE_COLS = ["trade_date", "open", "high", "low", "close", "volume"]
 KLINE_5MIN_COLS = ["bar_time", "open", "high", "low", "close", "volume", "amount", "turnover_rate"]
 
+T = TypeVar("T")
+
+_THREAD_POOL: dict[int, threading.Thread] = {}
+
+
+def call_with_timeout(
+    func: Callable[..., T],
+    *args: Any,
+    timeout: float = 20.0,
+    on_timeout: T | None = None,
+    description: str = "",
+    **kwargs: Any,
+) -> T | None:
+    """用 daemon 线程跑 `func(*args, **kwargs)`,超时返回 `on_timeout`。
+
+    Why: 之前用 ThreadPoolExecutor，但超时后 `future.cancel()` 无法终止
+    已在运行的线程，pool 的 `shutdown(wait=True)` 会等待卡死的函数返回，
+    把 20s 超时变成无限阻塞——这是 kline_daily 3406 只股票批量失败的根因。
+
+    改用 daemon 线程：超时后直接返回，线程成为孤儿进程，不会阻塞任何调用方。
+    daemon=True 确保进程退出时不会等待这些孤儿线程。
+
+    Args:
+        func:        要跑的同步函数
+        *args/**kwargs: 透传给 func
+        timeout:     秒数,默认 20s(对标 mairui/sina 的 urlopen timeout)
+        on_timeout:  超时返回的占位值(默认 None,调用方需处理 None)
+        description: 日志里的调用描述(如 'xueqiu.kline SH600519')
+
+    Returns: func 的返回值,或 on_timeout(超时时)
+    """
+    result: list[T | None | BaseException] = [None]
+    done = threading.Event()
+
+    def _wrapper() -> None:
+        try:
+            result[0] = func(*args, **kwargs)
+        except BaseException as e:
+            result[0] = e
+        finally:
+            done.set()
+
+    t = threading.Thread(
+        target=_wrapper,
+        daemon=True,
+        name=f"ds-timeout-{description or func.__name__}",
+    )
+    t.start()
+
+    if not done.wait(timeout=timeout):
+        logger.warning(
+            "[%s] %s 超时 (>%ss), 返回 on_timeout",
+            description or func.__name__, description or func.__name__, timeout,
+        )
+        return on_timeout
+
+    if isinstance(result[0], BaseException):
+        raise result[0]  # type: ignore[misc]
+    return result[0]
+
+
+def effective_market_date(
+    now: datetime | None = None,
+    holidays: set[str] | None = None,
+) -> str:
+    """返回"数据生效日"(最近已完成交易日)。
+
+    15:30 是 A 股收盘时刻;之后今天的数据才完整,之前应取昨天。
+    然后向前回退直到找到一个交易日（跳过周末和法定假期）。
+    用于数据源给"无明确来源日期"的数据(snowball quote_detail 的股本快照等)
+    打 trade_date 时用 — 永远不要用 datetime.now() 直接当 trade_date。
+
+    Why: 任务运行日不等于数据交易日。例如 7月9日 早盘 9:30 跑 share_snapshot,
+    此时拿到的股本快照实际对应 7月8日 收盘,不能标 7月9日。早期代码用
+    `datetime.now().strftime("%Y-%m-%d")` 导致 share 表里 7月9日 跑出来的
+    行标成 7月9日,与实际数据日不符,统计/回溯会出错。
+
+    2026-07-12 增强: 加入交易日历回退逻辑。如果 15:30 后是法定节假日
+    （如春节/国庆），会向前回退到最后一个交易日，不会把节假日当天当 trade_date。
+
+    Args:
+        now:      测试用注入点(默认 datetime.now())
+        holidays: 已知的 A 股休市日集合（YYYY-MM-DD 格式）。
+                  默认从 config 表 trading_calendar_holidays 加载。
+
+    Returns: 'YYYY-MM-DD' 字符串（永远是合法的交易日）
+    """
+    t = now or datetime.now()
+    if t.time() >= dtime(15, 0):
+        candidate = t
+    else:
+        candidate = t - timedelta(days=1)
+
+    # 加载交易日历（如果调用方未提供）
+    if holidays is None:
+        try:
+            from app.core.db import ops as db_ops
+            h_row = db_ops.fetch_config_by_key("trading_calendar_holidays")
+            if h_row and h_row.get("category") == "general":
+                import json
+                vals = json.loads(h_row.get("value", "") or "[]")
+                if isinstance(vals, list):
+                    holidays = {h for h in vals if isinstance(h, str)}
+        except Exception:
+            pass
+
+    if holidays is None:
+        holidays = set()
+
+    # 向前回退直到找到交易日
+    max_iter = 30  # 安全上限，避免死循环
+    while max_iter > 0:
+        date_str = candidate.strftime("%Y-%m-%d")
+        if candidate.weekday() < 5 and date_str not in holidays:
+            return date_str
+        candidate -= timedelta(days=1)
+        max_iter -= 1
+
+    # 兜底（正常情况下不会走到这里）
+    t2 = now or datetime.now()
+    if t2.time() >= dtime(15, 30):
+        return t2.strftime("%Y-%m-%d")
+    return (t2 - timedelta(days=1)).strftime("%Y-%m-%d")
+
 
 def to_code6(code: str) -> str:
     """'SH600000' / 'sh.600000' / '600000' / 'SH600000.SH' → '600000'"""

app/core/db/models.py

@@ -141,7 +141,6 @@ class Stock(ORMBase):
     exchange: Mapped[str] = mapped_column(String(8), nullable=False, default="")
     list_date: Mapped[Optional[date]] = mapped_column(Date)
     listing_status: Mapped[str] = mapped_column(String(16), nullable=False, default="normal")
-    industry: Mapped[Optional[str]] = mapped_column(String(64), default="")
     total_share: Mapped[float] = mapped_column(Float, default=0)
     float_share: Mapped[float] = mapped_column(Float, default=0)
     share_updated_at: Mapped[Optional[date]] = mapped_column(Date)
@@ -258,92 +257,7 @@ class Share(ORMBase):
     float_share: Mapped[float] = mapped_column(Float, nullable=False)
 
 
-# ──────────────────────── 10. sectors ────────────────────────
+# ──────────────────────── 12. market_regime_daily ────────────────────────
-class Sectors(ORMBase):
-    """行业字典（baostock 行业分类 / 申万 / 中证 等多种 taxonomy）。"""
-    __tablename__ = "sectors"
-    __table_args__ = (
-        Index("idx_sectors_sector_name", "sector_name"),
-        {"schema": "market_data"},
-    )
-
-    sector_key: Mapped[str] = mapped_column(String(64), primary_key=True)
-    sector_name: Mapped[str] = mapped_column(String(64), nullable=False, default="")
-    taxonomy: Mapped[Optional[str]] = mapped_column(String(64), default="")
-    level: Mapped[Optional[str]] = mapped_column(String(16), default="")
-    source: Mapped[Optional[str]] = mapped_column(String(32), default="")
-    enabled: Mapped[int] = mapped_column(Integer, default=1)
-    updated_at: Mapped[Optional[datetime]] = _updated_at()
-
-
-# ──────────────────────── 11. stock_sector_map ────────────────────────
-class StockSectorMap(ORMBase):
-    """股票-行业多对多映射（一张股票可对应多个行业 / 概念板块）。"""
-    __tablename__ = "stock_sector_map"
-    __table_args__ = (
-        Index("idx_stock_sector_map_sector_key", "sector_key"),
-        {"schema": "market_data"},
-    )
-
-    stock_code: Mapped[str] = mapped_column(String(10), primary_key=True)
-    sector_key: Mapped[str] = mapped_column(String(64), nullable=False, default="")
-    updated_at: Mapped[Optional[datetime]] = _updated_at()
-
-
-# ──────────────────────── 12. industry ────────────────────────
-class Industry(ORMBase):
-    """股票-行业映射（baostock 源，证监会分类标准）。"""
-    __tablename__ = "industry"
-    __table_args__ = (
-        Index("idx_industry_industry_name", "industry_name"),
-        {"schema": "market_data"},
-    )
-
-    code: Mapped[str] = mapped_column(String(10), primary_key=True)
-    industry_name: Mapped[Optional[str]] = mapped_column(String(64), default="")
-    industry_classification: Mapped[Optional[str]] = mapped_column(String(32), default="")
-    update_date: Mapped[Optional[date]] = mapped_column(Date)
-    updated_at: Mapped[Optional[datetime]] = _updated_at()
-
-
-# ──────────────────────── 13. sector_indices ────────────────────────
-class SectorIndices(ORMBase):
-    """行业日线（基点 100，复合收益）。"""
-    __tablename__ = "sector_indices"
-    __table_args__ = (
-        PrimaryKeyConstraint("trade_date", "sector_name"),
-        Index("idx_sector_indices_sector_name", "sector_name"),
-        {"schema": "market_data"},
-    )
-
-    trade_date: Mapped[date] = mapped_column(Date, nullable=False)
-    sector_name: Mapped[str] = mapped_column(String(64), nullable=False)
-    close: Mapped[float] = mapped_column(Float, default=0)
-    sector_amplitude: Mapped[float] = mapped_column(Float, default=0)
-
-
-# ──────────────────────── 14. sector_features_daily ────────────────────────
-class SectorFeaturesDaily(ORMBase):
-    """行业日特征：sector_ret / sector_amplitude / close / EMA / score。"""
-    __tablename__ = "sector_features_daily"
-    __table_args__ = (
-        PrimaryKeyConstraint("trade_date", "sector_name"),
-        Index("idx_sector_features_sector_name", "sector_name"),
-        {"schema": "market_data"},
-    )
-
-    trade_date: Mapped[date] = mapped_column(Date, nullable=False)
-    sector_name: Mapped[str] = mapped_column(String(64), nullable=False)
-    sector_ret: Mapped[float] = mapped_column(Float, default=0)
-    sector_amplitude: Mapped[float] = mapped_column(Float, default=0)
-    close: Mapped[float] = mapped_column(Float, default=0)
-    ema10: Mapped[float] = mapped_column(Float, default=0)
-    ema20: Mapped[float] = mapped_column(Float, default=0)
-    ema200: Mapped[float] = mapped_column(Float, default=0)
-    score: Mapped[int] = mapped_column(Integer, default=0)
-
-
-# ──────────────────────── 15. market_regime_daily ────────────────────────
 class MarketRegimeDaily(ORMBase):
     """市场情绪衍生指标（基于本地 kline 聚合）。"""
     __tablename__ = "market_regime_daily"
@@ -599,6 +513,80 @@ class StockNodeMap(ORMBase):
     updated_at: Mapped[Optional[datetime]] = _updated_at()
 
 
+# ──────────────────────── 22. kline_stock_macd_daily ────────────────────────
+class KlineStockMACDDaily(ORMBase):
+    """个股日 K 级别 MACD 指标（mairui /hsstock/history/macd 直拉）。
+
+    字段（mairui 原始字段名，与 K 线口径一致）：
+      diff  = DIF（快慢 EMA 差）
+      dea   = DEA（DIF 的 9 日 EMA，即 signal line）
+      macd  = MACD 柱（2 ×（DIF − DEA））
+      ema12 / ema26 = 计算 DIF 用的两条 EMA
+    stock_code 用 hermes 格式（SH600519），与项目其他表一致。
+    """
+    __tablename__ = "kline_stock_macd_daily"
+    __table_args__ = (
+        PrimaryKeyConstraint("stock_code", "trade_date"),
+        Index("idx_kline_stock_macd_date", "trade_date"),
+        {"schema": "market_data"},
+    )
+
+    stock_code: Mapped[str] = mapped_column(String(10), nullable=False)
+    trade_date: Mapped[date] = mapped_column(Date, nullable=False)
+    diff: Mapped[Optional[float]] = mapped_column(Float)
+    dea: Mapped[Optional[float]] = mapped_column(Float)
+    macd: Mapped[Optional[float]] = mapped_column(Float)
+    ema12: Mapped[Optional[float]] = mapped_column(Float)
+    ema26: Mapped[Optional[float]] = mapped_column(Float)
+    source: Mapped[Optional[str]] = mapped_column(String(32), default="mairui")
+    updated_at: Mapped[Optional[datetime]] = _updated_at()
+
+
+# ──────────────────────── 23. kline_stock_kdj_daily ────────────────────────
+class KlineStockKDJDaily(ORMBase):
+    """个股日 K 级别 KDJ 指标（mairui /hsstock/history/kdj 直拉）。
+
+    k / d / j 三条随机指标线。stock_code 用 hermes 格式（SH600519）。
+    """
+    __tablename__ = "kline_stock_kdj_daily"
+    __table_args__ = (
+        PrimaryKeyConstraint("stock_code", "trade_date"),
+        Index("idx_kline_stock_kdj_date", "trade_date"),
+        {"schema": "market_data"},
+    )
+
+    stock_code: Mapped[str] = mapped_column(String(10), nullable=False)
+    trade_date: Mapped[date] = mapped_column(Date, nullable=False)
+    k: Mapped[Optional[float]] = mapped_column(Float)
+    d: Mapped[Optional[float]] = mapped_column(Float)
+    j: Mapped[Optional[float]] = mapped_column(Float)
+    source: Mapped[Optional[str]] = mapped_column(String(32), default="mairui")
+    updated_at: Mapped[Optional[datetime]] = _updated_at()
+
+
+# ──────────────────────── 24. kline_stock_boll_daily ────────────────────────
+class KlineStockBOLLDaily(ORMBase):
+    """个股日 K 级别 BOLL 布林带指标（mairui /hsstock/history/boll 直拉）。
+
+    mairui 原始字段：u=上轨, m=中轨, d=下轨。落库列名统一为 upper/mid/lower。
+    stock_code 用 hermes 格式（SH600519）。
+    """
+    __tablename__ = "kline_stock_boll_daily"
+    __table_args__ = (
+        PrimaryKeyConstraint("stock_code", "trade_date"),
+        Index("idx_kline_stock_boll_date", "trade_date"),
+        {"schema": "market_data"},
+    )
+
+    stock_code: Mapped[str] = mapped_column(String(10), nullable=False)
+    trade_date: Mapped[date] = mapped_column(Date, nullable=False)
+    upper: Mapped[Optional[float]] = mapped_column(Float)
+    mid: Mapped[Optional[float]] = mapped_column(Float)
+    lower: Mapped[Optional[float]] = mapped_column(Float)
+    source: Mapped[Optional[str]] = mapped_column(String(32), default="mairui")
+    updated_at: Mapped[Optional[datetime]] = _updated_at()
+
+
 __all__ = [
     # 1-2
     "Config",
@@ -612,13 +600,7 @@ __all__ = [
     "Kline5Min",
     "Moneyflow",
     "Share",
-    # 10-12
+    # 10-11
-    "Sectors",
-    "StockSectorMap",
-    "Industry",
-    # 13-15
-    "SectorIndices",
-    "SectorFeaturesDaily",
     "MarketRegimeDaily",
     # 16
     "TickTrade",
@@ -631,4 +613,8 @@ __all__ = [
     "NodeCategory",
     "Node",
     "StockNodeMap",
+    # 22-24 (2026-07-08 mairui 技术指标 MACD/KDJ/BOLL - 日 K 级别)
+    "KlineStockMACDDaily",
+    "KlineStockKDJDaily",
+    "KlineStockBOLLDaily",
 ]

app/core/db/ops.py

@@ -22,10 +22,12 @@ from sqlalchemy.dialects.postgresql import insert as pg_insert
 from app.core.db.models import (
     Config,
     DatasetRegistry,
-    Industry,
     Kline5Min,
     KlineIndex,
     KlineStock,
+    KlineStockBOLLDaily,
+    KlineStockKDJDaily,
+    KlineStockMACDDaily,
     KlineStockMADaily,
     LonghubangDaily,
     LonghubangSeat,
@@ -34,13 +36,9 @@ from app.core.db.models import (
     Moneyflow,
     Node,
     NodeCategory,
-    SectorFeaturesDaily,
-    SectorIndices,
-    Sectors,
     Share,
     Stock,
     StockNodeMap,
-    StockSectorMap,
     SyncHistory,
     TickTrade,
 )
@@ -451,19 +449,16 @@ def upsert_stock(
     exchange: str,
     list_date: str = "",
     listing_status: str = "normal",
-    industry: str = "",
 ) -> None:
     """upsert 一只股票基础信息（单条）。
 
     注意：MySQL 9.7.0 在 ON DUPLICATE KEY UPDATE 阶段对 DATE 字段的 VALUES()/new.col
-    求值存在 bug（会强制把空串塞进去，触发 1292 严格模式错误）。所以这里：
+    求值存在 bug（会强制把空串塞进去，触发 1292 严格模式错误）。所以这里
-    - list_date 字段在首次 INSERT 时写入；UPDATE 分支不更新（保持原值或 NULL）
+    list_date 字段在首次 INSERT 时写入；UPDATE 分支不更新（保持原值或 NULL）。
-    - industry 字段也避开这个 bug
 
     批量写入请用 upsert_stocks_bulk()，性能高 10-20 倍。
     """
     del list_date  # 显式不接受 list_date 更新（旧值保留）
-    del industry   # 同上
     with get_session() as s:
         stmt = _pg_upsert(
             Stock,
@@ -477,7 +472,7 @@ def upsert_stock(
 def upsert_stocks_bulk(rows: list[dict[str, Any]], chunk_size: int = 500) -> int:
     """批量 upsert 股票基础信息。
 
-    每条 row 需有: code, name, exchange, listing_status（可选 list_date / industry）
+    每条 row 需有: code, name, exchange, listing_status（可选 list_date）
     性能：~5000 只股票从 ~50s 降到 ~3s。
     """
     if not rows:
@@ -499,7 +494,7 @@ def upsert_stocks_bulk(rows: list[dict[str, Any]], chunk_size: int = 500) -> int
             stmt = _pg_upsert(
                 Stock, values,
                 conflict_keys=["code"],
-                update_cols=["name", "exchange", "list_date", "listing_status", "industry"],
+                update_cols=["name", "exchange", "list_date", "listing_status"],
             )
             s.execute(stmt)
             total += len(chunk)
@@ -507,15 +502,25 @@ def upsert_stocks_bulk(rows: list[dict[str, Any]], chunk_size: int = 500) -> int
 
 
 def update_stock_share_snapshot(code: str, total_share: float, float_share: float, trade_date: str = "") -> None:
+    """更新 stocks 表的股本字段。
+
+    Args:
+        trade_date: 数据源返回的股本快照"as-of"交易日(用于 share 表写入)。
+                    注意:这个参数**不会**写到 stocks.share_updated_at 字段 —
+                    那个字段的语义是"我们什么时候拉到的"(运行日,date.today()),
+                    不是数据交易日。把 trade_date 写到 share_updated_at 会让
+                    task_share_snapshot / task_stocks_basic 的 dedup 检查
+                    (share_updated_at == today) 失效,见 task 层 fix。
+    """
+    del trade_date  # 显式不接受 — 见 docstring
     with get_session() as s:
-        snap_date = _to_date_str(trade_date) if trade_date else None
         s.execute(
             update(Stock)
             .where(Stock.code == code)
             .values(
                 total_share=float(total_share),
                 float_share=float(float_share),
-                share_updated_at=snap_date,
+                share_updated_at=date.today(),  # "拉取日",不是"数据日"
             )
         )
 
@@ -543,7 +548,6 @@ def _stock_row_to_dict(r: Stock) -> dict[str, Any]:
         "exchange": r.exchange or "",
         "list_date": _to_date_str(r.list_date),
         "listing_status": r.listing_status or "normal",
-        "industry": r.industry or "",
         "total_share": r.total_share if r.total_share is not None else 0,
         "float_share": r.float_share if r.float_share is not None else 0,
         "share_updated_at": _to_date_str(r.share_updated_at),
@@ -674,9 +678,13 @@ def get_kline_5min_snapshots() -> dict[str, Optional[datetime]]:
     返回：{stock_code (6位): max(bar_time) 或 None}
     用于 kline_5min 任务的"全量/增量"统一规划。
 
-    DB 里 stock_code 形如 "000001.SZ"（mairui 写入格式），函数剥掉
+    DB 里 stock_code 是 hermes 格式 "SH600000"（带 SH/SZ/BJ 前缀，无点），
-    交易所后缀统一为 6 位。bar_time 是 DATETIME 字段，SQLAlchemy 2.x
+    2026-07-09 之前误以为 mairui 原始格式 "000001.SZ"，用 split(".")[0] 剥
-    native 返回 datetime；如果是 str（方言边界情况）手动解析。
+    不到 → 全部股票被 skip → task 把整市场当"全量"重跑(实际是增量)。
+    修:用正则剥 SH/SZ/BJ 前缀。
+
+    bar_time 是 DATETIME 字段，SQLAlchemy 2.x native 返回 datetime；如果是
+    str（方言边界情况）手动解析。
     """
     _ensure_schema()
     out: dict[str, Optional[datetime]] = {}
@@ -686,7 +694,13 @@ def get_kline_5min_snapshots() -> dict[str, Optional[datetime]]:
             .group_by(Kline5Min.stock_code)
         ).all()
         for raw_code, latest in rows:
-            code6 = str(raw_code or "").strip().split(".")[0]
+            # 剥 SH/SZ/BJ 前缀(hermes 格式),容错 6 位裸码
+            code6 = str(raw_code or "").strip()
+            for prefix in ("SH", "SZ", "BJ"):
+                if code6.startswith(prefix):
+                    code6 = code6[len(prefix):]
+                    break
+            code6 = code6.split(".")[0]  # 兜底剥 "000001.SZ" 老格式
             if len(code6) != 6 or not code6.isdigit():
                 continue
             if isinstance(latest, datetime):
@@ -834,143 +848,6 @@ def replace_all_stock_node_map(rows: list[dict[str, Any]]) -> None:
             s.execute(stmt)
 
