perf(data): 四热路径日期区间SARGable化——substr(datetime,1,10)对索引列套函数打不进复合索引datetime列,每股扫全量日线史取短窗——08-25晨9:30生产实锤:momentum选股1/2(RPS池)174s未达<60s验收,阶段日志精确定罪(过滤段含双seek仅7.8s);病根=PanelFetcher的30天窗SQL每只股扫~5000行日线史取~22行,3226只≈1600万行=174s量级吻合。修=裸列datetime>=start AND datetime<end+1天排他上界(与按日期前10位比较在d裸日期/15m·5m时间戳两格式下语义严格等价,含end当日全部行排除次日),区间打进复合索引第4列每股只扫窗口行。四处同病同修:PanelFetcher宽表/PriceFetcher逐只get_price/limit-status近2根90天窗(原substr版90天下界同样打不进索引)/datareader CTA回测K线。+3测试:防回潮扫描(三文件钉死禁substr谓词)+双格式边界行为(d裸日期与15m时间戳end当日含次日排)+get_price同语义;portfolio+data_platform 644绿;待VPS探针终验计时 [vps]

sanguo_data/datareader.py

@@ -9,7 +9,7 @@ if _VNPY_SRC not in sys.path:
     sys.path.insert(0, _VNPY_SRC)
 
 import pandas as pd
-from datetime import datetime, date
+from datetime import datetime, date, timedelta
 from vnpy.trader.object import BarData
 from vnpy.trader.constant import Exchange, Interval
 from vnpy.trader.setting import SETTINGS
@@ -148,12 +148,16 @@ def read_index_daily(code: str, start, end, cfg) -> pd.DataFrame:
     conn = sqlite3.connect(db_path, timeout=30)
     conn.execute("PRAGMA busy_timeout = 30000")
     try:
-        # substr(datetime,1,10) 比日期规避混合格式（有纯日期有带时间，同 provider 模式）
+        # SARGable 日期区间(2026-08-25 同 provider P0 治本): 裸列 datetime>=/<,
+        # 右端 end+1 天排他——d 裸日期与 15m/5m 时间戳混合格式下与按日期前 10 位
+        # 比较语义严格等价(含 end 当日全部行);substr 版打不进复合索引 datetime
+        # 列, 短窗读每股扫全量日线史
+        end_excl = (datetime.strptime(end_str, "%Y-%m-%d") + timedelta(days=1)).strftime("%Y-%m-%d")
         df = pd.read_sql(
             "SELECT datetime, open_price, high_price, low_price, close_price, volume "
             "FROM dbbardata WHERE symbol=? AND exchange=? AND interval='d' "
-            "AND substr(datetime,1,10)>=? AND substr(datetime,1,10)<=? ORDER BY datetime",
+            "AND datetime>=? AND datetime<? ORDER BY datetime",
-            conn, params=(symbol, exchange, start_str, end_str),
+            conn, params=(symbol, exchange, start_str, end_excl),
         )
     finally:
         conn.close()

sanguo_portfolio/providers/fetchers/price.py

@@ -7,7 +7,7 @@ get_closes_panel(e7f9426 生产版,UNION ALL 340× 索引优化保留),合法参
 from __future__ import annotations
 
 import re
-from datetime import datetime
+from datetime import datetime, timedelta
 from typing import TYPE_CHECKING, Any, Dict, List, Optional, Sequence, Union
 
 import pandas as pd
@@ -28,6 +28,20 @@ def _safe_date_literal(s: str) -> str:
     return f"'{s}'"
 
