feat(data): 双源5yr全市场部署 + baostock 15min + 断点续传

- raw_redownload 加断点续传(exists/skip已存在, 扩范围重下覆盖)
- baostock_download: 15min双源(qfq+raw)下载器, NAS容器跑, 限速防封
- full_deploy_5yr.sh: 无人值守日线双源pipeline(qfq→raw→rsync→验证)
- verify_dual_source: 容器内双源部署验证(fetch_day/iter_bars/除权日)

实测: 日线双源5yr 29600文件×2(rsync NAS), 浦发除权日 raw-5.9%/qfq-0.7%,
容器内 fetch_day/iter_bars/全市场抽检5只全通过. 15min沪深300 baostock限流121只.

scripts/data_platform/baostock_download.py

@@ -0,0 +1,166 @@
+#!/usr/bin/env python3
+"""baostock 15min 下载（NAS 容器跑，本机零负载）。
+
+为何 baostock：新浪 stock_zh_a_minute 15min 历史仅半年；baostock 5.5 年 + 复权齐全
+（实测浦发 2021-01-04 起 21328 行；除权日 raw -5.9% vs qfq -0.7% 平滑）。
+
+双源：qfq(adjustflag=2) + raw(adjustflag=3)。断点续传（skip 已存在，扩范围重下）。
+数据落 NAS 本地盘（容器挂载 /volume1/stock → /stock，最快，不经网络写入）。
+
+约束：单线程 + SLEEP 限速（baostock 服务器温和限频，避免封）。
+
+用法（NAS 容器，本机编排）：
+  /var/packages/Docker/target/usr/bin/docker run --rm -d --name dl15 \\
+    --memory=512m -v /volume1/stock:/stock \\
+    sanguo_vnpy_v2:with-sqlite \\
+    sh -c "pip install baostock -q && python /stock/sanguo_vnpy/scripts/baostock_download.py --all --adjust both --start 2021-01-01"
+
+env:
+  DL_DIR       输出根（默认 /stock/A股数据/minute_kline）
+  STOCK_LIST   全市场 csv（默认 NAS stock_info csv）
+  SLEEP        每只每源间隔秒（默认 0.3）
+"""
+import argparse
+import csv
+import logging
+import os
+import sys
+import time
+
+import pandas as pd
+import baostock as bs
+
+logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s")
+log = logging.getLogger("dl_15min")
+
+DL_DIR = os.environ.get("DL_DIR", "/stock/A股数据/minute_kline")
+SLEEP = float(os.environ.get("SLEEP", "0.3"))
+DEFAULT_STOCK_LIST = "/stock/A股数据/stock_info/stock_basic_info_raw_20260326_113530.csv"
+
+
+def prefix(code: str) -> str:
+    return "sh" if code.startswith(("60", "68", "51", "56", "58")) else "sz"
+
+
+def bs_symbol(code: str) -> str:
+    return f"{prefix(code)}.{code}"
+
+
+def out_path(code: str, adjust: str) -> str:
+    d = os.path.join(DL_DIR, f"15min_{adjust}")
+    os.makedirs(d, exist_ok=True)
+    return os.path.join(d, f"{prefix(code)}{code}_15min.parquet")
+
+
+def exists(code: str, adjust: str, start_year: int) -> bool:
+    """断点续传：文件存在且含 start_year 数据（扩范围 start 更早 → 重下覆盖）。"""
+    f = out_path(code, adjust)
+    if not os.path.exists(f):
+        return False
+    try:
+        df = pd.read_parquet(f, columns=["date"])
+        return str(start_year) in df["date"].astype(str).str[:4].unique()
+    except Exception:
+        return False
+
+
+def download_one(code: str, start: str, end: str, adjustflag: str):
+    """baostock 15min，返回 (df, None) 或 (None, err)。"""
+    rs = bs.query_history_k_data_plus(
+        bs_symbol(code),
+        "date,time,open,high,low,close,volume,amount",
+        start_date=start, end_date=end, frequency="15", adjustflag=adjustflag,
+    )
+    rows, fields = [], rs.fields
+    while (rs.error_code == "0") & rs.next():
+        rows.append(rs.get_row_data())
+    if not rows:
+        return None, rs.error_msg
+    df = pd.DataFrame(rows, columns=fields)
+    for c in ("open", "high", "low", "close", "volume", "amount"):