 
-# ── 行业 / 概念板块 ─────────────────────────────────────────────────────
-
-
-def replace_all_industries(rows: list[dict[str, Any]]) -> None:
-    """全量替换 industry 表。rows: code, industry_name, industry_classification, update_date"""
-    if not rows:
-        return
-    with get_session() as s:
-        s.execute(delete(Industry))
-        values = [
-            {
-                "code": str(r.get("code", "")).zfill(6),
-                "industry_name": r.get("industry_name") or None,
-                "industry_classification": r.get("industry_classification") or None,
-                "update_date": _to_date_str(r.get("update_date")),
-            }
-            for r in rows
-        ]
-        if values:
-            stmt = _pg_upsert(Industry, values,
-                                   conflict_keys=["code"],
-                                   update_cols=["industry_name", "industry_classification", "update_date"])
-            s.execute(stmt)
-
-
-def fetch_all_industries() -> list[dict[str, Any]]:
-    _ensure_schema()
-    with get_session() as s:
-        rows = s.execute(
-            select(Industry.code, Industry.industry_name, Industry.industry_classification, Industry.update_date)
-        ).all()
-        return [
-            {
-                "code": r.code,
-                "industry_name": r.industry_name,
-                "industry_classification": r.industry_classification,
-                "update_date": _to_date_str(r.update_date),
-            }
-            for r in rows
-        ]
-
-
-def replace_all_sectors(rows: list[dict[str, Any]]) -> None:
-    """rows: sector_key, sector_name, taxonomy, level, source, enabled"""
-    if not rows:
-        return
-    with get_session() as s:
-        s.execute(delete(Sectors))
-        values = [
-            {
-                "sector_key": r.get("sector_key", ""),
-                "sector_name": r.get("sector_name", ""),
-                "taxonomy": r.get("taxonomy", ""),
-                "level": r.get("level", ""),
-                "source": r.get("source", ""),
-                "enabled": int(r.get("enabled", 1)),
-            }
-            for r in rows
-        ]
-        if values:
-            stmt = _pg_upsert(Sectors, values,
-                                   conflict_keys=["sector_key"],
-                                   update_cols=["sector_name", "taxonomy", "level", "source", "enabled"])
-            s.execute(stmt)
-
-
-def replace_all_stock_sector_map(rows: list[dict[str, Any]]) -> None:
-    """rows: stock_code, sector_key"""
-    if not rows:
-        return
-    with get_session() as s:
-        s.execute(delete(StockSectorMap))
-        values = [
-            {
-                "stock_code": str(r.get("stock_code", "")).zfill(6),
-                "sector_key": r.get("sector_key", ""),
-            }
-            for r in rows
-        ]
-        if values:
-            stmt = _pg_upsert(StockSectorMap, values,
-                                   conflict_keys=["stock_code"],
-                                   update_cols=["sector_key"])
-            s.execute(stmt)
-
-
-# ── 行业聚合（衍生）─────────────────────────────────────────────────────
-
-
-def replace_all_sector_indices(rows: list[dict[str, Any]]) -> None:
-    """rows: trade_date, sector_name, close, sector_amplitude"""
-    if not rows:
-        return
-    with get_session() as s:
-        values = [
-            {
-                "trade_date": _to_date_str(r.get("trade_date")),
-                "sector_name": r.get("sector_name", ""),
-                "close": float(r.get("close") or 0),
-                "sector_amplitude": float(r.get("sector_amplitude") or 0),
-            }
-            for r in rows
-        ]
-        if values:
-            stmt = _pg_upsert(SectorIndices, values,
-                                   conflict_keys=["trade_date", "sector_name"],
-                                   update_cols=["close", "sector_amplitude"])
-            s.execute(stmt)
-
-
-def replace_all_sector_features(rows: list[dict[str, Any]]) -> None:
-    """rows: trade_date, sector_name, sector_ret, sector_amplitude, close, ema10, ema20, ema200, score"""
-    if not rows:
-        return
-    with get_session() as s:
-        values = [
-            {
-                "trade_date": _to_date_str(r.get("trade_date")),
-                "sector_name": r.get("sector_name", ""),
-                "sector_ret": float(r.get("sector_ret") or 0),
-                "sector_amplitude": float(r.get("sector_amplitude") or 0),
-                "close": float(r.get("close") or 0),
-                "ema10": float(r.get("ema10") or 0),
-                "ema20": float(r.get("ema20") or 0),
-                "ema200": float(r.get("ema200") or 0),
-                "score": int(r.get("score") or 0),
-            }
-            for r in rows
-        ]
-        if values:
-            stmt = _pg_upsert(SectorFeaturesDaily, values,
-                                   conflict_keys=["trade_date", "sector_name"],
-                                   update_cols=["sector_ret", "sector_amplitude", "close",
-                                                "ema10", "ema20", "ema200", "score"])
-            s.execute(stmt)
-
-
 # ── 市场情绪（衍生）─────────────────────────────────────────────────────
 
 
@@ -1031,6 +908,51 @@ def upsert_kline_stock_ma_daily_rows(rows: list[dict[str, Any]]) -> None:
         s.execute(stmt)
 
 
+def upsert_kline_stock_macd_daily_rows(rows: list[dict[str, Any]]) -> int:
+    """rows: stock_code, trade_date, diff, dea, macd, ema12, ema26, source"""
+    if not rows:
+        return 0
+    with get_session() as s:
+        return _bulk_upsert_orm(s, KlineStockMACDDaily, rows)
+
+
+def upsert_kline_stock_kdj_daily_rows(rows: list[dict[str, Any]]) -> int:
+    """rows: stock_code, trade_date, k, d, j, source"""
+    if not rows:
+        return 0
+    with get_session() as s:
+        return _bulk_upsert_orm(s, KlineStockKDJDaily, rows)
+
+
+def upsert_kline_stock_boll_daily_rows(rows: list[dict[str, Any]]) -> int:
+    """rows: stock_code, trade_date, upper, mid, lower, source"""
+    if not rows:
+        return 0
+    with get_session() as s:
+        return _bulk_upsert_orm(s, KlineStockBOLLDaily, rows)
+
+
+_INDICATOR_MODEL = {
+    "macd": KlineStockMACDDaily,
+    "kdj": KlineStockKDJDaily,
+    "boll": KlineStockBOLLDaily,
+}
+
+
+def get_indicator_max_date(indicator: str, code: str) -> Optional[str]:
+    """某只票某指标已入库的最大 trade_date（YYYY-MM-DD），无则 None。"""
+    model = _INDICATOR_MODEL.get(indicator.lower())
+    if model is None:
+        raise ValueError(f"未知指标: {indicator!r}")
+    _ensure_schema()
+    with get_session() as s:
+        d = s.execute(
+            select(func.max(model.trade_date))
+            .where(model.stock_code == code)
+        ).scalar()
+        return _to_date_str(d) if d else None
+
+
 def pd_isna(v: Any) -> bool:
     """避免直接 import pandas（开销大），手写 nan/None 检查。"""
     if v is None:
@@ -1042,11 +964,26 @@ def pd_isna(v: Any) -> bool:
 
 
 def get_stock_kline_max_date(code: str) -> Optional[str]:
+    """某只票 kline_stock 已入库的最大 trade_date (YYYY-MM-DD),无则 None。
+
+    code 接受 6 位 ("600000") 或 hermes 格式 ("SH600000") — DB 里实际存的是
+    hermes 格式,2026-07-09 之前用 `where KlineStock.stock_code == '600000'`
+    永远查不到(hermes 有 SH/SZ/BJ 前缀),导致 incremental 把全市场当"无数据"
+    重跑。修:同时查 6 位和 hermes 两种格式,取较新结果。
+    """
     _ensure_schema()
+    c6 = str(code or "").strip()
+    hermes = f"SH{c6}" if c6.startswith(("5", "6", "9")) else (
+        f"SZ{c6}" if c6.startswith(("0", "2", "3")) else (
+        f"BJ{c6}" if c6.startswith(("4", "8")) else c6
+    ))
+    candidates = [c6]
+    if hermes != c6:
+        candidates.append(hermes)
     with get_session() as s:
         d = s.execute(
             select(func.max(KlineStock.trade_date))
-            .where(KlineStock.stock_code == code)
+            .where(KlineStock.stock_code.in_(candidates))
         ).scalar()
         return _to_date_str(d) if d else None
 

app/core/scheduler/scheduler.py

@@ -323,17 +323,6 @@ DEFAULT_SCHEDULES: list[tuple[str, dict, str]] = [
         },
         "每个交易日 21:35 拉取个股资金流（mairui 21:30 发布）",
     ),
-    (
-        "schedule_industry_sector",
-        {
-            "name": "周一行业映射",
-            "time": "09:30",
-            "condition": "trading_day",
-            "job": "industry_sector",
-            "enabled": True,
-        },
-        "每个交易日 09:30 全量更新股票-行业映射（开销大，可改为 weekly）",
-    ),
     (
         "schedule_market_regime",
         {
@@ -345,17 +334,6 @@ DEFAULT_SCHEDULES: list[tuple[str, dict, str]] = [
         },
         "每个交易日 16:15 聚合市场情绪（基于本地日K线）",
     ),
-    (
-        "schedule_share_snapshot",
-        {
-            "name": "盘后股本快照",
-            "time": "16:20",
-            "condition": "trading_day",
-            "job": "share_snapshot",
-            "enabled": True,
-        },
-        "每个交易日 16:20 刷新个股总股本/流通股本快照",
-    ),
     (
         "schedule_longhubang",
         {
@@ -389,6 +367,17 @@ DEFAULT_SCHEDULES: list[tuple[str, dict, str]] = [
         },
         "每个交易日 16:30 基于 kline_stock 计算 MA5/10/20/60 (本地派生，~30s @5213 只)",
     ),
+    (
+        "schedule_mairui_indicators",
+        {
+            "name": "日 K 技术指标 MACD/KDJ/BOLL",
+            "time": "16:40",
+            "condition": "trading_day",
+            "job": "mairui_indicators",
+            "enabled": True,
+        },
+        "每个交易日 16:40 从 mairui 直拉 MACD/KDJ/BOLL 增量 (3 指标 × 全市场，增量约数分钟)",
+    ),
 ]
 
 
@@ -428,8 +417,8 @@ def register_sync_jobs() -> None:
 
     for dataset_id in [
         "stock_basic", "kline_daily", "kline_index", "kline_5min",
-        "moneyflow", "industry_sector", "share_snapshot", "market_regime",
+        "moneyflow", "share_snapshot", "market_regime",
-        "longhubang", "stock_node", "mairui_ma_daily",
+        "longhubang", "stock_node", "mairui_ma_daily", "mairui_indicators",
     ]:
 
         def _make_job(did=dataset_id):

app/core/sync/base.py

@@ -34,13 +34,13 @@ def _is_market_closed() -> bool:
 
 
 def _effective_sync_end() -> str:
-    """若盘后则今天，否则昨天。"""
+    """若盘后则今天，否则昨天。
-    today = datetime.now()
+
-    if today.time() >= dtime(15, 30):
+    委派到 app.core.datasource.utils.effective_market_date() 统一实现，
-        return today.strftime("%Y-%m-%d")
+    该函数加载交易日历，会跳过节假日返回最近交易日。
-    # 昨天
+    """
-    from datetime import timedelta
+    from app.core.datasource.utils import effective_market_date
-    return (today - timedelta(days=1)).strftime("%Y-%m-%d")
+    return effective_market_date()
 
 
 class SyncTask:

app/core/sync/registry.py

@@ -94,30 +94,6 @@ SYNC_DEFINITIONS: list[dict[str, Any]] = [
         "dependency_ids": ["stock_basic"],
         "sort_order": 50,
     },
-    {
-        "dataset_id": "industry_sector",
-        "name": "股票-行业映射",
-        "description": "股票-行业映射 + 行业字典（Baostock）",
-        "storage_uri": "PG market_data.industry + sectors + stock_sector_map",
-        "storage_layer": "pg",
-        "management_role": "基础字典",
-        "source": "Baostock query_stock_industry",
-        "sync_script": "app.tasks.task_industry_sector:run",
-        "dependency_ids": ["stock_basic"],
-        "sort_order": 60,
-    },
-    {
-        "dataset_id": "sector_features",
-        "name": "行业聚合特征",
-        "description": "由 kline_stock + industry 衍生：行业日收益 / 行业指数 close / EMA10-20-200 / score。纯本地计算，无外部 API。",
-        "storage_uri": "PG market_data.sector_indices + sector_features_daily",
-        "storage_layer": "pg",
-        "management_role": "衍生源",
-        "source": "本地计算（kline_stock + industry）",
-        "sync_script": "app.tasks.task_sector_features:run",
-        "dependency_ids": ["kline_daily", "industry_sector"],
-        "sort_order": 65,
-    },
     {
         "dataset_id": "share_snapshot",
         "name": "股本快照",
@@ -178,6 +154,18 @@ SYNC_DEFINITIONS: list[dict[str, Any]] = [
         "dependency_ids": ["kline_daily"],
         "sort_order": 90,
     },
+    {
+        "dataset_id": "mairui_indicators",
+        "name": "日 K 技术指标 MACD/KDJ/BOLL (mairui)",
+        "description": "mairui /hsstock/history/{macd,kdj,boll} 直拉，写 3 张表；每交易日 16:40 增量。",
+        "storage_uri": "PG market_data.kline_stock_macd_daily + kdj_daily + boll_daily",
+        "storage_layer": "pg",
+        "management_role": "原始源",
+        "source": "mairui /hsstock/history/{macd,kdj,boll}/{symbol}/d/n",
+        "sync_script": "app.tasks.task_mairui_indicators:run",
+        "dependency_ids": ["stock_basic"],
+        "sort_order": 91,
+    },
 ]
 
 

app/entrypoints/worker.py

@@ -30,7 +30,15 @@ def start_scheduler_thread() -> None:
     给 service_run.sh 用：在 uvicorn 同进程跑后台 scheduler，避免
     docker 一个容器跑两个进程（supervisord / s6 那种方案）。
     幂等：多次调用只启动一次。
+
+    2026-07-12: 增加 RUNTIME_MODE 感知。systemd 模式下不启动 scheduler，
+    因为同步由 systemd timer 触发，避免双调度器重叠。
     """
+    logger.info(f"[worker] runtime_mode={settings.runtime_mode}")
+    if settings.runtime_mode == "systemd":
+        logger.info("[worker] systemd 模式：不启动进程内 scheduler，同步由 systemd timer 触发")
+        return
+
     # 1. 注册数据源 + seed config
     from app.core.datasource.registry import (
         build_default_registry,
@@ -74,6 +82,22 @@ def start_scheduler_thread() -> None:
 
 def main():
     logger.info("[worker] market_data_sync worker 启动")
+    logger.info(f"[worker] runtime_mode={settings.runtime_mode}")
+
+    if settings.runtime_mode == "systemd":
+        logger.info("[worker] systemd 模式：不启动进程内 scheduler，同步由 systemd timer 触发")
+        logger.info("[worker] 仅启动健康监控（数据源心跳），按 Ctrl+C 退出")
+        from app.core.datasource.registry import build_default_registry, seed_datasource_configs, start_health_monitor
+        build_default_registry()
+        seed_datasource_configs()
+        start_health_monitor()
+        try:
+            while True:
+                time.sleep(60)
+        except KeyboardInterrupt:
+            logger.info("[worker] 收到 SIGINT，退出")
+            sys.exit(0)
+        return
 
     # 1. 注册数据源 + seed config
     from app.core.datasource.registry import build_default_registry, seed_datasource_configs

app/mcp_server.py

@@ -34,6 +34,7 @@ if str(_PROJECT_ROOT) not in sys.path:
 
 # ── MCP server bootstrap ────────────────────────────────────────────
 from mcp.server import Server
+from mcp.server.sse import SseServerTransport
 from mcp.server.stdio import stdio_server
 from mcp.types import Tool, TextContent
 
@@ -125,9 +126,10 @@ def list_datasets() -> list[dict]:
     from sqlalchemy import func, select
     from app.core.db.models import (
         KlineStock, KlineIndex, Kline5Min, Moneyflow, Share, TickTrade,
-        SectorIndices, SectorFeaturesDaily, MarketRegimeDaily,
+        MarketRegimeDaily,
-        LonghubangDaily, LonghubangSeat, Stock, Industry,
+        LonghubangDaily, LonghubangSeat, Stock,
         NodeCategory, Node, StockNodeMap, KlineStockMADaily,
+        KlineStockMACDDaily, KlineStockKDJDaily, KlineStockBOLLDaily,
     )
     from app.core.db.orm import SessionLocal
 
@@ -140,15 +142,15 @@ def list_datasets() -> list[dict]:
         ("moneyflow", Moneyflow),
         ("share", Share),
         ("tick_trade", TickTrade),
-        ("sector_indices", SectorIndices),
-        ("sector_features_daily", SectorFeaturesDaily),
         ("market_regime_daily", MarketRegimeDaily),
         ("longhubang_daily", LonghubangDaily),
         ("longhubang_seat", LonghubangSeat),
-        ("industry", Industry),
         ("node_categories", NodeCategory),
         ("nodes", Node),
         ("stock_node_map", StockNodeMap),
+        ("kline_stock_macd_daily", KlineStockMACDDaily),
+        ("kline_stock_kdj_daily", KlineStockKDJDaily),
+        ("kline_stock_boll_daily", KlineStockBOLLDaily),
     ]
     out = []
     with SessionLocal() as s:
@@ -175,9 +177,10 @@ def get_dataset_info(table_name: str) -> dict:
     allowed = {
         "stocks", "kline_stock", "kline_index", "kline_5min",
         "kline_stock_ma_daily", "moneyflow", "share", "tick_trade",
-        "sector_indices", "sector_features_daily", "market_regime_daily",
+        "market_regime_daily",
-        "longhubang_daily", "longhubang_seat", "industry",
+        "longhubang_daily", "longhubang_seat",
         "node_categories", "nodes", "stock_node_map",
+        "kline_stock_macd_daily", "kline_stock_kdj_daily", "kline_stock_boll_daily",
         "dataset_registry", "sync_history", "config",
     }
     if table_name not in allowed:
@@ -291,11 +294,51 @@ async def call_tool(name: str, arguments: dict):
 
 # ── main ────────────────────────────────────────────────────────────
 
-async def main():
+async def main_sse(host: str = "127.0.0.1", port: int = 8101):
-    # 启动时注册数据源（让 datasource.health_check 之类方法可用）
+    """启动 MCP 的 SSE (HTTP) 服务，供持久运行。
+    
+    客户端连接: http://{host}:{port}/sse
+    """
+    sse_transport = SseServerTransport("/messages")
+
+    async def app(scope, receive, send):
+        if scope["type"] != "http":
+            return
+        path = scope.get("path", "")
+        if path == "/sse":
+            async with sse_transport.connect_sse(scope, receive, send) as (read, write):
+                await server.run(read, write, server.create_initialization_options())
+        elif path == "/messages" and scope.get("method") == "POST":
+            await sse_transport.handle_post_message(scope, receive, send)
+
+    import uvicorn
+    from app.core.utils.logging import get_logger
+    logger = get_logger("mcp")
+    logger.info(f"MCP SSE server listening on http://{host}:{port}/sse")
+    config = uvicorn.Config(app, host=host, port=port, log_level="info")
+    srv = uvicorn.Server(config)
+    await srv.serve()
+
+
+def main():
     from app.core.datasource.registry import build_default_registry
     build_default_registry()
 
+    import sys
+    if "--sse" in sys.argv:
+        host = "127.0.0.1"
+        port = 8101
+        for i, arg in enumerate(sys.argv):
+            if arg == "--host" and i + 1 < len(sys.argv):
+                host = sys.argv[i + 1]
+            if arg == "--port" and i + 1 < len(sys.argv):
+                port = int(sys.argv[i + 1])
+        asyncio.run(main_sse(host, port))
+    else:
+        asyncio.run(_stdio_main())
+
+
+async def _stdio_main():
     async with stdio_server() as (read_stream, write_stream):
         await server.run(
             read_stream,
@@ -305,4 +348,4 @@ async def main():
 
 
 if __name__ == "__main__":
-    asyncio.run(main())
+    main()

app/sources/mairui.py

@@ -32,7 +32,7 @@ from app.core.datasource.utils import (
 class MairuiSource(DataSource):
     key = "datasource_mairui"
     name = "麦蕊智数（mairui.club）"
-    provides = ["kline_daily", "kline_5min", "index_daily", "stock_basic", "moneyflow", "tick_trade", "stock_node"]
+    provides = ["kline_daily", "kline_5min", "index_daily", "stock_basic", "moneyflow", "tick_trade", "stock_node", "indicator_daily"]
     requires_credential = True
     credential_key = "MAIRUI_LICENCE"
 
@@ -285,6 +285,68 @@ class MairuiSource(DataSource):
             df = df[df["bar_time"] < pd.Timestamp(end) + pd.Timedelta(days=1)]
         return df
 
+    # ── 技术指标 fetch（MACD / KDJ / BOLL）─────────────────────
+    # mairui 端点：/hsstock/history/{indicator}/{symbol}/d/n/{licence}
+    #   indicator ∈ {macd, kdj, boll}；日 K 级别 "/d/"；不复权 "/n/"
+    # 各指标字段（除公共 "t" 外）：
+    #   macd → diff, dea, macd, ema12, ema26
+    #   kdj  → k, d, j
+    #   boll → u(上轨), d(下轨), m(中轨)   ← 注意 mairui 用 u/d/m
+    _INDICATOR_FIELDS = {
+        "macd": ["diff", "dea", "macd", "ema12", "ema26"],
+        "kdj": ["k", "d", "j"],
+        "boll": ["u", "d", "m"],
+    }
+
+    def fetch_indicator_daily(
+        self, indicator: str, code6: str, start: str = "", end: str = ""
+    ) -> pd.DataFrame:
+        """拉单只股票某个技术指标的日 K 序列。
+
+        返回列：trade_date + 该指标字段（macd/kdj/boll 各自的原始字段名）。
+        start/end 为 "YYYY-MM-DD"（可空 → 全历史）。
+        """
+        indicator = indicator.lower()
+        fields = self._INDICATOR_FIELDS.get(indicator)
+        if fields is None:
+            raise ValueError(f"未知指标: {indicator!r}，可选 {list(self._INDICATOR_FIELDS)}")
+        lic = self._get_licence()
+        if not lic:
+            return pd.DataFrame()
+        symbol = code6_to_mairui(code6)
+        path = f"/hsstock/history/{indicator}/{symbol}/d/n/{lic}"
+        params = {
+            "st": (start or "").replace("-", ""),
+            "et": (end or "").replace("-", ""),
+        }
+        data = self._fetch(path, params)
+        if not isinstance(data, list):
+            return pd.DataFrame()
+        records = []
+        for item in data:
+            try:
+                t = item.get("t", "")
+                d = t.split(" ")[0] if isinstance(t, str) else ""
+                if not d:
+                    continue
+                row = {"trade_date": d}
+                for f in fields:
+                    v = item.get(f)
+                    row[f] = None if v is None else float(v)
+                records.append(row)
+            except (KeyError, ValueError, TypeError):
+                continue
+        if not records:
+            return pd.DataFrame()
+        df = pd.DataFrame(records)
+        df["trade_date"] = pd.to_datetime(df["trade_date"], errors="coerce")
+        df = df.dropna(subset=["trade_date"]).sort_values("trade_date").reset_index(drop=True)
+        if start:
+            df = df[df["trade_date"] >= pd.Timestamp(start)]
+        if end:
+            df = df[df["trade_date"] <= pd.Timestamp(end)]
+        return df
+
     def fetch_index_daily(self, index_code: str, start: str, end: str) -> pd.DataFrame:
         """指数日 K。index_code 用 mairui 格式（如 '000300.SH'）。"""
         lic = self._get_licence()

app/sources/xueqiu.py

@@ -16,7 +16,7 @@ from typing import Any, Optional
 import pandas as pd
 
 from app.core.datasource.base import DataSource
-from app.core.datasource.utils import code6_to_xueqiu
+from app.core.datasource.utils import call_with_timeout, code6_to_xueqiu, effective_market_date
 
 logger = logging.getLogger("sync.xueqiu")
 