 
+def end_exclusive_str(end_str: str) -> str:
+    """end 日期 → 排他上界(end+1 天)字符串。SARGable 日期区间右端。
+
+    2026-08-25 P0 治本: 旧「取日期前 10 位再比较」写法对索引列套函数,日期区间
+    打不进复合索引 datetime 列 → 每只股扫全量日线史取 30 天窗(momentum RPS 池
+    3226 只实测 174s)。裸列 ``datetime<end+1天`` 与旧写法语义严格等价:
+    d 裸日期('2026-08-22'<…23')与 15m/5m 时间戳('2026-08-22 15:00:00'<…23')
+    均含 end 当日全部行、排除次日。
+    """
+    return (
+        datetime.strptime(end_str[:10], "%Y-%m-%d").date() + timedelta(days=1)
+    ).isoformat()
+
+
 def _safe_interval_literal(s: str) -> str:
     if not isinstance(s, str) or not _INTERVAL_RE.match(s) or len(s) > 16:
         raise ValueError(f"Invalid interval: {s!r}")
@@ -55,10 +69,13 @@ class PriceFetcher:
         conn = ctx._connect()
         start_str = query.start_date or "1990-01-01"
         end_str = query.end_date or datetime.now().strftime("%Y-%m-%d")
+        # SARGable 日期区间(2026-08-25 P0): 裸列 datetime>=/<,右端 end+1 天排他
+        # ——substr 版每只股扫全量日线史,详见 end_exclusive_str
+        end_excl = end_exclusive_str(end_str)
         q = (
             "SELECT datetime, open_price, high_price, low_price, close_price, "
             "volume, turnover FROM dbbardata WHERE symbol=? AND exchange=? "
-            "AND interval='d' AND substr(datetime,1,10)>=? AND substr(datetime,1,10)<=? "
+            "AND interval='d' AND datetime>=? AND datetime<? "
             "ORDER BY datetime"
         )
         need_qfq = query.fq in ("qfq", "pre", "前复权")
@@ -67,7 +84,7 @@ class PriceFetcher:
         for jq_code in query.security:
             sym, exc = jq_to_dbbardata(jq_code)
             frames[jq_code] = pd.read_sql(
-                q, conn, params=(sym, exc, start_str, end_str)
+                q, conn, params=(sym, exc, start_str, end_excl)
             )
             if need_qfq:
                 qfq_rows[jq_code] = _read_qfq_rows(_jq_to_bs_code(jq_code), conn)
@@ -188,18 +205,22 @@ class PanelFetcher:
 
         conn = ctx._connect()
         start_lit = _safe_date_literal(query.start or "1990-01-01")
-        end_lit = _safe_date_literal(query.end or datetime.now().strftime("%Y-%m-%d"))
+        end_raw = query.end or datetime.now().strftime("%Y-%m-%d")
+        _safe_date_literal(end_raw)  # 校验 YYYY-MM-DD(fail-fast 同老契约)
+        end_excl_lit = f"'{end_exclusive_str(end_raw)}'"
         interval_lit = _safe_interval_literal(query.interval)
 
         CHUNK_SIZE = 400
         chunk_frames: List[pd.DataFrame] = []
         for i in range(0, len(pairs), CHUNK_SIZE):
             chunk = pairs[i:i + CHUNK_SIZE]
+            # SARGable 日期区间(2026-08-25 P0): 裸列区间打进复合索引 datetime 列,
+            # 每股只扫窗口行(30 天窗 ~22 行);substr 版扫全量日线史(3226 只=174s)
             sub_template = (
                 "SELECT datetime, symbol, close_price FROM dbbardata "
                 f"WHERE symbol=? AND exchange=? AND interval={interval_lit} "
-                f"AND substr(datetime,1,10)>={start_lit} "
+                f"AND datetime>={start_lit} "
-                f"AND substr(datetime,1,10)<={end_lit}"
+                f"AND datetime<{end_excl_lit}"
             )
             q = " UNION ALL ".join([sub_template] * len(chunk))
             params: List[Any] = []