+        df[c] = pd.to_numeric(df[c], errors="coerce")
+    t = df["time"].astype(str)
+    df["datetime"] = pd.to_datetime(
+        df["date"] + " " + t.str[8:10] + ":" + t.str[10:12], errors="coerce"
+    )
+    df = df.dropna(subset=["close", "datetime"]).sort_values("datetime")
+    return df, None
+
+
+def save_one(code, df, adjust):
+    df.to_parquet(out_path(code, adjust))
+    return len(df)
+
+
+def load_all_codes(stock_list):
+    codes = []
+    with open(stock_list, encoding="utf-8", errors="replace") as f:
+        for row in csv.DictReader(f):
+            for k in ("code", "symbol", "ts_code", "代码", "股票代码", "成分券代码"):
+                if k in row and row[k]:
+                    c = str(row[k]).strip().split(".")[0]
+                    if c.isdigit() and len(c) == 6:
+                        codes.append(c)
+                        break
+    return codes
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--symbols")
+    ap.add_argument("--all", action="store_true")
+    ap.add_argument("--start", default="2021-01-01")
+    ap.add_argument("--end", default=None)
+    ap.add_argument("--adjust", default="qfq", help="qfq / raw / both")
+    ap.add_argument("--force", action="store_true")
+    args = ap.parse_args()
+
+    end = args.end or time.strftime("%Y-%m-%d")
+    start_year = int(args.start[:4])
+    adjusts = ["qfq", "raw"] if args.adjust == "both" else [args.adjust]
+    flag = {"qfq": "2", "raw": "3"}
+
+    if args.symbols:
+        codes = [c.strip() for c in args.symbols.split(",") if c.strip()]
+    elif args.all:
+        sl = os.environ.get("STOCK_LIST", DEFAULT_STOCK_LIST)
+        if not os.path.exists(sl):
+            log.error("STOCK_LIST 不存在: %s", sl)
+            sys.exit(1)
+        codes = load_all_codes(sl)
+        log.info("--all 读到 %d 只", len(codes))
+    else:
+        ap.error("需 --symbols 或 --all")
+
+    lg = bs.login()
+    log.info("baostock login: %s %s", lg.error_code, lg.error_msg)
+
+    ok = fail = rows = skipped = 0
+    for i, code in enumerate(codes, 1):
+        for adj in adjusts:
+            if not args.force and exists(code, adj, start_year):
+                skipped += 1
+                continue
+            df, err = download_one(code, args.start, end, flag[adj])
+            if df is None or df.empty:
+                fail += 1
+                if i <= 20 or i % 100 == 0:
+                    log.warning("[%d/%d] %s %s empty %s", i, len(codes), code, adj, err)
+            else:
+                try:
+                    n = save_one(code, df, adj)
+                    ok += 1
+                    rows += n
+                    if i <= 20 or i % 100 == 0:
+                        log.info("[%d/%d] %s %s ok %d", i, len(codes), code, adj, n)
+                except Exception as e:  # noqa: BLE001
+                    fail += 1
+                    log.error("save %s %s: %s", code, adj, e)
+            time.sleep(SLEEP)
+    bs.logout()
+    log.info("=== 完成 ok=%d skip=%d fail=%d rows=%d dir=%s ===",
+             ok, skipped, fail, rows, DL_DIR)
+
+
+if __name__ == "__main__":
+    main()