@@ -94,7 +94,11 @@ class XueqiuSource(DataSource):
         try:
             ball = self._import_ball()
             self._set_token_once()
-            result = ball.quote_detail("SH600036")
+            result = call_with_timeout(
+                ball.quote_detail, "SH600036",
+                timeout=10.0, on_timeout=None,
+                description="xueqiu.health_check",
+            )
             if result is None or result.get("error_code") != 0:
                 return {
                     "success": False,
@@ -123,12 +127,23 @@ class XueqiuSource(DataSource):
             count = min(max((end_dt - start_dt).days + 60, 10), 5000)
 
             # 加重试：雪球风控偶尔返回空/错误
+            # 用 call_with_timeout 包一层 — 2026-07-08 kline_daily 卡死 4h 根因
+            # 就是 pysnowball ball.kline() 底层 requests 无 read timeout,雪球
+            # 服务端卡住时这里永久阻塞。timeout=20s 与 mairui/sina 对齐。
             result = None
             for attempt in range(3):
-                result = ball.kline(symbol, period="day", count=count)
+                result = call_with_timeout(
+                    ball.kline, symbol, period="day", count=count,
+                    timeout=20.0, on_timeout=None,
+                    description=f"xueqiu.kline {symbol}",
+                )
                 if result and result.get("error_code") == 0:
                     break
-                time.sleep(0.3 * (attempt + 1))
+                if result is None and attempt < 2:
+                    time.sleep(0.3 * (attempt + 1))
+                    continue
+                if result and result.get("error_code") != 0:
+                    time.sleep(0.3 * (attempt + 1))
             if not result or result.get("error_code") != 0:
                 logger.warning(
                     "[kline %s] 雪球 kline 失败: error_code=%s desc=%s",
@@ -177,7 +192,12 @@ class XueqiuSource(DataSource):
             self._wait_rps()
 
             symbol = code6_to_xueqiu(code6)
-            result = ball.quote_detail(symbol)
+            # ball.quote_detail 同样有 pysnowball 无 read timeout 风险,包一层
+            result = call_with_timeout(
+                ball.quote_detail, symbol,
+                timeout=15.0, on_timeout=None,
+                description=f"xueqiu.quote_detail {symbol}",
+            )
             if not result or result.get("error_code") != 0:
                 logger.warning(
                     "[share %s] 雪球 quote_detail 失败: error_code=%s desc=%s",
@@ -187,14 +207,18 @@ class XueqiuSource(DataSource):
                 )
                 return None
             quote = result.get("data", {}).get("quote", {})
-            today_str = datetime.now().strftime("%Y-%m-%d")
+            # 雪球 quote_detail 不返回明确"股本变动日",time 字段是 quote 当前
+            # 时间,不是交易日。股本数据本质是慢变快照,这里标"最近已完成交易日"
+            # (15:30 前=昨天,15:30 后=今天) — 比 datetime.now() 精确,避免盘前
+            # 拉到的快照被错标为"今天"。
+            as_of = effective_market_date()
             total = round(float(quote.get("total_shares") or 0) / 1e8, 4)
             flt = round(float(quote.get("float_shares") or 0) / 1e8, 4)
             if total <= 0 or flt <= 0:
                 logger.warning("[share %s] 雪球返回成功但 total/float 为 0", code6)
                 return None
             return {
-                "trade_date": today_str,
+                "trade_date": as_of,
                 "total_share": total,
                 "float_share": flt,
             }

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:
+        # 不用 `with ThreadPoolExecutor(...) as pool:` — 它的 __exit__ 默认 wait=True,
-            futures = {pool.submit(self._sync_one, primary, p[0], p[1], p[2]): p[0] for p in plans}
+        # 一旦某个 worker 卡在 xueqiu IO,主线程会在 as_completed 触发 60s timeout 后
-            for i, future in enumerate(as_completed(futures), 1):
+        # 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
+# 源优先级：麦蕊 → 雪球 → 新浪
-# 经验：sina 在单日窗口 (start=end=某天) 会返回空，雪球更可靠
+# 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_xueqiu", "datasource_mairui", "datasource_xinlang"]
+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
+        # 1. 获取所有可用源（按优先级），支持 per-stock fallback
-        primary = self._pick_primary()
+        sources = self._get_available_sources()
-        if primary is None:
+        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})...",
+            message=f"使用 [{source_names}] 同步日K线 (fetcher={max_workers}, batch={BATCH_SIZE})...",
-            current_step=primary.key,
+            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]:
+        def _rows_from_df(code6: str, df: pd.DataFrame) -> tuple[list[dict], str]:
-            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。
             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:
+        # 不用 `with ThreadPoolExecutor(...) as pool:` — 它的 __exit__ 默认 wait=True,
-            futures = {pool.submit(_fetch_one, c6, fs): c6 for c6, fs in jobs}
+        # 一旦某个 worker 卡在 xueqiu IO,主线程会在 as_completed 触发 timeout 后
-            done_cnt = 0
+        # 仍被 __exit__ 阻塞等 worker 退出 → 进程挂死(2026-07-09 13:21 教训)。
-            for future in as_completed(futures):
+        # 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)
+                available.append(ds_registry.get(key))
-        return None
+        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:
+        pool = ThreadPoolExecutor(max_workers=max_workers)
-            futures = {pool.submit(self._sync_one, mr, c6): c6 for c6 in stock_codes}
+        futures = {pool.submit(self._sync_one, mr, c6): c6 for c6 in stock_codes}
-            for i, future in enumerate(as_completed(futures), 1):
+        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 {

bin/backfill_share_history.py

@@ -0,0 +1,409 @@
+#!/usr/bin/env python3
+"""历史股本回填脚本。
+
+目标：把 2023-01-01 ~ 2026-07-15 期间每只 A 股的股本，按“每周一条”
+写入 market_data.share 表。
+
+数据源：
+  1. akshare stock_share_change_cninfo（巨潮资讯-公司股本变动）
+     覆盖 2023 及更早的历史股本变动记录，单位：万股。
+  2. 现有 market_data.share 快照（2026 年雪球 quote_detail 已写入）
+     作为 2024 年底之后最新股本的补充。
+
+策略：
+  - 每周五作为本周代表日（若周五休市则仍取周五，因为股本变动是事件日，
+    不依赖当天是否交易；最后一周如果结束日不是周五，额外补一个结束日）。
+  - 对每只股票，把历史股本变动 + 现有快照合并成一条时间线，按日期前向填充，
+    得到每个采样日的 total_share / float_share。
+  - 最终批量 upsert 到 share 表。
+  - 默认使用巨潮“总股本 / 已流通股份”口径；加 --a-share-only 时只取 A 股部分
+    （total_share = 人民币普通股，float_share = 人民币普通股 - 流通受限股份），
+    与雪球 quote_detail 的 float_shares 口径更一致。
+
+用法：
+  .venv/bin/python bin/backfill_share_history.py
+  .venv/bin/python bin/backfill_share_history.py --dry-run --limit 50
+  .venv/bin/python bin/backfill_share_history.py --max-workers 10 --start 2023-01-01 --end 2026-07-15
+  .venv/bin/python bin/backfill_share_history.py --a-share-only
+"""
+from __future__ import annotations
+
+import argparse
+import os
+import sys
+import time
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from datetime import date, datetime, timedelta
+from typing import Any, Optional
+
+import pandas as pd
+from sqlalchemy import text
+
+# 项目根目录导入
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from app.core.db import ops as db_ops
+from app.core.db.orm import get_session
+from app.core.datasource.utils import is_a_share_code, to_hermes
+from app.core.utils.logging import get_logger
+
+logger = get_logger("backfill_share_history")
+
+# 默认回填区间
+DEFAULT_START = "2023-01-01"
+DEFAULT_END = "2026-07-15"
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="回填历史股本到 market_data.share")
+    parser.add_argument("--start", default=DEFAULT_START, help="开始日期 (YYYY-MM-DD)")
+    parser.add_argument("--end", default=DEFAULT_END, help="结束日期 (YYYY-MM-DD)")
+    parser.add_argument("--max-workers", type=int, default=5, help="并发 worker 数")
+    parser.add_argument("--limit", type=int, default=0, help="仅处理前 N 只股票（测试用）")
+    parser.add_argument(
+        "--codes",
+        default="",
+        help="指定股票代码，逗号分隔（如 000001,600519）；空则处理全市场",
+    )
+    parser.add_argument("--dry-run", action="store_true", help="只统计不写入数据库")
+    parser.add_argument(
+        "--a-share-only",
+        action="store_true",
+        help="只取 A 股股本口径（去掉 H 股 / B 股）",
+    )
+    parser.add_argument(
+        "--upsert-chunk",
+        type=int,
+        default=2000,
+        help="每次 upsert 行数",
+    )
+    return parser.parse_args()
+
+
+def load_stock_codes(limit: int = 0, codes_str: str = "") -> list[str]:
+    """返回 6 位代码列表（已过滤 A 股）。"""
+    if codes_str:
+        raw = [c.strip() for c in codes_str.split(",") if c.strip()]
+        return [c for c in raw if is_a_share_code(c)]
+
+    codes = [
+        c
+        for c in db_ops.iter_stock_codes(active_only=True)
+        if is_a_share_code(c)
+    ]
+    if limit > 0:
+        codes = codes[:limit]
+    return codes
+
+
+def load_existing_share_snapshots(code6s: list[str]) -> dict[str, dict[str, float]]:
+    """读取每只股票的最新 share 快照（ trade_date -> total_share/float_share ）。
+
+    返回：{code6: {trade_date_str: {"total_share": float, "float_share": float}}}
+    """
+    if not code6s:
+        return {}
+    hermes_codes = [to_hermes(c) for c in code6s]
+    out: dict[str, dict[str, dict[str, float]]] = {c: {} for c in code6s}
+
+    # 按 hermes 前缀分组查询，避免 SQL IN 子句过长
+    batch_size = 500
+    for i in range(0, len(hermes_codes), batch_size):
+        batch = hermes_codes[i : i + batch_size]
+        placeholders = ",".join([f"'{c}'" for c in batch])
+        sql = f"""
+            SELECT stock_code, trade_date, total_share, float_share
+            FROM market_data.share
+            WHERE stock_code IN ({placeholders})
+              AND trade_date >= '2023-01-01'
+            ORDER BY stock_code, trade_date
+        """
+        with get_session() as s:
+            rows = s.execute(text(sql)).fetchall()
+        for r in rows:
+            hermes = r[0]
+            code6 = hermes[2:] if hermes[:2] in ("SH", "SZ", "BJ") else hermes
+            if code6 not in out:
+                out[code6] = {}
+            out[code6][r[1].strftime("%Y-%m-%d")] = {
+                "total_share": float(r[2] or 0),
+                "float_share": float(r[3] or 0),
+            }
+    return out
+
+
+def fetch_cninfo_changes(code6: str, a_share_only: bool = False) -> pd.DataFrame:
+    """调用 akshare 巨潮接口，返回标准化后的股本变动 DataFrame。
+
+    列：change_date, total_share, float_share（单位：亿股）
+
+    Args:
+        a_share_only: True 时只取 A 股口径：
+            total_share = 人民币普通股（缺失时用 总股本 - H股 - B股）
+            float_share = 人民币普通股 - 流通受限股份
+    """
+    try:
+        import akshare as ak
+
+        df = ak.stock_share_change_cninfo(symbol=code6)
+    except Exception as e:
+        logger.warning("[%s] 巨潮接口调用失败: %s", code6, e)
+        return pd.DataFrame()
+
+    if df is None or df.empty:
+        return pd.DataFrame()
+
+    # 只保留我们需要的列
+    if a_share_only:
+        rename_map = {
+            "变动日期": "change_date",
+            "总股本": "total_share_all",
+            "人民币普通股": "a_share_total",
+            "流通受限股份": "restricted",
+            "境外上市外资股-H股": "h_share",
+            "境内上市外资股-B股": "b_share",
+        }
+    else:
+        rename_map = {
+            "变动日期": "change_date",
+            "总股本": "total_share",
+            "已流通股份": "float_share",
+        }
+
+    missing = [k for k in rename_map if k not in df.columns]
+    if missing:
+        logger.warning("[%s] 巨潮返回缺少字段: %s", code6, missing)
+        return pd.DataFrame()
+
+    df = df[list(rename_map.keys())].rename(columns=rename_map)
+    df["change_date"] = pd.to_datetime(df["change_date"], errors="coerce")
+    for col in rename_map.values():
+        if col != "change_date":
+            df[col] = pd.to_numeric(df[col], errors="coerce")
+
+    # 万股 -> 亿股
+    for col in rename_map.values():
+        if col != "change_date":
+            df[col] = df[col] / 10000.0
+
+    if a_share_only:
+        # A 股总股本：优先用“人民币普通股”，缺失时从总股本中扣除 H 股 / B 股
+        a_total = df["a_share_total"].where(
+            df["a_share_total"].notna(),
+            df["total_share_all"] - df["h_share"].fillna(0) - df["b_share"].fillna(0),
+        )
+        # A 股流通股本：A 股总股本 - 流通受限股份
+        a_float = a_total - df["restricted"].fillna(0)
+        df["total_share"] = a_total.round(4)
+        df["float_share"] = a_float.round(4)
+
+    df = df.dropna(subset=["change_date", "total_share", "float_share"])
+    if df.empty:
+        return pd.DataFrame()
+
+    # 剔除异常：A 股流通股本不能为负，也不能超过 A 股总股本
+    df = df[df["float_share"] >= 0]
+    df = df[df["float_share"] <= df["total_share"] * 1.001]
+    if df.empty:
+        return pd.DataFrame()
+
+    df = df[["change_date", "total_share", "float_share"]].sort_values("change_date").reset_index(drop=True)
+    return df
+
+
+def build_weekly_snapshots(
+    code6: str,
+    cninfo_df: pd.DataFrame,
+    existing: dict[str, dict[str, float]],
+    sample_dates: list[date],
+) -> list[dict[str, Any]]:
+    """合并历史变动 + 现有快照，生成该股票每周股本记录。"""
+    # 构建时间线：change_date -> 最新股本
+    timeline: dict[str, dict[str, float]] = {}
+
+    # 1) 巨潮历史变动
+    for _, row in cninfo_df.iterrows():
+        d = row["change_date"].strftime("%Y-%m-%d")
+        timeline[d] = {
+            "total_share": float(row["total_share"]),
+            "float_share": float(row["float_share"]),
+        }
+
+    # 2) 现有 share 快照（仅用于补充巨潮未覆盖的 2025 年及之后最新日期，
+    #    避免旧口径数据覆盖 2024 及更早的历史回填）
+    if existing:
+        for d, v in existing.items():
+            if d >= "2025-01-01":
+                timeline[d] = v
+
+    if not timeline:
+        return []
+
+    # 按日期排序并做前向填充
+    dates = sorted(timeline.keys())
+    rows = []
+    for sd in sample_dates:
+        sd_str = sd.strftime("%Y-%m-%d")
+        # 找到 <= sd 的最新一条
+        latest = None
+        for d in dates:
+            if d > sd_str:
+                break
+            latest = d
+        if latest is None:
+            continue
+        v = timeline[latest]
+        if v["total_share"] <= 0 or v["float_share"] <= 0:
+            continue
+        rows.append(
+            {
+                "stock_code": to_hermes(code6),
+                "trade_date": sd_str,
+                "total_share": v["total_share"],
+                "float_share": v["float_share"],
+            }
+        )
+    return rows
+
+
+def generate_sample_dates(start: str, end: str) -> list[date]:
+    """生成每周五采样日，若结束日不是周五则额外包含结束日。"""
+    start_dt = datetime.strptime(start, "%Y-%m-%d").date()
+    end_dt = datetime.strptime(end, "%Y-%m-%d").date()
+
+    # pandas 每周五
+    fridays = pd.date_range(start=start, end=end, freq="W-FRI")
+    dates = [d.date() for d in fridays]
+
+    # 若结束日不是周五，补一个结束日本身
+    if end_dt not in dates:
+        dates.append(end_dt)
+        dates.sort()
+
+    return dates
+
+
+def process_one_stock(
+    code6: str,
+    existing_map: dict[str, dict[str, float]],
+    sample_dates: list[date],
+    a_share_only: bool = False,
+) -> list[dict[str, Any]]:
+    cninfo_df = fetch_cninfo_changes(code6, a_share_only=a_share_only)
+    existing = existing_map.get(code6, {})
+    return build_weekly_snapshots(code6, cninfo_df, existing, sample_dates)
+
+
+def main() -> int:
+    args = parse_args()
+
+    logger.info("开始回填历史股本: %s ~ %s", args.start, args.end)
+    logger.info(
+        "参数: workers=%s, dry_run=%s, limit=%s, a_share_only=%s",
+        args.max_workers, args.dry_run, args.limit, args.a_share_only,
+    )
+
+    # 1. 股票代码
+    code6s = load_stock_codes(limit=args.limit, codes_str=args.codes)
+    if not code6s:
+        logger.error("无可用股票代码")
+        return 1
+    logger.info("共 %s 只股票待处理", len(code6s))
+
+    # 2. 采样日期
+    sample_dates = generate_sample_dates(args.start, args.end)
+    logger.info("采样日期: %s 个（%s ~ %s）", len(sample_dates), sample_dates[0], sample_dates[-1])
+
+    # 3. 预读现有 share 快照
+    logger.info("预读现有 share 快照...")
+    t0 = time.time()
+    existing_map = load_existing_share_snapshots(code6s)
+    logger.info("预读完成，耗时 %.1fs", time.time() - t0)
+
+    # 4. 多线程拉取 + 合并
+    all_rows: list[dict[str, Any]] = []
+    ok = fail = 0
+    t0 = time.time()
+    with ThreadPoolExecutor(max_workers=args.max_workers) as pool:
+        futures = {
+            pool.submit(
+                process_one_stock, c, existing_map, sample_dates, args.a_share_only
+            ): c
+            for c in code6s
+        }
+        for i, future in enumerate(as_completed(futures), 1):
+            c6 = futures[future]
+            try:
+                rows = future.result()
+            except Exception as e:
+                fail += 1
+                logger.warning("[%s] 处理异常: %s", c6, e)
+                continue
+            if rows:
+                all_rows.extend(rows)
+                ok += 1
+            else:
+                fail += 1
+            if i % 100 == 0 or i == len(code6s):
+                logger.info(
+                    "进度 %s/%s, OK=%s, FAIL=%s, rows=%s",
+                    i,
+                    len(code6s),
+                    ok,
+                    fail,
+                    len(all_rows),
+                )
+
+    elapsed = time.time() - t0
+    logger.info(
+        "拉取完成: OK=%s, FAIL=%s, 总条数=%s, 耗时=%.1fs",
+        ok,
+        fail,
+        len(all_rows),
+        elapsed,
+    )
+
+    if not all_rows:
+        logger.warning("无数据可写入")
+        return 0
+
+    # 5. 写入
+    if args.dry_run:
+        logger.info("DRY RUN: 本应写入 %s 行", len(all_rows))
+        # 打印几只样本（首尾各 3 行，方便看后段是否被现有快照覆盖）
+        df = pd.DataFrame(all_rows)
+        sample_codes = df["stock_code"].unique()[:3]
+        for sc in sample_codes:
+            sub = df[df["stock_code"] == sc]
+            preview = pd.concat([sub.head(3), sub.tail(3)])
+            logger.info("样本 %s:\n%s", sc, preview.to_string(index=False))
+        return 0
+
+    # 5. 写入前先清理目标区间内非采样日的旧记录，保证“每周一条”
+    keep_dates = {d.strftime("%Y-%m-%d") for d in sample_dates}
+    keep_str = ",".join([f"'{d}'" for d in keep_dates])
+    delete_sql = f"""
+        DELETE FROM market_data.share
+        WHERE trade_date BETWEEN '{args.start}' AND '{args.end}'
+          AND trade_date NOT IN ({keep_str})
+    """
+    logger.info("清理非采样日旧记录...")
+    with get_session() as s:
+        deleted = s.execute(text(delete_sql)).rowcount
+        s.commit()
+    logger.info("清理完成: 删除 %s 行", deleted)
+
+    logger.info("开始写入 share 表，chunk=%s", args.upsert_chunk)
+    t0 = time.time()
+    total_written = 0
+    for i in range(0, len(all_rows), args.upsert_chunk):
+        chunk = all_rows[i : i + args.upsert_chunk]
+        total_written += db_ops.upsert_share(chunk)
+        logger.info("已写入 %s/%s", total_written, len(all_rows))
+    logger.info("写入完成: %s 行, 耗时 %.1fs", total_written, time.time() - t0)
+
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

bin/daily_sync_check.py

@@ -29,6 +29,7 @@ from pathlib import Path
 from typing import Any, Optional
 
 import requests
+import sqlalchemy as sa
 
 
 # ── 路径与日志 ─────────────────────────────────────────────────────────────
@@ -42,21 +43,43 @@ SECRETS_FILE = PROJECT_ROOT / "config" / "daily_check_secrets.env"
 LOG = logging.getLogger("daily-sync-check")
 
 
+# ── 运行模式 ───────────────────────────────────────────────────────────────
+def _runtime_mode() -> str:
+    """读取当前运行模式：docker | systemd。
+
+    优先读环境变量 RUNTIME_MODE，未设置时尝试从 .env 读取，默认 docker。
+    """
+    mode = os.environ.get("RUNTIME_MODE", "")
+    if not mode:
+        env_file = PROJECT_ROOT / ".env"
+        if env_file.exists():
+            try:
+                for line in env_file.read_text(encoding="utf-8").splitlines():
+                    line = line.strip()
+                    if line.startswith("RUNTIME_MODE="):
+                        mode = line.split("=", 1)[1].strip().strip('"').strip("'")
+                        break
+            except Exception:
+                pass
+    return (mode or "docker").lower()
+
+
 # ── 预期调度表 ─────────────────────────────────────────────────────────────
 # 与 [[market-data-overview]] 保持一致；同步触发点（HH:MM）按 Asia/Shanghai 解释。
 # weekend_only=True 表示只在周六/周日期望运行（实际仅 stock_node 周六一次）。
 SCHEDULE: dict[str, dict[str, Any]] = {
     "stock_basic":     {"window_end": "09:30", "weekend_only": False},
-    "industry_sector": {"window_end": "09:45", "weekend_only": False},
     "kline_index":     {"window_end": "15:35", "weekend_only": False},
     "kline_daily":     {"window_end": "15:45", "weekend_only": False},
     "kline_5min":      {"window_end": "16:05", "weekend_only": False},
     "market_regime":   {"window_end": "16:20", "weekend_only": False},
-    "share_snapshot":  {"window_end": "16:25", "weekend_only": False},
+    "share_snapshot":  {"window_end": "12:00", "weekend_only": True},   # 周六
+
     "moneyflow":       {"window_end": "21:40", "weekend_only": False},
     "longhubang":      {"window_end": "22:05", "weekend_only": False},
     "stock_node":      {"window_end": "12:00", "weekend_only": True},   # 周六
     "mairui_ma_daily": {"window_end": "17:00", "weekend_only": False},  # 本地派生
+    "mairui_indicators": {"window_end": "17:10", "weekend_only": False},  # mairui MACD/KDJ/BOLL
 }
 
 
@@ -245,25 +268,29 @@ def _classify_task(
 
 
 # ── systemd unit drift 检测 ────────────────────────────────────────────────
-SYSTEMD_REPO_DIR = PROJECT_ROOT / "bin" / "systemd"
+SYSTEMD_REPO_DIR = PROJECT_ROOT / "deploy" / "systemd" / "units"
+SYSTEMD_LEGACY_REPO_DIR = PROJECT_ROOT / "bin" / "systemd"
 SYSTEMD_ETC_DIR = Path("/etc/systemd/system")
 
 
 def _check_systemd_drift() -> dict[str, Any]:
-    """对比 bin/systemd/ 与 /etc/systemd/system/ 下的 market-sync* unit。
+    """对比 deploy/systemd/units/ 与 /etc/systemd/system/ 下的 market-sync* unit。
+
+    兼容旧路径 bin/systemd/：如果 deploy/systemd/units/ 不存在则回退。
 
     返回：
       missing_in_etc: repo 有但 /etc 没装（drift 1：未部署）
       drifted:        两边都有但内容不一致（drift 2：部署的版本过期）
       orphan_in_etc:  /etc 有但 repo 没有（drift 3：临时/僵尸 unit）
     """
+    repo_dir = SYSTEMD_REPO_DIR if SYSTEMD_REPO_DIR.exists() else SYSTEMD_LEGACY_REPO_DIR
     missing_in_etc: list[str] = []
     drifted: list[dict[str, str]] = []
     orphan_in_etc: list[str] = []
 
     repo_units: set[str] = set()
-    if SYSTEMD_REPO_DIR.exists():
+    if repo_dir.exists():
-        for f in SYSTEMD_REPO_DIR.iterdir():
+        for f in repo_dir.iterdir():
             if f.suffix in {".service", ".timer"} and f.name.startswith("market-sync"):
                 repo_units.add(f.name)
 
@@ -279,8 +306,12 @@ def _check_systemd_drift() -> dict[str, Any]:
 
     # drift 2: 两边都有但内容不一致
     for name in sorted(repo_units & etc_units):
-        repo_path = SYSTEMD_REPO_DIR / name
+        repo_path = repo_dir / name
         etc_path = SYSTEMD_ETC_DIR / name
+        # 跳过被 mask 的 unit：/etc 下是 /dev/null 的符号链接，表示 systemd 故意禁用
+        # 这是 deploy.sh 对 market-sync-worker.service 的标准操作，不应报 drift
+        if etc_path.is_symlink() and etc_path.resolve() == Path("/dev/null"):
+            continue
         try:
             if repo_path.read_bytes() != etc_path.read_bytes():
                 drifted.append({
@@ -308,6 +339,87 @@ def _check_systemd_drift() -> dict[str, Any]:
     }
 
 
+# ── 数据一致性检查 ─────────────────────────────────────────────────────────
+
+def _check_data_consistency() -> dict[str, Any]:
+    """检查上游 kline_stock 与下游衍生表是否同步。
+
+    当前覆盖：
+      - kline_stock_ma_daily 是否比 kline_stock 缺行/缺日期
+      - market_regime_daily 是否比 kline_stock 缺最近交易日
+
+    返回：
+      alerts: 可读告警列表
+      details: 原始指标
+    """
+    from app.core.config import settings
+
+    alerts: list[dict[str, str]] = []
+    details: dict[str, Any] = {"checked": False}
+
+    try:
+        engine = sa.create_engine(settings.pg_sqlalchemy_url())
+        with engine.connect() as conn:
+            # 1) kline_stock vs kline_stock_ma_daily
+            kline_max = conn.execute(
+                sa.text("SELECT MAX(trade_date) FROM market_data.kline_stock")
+            ).scalar()
+            ma_max = conn.execute(
+                sa.text("SELECT MAX(trade_date) FROM market_data.kline_stock_ma_daily")
+            ).scalar()
+            missing_rows = conn.execute(
+                sa.text("""
+                    SELECT COUNT(*) FROM (
+                        SELECT stock_code, trade_date FROM market_data.kline_stock
+                        EXCEPT
+                        SELECT stock_code, trade_date FROM market_data.kline_stock_ma_daily
+                    ) t
+                """)
+            ).scalar() or 0
+
+            details["kline_stock_max_date"] = str(kline_max) if kline_max else None
+            details["kline_stock_ma_daily_max_date"] = str(ma_max) if ma_max else None
+            details["kline_ma_missing_rows"] = missing_rows
+
+            if kline_max and ma_max and ma_max < kline_max:
+                alerts.append({
+                    "dataset_id": "data_consistency/kline_ma_daily",
+                    "status": "stale",
+                    "reason": f"kline_stock_ma_daily 最新日期 {ma_max} 落后于 kline_stock {kline_max}",
+                })
+            elif missing_rows and missing_rows > 0:
+                alerts.append({
+                    "dataset_id": "data_consistency/kline_ma_daily",
+                    "status": "gap",
+                    "reason": f"kline_stock 有 {missing_rows} 行未覆盖到 kline_stock_ma_daily",
+                })
+
+            # 2) kline_stock vs market_regime_daily（最近 3 个交易日）
+            regime_max = conn.execute(
+                sa.text("SELECT MAX(trade_date) FROM market_data.market_regime_daily")
+            ).scalar()
+
+            details["market_regime_daily_max_date"] = str(regime_max) if regime_max else None
+
+            if kline_max and regime_max and regime_max < kline_max:
+                alerts.append({
+                    "dataset_id": "data_consistency/market_regime",
+                    "status": "stale",
+                    "reason": f"market_regime_daily 最新日期 {regime_max} 落后于 kline_stock {kline_max}",
+                })
+
+        details["checked"] = True
+    except Exception as e:
+        details["error"] = str(e)
+        alerts.append({
+            "dataset_id": "data_consistency/check_failed",
+            "status": "error",
+            "reason": f"数据一致性检查异常: {e}",
+        })
+
+    return {"alerts": alerts, "details": details}
+
+
 # ── 报告生成 ───────────────────────────────────────────────────────────────
 def build_report(check_date: date) -> dict[str, Any]:
     """生成当日巡检报告。"""
@@ -338,7 +450,7 @@ def build_report(check_date: date) -> dict[str, Any]:
                 alerts.append({
                     "dataset_id": f"systemd/{u}",
                     "status": "not_deployed",
-                    "reason": f"unit {u} 在 bin/systemd/ 有但 /etc/systemd/system/ 没装",
+                    "reason": f"unit {u} 在 deploy/systemd/units/ 有但 /etc/systemd/system/ 没装",
                 })
         if drift["drifted"]:
             for d in drift["drifted"]:
@@ -355,6 +467,10 @@ def build_report(check_date: date) -> dict[str, Any]:
                     "reason": f"unit {u} 在 /etc/systemd/system/ 但 repo 没维护",
                 })
 
+    # 数据一致性检查（上游回填后下游未跟进）
+    consistency = _check_data_consistency()
+    alerts.extend(consistency["alerts"])
+
     overall = "ok" if not alerts else (
         "warning" if any(a["status"] in {"missed", "never_run", "not_deployed", "drifted"} for a in alerts) else "error"
     )
@@ -363,11 +479,13 @@ def build_report(check_date: date) -> dict[str, Any]:
         "schema_version": 2,
         "generated_at": now.strftime("%Y-%m-%d %H:%M:%S"),
         "check_date": check_date.isoformat(),
+        "runtime_mode": _runtime_mode(),
         "overall": overall,
         "counts": counts,
         "alerts": alerts,
         "tasks": classified,
         "systemd_drift": drift,
+        "data_consistency": consistency["details"],
     }
 
 
@@ -383,7 +501,7 @@ def write_report(report: dict[str, Any]) -> Path:
 def print_summary(report: dict[str, Any]) -> None:
     """stdout 一眼看懂的汇总。"""
     print("=" * 78)
-    print(f"每日同步巡检  date={report['check_date']}  overall={report['overall']}")
+    print(f"每日同步巡检  date={report['check_date']}  runtime_mode={report.get('runtime_mode', 'unknown')}  overall={report['overall']}")
     print("=" * 78)
     print(f"counts: {report['counts']}")
     if report["alerts"]:

bin/deploy_systemd.sh

@@ -0,0 +1 @@
+deploy/systemd/deploy.sh

bin/market_sync_kline_5min_run.sh

@@ -0,0 +1,17 @@
+#!/bin/bash
+# systemd 调度入口：单跑 kline_5min（评分核心依赖）
+set -u
+
+PROJECT_ROOT="/home/gao/Development/quant_home/market_sync"
+LOG_DIR="$PROJECT_ROOT/logs"
+mkdir -p "$LOG_DIR"
+
+TS=$(date +%Y%m%d_%H%M%S)
+LOG="$LOG_DIR/kline_5min_${TS}.log"
+
+echo "[market-sync-kline-5min] start ts=$TS log=$LOG" | tee -a "$LOG"
+cd "$PROJECT_ROOT"
+.venv/bin/python -m app.entrypoints.cli sync kline_5min 2>&1 | tee -a "$LOG"
+EXIT_CODE=${PIPESTATUS[0]}
+echo "[market-sync-kline-5min] done exit_code=$EXIT_CODE" | tee -a "$LOG"
+exit $EXIT_CODE

bin/market_sync_kline_daily_run.sh

@@ -0,0 +1,17 @@
+#!/bin/bash
+# systemd 调度入口：单跑 kline_daily（评分核心依赖）
+set -u
+
+PROJECT_ROOT="/home/gao/Development/quant_home/market_sync"
+LOG_DIR="$PROJECT_ROOT/logs"
+mkdir -p "$LOG_DIR"
+
+TS=$(date +%Y%m%d_%H%M%S)
+LOG="$LOG_DIR/kline_daily_${TS}.log"
+
+echo "[market-sync-kline-daily] start ts=$TS log=$LOG" | tee -a "$LOG"
+cd "$PROJECT_ROOT"
+.venv/bin/python -m app.entrypoints.cli sync kline_daily 2>&1 | tee -a "$LOG"
+EXIT_CODE=${PIPESTATUS[0]}
+echo "[market-sync-kline-daily] done exit_code=$EXIT_CODE" | tee -a "$LOG"
+exit $EXIT_CODE

bin/market_sync_kline_index_run.sh

@@ -0,0 +1,17 @@
+#!/bin/bash
+# systemd 调度入口：单跑 kline_index
+set -u
+
+PROJECT_ROOT="/home/gao/Development/quant_home/market_sync"
+LOG_DIR="$PROJECT_ROOT/logs"
+mkdir -p "$LOG_DIR"
+
+TS=$(date +%Y%m%d_%H%M%S)
+LOG="$LOG_DIR/kline_index_${TS}.log"
+
+echo "[market-sync-kline-index] start ts=$TS log=$LOG" | tee -a "$LOG"
+cd "$PROJECT_ROOT"
+.venv/bin/python -m app.entrypoints.cli sync kline_index 2>&1 | tee -a "$LOG"
+EXIT_CODE=${PIPESTATUS[0]}
+echo "[market-sync-kline-index] done exit_code=$EXIT_CODE" | tee -a "$LOG"
+exit $EXIT_CODE

bin/market_sync_mairui_indicators_run.sh

@@ -0,0 +1,25 @@
+#!/bin/bash
+# systemd 调度的入口（专跑 mairui_indicators）
+#
+# 设计：
+#   - 交易日 16:40 触发，从 mairui 拉 MACD/KDJ/BOLL 增量
+#   - 数据量大（3 指标 × 全市场），单独跑避免拖垮 runall 链
+#   - 单独日志到 logs/mairui_indicators_<ts>.log
+#   - exit code 透传
+set -u
+
+PROJECT_ROOT="/home/gao/Development/quant_home/market_sync"
+LOG_DIR="$PROJECT_ROOT/logs"
+mkdir -p "$LOG_DIR"
+
+TS=$(date +%Y%m%d_%H%M%S)
+LOG="$LOG_DIR/mairui_indicators_${TS}.log"
+
+echo "[market-sync-mairui-indicators] start ts=$TS log=$LOG" | tee -a "$LOG"
+
+# ── 跑 mairui_indicators ──
+cd "$PROJECT_ROOT"
+.venv/bin/python -m app.entrypoints.cli sync mairui_indicators 2>&1 | tee -a "$LOG"
+EXIT_CODE=${PIPESTATUS[0]}
+echo "[market-sync-mairui-indicators] done exit_code=$EXIT_CODE" | tee -a "$LOG"
+exit $EXIT_CODE

bin/market_sync_mairui_ma_daily_run.sh

@@ -0,0 +1,25 @@
+#!/bin/bash
+# systemd 调度的入口（专跑 mairui_ma_daily）
+#
+# 设计：
+#   - 交易日 16:30 触发，基于 kline_stock 计算全市场 MA5/10/20/60
+#   - 本地派生任务，无外部 API 依赖
+#   - 单独日志到 logs/mairui_ma_daily_<ts>.log
+#   - exit code 透传
+set -u
+
+PROJECT_ROOT="/home/gao/Development/quant_home/market_sync"
+LOG_DIR="$PROJECT_ROOT/logs"
+mkdir -p "$LOG_DIR"
+
+TS=$(date +%Y%m%d_%H%M%S)
+LOG="$LOG_DIR/mairui_ma_daily_${TS}.log"
+
+echo "[market-sync-mairui-ma-daily] start ts=$TS log=$LOG" | tee -a "$LOG"
+
+# ── 跑 mairui_ma_daily ──
+cd "$PROJECT_ROOT"
+.venv/bin/python -m app.entrypoints.cli sync mairui_ma_daily 2>&1 | tee -a "$LOG"
+EXIT_CODE=${PIPESTATUS[0]}
+echo "[market-sync-mairui-ma-daily] done exit_code=$EXIT_CODE" | tee -a "$LOG"
+exit $EXIT_CODE

bin/market_sync_market_regime_run.sh

@@ -0,0 +1,17 @@
+#!/bin/bash
+# systemd 调度入口：单跑 market_regime（本地计算，基于 kline_stock）
+set -u
+
+PROJECT_ROOT="/home/gao/Development/quant_home/market_sync"
+LOG_DIR="$PROJECT_ROOT/logs"
+mkdir -p "$LOG_DIR"
+
+TS=$(date +%Y%m%d_%H%M%S)
+LOG="$LOG_DIR/market_regime_${TS}.log"
+
+echo "[market-sync-market-regime] start ts=$TS log=$LOG" | tee -a "$LOG"
+cd "$PROJECT_ROOT"
+.venv/bin/python -m app.entrypoints.cli sync market_regime 2>&1 | tee -a "$LOG"
+EXIT_CODE=${PIPESTATUS[0]}
+echo "[market-sync-market-regime] done exit_code=$EXIT_CODE" | tee -a "$LOG"
+exit $EXIT_CODE

bin/market_sync_morning_run.sh

@@ -1,8 +1,6 @@
 #!/bin/bash
-# 早盘前同步：stock_basic + industry_sector
+# 早盘前同步：stock_basic（仅）
 # - stock_basic：雪球 quote_detail 拉全市场股本快照（~9min @ 5 RPS）
-# - industry_sector：baostock 拉股票-行业映射（~16min @ 5 RPS）
-# - 串行执行，industry_sector 依赖 stock_basic
 # - 与 runall_once.py 区别：不拉 kline_daily / kline_5min 等耗时长任务
 
 set -u
@@ -27,7 +25,7 @@ seed_sync_registry()
 from app.tasks import get_task
 
 results = []
-for tid in ['stock_basic', 'industry_sector']:
+for tid in ['stock_basic']:
     print(f'[morning-run] >>> 开始 {tid}', flush=True)
     t0 = time.time()
     try:

bin/market_sync_share_run.sh

@@ -0,0 +1,24 @@
+#!/bin/bash
+# systemd 调度的入口（专跑股本快照）
+#
+# 设计：
+#   - 与 market_sync_stock_node_run.sh 同构,只跑 share_snapshot 一个 task
+#   - 每周六 11:30 触发（周末收盘后股本数据稳定，与 stock_node 同时段）
+#   - 单独日志到 logs/share_<ts>.log
+#   - exit code 透传
+set -u
+
+PROJECT_ROOT="/home/gao/Development/quant_home/market_sync"
+LOG_DIR="$PROJECT_ROOT/logs"
+mkdir -p "$LOG_DIR"
+
+TS=$(date +%Y%m%d_%H%M%S)
+LOG="$LOG_DIR/share_${TS}.log"
+
+echo "[market-sync-share] start ts=$TS log=$LOG" | tee -a "$LOG"
+
+cd "$PROJECT_ROOT"
+.venv/bin/python -m app.entrypoints.cli sync share_snapshot 2>&1 | tee -a "$LOG"
+EXIT_CODE=${PIPESTATUS[0]}
+echo "[market-sync-share] done exit_code=$EXIT_CODE" | tee -a "$LOG"
+exit $EXIT_CODE

bin/run_remaining.sh

@@ -1,5 +1,6 @@
 #!/bin/bash
-# 跑剩余任务：tick_trade → moneyflow → share_snapshot → market_regime
+# 跑剩余任务：tick_trade → moneyflow → market_regime
+# share_snapshot 已改为每周单独 timer（market-sync-share.timer），不在此处触发
 # 跳过 kline_5min（全量回填 6 年太慢，下次有空再补）
 # tick_trade 用 --force 跳过 21:00 门控（人工补跑场景）
 set -e
@@ -10,7 +11,7 @@ mkdir -p logs
 # 先清理上次可能留下的卡死记录（recover_interrupted_syncs 启动时自动处理，但显式更稳）
 echo "=== 启动时间: $(date) ===" | tee logs/runall_remaining.log
 
-for TASK in "tick_trade --force" moneyflow share_snapshot market_regime; do
+for TASK in "tick_trade --force" moneyflow market_regime; do
   echo "" | tee -a logs/runall_remaining.log
   echo "=== [$TASK] 开始 $(date) ===" | tee -a logs/runall_remaining.log
   T0=$(date +%s)

bin/runall_once.py

@@ -9,13 +9,26 @@
 注：以下 task **不**在 runall 链中 —— 各自有独立的 systemd timer：
   - tick_trade   : mairui 21:00 发布 → market-sync-tick.timer (21:05)
   - moneyflow    : mairui 21:30 发布 → market-sync-moneyflow.timer (21:35)
+
+2026-07-08 教训: 之前 runall 是顺序串行,任一 task 永久卡住(kline_daily 在
+5200/5204 卡死 4h)→ 后面 5 个 task 全部没机会跑,runall 进程被 systemd 4h
+超时杀, dataset_registry 卡在 running 状态无人清理。修复:
+  1. 启动时调 recover_interrupted_dataset_registry() — 清上次被中断的卡死
+  2. 每个 task 用 multiprocessing.Process 跑 + 硬上限 PER_TASK_TIMEOUT_SEC,
+     超时直接 kill 子进程,继续下一个 task (不卡整条 runall)
+  3. 数据源/任务层也加了 timeout(xueqiu 20s + kline_daily as_completed 30s),
+     这里是最后一道防线
 """
 from __future__ import annotations
 
 import json
+import multiprocessing as mp
+import os
+import signal
 import sys
 import time
 from pathlib import Path
+from typing import Any
 
 _PROJECT_ROOT = Path(__file__).resolve().parent.parent
 if str(_PROJECT_ROOT) not in sys.path:
@@ -30,8 +43,19 @@ from app.core.sync.registry import seed_sync_registry
 build_default_registry()
 seed_sync_registry()
 
+from app.core.db import ops as db_ops
 from app.tasks import get_task
 
+# 启动时清理上一轮被中断的卡死状态(2026-07-08 教训)
+# 注意: 这里清的是"上次 runall 被杀时的 running",如果当前 runall 自己卡死,
+# 子进程被 kill 后 recover 仍要再调一次 — 见末尾的 finally。
+try:
+    n_recovered = db_ops.recover_interrupted_dataset_registry()
+    if n_recovered:
+        logger.warning("[runall] 启动时清掉 %d 个上轮卡死的 running 任务", n_recovered)
+except Exception as e:
+    logger.warning("[runall] 启动清理失败(非致命): %s", e)
+
 # 顺序执行，按 sort_order 走（便于排查 + 避免外部 API 限流）
 # 注：tick_trade / moneyflow 由独立 timer 跑（21:05 / 21:35）
 TASKS = [
@@ -39,48 +63,111 @@ TASKS = [
     "kline_index",
     "kline_daily",
     "kline_5min",
-    "industry_sector",
-    "sector_features",
-    "share_snapshot",
     "market_regime",
 ]
 