scripts/data_platform/full_deploy_5yr.sh

@@ -0,0 +1,38 @@
+#!/bin/bash
+# 无人值守日线双源 5yr 全市场部署 pipeline（本机串行，qfq→raw→rsync→验证）。
+# 不并发打新浪源（封IP）：qfq 进程结束才下 raw。
+# 15min baostock 走 NAS 容器，与本机并行（不同源不同机）。
+set -uo pipefail
+cd /Users/chufeng/.openclaw/sanguo_projects/sanguo_vnpy_v2
+LOG=/tmp/full_deploy.log
+SL=data_cache/stock_info/stock_basic_info_raw_20260326_113530.csv
+log(){ echo "[$(date '+%m-%d %H:%M:%S')] $*" | tee -a "$LOG"; }
+
+log "=== STAGE1 等 qfq 5yr 进程结束 ==="
+while pgrep -f "raw_redownload.py --all --adjust qfq" >/dev/null; do sleep 60; done
+log "qfq 完成: $(find data_cache/qfq -name '*.parquet'|wc -l) 文件"
+
+log "=== STAGE2 raw 5yr 全市场（断点续传，单线程限速）==="
+STOCK_LIST="$SL" RAW_DIR=data_cache/raw SLEEP=1.0 \
+  python3 scripts/data_platform/raw_redownload.py --all --start 2021-01-01 --adjust "" >> "$LOG" 2>&1
+log "raw 完成: $(find data_cache/raw -name '*.parquet'|wc -l) 文件"
+
+log "=== STAGE3 rsync qfq 全量(--delete覆盖NAS旧5验证股) → NAS ==="
+rsync -az --delete \
+  data_cache/qfq/ \
+  sanguo-nas:/volume1/stock/A股数据/日线数据/qfq/ >> "$LOG" 2>&1
+log "qfq rsync done"
+
+log "=== STAGE4 rsync raw 2021-2023 → NAS（不覆盖NAS已有24-26）==="
+for y in 2021 2022 2023; do
+  rsync -az \
+    "data_cache/raw/$y/" \
+    "sanguo-nas:/volume1/stock/A股数据/日线数据/raw/$y/" >> "$LOG" 2>&1
+  log "  raw/$y done"
+done
+
+log "=== STAGE5 验证 NAS 文件数 ==="
+ssh sanguo-nas 'for d in qfq raw; do for y in 2021 2022 2023 2024 2025 2026; do echo "$d/$y=$(ls /volume1/stock/A股数据/日线数据/$d/$y/*.parquet 2>/dev/null|wc -l)"; done; done' >> "$LOG" 2>&1
+
+log "=== ALL DONE ==="
+echo "ALL_DONE_MARKER" >> "$LOG"

scripts/data_platform/raw_redownload.py

@@ -79,6 +79,17 @@ def save_one(code: str, df) -> int:
     return n
 
 
+def exists(code: str, start_year: int) -> bool:
+    """symbol 在 start_year 是否已有 parquet（断点续传）。
+
+    同范围续跑 → skip；扩范围（start 更早）→ 新 start_year 不存在 → 重下全量覆盖。
+    """
+    pref = prefix_for(code)
+    return os.path.exists(
+        os.path.join(RAW_DIR, str(start_year), f"{pref}{code}_daily.parquet")
+    )
+
+
 def load_all_codes(stock_list: str) -> list[str]:
     """从 stock_basic_info csv 读代码列表（容错列名）。"""
     codes = []
@@ -101,6 +112,7 @@ def main():
     ap.add_argument("--start", default="2024-01-01")
     ap.add_argument("--end", default=None, help="默认今天")
     ap.add_argument("--adjust", default="", help="复权: '' raw / 'qfq' 前复权（双源用）")
+    ap.add_argument("--force", action="store_true", help="强制重下（默认 skip 已存在=断点续传）")
     args = ap.parse_args()
 
     end = args.end or time.strftime("%Y-%m-%d")
@@ -121,8 +133,14 @@ def main():
     import warnings
     warnings.filterwarnings("ignore")
 
-    ok = fail = rows = 0
+    start_year = int(args.start[:4])
+    ok = fail = rows = skipped = 0
     for i, code in enumerate(codes, 1):
+        if not args.force and exists(code, start_year):
+            skipped += 1
+            if skipped % 500 == 0:
+                log.info("[%d/%d] ... skipped %d 已存在", i, len(codes), skipped)
+            continue
         df, err = download_one(ak, code, args.start, end, args.adjust)
         if df is None:
             fail += 1
@@ -138,7 +156,7 @@ def main():
                 log.error("[%d/%d] %s SAVE FAIL %s", i, len(codes), code, e)
         time.sleep(SLEEP)  # 限速（用户约束）
 