 # 不跳任何 task — 全量跑
 SKIP: set[str] = set()
 
-results = {}
+# 每个 task 的硬上限（秒）。覆盖数据源/任务层都失败的最坏情况:
-t_all = time.time()
+#   kline_daily:  35 min (正常 18 min, 留余量)
-for tid in TASKS:
+#   kline_5min:  75 min (全量首次跑很久, 增量 ~30 min)
-    if tid in SKIP:
+#   share_snapshot: 15 min
-        results[tid] = {"status": "skipped", "message": "已 ok"}
+#   其他小表: 10 min
-        logger.info(f"[runall] {tid} 跳过（已 ok）")
+PER_TASK_TIMEOUT_SEC = {
-        continue
+    "kline_daily": 35 * 60,
-    logger.info(f"[runall] >>> 开始 {tid}")
+    "kline_5min": 75 * 60,
-    t0 = time.time()
+    "share_snapshot": 15 * 60,
+    "market_regime": 5 * 60,
+    "stock_basic": 15 * 60,
+    "kline_index": 5 * 60,
+}
+DEFAULT_TASK_TIMEOUT = 20 * 60
+
+
+def _run_task_in_subprocess(tid: str, result_queue: mp.Queue) -> None:
+    """子进程入口: 跑 task.run() 并把 result 通过 queue 返回。
+
+    单独进程是为了父进程能用 SIGKILL 干掉它(timeout 时)而不污染主 runall 状态。
+    """
     try:
         task = get_task(tid)
-        # 大表 task 限小批量股票，避免跑爆；其他 task 跑全量
         kwargs = {"max_workers": 5}
-        if tid in {"kline_daily", "kline_5min", "moneyflow", "share_snapshot"}:
-            # 这些 task 支持 MARKET_DATA_STOCK_LIMIT 环境变量，但通过 cli 跑可 --codes 限
-            # 测全量费时，先跑全量
-            pass
         r = task.run(trigger_source="runall", **kwargs)
-        elapsed = round(time.time() - t0, 1)
+        result_queue.put(r)
-        r["elapsed_sec"] = elapsed
-        results[tid] = r
-        logger.info(f"[runall] <<< {tid} 完成 status={r.get('status')} elapsed={elapsed}s msg={r.get('message')}")
     except Exception as e:
+        result_queue.put({"status": "error", "message": f"{type(e).__name__}: {e}"})
+
+
+results: dict[str, Any] = {}
+t_all = time.time()
+
+# 注意: 主执行块必须在 __name__ == '__main__' 内部，否则 multiprocessing spawn
+# 子进程重新 import 本模块时会再次执行进程启动逻辑，导致 RuntimeError:
+# "An attempt has been made to start a new process before the current process
+# has finished its bootstrapping phase." (2026-07-13 15:30 runall 全部失败)
+def _run_tasks():
+    global results
+    for tid in TASKS:
+        if tid in SKIP:
+            results[tid] = {"status": "skipped", "message": "已 ok"}
+            logger.info(f"[runall] {tid} 跳过（已 ok）")
+            continue
+        timeout_sec = PER_TASK_TIMEOUT_SEC.get(tid, DEFAULT_TASK_TIMEOUT)
+        logger.info(f"[runall] >>> 开始 {tid} (hard timeout {timeout_sec}s)")
+        t0 = time.time()
+        # 子进程跑 task,父进程用 .join(timeout) 守门
+        ctx = mp.get_context("spawn")
+        result_queue: mp.Queue = ctx.Queue()
+        proc = ctx.Process(target=_run_task_in_subprocess, args=(tid, result_queue), name=f"runall-{tid}")
+        proc.start()
+        proc.join(timeout=timeout_sec)
         elapsed = round(time.time() - t0, 1)
-        logger.exception(f"[runall] !!! {tid} 异常: {e}")
-        results[tid] = {"status": "error", "message": str(e), "elapsed_sec": elapsed}
 
-    # 每个 task 之间 sleep 5s 让健康监控跑
+        if proc.is_alive():
-    time.sleep(5)
+            logger.error(
+                f"[runall] !!! {tid} 超时 (>{timeout_sec}s, elapsed={elapsed}s), 强制 kill"
+            )
+            proc.terminate()
+            proc.join(5)
+            if proc.is_alive():
+                proc.kill()
+                proc.join(2)
+            results[tid] = {
+                "status": "error",
+                "message": f"runall timeout (>={timeout_sec}s), killed",
+                "elapsed_sec": elapsed,
+            }
+            try:
+                db_ops.recover_interrupted_dataset_registry()
+            except Exception as e:
+                logger.warning("[runall] 超时后清理 dataset_registry 失败: %s", e)
+        else:
+            try:
+                r = result_queue.get(timeout=5)
+            except Exception as e:
+                r = {"status": "error", "message": f"无法获取子进程结果: {e}"}
+            r["elapsed_sec"] = elapsed
+            results[tid] = r
+            logger.info(
+                f"[runall] <<< {tid} 完成 status={r.get('status')} elapsed={elapsed}s msg={r.get('message')}"
+            )
 
-elapsed_total = round(time.time() - t_all, 1)
+        time.sleep(5)
-summary = {"total_elapsed_sec": elapsed_total, "tasks": results}
+
-out = _PROJECT_ROOT / "logs" / "runall_summary.json"
+
-out.write_text(json.dumps(summary, ensure_ascii=False, indent=2))
+if __name__ == '__main__':
-logger.info(f"[runall] 全部完成 total={elapsed_total}s summary={out}")
+    _run_tasks()
-print(json.dumps(summary, ensure_ascii=False, indent=2))
+
+    # 全部完成清一次 stuck 状态
+    try:
+        db_ops.recover_interrupted_dataset_registry()
+    except Exception as e:
+        logger.warning("[runall] 末尾清理失败(非致命): %s", e)
+
+    elapsed_total = round(time.time() - t_all, 1)
+    summary: dict[str, Any] = {"total_elapsed_sec": elapsed_total, "tasks": results}
+    out = _PROJECT_ROOT / "logs" / "runall_summary.json"
+    out.write_text(json.dumps(summary, ensure_ascii=False, indent=2))
+    logger.info(f"[runall] 全部完成 total={elapsed_total}s summary={out}")
+    print(json.dumps(summary, ensure_ascii=False, indent=2))

bin/service_run.sh

@@ -10,6 +10,15 @@
 
 set -eu
 
+RUNTIME_MODE="${RUNTIME_MODE:-docker}"
+
+if [ "$RUNTIME_MODE" != "docker" ]; then
+    echo "[service] ❌ ERROR: service_run.sh 是 Docker 入口，只在 RUNTIME_MODE=docker 时可用。" >&2
+    echo "[service]    当前 RUNTIME_MODE=$RUNTIME_MODE" >&2
+    echo "[service]    Linux 宿主机部署请使用 systemd timer/service，不要运行此脚本。" >&2
+    exit 1
+fi
+
 # 等 PG 就绪（避免 race：market_sync 比 postgres 早启）
 if [ -n "${PG_HOST:-}" ]; then
     echo "[service] wait for PG ${PG_HOST}:${PG_PORT:-5432}..."

bin/systemd/market-sync-mairui-indicators.service

@@ -0,0 +1,32 @@
+[Unit]
+Description=Market data MACD/KDJ/BOLL indicators sync (mairui, daily 16:40)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_mairui_indicators_run.sh
+After=network-online.target market-sync.service
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 入口脚本：单独跑 mairui_indicators 一个 task
+# mairui 16:30 后发布指标数据 → 16:40 触发留 buffer
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_mairui_indicators_run.sh
+
+# 硬上限 3h（5204 只 × 3 指标，正常 ~20min，全量时可能更久）
+TimeoutStartSec=10800
+
+# 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑）
+# Restart=no 是 oneshot 默认值
+
+# 日志走 journald
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-mairui-indicators
+
+# 环境
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

bin/systemd/market-sync-mairui-indicators.timer

@@ -0,0 +1,14 @@
+[Unit]
+Description=Schedule MACD/KDJ/BOLL indicators sync — Mon..Fri 16:40 Asia/Shanghai
+# market-sync-mairui-indicators.service 是这个 timer 的执行单元
+
+[Timer]
+# mairui 16:30 后发布指标数据 → 16:40 触发留 buffer
+OnCalendar=Mon..Fri 16:40:00 Asia/Shanghai
+# 系统关机 / 错过执行时，下次开机补跑一次（避免漏数据）
+Persistent=true
+# 单位（service）的精确名称
+Unit=market-sync-mairui-indicators.service
+
+[Install]
+WantedBy=timers.target

bin/systemd/market-sync-mairui-ma-daily.service

@@ -0,0 +1,32 @@
+[Unit]
+Description=Market data MA daily sync (local derivative, daily 16:30)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_mairui_ma_daily_run.sh
+After=network-online.target market-sync.service
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 入口脚本：单独跑 mairui_ma_daily 一个 task
+# 本地派生任务，依赖 kline_stock 已更新
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_mairui_ma_daily_run.sh
+
+# 硬上限 2h（正常 ~30min，留 buffer）
+TimeoutStartSec=7200
+
+# 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑）
+# Restart=no 是 oneshot 默认值
+
+# 日志走 journald
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-mairui-ma-daily
+
+# 环境
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

bin/systemd/market-sync-mairui-ma-daily.timer

@@ -0,0 +1,14 @@
+[Unit]
+Description=Schedule MA daily sync — Mon..Fri 16:30 Asia/Shanghai
+# market-sync-mairui-ma-daily.service 是这个 timer 的执行单元
+
+[Timer]
+# kline_daily 15:40 / kline_index 15:30 跑完后，16:30 计算 MA
+OnCalendar=Mon..Fri 16:30:00 Asia/Shanghai
+# 系统关机 / 错过执行时，下次开机补跑一次（避免漏数据）
+Persistent=true
+# 单位（service）的精确名称
+Unit=market-sync-mairui-ma-daily.service
+
+[Install]
+WantedBy=timers.target

bin/systemd/market-sync-morning.service

@@ -1,5 +1,5 @@
 [Unit]
-Description=Market sync morning pre-market — stock_basic + industry_sector
+Description=Market sync morning pre-market — stock_basic
 Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_morning_run.sh
 After=network-online.target
 Wants=network-online.target
@@ -10,10 +10,10 @@ WorkingDirectory=/home/gao/Development/quant_home/market_sync
 User=gao
 Group=gao
 
-# 早盘前只跑 stock_basic + industry_sector（避免和 15:30 runall 重复）
+# 早盘前只跑 stock_basic（确保最新股票列表+股本快照）
 ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_morning_run.sh
 
-# stock_basic ~9min + industry_sector ~16min + sleep 5s + buffer = 1h 够用
+# stock_basic ~9min + buffer = 1h 够用
 TimeoutStartSec=3600
 
 # 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑）

bin/systemd/market-sync-morning.timer

@@ -1,10 +1,10 @@
 [Unit]
 Description=Schedule market sync morning — Mon..Fri 09:00 Asia/Shanghai
 # market-sync-morning.service 是这个 timer 的执行单元
-# 早盘前 09:00 触发：跑 stock_basic + industry_sector，给 09:30 开盘留 30min buffer
+# 早盘前 09:00 触发：跑 stock_basic，给 09:30 开盘留 30min buffer
 
 [Timer]
-# A 股开盘 09:30，09:00 跑 stock_basic (~9min) + industry_sector (~16min) = ~25min
+# A 股开盘 09:30，09:00 跑 stock_basic (~9min) 即可
 OnCalendar=Mon..Fri 09:00:00 Asia/Shanghai
 # 关机/错过时下次开机补跑
 Persistent=true

bin/systemd/market-sync-share.service

@@ -0,0 +1,31 @@
+[Unit]
+Description=Market data 股本快照 sync (雪球源, weekly Sat 11:30)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_share_run.sh
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 入口脚本：单独跑 share_snapshot 一个 task
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_share_run.sh
+
+# 硬上限 30min（全市场 5200 只雪球 quote_detail，10 worker 约 3-8min，留 buffer）
+TimeoutStartSec=1800
+
+# 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑）
+# Restart=no 是 oneshot 默认值
+
+# 日志走 journald（journalctl -u market-sync-share.service -f）
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-share
+
+# 环境（不读 /etc/environment，只带这几个；.env 由 cli 内部 load_dotenv）
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

bin/systemd/market-sync-share.timer

@@ -0,0 +1,16 @@
+[Unit]
+Description=Schedule share_snapshot sync — Sat 11:30 Asia/Shanghai
+# market-sync-share.service 是这个 timer 的执行单元
+# 股本快照每周跑一次即可（雪球 quote_detail，周末收盘后数据稳定）
+
+[Timer]
+# 周六 11:30（和 stock_node 同时段）
+OnCalendar=Sat *-*-* 11:30:00 Asia/Shanghai
+# 系统关机 / 错过执行时,下次开机补跑一次(避免漏数据)
+Persistent=true
+# 单位(service)的精确名称
+Unit=market-sync-share.service
+# 不要 AccuracySec(默认 1min 漂移够用)
+
+[Install]
+WantedBy=timers.target

bin/systemd/market-sync-stock-node.service

@@ -1,7 +1,7 @@
 [Unit]
 Description=Market data 股票-节点映射 sync (mairui /hszg, weekly Sat 11:30)
 Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_stock_node_run.sh
-After=network-online.target
+After=network-online.target market-sync-morning.service
 Wants=network-online.target
 
 [Service]

bin/systemd/market-sync-worker.service

@@ -0,0 +1,25 @@
+[Unit]
+Description=Market Data Sync Worker (PostgreSQL, scheduler + health monitor)
+Documentation=file:///home/gao/Development/quant_home/market_sync/app/entrypoints/worker.py
+# 依赖 network 和 PG
+After=network-online.target postgresql.service
+Wants=network-online.target
+
+[Service]
+Type=simple
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 入口：启动进程内 scheduler + health monitor + recover 卡死任务
+# 注意：实际入口是 app/entrypoints/worker.py，不是 app/worker.py
+# 2026-07-12 修复: 把 python -m app.worker 改为 python -m app.entrypoints.worker
+ExecStart=/home/gao/Development/quant_home/market_sync/.venv/bin/python -m app.entrypoints.worker
+
+# 明确标记运行模式，让 worker.py 知道不要启动 scheduler
+# （systemd 模式下同步由 timer 触发，避免双调度器重叠）
+Environment=PYTHONUNBUFFERED=1
+Environment="RUNTIME_MODE=systemd"
+
+[Install]
+WantedBy=multi-user.target

bin/systemd/market-sync.service

@@ -13,10 +13,11 @@ Group=gao
 # 入口脚本（runall_once.py + structured_watch.sh 二合一）
 ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_run.sh
 
-# 硬上限 4h（实际 ~80min + kline_5min 增量 70min, 总 ~2.5h, 留 buffer 给 retry / 慢任务）
+# 硬上限 3h（实际 ~80min + kline_5min 增量 70min, 总 ~2.5h, 留 30min buffer）
-# 历史 bug（2026-07-02）：2h 超时截断，share_snapshot / market_regime 没跑成
+# 2026-07-08 教训: 之前 4h 让 kline_daily 单点卡死拖 4h,runall 链全部受影响。
-# 详见 docs/works/2026-07-03-01-market-sync-timeout.md
+# runall_once.py 已加 per-task multiprocessing timeout(单 task 最长 75min),
-TimeoutStartSec=14400
+# 这里 3h 是 systemd 层兜底。
+TimeoutStartSec=10800
 
 # 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑——
 # 重跑会撞外部 API 限流窗口，浪费 1.5h）

bin/systemd_deploy.sh

@@ -1,185 +0,0 @@
-#!/bin/bash
-# market_sync systemd 一键部署脚本
-#
-# 用途：把 bin/systemd/*.service 和 *.timer 同步到 /etc/systemd/system/，
-#       解决 work #04 检测到的 5 处 drift。
-#
-# 运行：sudo bash bin/systemd_deploy.sh
-#
-# 设计原则：
-#   - 必须 sudo（写 /etc/systemd/system/ + daemon-reload）
-#   - 幂等：重复运行无副作用
-#   - 显示 plan → 等待确认 → 执行
-#   - 失败立即中止，不留半套状态
-
-set -e
-
-PROJECT_ROOT="/home/gao/Development/quant_home/market_sync"
-REPO_UNITS_DIR="$PROJECT_ROOT/bin/systemd"
-ETC_UNITS_DIR="/etc/systemd/system"
-
-# ── 0. 前置检查 ─────────────────────────────────────────────────────────────
-if [ "$(id -u)" -ne 0 ]; then
-  echo "❌ 必须 sudo 运行：sudo bash $0"
-  exit 1
-fi
-
-if [ ! -d "$REPO_UNITS_DIR" ]; then
-  echo "❌ repo unit 目录不存在: $REPO_UNITS_DIR"
-  exit 1
-fi
-
-# ── 1. 计算 plan ────────────────────────────────────────────────────────────
-declare -a TO_COPY=()        # repo→etc 拷贝
-declare -a TO_REMOVE=()      # etc 中孤儿（repo 已删）
-declare -a TO_RESTART_SVC=() # service 文件变了需要 daemon-reload + restart
-
-# repo 中所有 market-sync* unit
-mapfile -t REPO_UNITS < <(find "$REPO_UNITS_DIR" -maxdepth 1 -type f \( -name "*.service" -o -name "*.timer" \) -printf '%f\n' | sort)
-# /etc 中所有 market-sync* unit
-mapfile -t ETC_UNITS < <(find "$ETC_UNITS_DIR" -maxdepth 1 -type f \( -name "market-sync*.service" -o -name "market-sync*.timer" \) -printf '%f\n' | sort 2>/dev/null || true)
-
-# drift 1: repo 有 /etc 没装
-for u in "${REPO_UNITS[@]}"; do
-  if [ ! -f "$ETC_UNITS_DIR/$u" ]; then
-    TO_COPY+=("$u")
-  fi
-done
-
-# drift 2: 两边都有但内容不一致
-for u in "${REPO_UNITS[@]}"; do
-  if [ -f "$ETC_UNITS_DIR/$u" ]; then
-    if ! diff -q "$REPO_UNITS_DIR/$u" "$ETC_UNITS_DIR/$u" > /dev/null 2>&1; then
-      TO_COPY+=("$u")
-      if [[ "$u" == *.service ]]; then
-        TO_RESTART_SVC+=("${u%.service}")
-      fi
-    fi
-  fi
-done
-
-# drift 3: /etc 有 repo 没维护（孤儿）
-for u in "${ETC_UNITS[@]}"; do
-  found=0
-  for r in "${REPO_UNITS[@]}"; do
-    if [ "$r" = "$u" ]; then found=1; break; fi
-  done
-  if [ $found -eq 0 ]; then
-    TO_REMOVE+=("$u")
-  fi
-done
-
-# ── 2. 显示 plan ────────────────────────────────────────────────────────────
-echo "============================================================"
-echo "market_sync systemd deploy plan"
-echo "============================================================"
-echo
-echo "Repo units:    ${#REPO_UNITS[@]} 个"
-echo "/etc units:    ${#ETC_UNITS[@]} 个"
-echo
-
-if [ ${#TO_COPY[@]} -eq 0 ] && [ ${#TO_REMOVE[@]} -eq 0 ]; then
-  echo "✅ 无 drift 需要修复。无需执行。"
-  exit 0
-fi
-
-if [ ${#TO_COPY[@]} -gt 0 ]; then
-  echo "📋 将要拷贝 ($(echo "${TO_COPY[@]}" | tr ' ' '\n' | wc -l) 个):"
-  for u in "${TO_COPY[@]}"; do
-    echo "    cp bin/systemd/$u  →  /etc/systemd/system/$u"
-  done
-  echo
-fi
-
-if [ ${#TO_RESTART_SVC[@]} -gt 0 ]; then
-  echo "🔄 将要重启 service ($(echo "${TO_RESTART_SVC[@]}" | tr ' ' '\n' | wc -l) 个):"
-  for s in "${TO_RESTART_SVC[@]}"; do
-    echo "    systemctl reset-failed $s.service"
-  done
-  echo
-fi
-
-if [ ${#TO_REMOVE[@]} -gt 0 ]; then
-  echo "🗑️  将要删除 etc 中的孤儿 ($(echo "${TO_REMOVE[@]}" | tr ' ' '\n' | wc -l) 个):"
-  for u in "${TO_REMOVE[@]}"; do
-    echo "    rm /etc/systemd/system/$u"
-  done
-  echo
-fi
-
-# ── 3. 等待确认 ────────────────────────────────────────────────────────────
-read -p "确认执行？[y/N] " -n 1 -r
-echo
-if [[ ! $REPLY =~ ^[Yy]$ ]]; then
-  echo "已取消。"
-  exit 0
-fi
-
-# ── 4. 执行 ─────────────────────────────────────────────────────────────────
-echo
-echo "=== 执行 ==="
-
-# 4a. cp 漂移的 unit
-for u in "${TO_COPY[@]}"; do
-  cp -v "$REPO_UNITS_DIR/$u" "$ETC_UNITS_DIR/$u"
-done
-
-# 4b. 删孤儿
-for u in "${TO_REMOVE[@]}"; do
-  rm -v "$ETC_UNITS_DIR/$u"
-done
-
-# 4c. reload systemd
-echo
-echo "=== systemctl daemon-reload ==="
-systemctl daemon-reload
-
-# 4d. reset-failed + restart 修改过的 service
-for s in "${TO_RESTART_SVC[@]}"; do
-  if systemctl is-enabled "${s}.service" 2>/dev/null | grep -q enabled; then
-    systemctl reset-failed "${s}.service" || true
-    echo "  reset-failed ${s}.service (没自动 restart — 由 timer 自然触发)"
-  fi
-done
-
-# 4e. 启用新 timer（如果有）
-for u in "${TO_COPY[@]}"; do
-  if [[ "$u" == *.timer ]]; then
-    timer="${u%.timer}"
-    if systemctl list-unit-files "${timer}.timer" 2>/dev/null | grep -q "${timer}.timer"; then
-      if ! systemctl is-enabled "${timer}.timer" 2>/dev/null | grep -q enabled; then
-        systemctl enable "${timer}.timer"
-        echo "  enable ${timer}.timer"
-      fi
-    fi
-  fi
-done
-
-# ── 5. 验证 ────────────────────────────────────────────────────────────────
-echo
-echo "=== 验证 ==="
-echo "Repo units:    ${#REPO_UNITS[@]}"
-echo "/etc units:    $(find "$ETC_UNITS_DIR" -maxdepth 1 -type f -name 'market-sync*.service' -o -name 'market-sync*.timer' | wc -l)"
-
-drift_count=0
-for u in "${REPO_UNITS[@]}"; do
-  if [ -f "$ETC_UNITS_DIR/$u" ]; then
-    if ! diff -q "$REPO_UNITS_DIR/$u" "$ETC_UNITS_DIR/$u" > /dev/null 2>&1; then
-      echo "  ❌ $u 仍有 drift"
-      drift_count=$((drift_count + 1))
-    fi
-  else
-    echo "  ❌ $u 仍未部署"
-    drift_count=$((drift_count + 1))
-  fi
-done
-
-echo
-if [ $drift_count -eq 0 ]; then
-  echo "✅ 所有 unit 已对齐，0 drift"
-  echo
-  echo "下一步：跑 .venv/bin/python bin/daily_sync_check.py --report-only 验证"
-else
-  echo "❌ 还有 $drift_count 处 drift，请检查"
-  exit 1
-fi

bin/systemd_deploy.sh

@@ -0,0 +1 @@
+deploy/systemd/deploy.sh

bin/task_watch.py

@@ -0,0 +1,218 @@
+"""自适应任务跟踪监视器。
+
+用法:
+  .venv/bin/python bin/task_watch.py          # 单次检查
+  .venv/bin/python bin/task_watch.py --watch   # 持续监视（自动调整检查间隔）
+
+设计：
+  1. 查询 dataset_registry 中 status='running' 的任务
+  2. 从 config 表中解析历史运行耗时（解析 lastMessage 中的 elapsed_sec）
+  3. 估算剩余时间 = 历史平均耗时 - 已用时间
+  4. 自适应下次检查间隔 = clamp(剩余 * 0.15, 15s, 300s)
+     无历史数据时默认 60s
+  5. 持续监视模式下，每次检查后动态调整下一次的 sleep 时间
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import re
+import sys
+import time
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any
+
+_PROJECT_ROOT = Path(__file__).resolve().parent.parent
+if str(_PROJECT_ROOT) not in sys.path:
+    sys.path.insert(0, str(_PROJECT_ROOT))
+
+from sqlalchemy import create_engine, text
+
+from app.core.config import settings
+from app.core.utils.logging import setup_logging, get_logger
+
+setup_logging()
+logger = get_logger("task_watch")
+
+# ── 历史耗时解析 ─────────────────────────────────────────────────────
+
+_ELAPSED_RE = re.compile(r"elapsed[=_ ]sec[= ]?(\d+\.?\d*)", re.IGNORECASE)
+_ELAPSED_SUFFIX_RE = re.compile(r"(\d+\.?\d*)s")
+
+
+def _parse_elapsed_sec(message: str) -> float | None:
+    for pattern in (_ELAPSED_RE,):
+        m = pattern.search(message)
+        if m:
+            return float(m.group(1))
+    return None
+
+
+# ── 历史耗时数据库（从 config 表 lastMessage 解析）───────────────────
+
+
+def _load_history(conn: Any) -> dict[str, list[float]]:
+    history: dict[str, list[float]] = {}
+    rows = conn.execute(
+        text("""
+            SELECT key, value
+            FROM market_data.config
+            WHERE category = 'schedule'
+        """)
+    ).fetchall()
+    for key, value_json in rows:
+        try:
+            val = json.loads(value_json)
+        except (json.JSONDecodeError, TypeError):
+            continue
+        job = val.get("job") or key.removeprefix("schedule_")
+        msg = val.get("lastMessage") or ""
+        elapsed = _parse_elapsed_sec(msg)
+        if elapsed and elapsed > 0:
+            history.setdefault(job, []).append(elapsed)
+    return history
+
+
+# ── 活跃任务查询 ──────────────────────────────────────────────────────
+
+
+def _fetch_running_tasks(conn: Any) -> list[dict[str, Any]]:
+    rows = conn.execute(
+        text("""
+            SELECT dataset_id, status, started_at, message
+            FROM market_data.dataset_registry
+            WHERE status = 'running'
+            ORDER BY started_at NULLS LAST
+        """)
+    ).fetchall()
+    tasks = []
+    for row in rows:
+        tasks.append({
+            "dataset_id": row[0],
+            "status": row[1],
+            "started_at": row[2],
+            "message": row[3] or "",
+        })
+    return tasks
+
+
+# ── 估算 ──────────────────────────────────────────────────────────────
+
+
+def _estimate(
+    task: dict[str, Any],
+    history: dict[str, list[float]],
+) -> tuple[float | None, float | None, float | None]:
+    """返回 (历史平均耗时秒, 已用秒, 预估剩余秒)。"""
+    tid = task["dataset_id"]
+    starts = task["started_at"]
+    if starts is None:
+        return None, None, None
+
+    started_wall = starts.replace(tzinfo=None)
+    elapsed = (datetime.now() - started_wall).total_seconds()
+    elapsed = max(round(elapsed, 1), 0)
+
+    durations = history.get(tid, [])
+    if not durations:
+        return None, elapsed, None
+
+    avg_duration = sum(durations) / len(durations)
+    remaining = max(round(avg_duration - elapsed, 1), 0)
+    return round(avg_duration, 1), elapsed, remaining
+
+
+# ── 自适应检查间隔 ────────────────────────────────────────────────────
+
+
+def _next_interval(remaining: float | None) -> float:
+    if remaining is None:
+        return 60.0
+    interval = remaining * 0.15
+    return max(15.0, min(interval, 300.0))
+
+
+# ── 输出 ──────────────────────────────────────────────────────────────
+
+
+def _fmt_sec(sec: float | None) -> str:
+    if sec is None:
+        return "N/A"
+    if sec < 60:
+        return f"{sec:.0f}s"
+    if sec < 3600:
+        return f"{sec/60:.1f}min"
+    return f"{sec/3600:.1f}h"
+
+
+def _print_report(
+    tasks: list[dict[str, Any]],
+    history: dict[str, list[float]],
+    next_interval: float,
+) -> None:
+    now = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+    header = f"── Task Watch [{now}] ──"
+    print(header)
+    if not tasks:
+        print("  无运行中的任务")
+        print(f"  下次检查: {_fmt_sec(next_interval)}")
+        return
+
+    lines = []
+    for t in tasks:
+        avg, elapsed, remaining = _estimate(t, history)
+        durations = history.get(t["dataset_id"], [])
+        progress = ""
+        if avg and elapsed:
+            pct = min(elapsed / avg * 100, 99.9)
+            progress = f" {pct:.0f}%"
+        line = (
+            f"  {t['dataset_id']:20s}"
+            f" elapsed={_fmt_sec(elapsed):>8s}"
+            f" avg={_fmt_sec(avg):>8s}"
+            f" remain={_fmt_sec(remaining):>8s}"
+            f"{progress}"
+            f"  (历史 {len(durations)} 次)"
+        )
+        lines.append(line)
+
+    for l in lines:
+        print(l)
+    print(f"  下次检查: {_fmt_sec(next_interval)}")
+    print("─" * len(header))
+
+
+# ── 主循环 ────────────────────────────────────────────────────────────
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="自适应任务跟踪监视器")
+    parser.add_argument("--watch", action="store_true", help="持续监视模式（自动调整检查间隔）")
+    args = parser.parse_args()
+
+    engine = create_engine(settings.pg_sqlalchemy_url())
+
+    while True:
+        with engine.connect() as conn:
+            history = _load_history(conn)
+            tasks = _fetch_running_tasks(conn)
+
+        interval = 60.0
+        if tasks:
+            intervals = []
+            for t in tasks:
+                _, _, remaining = _estimate(t, history)
+                intervals.append(_next_interval(remaining))
+            interval = min(intervals) if intervals else 60.0
+
+        _print_report(tasks, history, interval)
+
+        if not args.watch:
+            break
+
+        time.sleep(interval)
+
+
+if __name__ == "__main__":
+    main()

deploy/docker/Dockerfile

@@ -0,0 +1,65 @@
+# syntax=docker/dockerfile:1.6
+# ─────────────────────────────────────────────────────────────
+# market_sync 数据同步服务
+# 单镜像同时跑：FastAPI dashboard + 进程内 scheduler（worker）
+# 部署：docker compose up -d
+# ─────────────────────────────────────────────────────────────
+
+# --- builder stage: 装依赖到 venv ---
+FROM python:3.11-slim AS builder
+
+ENV PYTHONUNBUFFERED=1 \
+    PYTHONDONTWRITEBYTECODE=1 \
+    PIP_NO_CACHE_DIR=1 \
+    PIP_DISABLE_PIP_VERSION_CHECK=1
+
+# 仅装编译期需要的系统包（很多 wheel 已预编译，但 psycopg2 / cryptography 可能要）
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        build-essential gcc libpq-dev \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+
+# 先 copy requirements 单独一层（利用 Docker 缓存：依赖不变就不重装）
+COPY requirements.txt .
+RUN python -m venv /app/.venv \
+    && /app/.venv/bin/pip install --upgrade pip \
+    && /app/.venv/bin/pip install -r requirements.txt
+
+
+# --- runtime stage: 极简基础镜像 + 仅复制 venv 与代码 ---
+FROM python:3.11-slim AS runtime
+
+ENV PYTHONUNBUFFERED=1 \
+    PYTHONDONTWRITEBYTECODE=1 \
+    PATH="/app/.venv/bin:${PATH}" \
+    PYTHONPATH=/app \
+    # service 模式默认监听
+    API_HOST=0.0.0.0 \
+    API_PORT=8100 \
+    # Docker 模式：单容器内使用进程内 scheduler
+    RUNTIME_MODE=docker
+
+# runtime 只需要 psycopg2 的运行时库 + tzdata（pandas tz aware 需要）
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        libpq5 tzdata curl \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+
+# 从 builder 复制 venv（已编译好的依赖）
+COPY --from=builder /app/.venv /app/.venv
+
+# 复制应用代码
+COPY app/ ./app/
+COPY bin/ ./bin/
+
+# 默认启动 service 入口（FastAPI + 进程内 scheduler）
+# 也可覆盖为 cli / worker 单跑某个 task：
+#   docker compose run --rm market_sync python -m app.entrypoints.cli sync kline_daily
+EXPOSE 8100
+
+HEALTHCHECK --interval=60s --timeout=5s --start-period=30s --retries=3 \
+    CMD curl -fsS http://localhost:8100/api/health || exit 1
+
+ENTRYPOINT ["/app/bin/service_run.sh"]

deploy/docker/docker-compose.yml

@@ -0,0 +1,86 @@
+# docker-compose.yml — 一键起 market_sync 数据同步服务
+#
+# 包含：
+#   - postgres        : PostgreSQL 16（market_data schema 的主库）
+#   - market_sync     : 数据同步 + dashboard + scheduler（单进程）
+#
+# 用法：
+#   docker compose up -d                 # 后台启动
+#   docker compose logs -f market_sync   # 看同步日志
+#   docker compose exec market_sync \
+#       python -m app.entrypoints.cli sync stock_node --type2 2,3
+#
+version: "3.9"
+
+services:
+  postgres:
+    image: postgres:16-alpine
+    restart: unless-stopped
+    environment:
+      POSTGRES_USER: market_sync
+      POSTGRES_PASSWORD: market_sync
+      POSTGRES_DB: market_data
+    ports:
+      - "5432:5432"
+    volumes:
+      - pgdata:/var/lib/postgresql/data
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -U market_sync -d market_data"]
+      interval: 10s
+      timeout: 3s
+      retries: 5
+
+  market_sync:
+    build:
+      context: .
+      dockerfile: Dockerfile
+    image: market_sync:latest
+    restart: unless-stopped
+    depends_on:
+      postgres:
+        condition: service_healthy
+    environment:
+      # 运行模式：Docker 模式使用进程内 scheduler
+      RUNTIME_MODE: docker
+      # PG 连接（指向同 compose 的 postgres 服务）
+      PG_HOST: postgres
+      PG_PORT: "5432"
+      PG_USER: market_sync
+      PG_PASSWORD: market_sync
+      PG_DB_NAME: market_data
+      DB_BACKEND: pg
+      # 调度
+      SCHEDULER_TIMEZONE: Asia/Shanghai
+      SCHEDULER_TICK_SECONDS: "5"
+      SCHEDULER_AUTO_SEED: "true"
+      # API
+      API_HOST: 0.0.0.0
+      API_PORT: "8100"
+      # 日志
+      LOG_DIR: /app/logs
+      LOG_LEVEL: INFO
+      # 数据源开关
+      DS_BAOSTOCK_ENABLED: "true"
+      DS_SINA_ENABLED: "true"
+      # 凭证（可放进 .env / secrets 文件，避免明文）
+      MAIRUI_LICENCE: ${MAIRUI_LICENCE:-}
+      XUEQIU_TOKEN: ${XUEQIU_TOKEN:-}
+      # 麦蕊限速（钻石 100 RPS 留 buffer；服务化后保守点）
+      MAIRUI_RPS_LIMIT: "10"
+      # 节假日
+      TRADING_HOLIDAYS: ${TRADING_HOLIDAYS:-}
+    ports:
+      - "8100:8100"
+    volumes:
+      # 持久化日志（任务失败时方便排查）
+      - sync_logs:/app/logs
+    healthcheck:
+      test: ["CMD", "curl", "-fsS", "http://localhost:8100/api/health"]
+      interval: 60s
+      timeout: 5s
+      retries: 3
+      start_period: 30s
+
+volumes:
+  pgdata:
+  sync_logs:

deploy/systemd/deploy.sh

@@ -0,0 +1,211 @@
+#!/bin/bash
+# market_sync systemd 一键部署脚本
+#
+# 用途：把 deploy/systemd/units/*.service 和 *.timer 同步到 /etc/systemd/system/，
+#       并自动启用 timer、处理 worker 服务状态。
+#
+# 运行：sudo bash deploy/systemd/deploy.sh
+#      sudo bash deploy/systemd/deploy.sh --yes    # 跳过确认
+#
+# 设计原则：
+#   - 必须 sudo（写 /etc/systemd/system/ + daemon-reload）
+#   - 幂等：重复运行无副作用
+#   - 显示 plan → 等待确认 → 执行
+#   - 失败立即中止，不留半套状态
+
+set -e
+
+PROJECT_ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+REPO_UNITS_DIR="$PROJECT_ROOT/deploy/systemd/units"
+ETC_UNITS_DIR="/etc/systemd/system"
+SKIP_CONFIRM=false
+
+# ── 0. 前置检查 ─────────────────────────────────────────────────────────────
+if [ "$(id -u)" -ne 0 ]; then
+  echo "❌ 必须 sudo 运行：sudo bash $0"
+  exit 1
+fi
+
+if [ "${1:-}" = "--yes" ]; then
+  SKIP_CONFIRM=true
+fi
+
+if [ ! -d "$REPO_UNITS_DIR" ]; then
+  echo "❌ repo unit 目录不存在: $REPO_UNITS_DIR"
+  exit 1
+fi
+
+# ── 1. 计算 plan ────────────────────────────────────────────────────────────
+declare -a TO_COPY=()        # repo→etc 拷贝
+declare -a TO_REMOVE=()      # etc 中孤儿（repo 已删）
+declare -a TO_RESTART_SVC=() # service 文件变了需要 daemon-reload + restart
+
+# repo 中所有 market-sync* unit
+mapfile -t REPO_UNITS < <(find "$REPO_UNITS_DIR" -maxdepth 1 -type f \( -name "*.service" -o -name "*.timer" \) -printf '%f\n' | sort)
+# /etc 中所有 market-sync* unit
+mapfile -t ETC_UNITS < <(find "$ETC_UNITS_DIR" -maxdepth 1 -type f \( -name "market-sync*.service" -o -name "market-sync*.timer" \) -printf '%f\n' | sort 2>/dev/null || true)
+
+# drift 1: repo 有 /etc 没装
+for u in "${REPO_UNITS[@]}"; do
+  if [ ! -f "$ETC_UNITS_DIR/$u" ]; then
+    TO_COPY+=("$u")
+  fi
+done
+
+# drift 2: 两边都有但内容不一致
+for u in "${REPO_UNITS[@]}"; do
+  if [ -f "$ETC_UNITS_DIR/$u" ]; then
+    if ! diff -q "$REPO_UNITS_DIR/$u" "$ETC_UNITS_DIR/$u" > /dev/null 2>&1; then
+      TO_COPY+=("$u")
+      if [[ "$u" == *.service ]]; then
+        TO_RESTART_SVC+=("${u%.service}")
+      fi
+    fi
+  fi
+done
+
+# drift 3: /etc 有 repo 没维护（孤儿）
+for u in "${ETC_UNITS[@]}"; do
+  found=0
+  for r in "${REPO_UNITS[@]}"; do
+    if [ "$r" = "$u" ]; then found=1; break; fi
+  done
+  if [ $found -eq 0 ]; then
+    TO_REMOVE+=("$u")
+  fi
+done
+
+# ── 2. 显示 plan ────────────────────────────────────────────────────────────
+echo "============================================================"
+echo "market_sync systemd deploy plan"
+echo "============================================================"
+echo "Repo units dir: $REPO_UNITS_DIR"
+echo
+echo "Repo units:    ${#REPO_UNITS[@]} 个"
+echo "/etc units:    ${#ETC_UNITS[@]} 个"
+echo
+
+if [ ${#TO_COPY[@]} -eq 0 ] && [ ${#TO_REMOVE[@]} -eq 0 ]; then
+  echo "✅ 无 drift 需要修复。"
+else
+  if [ ${#TO_COPY[@]} -gt 0 ]; then
+    echo "📋 将要拷贝 (${#TO_COPY[@]} 个):"
+    for u in "${TO_COPY[@]}"; do
+      echo "    cp deploy/systemd/units/$u  →  /etc/systemd/system/$u"
+    done
+    echo
+  fi
+
+  if [ ${#TO_RESTART_SVC[@]} -gt 0 ]; then
+    echo "🔄 将要处理 service (${#TO_RESTART_SVC[@]} 个):"
+    for s in "${TO_RESTART_SVC[@]}"; do
+      echo "    systemctl reset-failed $s.service"
+    done
+    echo
+  fi
+
+  if [ ${#TO_REMOVE[@]} -gt 0 ]; then
+    echo "🗑️  将要删除 etc 中的孤儿 (${#TO_REMOVE[@]} 个):"
+    for u in "${TO_REMOVE[@]}"; do
+      echo "    rm /etc/systemd/system/$u"
+    done
+    echo
+  fi
+fi
+
+# ── 3. 等待确认 ────────────────────────────────────────────────────────────
+if [ "$SKIP_CONFIRM" = false ]; then
+  read -p "确认执行？[y/N] " -n 1 -r
+  echo
+  if [[ ! $REPLY =~ ^[Yy]$ ]]; then
+    echo "已取消。"
+    exit 0
+  fi
+fi
+
+# ── 4. 执行 ─────────────────────────────────────────────────────────────────
+echo
+echo "=== 执行 ==="
+
+# 4a. cp 漂移的 unit
+for u in "${TO_COPY[@]}"; do
+  cp -v "$REPO_UNITS_DIR/$u" "$ETC_UNITS_DIR/$u"
+done
+
+# 4b. 删孤儿
+for u in "${TO_REMOVE[@]}"; do
+  rm -v "$ETC_UNITS_DIR/$u"
+done
+
+# 4c. reload systemd
+echo
+echo "=== systemctl daemon-reload ==="
+systemctl daemon-reload
+
+# 4d. 停用并 mask worker 服务（systemd 模式下不需要进程内 scheduler）
+if systemctl is-enabled market-sync-worker.service >/dev/null 2>&1 || \
+   systemctl is-active market-sync-worker.service >/dev/null 2>&1; then
+  echo "=== 关闭 market-sync-worker.service（systemd 模式下由 timer 触发任务）==="
+  systemctl stop market-sync-worker.service || true
+  systemctl disable market-sync-worker.service || true
+  systemctl mask market-sync-worker.service || true
+fi
+
+# 4e. reset-failed + restart 修改过的 service
+for s in "${TO_RESTART_SVC[@]}"; do
+  if systemctl is-enabled "${s}.service" 2>/dev/null | grep -q enabled; then
+    systemctl reset-failed "${s}.service" || true
+    echo "  reset-failed ${s}.service (由 timer 自然触发)"
+  fi
+done
+
+# 4f. 启用所有 timer（幂等）
+echo
+echo "=== 启用 timers ==="
+for u in "${REPO_UNITS[@]}"; do
+  if [[ "$u" == *.timer ]]; then
+    timer="${u%.timer}"
+    if ! systemctl is-enabled "${timer}.timer" >/dev/null 2>&1; then
+      systemctl enable "${timer}.timer"
+      echo "  enable ${timer}.timer"
+    else
+      echo "  ${timer}.timer 已启用"
+    fi
+  fi
+done
+
+# ── 5. 验证 ────────────────────────────────────────────────────────────────
+echo
+echo "=== 验证 ==="
+echo "Repo units:    ${#REPO_UNITS[@]}"
+echo "/etc units:    $(find "$ETC_UNITS_DIR" -maxdepth 1 -type f \( -name 'market-sync*.service' -o -name 'market-sync*.timer' \) | wc -l)"
+
+drift_count=0
+for u in "${REPO_UNITS[@]}"; do
+  if [ -f "$ETC_UNITS_DIR/$u" ]; then
+    if ! diff -q "$REPO_UNITS_DIR/$u" "$ETC_UNITS_DIR/$u" > /dev/null 2>&1; then
+      echo "  ❌ $u 仍有 drift"
+      drift_count=$((drift_count + 1))
+    fi
+  else
+    echo "  ❌ $u 仍未部署"
+    drift_count=$((drift_count + 1))
+  fi
+done
+
+# 验证 worker 已被 mask
+if systemctl is-enabled market-sync-worker.service 2>/dev/null | grep -q masked; then
+  echo "  ✅ market-sync-worker.service 已 mask（systemd 模式下不启用进程内 scheduler）"
+else
+  echo "  ⚠️  market-sync-worker.service 未被 mask，systemd 模式下可能双调度器重叠"
+fi
+
+echo
+if [ $drift_count -eq 0 ]; then
+  echo "✅ 所有 unit 已对齐，0 drift"
+  echo
+  echo "下一步：跑 .venv/bin/python bin/daily_sync_check.py --report-only 验证"
+else
+  echo "❌ 还有 $drift_count 处 drift，请检查"
+  exit 1
+fi

deploy/systemd/deploy_split.sh

@@ -0,0 +1,28 @@
+#!/bin/bash
+# ── 停止旧 market-sync ──
+sudo systemctl stop market-sync.timer
+sudo systemctl disable market-sync.timer
+sudo systemctl stop market-sync.service
+sudo systemctl disable market-sync.service
+
+# ── 安装新独立 timer+service (4组) ──
+cd /home/gao/Development/quant_home/market_sync/deploy/systemd/units
+
+for f in market-sync-kline-index market-sync-kline-daily market-sync-kline-5min market-sync-market-regime; do
+  sudo cp ${f}.service /etc/systemd/system/
+  sudo cp ${f}.timer /etc/systemd/system/
+done
+
+# ── 更新 morning service：去掉已废弃的 industry_sector ──
+# （只用 morning 跑 stock_basic，不再跑 industry_sector）
+# 修改 morning run 脚本只跑 stock_basic
+sudo systemctl daemon-reload
+
+# ── 启动新 timer ──
+for f in market-sync-kline-index market-sync-kline-daily market-sync-kline-5min market-sync-market-regime; do
+  sudo systemctl enable ${f}.timer
+  sudo systemctl start ${f}.timer
+done
+
+echo "=== 完成 ==="
+systemctl list-timers --all 2>/dev/null | grep market-sync