-    log.info("=== 完成: ok=%d fail=%d rows=%d，raw_dir=%s ===", ok, fail, rows, RAW_DIR)
+    log.info("=== 完成: ok=%d skip=%d fail=%d rows=%d，raw_dir=%s ===", ok, skipped, fail, rows, RAW_DIR)
 
 
 if __name__ == "__main__":

scripts/verify_dual_source.py

@@ -0,0 +1,64 @@
+#!/usr/bin/env python3
+"""双源（raw/qfq）部署验证：容器内读 NAS data_platform.yaml 配置的双源，
+确认 fetch_day（C-S3 实走接口）+ iter_bars 在 raw_dir/qfq_dir 正确路由，
+且除权日 raw 有缺口 / qfq 平滑（分红除权准确）。
+
+用法（容器）：
+  docker run --rm -v /volume1/stock:/volume1/stock \
+    -v /volume1/homes/admin/.sanguo_projects/sanguo_vnpy_v2:/app \
+    --entrypoint sh sanguo_vnpy_v2:with-sqlite \
+    -c "cd /app && python scripts/verify_dual_source.py"
+"""
+import sys
+sys.path.insert(0, "/app")
+from sanguo_data.config import load_config, find_config_path
+from sanguo_trader.data_source import fetch_day, iter_bars
+
+cfg = load_config(find_config_path())
+print(f"config: {find_config_path()}")
+print(f"data_paths: raw_dir={cfg.data_paths.get('raw_dir')} qfq_dir={cfg.data_paths.get('qfq_dir')}")
+
+OK = True
+
+# 1) fetch_day（C-S3 实走接口）除权日双源
+print("\n=== fetch_day 除权日双源（浦发 2022-07-21 除权）===")
+res = {}
+for adj in ["qfq", "raw"]:
+    b  = fetch_day("600000", "2022-07-21", "d", adj, cfg)
+    bp = fetch_day("600000", "2022-07-20", "d", adj, cfg)
+    if b is None or bp is None:
+        print(f"  ❌ {adj}: fetch_day 返回 None（数据缺失或路由失败）")
+        OK = False
+        continue
+    drop = ((b.close_price / bp.close_price) - 1) * 100
+    res[adj] = drop
+    print(f"  {adj}: 07-20={bp.close_price:.4f} → 07-21={b.close_price:.4f} ({drop:+.1f}%)")
+
+# 除权日 raw 应有明显缺口（< -3%），qfq 平滑（> -3%）
+if "raw" in res and "qfq" in res:
+    if res["raw"] < -3 and res["qfq"] > -3:
+        print("  ✅ 除权处理正确：raw 含缺口（撮合真实价）/ qfq 平滑（信号准）")
+    else:
+        print(f"  ⚠️ 除权差异不明显 raw={res['raw']:.1f}% qfq={res['qfq']:.1f}%")
+
+# 2) iter_bars cross-section（symbols=list）
+print("\n=== iter_bars cross-section（4 天）===")
+for adj in ["qfq", "raw"]:
+    days = list(iter_bars(["600000"], "2022-07-19", "2022-07-22", "d", adj, cfg))
+    if not days:
+        print(f"  ❌ {adj}: iter_bars 0 days")
+        OK = False
+    else:
+        last = days[-1]
+        print(f"  {adj}: {len(days)} days, 末日={last[0]} close={last[1]['600000'].close_price:.4f}")
+
+# 3) 全市场可读性抽检（5 只）
+print("\n=== 全市场抽检（5 只 fetch_day）===")
+for sym in ["600000", "000001", "000858", "300750", "688981"]:
+    b = fetch_day(sym, "2026-07-07", "d", "qfq", cfg)
+    print(f"  {sym}: {'✅ close='+format(b.close_price,'.4f') if b else '❌ None'}")
+    if b is None:
+        OK = False
+
+print("\n" + ("✅ 双源部署验证通过" if OK else "❌ 双源验证有失败项"))
+sys.exit(0 if OK else 1)