deploy/systemd/scripts/deploy_systemd.sh

@@ -0,0 +1,59 @@
+#!/bin/bash
+# 部署 market-sync systemd units 到 /etc/systemd/system/
+# 2026-07-09 新增: market-sync-worker.service(长驻进程, 调度 schedule_* 任务)
+#
+# 用法: sudo bash bin/deploy_systemd.sh
+#
+# 跑这个脚本会:
+#   1. 复制 bin/systemd/market-sync*.{service,timer} 到 /etc/systemd/system/
+#   2. systemctl daemon-reload
+#   3. enable + start market-sync-worker.service
+#   4. 不动已存在的 .timer (它们都已 enable, 重复 enable 无副作用)
+set -euo pipefail
+
+REPO_DIR="$(cd "$(dirname "$0")/.." && pwd)"
+SYSTEMD_SRC="$REPO_DIR/bin/systemd"
+SYSTEMD_DST="/etc/systemd/system"
+
+if [ "$(id -u)" -ne 0 ]; then
+    echo "❌ 必须用 sudo 跑: sudo bash $0" >&2
+    exit 1
+fi
+
+echo "=== 1. 复制 unit files ==="
+for f in "$SYSTEMD_SRC"/market-sync*.{service,timer}; do
+    [ -f "$f" ] || continue
+    name="$(basename "$f")"
+    cp -v "$f" "$SYSTEMD_DST/$name"
+    chmod 644 "$SYSTEMD_DST/$name"
+done
+
+echo ""
+echo "=== 2. daemon-reload ==="
+systemctl daemon-reload
+
+echo ""
+echo "=== 3. 启用 + 启动 worker 长驻进程(关键!) ==="
+if ! systemctl is-enabled --quiet market-sync-worker.service; then
+    systemctl enable market-sync-worker.service
+fi
+# 如果已通过 nohup 启动了, 先 kill 掉避免双跑
+if pgrep -f "app\.entrypoints\.worker" >/dev/null; then
+    echo "  ⚠️  检测到手动启动的 worker 进程, 先 kill"
+    pkill -TERM -f "app\.entrypoints\.worker" || true
+    sleep 2
+    pkill -KILL -f "app\.entrypoints\.worker" 2>/dev/null || true
+fi
+systemctl restart market-sync-worker.service
+sleep 2
+systemctl status market-sync-worker.service --no-pager | head -10
+
+echo ""
+echo "=== 4. 列出所有 market-sync timer (验证) ==="
+systemctl list-timers market-sync* --no-pager 2>&1 | head -20
+
+echo ""
+echo "✅ 部署完成"
+echo "  - 长驻 worker: systemctl status market-sync-worker.service"
+echo "  - 看任务执行: tail -f logs/worker_manual_*.log  (本进程) 或 journalctl -u market-sync-worker -f  (systemd 接管后)"
+echo "  - 巡检:        .venv/bin/python bin/daily_sync_check.py"

deploy/systemd/scripts/systemd_deploy.sh

@@ -0,0 +1,185 @@
+#!/bin/bash
+# market_sync systemd 一键部署脚本
+#
+# 用途：把 bin/systemd/*.service 和 *.timer 同步到 /etc/systemd/system/，
+#       解决 work #04 检测到的 5 处 drift。
+#
+# 运行：sudo bash bin/systemd_deploy.sh
+#
+# 设计原则：
+#   - 必须 sudo（写 /etc/systemd/system/ + daemon-reload）
+#   - 幂等：重复运行无副作用
+#   - 显示 plan → 等待确认 → 执行
+#   - 失败立即中止，不留半套状态
+
+set -e
+
+PROJECT_ROOT="/home/gao/Development/quant_home/market_sync"
+REPO_UNITS_DIR="$PROJECT_ROOT/bin/systemd"
+ETC_UNITS_DIR="/etc/systemd/system"
+
+# ── 0. 前置检查 ─────────────────────────────────────────────────────────────
+if [ "$(id -u)" -ne 0 ]; then
+  echo "❌ 必须 sudo 运行：sudo bash $0"
+  exit 1
+fi
+
+if [ ! -d "$REPO_UNITS_DIR" ]; then
+  echo "❌ repo unit 目录不存在: $REPO_UNITS_DIR"
+  exit 1
+fi
+
+# ── 1. 计算 plan ────────────────────────────────────────────────────────────
+declare -a TO_COPY=()        # repo→etc 拷贝
+declare -a TO_REMOVE=()      # etc 中孤儿（repo 已删）
+declare -a TO_RESTART_SVC=() # service 文件变了需要 daemon-reload + restart
+
+# repo 中所有 market-sync* unit
+mapfile -t REPO_UNITS < <(find "$REPO_UNITS_DIR" -maxdepth 1 -type f \( -name "*.service" -o -name "*.timer" \) -printf '%f\n' | sort)
+# /etc 中所有 market-sync* unit
+mapfile -t ETC_UNITS < <(find "$ETC_UNITS_DIR" -maxdepth 1 -type f \( -name "market-sync*.service" -o -name "market-sync*.timer" \) -printf '%f\n' | sort 2>/dev/null || true)
+
+# drift 1: repo 有 /etc 没装
+for u in "${REPO_UNITS[@]}"; do
+  if [ ! -f "$ETC_UNITS_DIR/$u" ]; then
+    TO_COPY+=("$u")
+  fi
+done
+
+# drift 2: 两边都有但内容不一致
+for u in "${REPO_UNITS[@]}"; do
+  if [ -f "$ETC_UNITS_DIR/$u" ]; then
+    if ! diff -q "$REPO_UNITS_DIR/$u" "$ETC_UNITS_DIR/$u" > /dev/null 2>&1; then
+      TO_COPY+=("$u")
+      if [[ "$u" == *.service ]]; then
+        TO_RESTART_SVC+=("${u%.service}")
+      fi
+    fi
+  fi
+done
+
+# drift 3: /etc 有 repo 没维护（孤儿）
+for u in "${ETC_UNITS[@]}"; do
+  found=0
+  for r in "${REPO_UNITS[@]}"; do
+    if [ "$r" = "$u" ]; then found=1; break; fi
+  done
+  if [ $found -eq 0 ]; then
+    TO_REMOVE+=("$u")
+  fi
+done
+
+# ── 2. 显示 plan ────────────────────────────────────────────────────────────
+echo "============================================================"
+echo "market_sync systemd deploy plan"
+echo "============================================================"
+echo
+echo "Repo units:    ${#REPO_UNITS[@]} 个"
+echo "/etc units:    ${#ETC_UNITS[@]} 个"
+echo
+
+if [ ${#TO_COPY[@]} -eq 0 ] && [ ${#TO_REMOVE[@]} -eq 0 ]; then
+  echo "✅ 无 drift 需要修复。无需执行。"
+  exit 0
+fi
+
+if [ ${#TO_COPY[@]} -gt 0 ]; then
+  echo "📋 将要拷贝 ($(echo "${TO_COPY[@]}" | tr ' ' '\n' | wc -l) 个):"
+  for u in "${TO_COPY[@]}"; do
+    echo "    cp bin/systemd/$u  →  /etc/systemd/system/$u"
+  done
+  echo
+fi
+
+if [ ${#TO_RESTART_SVC[@]} -gt 0 ]; then
+  echo "🔄 将要重启 service ($(echo "${TO_RESTART_SVC[@]}" | tr ' ' '\n' | wc -l) 个):"
+  for s in "${TO_RESTART_SVC[@]}"; do
+    echo "    systemctl reset-failed $s.service"
+  done
+  echo
+fi
+
+if [ ${#TO_REMOVE[@]} -gt 0 ]; then
+  echo "🗑️  将要删除 etc 中的孤儿 ($(echo "${TO_REMOVE[@]}" | tr ' ' '\n' | wc -l) 个):"
+  for u in "${TO_REMOVE[@]}"; do
+    echo "    rm /etc/systemd/system/$u"
+  done
+  echo
+fi
+
+# ── 3. 等待确认 ────────────────────────────────────────────────────────────
+read -p "确认执行？[y/N] " -n 1 -r
+echo
+if [[ ! $REPLY =~ ^[Yy]$ ]]; then
+  echo "已取消。"
+  exit 0
+fi
+
+# ── 4. 执行 ─────────────────────────────────────────────────────────────────
+echo
+echo "=== 执行 ==="
+
+# 4a. cp 漂移的 unit
+for u in "${TO_COPY[@]}"; do
+  cp -v "$REPO_UNITS_DIR/$u" "$ETC_UNITS_DIR/$u"
+done
+
+# 4b. 删孤儿
+for u in "${TO_REMOVE[@]}"; do
+  rm -v "$ETC_UNITS_DIR/$u"
+done
+
+# 4c. reload systemd
+echo
+echo "=== systemctl daemon-reload ==="
+systemctl daemon-reload
+
+# 4d. reset-failed + restart 修改过的 service
+for s in "${TO_RESTART_SVC[@]}"; do
+  if systemctl is-enabled "${s}.service" 2>/dev/null | grep -q enabled; then
+    systemctl reset-failed "${s}.service" || true
+    echo "  reset-failed ${s}.service (没自动 restart — 由 timer 自然触发)"
+  fi
+done
+
+# 4e. 启用新 timer（如果有）
+for u in "${TO_COPY[@]}"; do
+  if [[ "$u" == *.timer ]]; then
+    timer="${u%.timer}"
+    if systemctl list-unit-files "${timer}.timer" 2>/dev/null | grep -q "${timer}.timer"; then
+      if ! systemctl is-enabled "${timer}.timer" 2>/dev/null | grep -q enabled; then
+        systemctl enable "${timer}.timer"
+        echo "  enable ${timer}.timer"
+      fi
+    fi
+  fi
+done
+
+# ── 5. 验证 ────────────────────────────────────────────────────────────────
+echo
+echo "=== 验证 ==="
+echo "Repo units:    ${#REPO_UNITS[@]}"
+echo "/etc units:    $(find "$ETC_UNITS_DIR" -maxdepth 1 -type f -name 'market-sync*.service' -o -name 'market-sync*.timer' | wc -l)"
+
+drift_count=0
+for u in "${REPO_UNITS[@]}"; do
+  if [ -f "$ETC_UNITS_DIR/$u" ]; then
+    if ! diff -q "$REPO_UNITS_DIR/$u" "$ETC_UNITS_DIR/$u" > /dev/null 2>&1; then
+      echo "  ❌ $u 仍有 drift"
+      drift_count=$((drift_count + 1))
+    fi
+  else
+    echo "  ❌ $u 仍未部署"
+    drift_count=$((drift_count + 1))
+  fi
+done
+
+echo
+if [ $drift_count -eq 0 ]; then
+  echo "✅ 所有 unit 已对齐，0 drift"
+  echo
+  echo "下一步：跑 .venv/bin/python bin/daily_sync_check.py --report-only 验证"
+else
+  echo "❌ 还有 $drift_count 处 drift，请检查"
+  exit 1
+fi

deploy/systemd/units/market-sync-daily-check.service

@@ -0,0 +1,31 @@
+[Unit]
+Description=Market sync daily check — reads dataset_registry, writes report, alerts on failures
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/daily_sync_check.py
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 巡检：读 PG dataset_registry，对比今日预期，写 logs/daily_check_*.json
+# 失败时 POST webhook 到 FAILURE_WEBHOOK_URL（HMAC-SHA256 签名）
+ExecStart=/home/gao/Development/quant_home/market_sync/.venv/bin/python /home/gao/Development/quant_home/market_sync/bin/daily_sync_check.py
+
+# 巡检应 < 30s 完成；超时 = PG 不通或代码 bug，不让卡死
+TimeoutStartSec=120
+
+# 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑）
+# 失败时 exit code = 1，可被 OnFailure= 钩住发通知
+
+# 日志走 journald（journalctl -u market-sync-daily-check -f）
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-daily-check
+
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-daily-check.timer

@@ -0,0 +1,15 @@
+[Unit]
+Description=Schedule market sync daily check — Daily 23:00 Asia/Shanghai
+# market-sync-daily-check.service 是这个 timer 的执行单元
+# 23:00 触发：周一~五最后一个任务是 longhubang @ 22:00，留 1h buffer；
+# 周六 23:00 给 stock_node @ 11:30 留 11h+ buffer
+
+[Timer]
+# 每天 23:00 触发；Persistent=true 确保关机/错过时下次开机补跑
+OnCalendar=*-*-* 23:00:00 Asia/Shanghai
+Persistent=true
+Unit=market-sync-daily-check.service
+# 不要 AccuracySec（默认 1min 漂移够用）
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-kline-5min.service

@@ -0,0 +1,20 @@
+[Unit]
+Description=Market sync kline_5min — 5分钟K (评分核心)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_kline_5min_run.sh
+After=network-online.target market-sync-kline-daily.service
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_kline_5min_run.sh
+TimeoutStartSec=7200
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-kline-5min
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-kline-5min.timer

@@ -0,0 +1,10 @@
+[Unit]
+Description=Schedule kline_5min sync — Mon..Fri 16:00 Asia/Shanghai
+
+[Timer]
+OnCalendar=Mon..Fri 16:00:00 Asia/Shanghai
+Persistent=true
+Unit=market-sync-kline-5min.service
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-kline-daily-early.timer

@@ -0,0 +1,11 @@
+[Unit]
+Description=Market sync kline_daily — 15:10 麦蕊优先抢跑
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_kline_daily_run.sh
+
+[Timer]
+OnCalendar=Mon..Fri 15:10:00
+Persistent=false
+RandomizedDelaySec=0
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-kline-daily.service

@@ -0,0 +1,20 @@
+[Unit]
+Description=Market sync kline_daily — 个股日K (评分核心)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_kline_daily_run.sh
+After=network-online.target market-sync-kline-index.service
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_kline_daily_run.sh
+TimeoutStartSec=3600
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-kline-daily
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-kline-daily.timer

@@ -0,0 +1,10 @@
+[Unit]
+Description=Schedule kline_daily sync — Mon..Fri 15:40 Asia/Shanghai
+
+[Timer]
+OnCalendar=Mon..Fri 15:40:00 Asia/Shanghai
+Persistent=true
+Unit=market-sync-kline-daily.service
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-kline-index.service

@@ -0,0 +1,20 @@
+[Unit]
+Description=Market sync kline_index — 指数日K (独立, 非评分依赖)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_kline_index_run.sh
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_kline_index_run.sh
+TimeoutStartSec=600
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-kline-index
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-kline-index.timer

@@ -0,0 +1,10 @@
+[Unit]
+Description=Schedule kline_index sync — Mon..Fri 15:30 Asia/Shanghai
+
+[Timer]
+OnCalendar=Mon..Fri 15:30:00 Asia/Shanghai
+Persistent=true
+Unit=market-sync-kline-index.service
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-lhb.service

@@ -0,0 +1,33 @@
+[Unit]
+Description=Market data 龙虎榜 sync (akshare 源, daily 22:00)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_lhb_run.sh
+After=network-online.target market-sync.service
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 入口脚本：单独跑 longhubang 一个 task
+# 龙虎榜数据 19:00~21:00 陆续出齐 → 22:00 触发（与 moneyflow 21:35 错开避免并发）
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_lhb_run.sh
+
+# 硬上限 1h（每日 ~200 次 akshare 调用 + DB upsert，3-5 分钟内完成，留 buffer）
+TimeoutStartSec=3600
+
+# 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑——
+# 重跑会撞外部 API 限流窗口，浪费资源）
+# Restart=no 是 oneshot 默认值
+
+# 日志走 journald（journalctl -u market-sync-lhb.service -f）
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-lhb
+
+# 环境（不读 /etc/environment，只带这几个；.env 由 cli 内部 load_dotenv）
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-lhb.timer

@@ -0,0 +1,16 @@
+[Unit]
+Description=Schedule 龙虎榜 sync — Mon..Fri 22:00 Asia/Shanghai
+# market-sync-lhb.service 是这个 timer 的执行单元
+# 龙虎榜数据 19:00~21:00 陆续出齐 → 22:00 触发（与 moneyflow 21:35 错开）
+
+[Timer]
+# 龙虎榜出齐 → 22:00 触发
+OnCalendar=Mon..Fri 22:00:00 Asia/Shanghai
+# 系统关机 / 错过执行时，下次开机补跑一次（避免漏数据）
+Persistent=true
+# 单位（service）的精确名称
+Unit=market-sync-lhb.service
+# 不要 AccuracySec（默认 1min 漂移够用，避免 22:00:00 整点打堆）
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-mairui-indicators.service

@@ -0,0 +1,32 @@
+[Unit]
+Description=Market data MACD/KDJ/BOLL indicators sync (mairui, daily 16:40)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_mairui_indicators_run.sh
+After=network-online.target market-sync.service
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 入口脚本：单独跑 mairui_indicators 一个 task
+# mairui 16:30 后发布指标数据 → 16:40 触发留 buffer
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_mairui_indicators_run.sh
+
+# 硬上限 3h（5204 只 × 3 指标，正常 ~20min，全量时可能更久）
+TimeoutStartSec=10800
+
+# 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑）
+# Restart=no 是 oneshot 默认值
+
+# 日志走 journald
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-mairui-indicators
+
+# 环境
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-mairui-indicators.timer

@@ -0,0 +1,14 @@
+[Unit]
+Description=Schedule MACD/KDJ/BOLL indicators sync — Mon..Fri 16:40 Asia/Shanghai
+# market-sync-mairui-indicators.service 是这个 timer 的执行单元
+
+[Timer]
+# mairui 16:30 后发布指标数据 → 16:40 触发留 buffer
+OnCalendar=Mon..Fri 16:40:00 Asia/Shanghai
+# 系统关机 / 错过执行时，下次开机补跑一次（避免漏数据）
+Persistent=true
+# 单位（service）的精确名称
+Unit=market-sync-mairui-indicators.service
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-mairui-ma-daily.service

@@ -0,0 +1,32 @@
+[Unit]
+Description=Market data MA daily sync (local derivative, daily 16:30)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_mairui_ma_daily_run.sh
+After=network-online.target market-sync.service
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 入口脚本：单独跑 mairui_ma_daily 一个 task
+# 本地派生任务，依赖 kline_stock 已更新
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_mairui_ma_daily_run.sh
+
+# 硬上限 2h（正常 ~30min，留 buffer）
+TimeoutStartSec=7200
+
+# 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑）
+# Restart=no 是 oneshot 默认值
+
+# 日志走 journald
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-mairui-ma-daily
+
+# 环境
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-mairui-ma-daily.timer

@@ -0,0 +1,14 @@
+[Unit]
+Description=Schedule MA daily sync — Mon..Fri 16:30 Asia/Shanghai
+# market-sync-mairui-ma-daily.service 是这个 timer 的执行单元
+
+[Timer]
+# kline_daily 15:40 / kline_index 15:30 跑完后，16:30 计算 MA
+OnCalendar=Mon..Fri 16:30:00 Asia/Shanghai
+# 系统关机 / 错过执行时，下次开机补跑一次（避免漏数据）
+Persistent=true
+# 单位（service）的精确名称
+Unit=market-sync-mairui-ma-daily.service
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-market-regime.service

@@ -0,0 +1,20 @@
+[Unit]
+Description=Market sync market_regime — 市场情绪 (本地计算, 非评分依赖)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_market_regime_run.sh
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_market_regime_run.sh
+TimeoutStartSec=600
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-market-regime
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-market-regime.timer

@@ -0,0 +1,10 @@
+[Unit]
+Description=Schedule market_regime sync — Mon..Fri 16:30 Asia/Shanghai
+
+[Timer]
+OnCalendar=Mon..Fri 16:30:00 Asia/Shanghai
+Persistent=true
+Unit=market-sync-market-regime.service
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-mcp.service

@@ -0,0 +1,22 @@
+[Unit]
+Description=market_sync MCP Server — Model Context Protocol for sync data
+Documentation=file:///home/gao/Development/quant_home/market_sync/app/mcp_server.py
+After=network-online.target postgres-container.service
+Wants=network-online.target
+
+[Service]
+Type=simple
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+ExecStart=/home/gao/Development/quant_home/market_sync/.venv/bin/python -m app.mcp_server --sse --host 127.0.0.1 --port 8101
+Restart=on-failure
+RestartSec=5
+TimeoutStartSec=30
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-mcp
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-moneyflow.service

@@ -0,0 +1,33 @@
+[Unit]
+Description=Market data moneyflow sync (mairui transaction, daily 21:35)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_moneyflow_run.sh
+After=network-online.target market-sync.service
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 入口脚本：单独跑 moneyflow 一个 task
+# mairui 文档：「更新：每日 21:30」—— 21:35 触发留 5min buffer
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_moneyflow_run.sh
+
+# 硬上限 2h（5206 只 × 50 RPS = 104s 理论上限，留 buffer 给重试 / 慢响应）
+TimeoutStartSec=7200
+
+# 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑——
+# 重跑会撞外部 API 限流窗口，浪费 1.5h）
+# Restart=no 是 oneshot 默认值
+
+# 日志走 journald（journalctl -u market-sync-moneyflow.service -f）
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-moneyflow
+
+# 环境（不读 /etc/environment，只带这几个；.env 由 cli 内部 load_dotenv）
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-moneyflow.timer

@@ -0,0 +1,16 @@
+[Unit]
+Description=Schedule moneyflow sync — Mon..Fri 21:35 Asia/Shanghai
+# market-sync-moneyflow.service 是这个 timer 的执行单元
+# mairui 文档：「更新：每日 21:30」—— 21:35 触发留 5 分钟 buffer
+
+[Timer]
+# mairui 21:30 发布数据 → 21:35 触发
+OnCalendar=Mon..Fri 21:35:00 Asia/Shanghai
+# 系统关机 / 错过执行时，下次开机补跑一次（避免漏数据）
+Persistent=true
+# 单位（service）的精确名称
+Unit=market-sync-moneyflow.service
+# 不要 AccuracySec（默认 1min 漂移够用，避免 21:35:00 整点打堆）
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-morning.service

@@ -0,0 +1,29 @@
+[Unit]
+Description=Market sync morning pre-market — stock_basic
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_morning_run.sh
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 早盘前只跑 stock_basic（确保最新股票列表+股本快照）
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_morning_run.sh
+
+# stock_basic ~9min + buffer = 1h 够用
+TimeoutStartSec=3600
+
+# 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑）
+
+# 日志走 journald（journalctl -u market-sync-morning -f）
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-morning
+
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-morning.timer

@@ -0,0 +1,14 @@
+[Unit]
+Description=Schedule market sync morning — Mon..Fri 09:00 Asia/Shanghai
+# market-sync-morning.service 是这个 timer 的执行单元
+# 早盘前 09:00 触发：跑 stock_basic，给 09:30 开盘留 30min buffer
+
+[Timer]
+# A 股开盘 09:30，09:00 跑 stock_basic (~9min) 即可
+OnCalendar=Mon..Fri 09:00:00 Asia/Shanghai
+# 关机/错过时下次开机补跑
+Persistent=true
+Unit=market-sync-morning.service
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-share.service

@@ -0,0 +1,31 @@
+[Unit]
+Description=Market data 股本快照 sync (雪球源, weekly Sat 11:30)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_share_run.sh
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 入口脚本：单独跑 share_snapshot 一个 task
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_share_run.sh
+
+# 硬上限 30min（全市场 5200 只雪球 quote_detail，10 worker 约 3-8min，留 buffer）
+TimeoutStartSec=1800
+
+# 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑）
+# Restart=no 是 oneshot 默认值
+
+# 日志走 journald（journalctl -u market-sync-share.service -f）
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-share
+
+# 环境（不读 /etc/environment，只带这几个；.env 由 cli 内部 load_dotenv）
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-share.timer

@@ -0,0 +1,16 @@
+[Unit]
+Description=Schedule share_snapshot sync — Sat 11:30 Asia/Shanghai
+# market-sync-share.service 是这个 timer 的执行单元
+# 股本快照每周跑一次即可（雪球 quote_detail，周末收盘后数据稳定）
+
+[Timer]
+# 周六 11:30（和 stock_node 同时段）
+OnCalendar=Sat *-*-* 11:30:00 Asia/Shanghai
+# 系统关机 / 错过执行时,下次开机补跑一次(避免漏数据)
+Persistent=true
+# 单位(service)的精确名称
+Unit=market-sync-share.service
+# 不要 AccuracySec(默认 1min 漂移够用)
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-stock-node.service

@@ -0,0 +1,33 @@
+[Unit]
+Description=Market data 股票-节点映射 sync (mairui /hszg, weekly Sat 11:30)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_stock_node_run.sh
+After=network-online.target market-sync-morning.service
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 入口脚本：单独跑 stock_node 一个 task
+# mairui /hszg 数据每周六 11:00 更新 → 11:30 触发,留 30 min buffer
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_stock_node_run.sh
+
+# 硬上限 2h(1100+ 调用 + 写库 + retry buffer)
+TimeoutStartSec=7200
+
+# 不要 Restart=(oneshot 失败就让 OnFailure= 发通知,别自动重跑——
+# 重跑会撞 mairui 配额窗口)
+# Restart=no 是 oneshot 默认值
+
+# 日志走 journald(journalctl -u market-sync-stock-node.service -f)
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-stock-node
+
+# 环境(不读 /etc/environment,只带这几个;.env 由 cli 内部 load_dotenv)
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-stock-node.timer

@@ -0,0 +1,16 @@
+[Unit]
+Description=Schedule stock_node sync — Sat 11:30 Asia/Shanghai
+# market-sync-stock-node.service 是这个 timer 的执行单元
+# mairui /hszg 数据每周六 11:00 更新 → 11:30 触发(30 min buffer)
+
+[Timer]
+# 周六 11:30(mairui 11:00 更新后)
+OnCalendar=Sat *-*-* 11:30:00 Asia/Shanghai
+# 系统关机 / 错过执行时,下次开机补跑一次(避免漏数据)
+Persistent=true
+# 单位(service)的精确名称
+Unit=market-sync-stock-node.service
+# 不要 AccuracySec(默认 1min 漂移够用,避免 11:30:00 整点打堆)
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-tick.service

@@ -0,0 +1,33 @@
+[Unit]
+Description=Market data tick-by-tick trade sync (mairui hsrl/zbjy, daily 21:05)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_tick_run.sh
+After=network-online.target market-sync.service
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 入口脚本：单独跑 tick_trade 一个 task（mairui 21:00 发布数据，
+# 这个 service 在 21:05 触发，留 5min buffer；force=True 绕过 task 内部 21:00 门控）
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_tick_run.sh
+
+# 硬上限 2h（5000 只 × 1 RPS ≈ 17min，留 buffer 给 retry / 慢任务）
+TimeoutStartSec=7200
+
+# 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑——
+# 重跑会撞外部 API 限流窗口，浪费 1.5h）
+# Restart=no 是 oneshot 默认值
+
+# 日志走 journald（journalctl -u market-sync-tick.service -f）
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync-tick
+
+# 环境（不读 /etc/environment，只带这几个；.env 由 cli 内部 load_dotenv）
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync-tick.timer

@@ -0,0 +1,17 @@
+[Unit]
+Description=Schedule tick_trade sync — Mon..Fri 21:05 Asia/Shanghai
+# market-sync-tick.service 是这个 timer 的执行单元
+# 触发顺序：market-sync.service (15:30) 之后 market-sync-tick.service (21:05)
+# 但 systemd timer 独立触发，After= 只是声明依赖关系（非强制）
+
+[Timer]
+# mairui 文档：「更新：每日 21:00」—— 21:05 触发留 5 分钟 buffer
+OnCalendar=Mon..Fri 21:05:00 Asia/Shanghai
+# 系统关机 / 错过执行时，下次开机补跑一次（避免漏数据）
+Persistent=true
+# 单位（service）的精确名称
+Unit=market-sync-tick.service
+# 不要 AccuracySec（默认 1min 漂移够用，避免 21:05:00 整点打堆）
+
+[Install]
+WantedBy=timers.target

deploy/systemd/units/market-sync-worker.service

@@ -0,0 +1,25 @@
+[Unit]
+Description=Market Data Sync Worker (PostgreSQL, scheduler + health monitor)
+Documentation=file:///home/gao/Development/quant_home/market_sync/app/entrypoints/worker.py
+# 依赖 network 和 PG
+After=network-online.target postgresql.service
+Wants=network-online.target
+
+[Service]
+Type=simple
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 入口：启动进程内 scheduler + health monitor + recover 卡死任务
+# 注意：实际入口是 app/entrypoints/worker.py，不是 app/worker.py
+# 2026-07-12 修复: 把 python -m app.worker 改为 python -m app.entrypoints.worker
+ExecStart=/home/gao/Development/quant_home/market_sync/.venv/bin/python -m app.entrypoints.worker
+
+# 明确标记运行模式，让 worker.py 知道不要启动 scheduler
+# （systemd 模式下同步由 timer 触发，避免双调度器重叠）
+Environment=PYTHONUNBUFFERED=1
+Environment="RUNTIME_MODE=systemd"
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync.service

@@ -0,0 +1,35 @@
+[Unit]
+Description=Market Data Full Sync (PostgreSQL, 9 sync tasks end-to-end)
+Documentation=file:///home/gao/Development/quant_home/market_sync/bin/market_sync_run.sh
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+WorkingDirectory=/home/gao/Development/quant_home/market_sync
+User=gao
+Group=gao
+
+# 入口脚本（runall_once.py + structured_watch.sh 二合一）
+ExecStart=/home/gao/Development/quant_home/market_sync/bin/market_sync_run.sh
+
+# 硬上限 3h（实际 ~80min + kline_5min 增量 70min, 总 ~2.5h, 留 30min buffer）
+# 2026-07-08 教训: 之前 4h 让 kline_daily 单点卡死拖 4h,runall 链全部受影响。
+# runall_once.py 已加 per-task multiprocessing timeout(单 task 最长 75min),
+# 这里 3h 是 systemd 层兜底。
+TimeoutStartSec=10800
+
+# 不要 Restart=（oneshot 失败就让 OnFailure= 发通知，别自动重跑——
+# 重跑会撞外部 API 限流窗口，浪费 1.5h）
+# Restart=no 是 oneshot 默认值
+
+# 日志走 journald（journalctl -u market-sync.service -f）
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=market-sync
+
+# 环境（不读 /etc/environment，只带这几个；.env 由 runall_once.py 内部 load_dotenv）
+Environment=PYTHONUNBUFFERED=1
+
+[Install]
+WantedBy=multi-user.target

deploy/systemd/units/market-sync.timer

@@ -0,0 +1,15 @@
+[Unit]
+Description=Schedule market data full sync — Mon..Fri 15:30 Asia/Shanghai
+# market-sync.service 是这个 timer 的执行单元
+
+[Timer]
+# A 股收盘时间 15:00，30 分钟 buffer 让最后一笔 tick 落地
+OnCalendar=Mon..Fri 15:30:00 Asia/Shanghai
+# 系统关机 / 错过执行时，下次开机补跑一次（避免漏数据）
+Persistent=true
+# 单位（service）的精确名称
+Unit=market-sync.service
+# 不要 AccuracySec（默认 1min 漂移够用，避免 15:30:00 整点打堆）
+
+[Install]
+WantedBy=timers.target

docker-compose.yml

@@ -1,84 +0,0 @@
-# docker-compose.yml — 一键起 market_sync 数据同步服务
-#
-# 包含：
-#   - postgres        : PostgreSQL 16（market_data schema 的主库）
-#   - market_sync     : 数据同步 + dashboard + scheduler（单进程）
-#
-# 用法：
-#   docker compose up -d                 # 后台启动
-#   docker compose logs -f market_sync   # 看同步日志
-#   docker compose exec market_sync \
-#       python -m app.entrypoints.cli sync stock_node --type2 2,3
-#
-version: "3.9"
-
-services:
-  postgres:
-    image: postgres:16-alpine
-    restart: unless-stopped
-    environment:
-      POSTGRES_USER: market_sync
-      POSTGRES_PASSWORD: market_sync
-      POSTGRES_DB: market_data
-    ports:
-      - "5432:5432"
-    volumes:
-      - pgdata:/var/lib/postgresql/data
-    healthcheck:
-      test: ["CMD-SHELL", "pg_isready -U market_sync -d market_data"]
-      interval: 10s
-      timeout: 3s
-      retries: 5
-
-  market_sync:
-    build:
-      context: .
-      dockerfile: Dockerfile
-    image: market_sync:latest
-    restart: unless-stopped
-    depends_on:
-      postgres:
-        condition: service_healthy
-    environment:
-      # PG 连接（指向同 compose 的 postgres 服务）
-      PG_HOST: postgres
-      PG_PORT: "5432"
-      PG_USER: market_sync
-      PG_PASSWORD: market_sync
-      PG_DB_NAME: market_data
-      DB_BACKEND: pg
-      # 调度
-      SCHEDULER_TIMEZONE: Asia/Shanghai
-      SCHEDULER_TICK_SECONDS: "5"
-      SCHEDULER_AUTO_SEED: "true"
-      # API
-      API_HOST: 0.0.0.0
-      API_PORT: "8100"
-      # 日志
-      LOG_DIR: /app/logs
-      LOG_LEVEL: INFO
-      # 数据源开关
-      DS_BAOSTOCK_ENABLED: "true"
-      DS_SINA_ENABLED: "true"
-      # 凭证（可放进 .env / secrets 文件，避免明文）
-      MAIRUI_LICENCE: ${MAIRUI_LICENCE:-}
-      XUEQIU_TOKEN: ${XUEQIU_TOKEN:-}
-      # 麦蕊限速（钻石 100 RPS 留 buffer；服务化后保守点）
-      MAIRUI_RPS_LIMIT: "10"
-      # 节假日
-      TRADING_HOLIDAYS: ${TRADING_HOLIDAYS:-}
-    ports:
-      - "8100:8100"
-    volumes:
-      # 持久化日志（任务失败时方便排查）
-      - sync_logs:/app/logs
-    healthcheck:
-      test: ["CMD", "curl", "-fsS", "http://localhost:8100/api/health"]
-      interval: 60s
-      timeout: 5s
-      retries: 3
-      start_period: 30s
-
-volumes:
-  pgdata:
-  sync_logs:

docker-compose.yml

@@ -0,0 +1 @@
+deploy/docker/docker-compose.yml

opencode.json

@@ -0,0 +1,10 @@
+{
+  "$schema": "https://opencode.ai/config.json",
+  "mcp": {
+    "market_sync": {
+      "type": "remote",
+      "url": "http://127.0.0.1:8101/sse",
+      "enabled": true
+    }
+  }
+}

scripts/run_all_sync.sh

@@ -1,81 +1,52 @@
 #!/bin/bash
-# 串行跑所有 6 个同步任务，后台模式，写到 logs/
+set -u
-# 用法：./run_all_sync.sh [STEP]
-#   STEP 为空 = 跑 1-6
-#   STEP=1 只跑 stock_basic，等等
 
-set -e
+PROJECT_ROOT="/home/gao/Development/quant_home/market_sync"
-SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+LOG_DIR="$PROJECT_ROOT/logs"
-PROJECT_DIR="$(cd "$SCRIPT_DIR/.." && pwd)"
+RUN_LOG="$LOG_DIR/run_all_sync_$(date +%Y%m%d_%H%M%S).log"
-cd "$PROJECT_DIR"
 
-mkdir -p logs
+cd "$PROJECT_ROOT"
-LOG_DIR="$PROJECT_DIR/logs"
-TIMESTAMP=$(date +%Y%m%d_%H%M%S)
-RUN_LOG="$LOG_DIR/run_all_$TIMESTAMP.log"
 
-# 读 token
+LOG_PREFIX="[run_all_sync]"
-if [ -f "$PROJECT_DIR/.env" ]; then
-    set -a
-    source "$PROJECT_DIR/.env"
-    set +a
-fi
-
-# 把 .env 里的 TRADING_HOLIDAYS 转成 -- 跳过
-# XUEQIU_TOKEN 在 .env 里
-
-source .venv/bin/activate
-export PYTHONPATH="$PROJECT_DIR"
-
-LOG_PREFIX="[$TIMESTAMP]"
 
 run_step() {
-    local step_name="$1"
+    local step=$1 task=$2 extra_args=${3:-}
-    local task_id="$2"
+    echo "$LOG_PREFIX === 步骤 $step: $task ===" | tee -a "$RUN_LOG"
-    local extra_args="$3"
+    if [ -n "$extra_args" ]; then
-    local step_log="$LOG_DIR/step_${step_name}_$TIMESTAMP.log"
+        .venv/bin/python -m app.entrypoints.cli sync "$task" $extra_args 2>&1 | tee -a "$RUN_LOG"
-
-    echo "$LOG_PREFIX === Step $step_name: $task_id ===" | tee -a "$RUN_LOG"
-    echo "$LOG_PREFIX   log: $step_log" | tee -a "$RUN_LOG"
-
-    # 跑任务（前台模式，方便顺序执行 + 立即看到结果）
-    python -m app.entrypoints.cli sync "$task_id" $extra_args 2>&1 | tee "$step_log" | tail -10
-    local exit_code=${PIPESTATUS[0]}
-
-    if [ $exit_code -eq 0 ]; then
-        echo "$LOG_PREFIX ✓ Step $step_name OK" | tee -a "$RUN_LOG"
     else
-        echo "$LOG_PREFIX ✗ Step $step_name FAIL (exit=$exit_code)" | tee -a "$RUN_LOG"
+        .venv/bin/python -m app.entrypoints.cli sync "$task" 2>&1 | tee -a "$RUN_LOG"
-        return $exit_code
     fi
+    local rc=${PIPESTATUS[0]}
+    if [ $rc -ne 0 ]; then
+        echo "$LOG_PREFIX ✗ 步骤 $step ($task) 失败 (exit=$rc)" | tee -a "$RUN_LOG"
+        return $rc
+    fi
+    echo "$LOG_PREFIX ✓ 步骤 $step ($task) 完成" | tee -a "$RUN_LOG"
 }
 
 case "${1:-all}" in
     1|stock_basic)        run_step "1_stock_basic" "stock_basic" ;;
-    2|industry_sector)    run_step "2_industry_sector" "industry_sector" ;;
+    2|kline_index)        run_step "2_kline_index" "kline_index" ;;
-    3|kline_index)        run_step "3_kline_index" "kline_index" ;;
+    3|kline_daily)        run_step "3_kline_daily" "kline_daily" "--workers 20" ;;
-    4|kline_daily)        run_step "4_kline_daily" "kline_daily" "--workers 20" ;;
+    4|share_snapshot)     run_step "4_share_snapshot" "share_snapshot" "--workers 10" ;;
-    5|share_snapshot)     run_step "5_share_snapshot" "share_snapshot" "--workers 10" ;;
+    5|market_regime)      run_step "5_market_regime" "market_regime" ;;
-    6|market_regime)      run_step "6_market_regime" "market_regime" ;;
+    6|kline_5min)         run_step "6_kline_5min" "kline_5min" "--workers 10" ;;
-    7|kline_5min)         run_step "7_kline_5min" "kline_5min" "--workers 10" ;;
+    7|moneyflow)          run_step "7_moneyflow" "moneyflow" "--workers 10" ;;
-    8|moneyflow)          run_step "8_moneyflow" "moneyflow" "--workers 10" ;;
+    8|tick_trade)         run_step "8_tick_trade" "tick_trade" "--workers 5 --force" ;;
-    9|sector_features)    run_step "9_sector_features" "sector_features" ;;
-    10|tick_trade)        run_step "10_tick_trade" "tick_trade" "--workers 5 --force" ;;
     all)
         run_step "1_stock_basic" "stock_basic" || exit 1
-        run_step "2_industry_sector" "industry_sector" || exit 1
+        run_step "2_kline_index" "kline_index" || exit 1
-        run_step "3_kline_index" "kline_index" || exit 1
+        run_step "3_kline_daily" "kline_daily" "--workers 20" || exit 1
-        run_step "4_kline_daily" "kline_daily" "--workers 20" || exit 1
+        run_step "4_share_snapshot" "share_snapshot" "--workers 10" || exit 1
-        run_step "5_share_snapshot" "share_snapshot" "--workers 10" || exit 1
+        run_step "5_market_regime" "market_regime" || exit 1
-        run_step "6_market_regime" "market_regime" || exit 1
+        run_step "6_kline_5min" "kline_5min" "--workers 10" || exit 1
-        run_step "7_kline_5min" "kline_5min" "--workers 10" || exit 1
+        run_step "7_moneyflow" "moneyflow" "--workers 10" || exit 1
-        run_step "8_moneyflow" "moneyflow" "--workers 10" || exit 1
+        run_step "8_tick_trade" "tick_trade" "--workers 5 --force" || exit 1
-        run_step "9_sector_features" "sector_features" || exit 1
+        echo "$LOG_PREFIX ✓ 全部 8 步完成" | tee -a "$RUN_LOG"
-        run_step "10_tick_trade" "tick_trade" "--workers 5 --force" || exit 1
-        echo "$LOG_PREFIX ✓ 全部 10 步完成" | tee -a "$RUN_LOG"
         ;;
     *)
-        echo "usage: $0 [1|2|3|4|5|6|7|8|9|10|all]"
+        echo "usage: $0 [1|2|3|4|5|6|7|8|all]"
         exit 1
         ;;
 esac

tests/test_schema_models.py

@@ -7,24 +7,25 @@ from __future__ import annotations
 
 
 def test_orm_metadata_registers_all_tables():
-    """ORMBase.metadata 应该注册 23 张 PG 表（项目主业务表）。"""
+    """ORMBase.metadata 应该注册 25 张 PG 表（项目主业务表）。"""
     from app.core.db import models  # 触发全部模型 import
     from app.core.db.orm import ORMBase
 
     # 注意：SA 2.x 的 MetaData.tables 既能按 bare name 也能按 schema-qualified name 索引
     # （keyed 容器），所以用 values() 拿 Table 对象再用 .name 拿 bare 名
     bare_names = {t.name for t in ORMBase.metadata.tables.values()}
-    assert len(bare_names) == 23, f"期望 23 张 ORM 表，实际 {len(bare_names)}: {bare_names}"
+    assert len(bare_names) == 21, f"期望 21 张 ORM 表，实际 {len(bare_names)}: {bare_names}"
     expected = {
         "config", "dataset_registry", "stocks", "indices",
         "kline_stock", "kline_index", "kline_5min", "moneyflow", "share",
-        "sectors", "stock_sector_map", "industry",
+        "market_regime_daily",
-        "sector_indices", "sector_features_daily", "market_regime_daily",
         "tick_trade",
         "longhubang_daily", "longhubang_seat",  # 2026-07-01 龙虎榜
         "kline_stock_ma_daily",  # 2026-07-04 mairui_ma_daily
         "node_categories", "nodes", "stock_node_map",  # 2026-07-02 股票-节点映射
         "sync_history",  # 2026-07-07 同步历史表
+        # 2026-07-08 mairui 技术指标 MACD/KDJ/BOLL
+        "kline_stock_macd_daily", "kline_stock_kdj_daily", "kline_stock_boll_daily",
     }
     assert bare_names == expected, f"ORM 表名集合与期望不符: {bare_names ^ expected}"
 

tests/test_smoke.py

@@ -35,10 +35,10 @@ def test_tasks_registry():
     from app.tasks import TASKS, get_task
     assert isinstance(TASKS, dict)
     # 注：每次新增 task 都要更新这里的数字
-    # 当前 13 个：stock_basic, kline_daily, kline_index, kline_5min, tick_trade,
+    # 当前 12 个：stock_basic, kline_daily, kline_index, kline_5min, tick_trade,
-    #          moneyflow, industry_sector, sector_features, share_snapshot, market_regime,
+    #          moneyflow, share_snapshot, market_regime,
-    #          longhubang, mairui_ma_daily, stock_node
+    #          longhubang, mairui_ma_daily, stock_node, mairui_indicators
-    assert len(TASKS) == 13
+    assert len(TASKS) == 12
     # get_task 应该返回实例
     task = get_task("kline_daily")
     assert task.dataset_id == "kline_daily"
@@ -46,10 +46,10 @@ def test_tasks_registry():
 
 def test_sync_definitions():
     from app.core.sync.registry import SYNC_DEFINITIONS
-    assert len(SYNC_DEFINITIONS) == 13
+    assert len(SYNC_DEFINITIONS) == 12
     # 所有 dataset_id 应唯一
     ids = [d["dataset_id"] for d in SYNC_DEFINITIONS]
-    assert len(set(ids)) == 13
+    assert len(set(ids)) == 12
     # tick_trade 应在其中且 sort_order=45
     tt = next(d for d in SYNC_DEFINITIONS if d["dataset_id"] == "tick_trade")
     assert tt["sort_order"] == 45