sanguo_portfolio/providers/local_unified_provider.py

@@ -782,14 +782,18 @@ class LocalUnifiedProvider(DataProvider):  # type: ignore[misc]
         from datetime import timedelta
         start_lim = (datetime.strptime(date_str, "%Y-%m-%d") - timedelta(days=90)).strftime("%Y-%m-%d")
         start_lit = _safe_date_literal(start_lim)
-        end_lit = _safe_date_literal(date_str)
+        # SARGable 日期区间(2026-08-25 P0): 裸列区间+end+1 天排他——substr 版的
+        # 90 天下界打不进索引, 每股仍扫全量日线史(语义等价, 见 fetchers.price)
+        end_excl_lit = _safe_date_literal(
+            (datetime.strptime(date_str, "%Y-%m-%d") + timedelta(days=1)).strftime("%Y-%m-%d")
+        )
         interval_lit = _safe_interval_literal("d")
         # 纯 SELECT UNION ALL(子查询带 ORDER BY/LIMIT 触发 SQLite compound 限制);
-        # 用 90 天下界限定范围(全历史 → 近 90 天), 走复合索引, pandas 端取最近 2 根。
+        # 90 天下界 + 裸列区间走复合索引(每股只扫窗口行), pandas 端取最近 2 根。
         sub = (
             "SELECT symbol, exchange, close_price, high_price, low_price, volume, datetime "
             f"FROM dbbardata WHERE symbol=? AND exchange=? AND interval={interval_lit} "
-            f"AND substr(datetime,1,10)>={start_lit} AND substr(datetime,1,10)<={end_lit}"
+            f"AND datetime>={start_lit} AND datetime<{end_excl_lit}"
         )
         bars: Dict[tuple, list] = {}
         sym_exc = list({p[1] for p in pairs})  # 去重 (sym,exc)

tests/portfolio/test_fetchers.py

@@ -242,3 +242,62 @@ class TestOptionalContract:
             fields=["close", "acc_net_value"],
         )
         assert df["acc_net_value"].isna().all()
+
+
+# ======================== SARGable 日期区间(2026-08-25 P0) ========================
+
+class TestSargableDateRange:
+    """substr(datetime,1,10) 对索引列套函数 → 日期区间打不进复合索引 datetime 列,
+    每股扫全量日线史取短窗(momentum RPS 池 3226 只 30 天窗 VPS 实测 174s;provider
+    /datareader 四热路径同病)。裸列 ``datetime>=start AND datetime<end+1天`` 与
+    substr 语义严格等价(d 裸日期/带时间戳两格式, 含 end 当日全部行、排除次日)。"""
+
+    def test_no_substr_datetime_regression(self):
+        """钉死四个热路径文件不再回潮非 SARGable 谓词。"""
+        import pathlib
+        root = pathlib.Path(__file__).resolve().parents[2]
+        for rel in (
+            "sanguo_portfolio/providers/fetchers/price.py",
+            "sanguo_portfolio/providers/local_unified_provider.py",
+            "sanguo_data/datareader.py",
+        ):
+            src = (root / rel).read_text(encoding="utf-8")
+            assert "substr(datetime" not in src, f"{rel} 回潮非 SARGable 谓词"
+
+    def test_panel_end_date_inclusive_bare_and_timestamp(self, unified_provider):
+        """end 当日全部行含(d 裸日期 + 15m 时间戳两格式), 次日排除。"""
+        conn = sqlite3.connect(unified_provider.db_path)
+        conn.executemany("INSERT INTO dbbardata VALUES(?,?,?,?,?,?,?,?,?,?,?)", [
+            # d 裸日期: end 当日含 / 次日排
+            ("600600", "SSE", "2024-06-19", "d", 100, 1e3, 0, 1.0, 1.0, 1.0, 10.0),
+            ("600600", "SSE", "2024-06-21", "d", 100, 1e3, 0, 1.0, 1.0, 1.0, 12.0),
+            # 15m 时间戳: end 当日含 / 次日排
+            ("600600", "SSE", "2024-06-20 14:35:00", "15m", 100, 1e3, 0, 1.0, 1.0, 1.0, 11.0),
+            ("600600", "SSE", "2024-06-21 09:35:00", "15m", 100, 1e3, 0, 1.0, 1.0, 1.0, 13.0),
+        ])
+        conn.commit()
+        conn.close()
+        # d 面: 窗口 06-18~06-20 → 只有裸日期 06-19 行
+        panel_d = unified_provider.get_closes_panel(
+            ["600600.XSHG"], "2024-06-18", "2024-06-20", fq="raw")
+        assert [str(d)[:10] for d in panel_d.index] == ["2024-06-19"]
+        assert panel_d.iloc[0, 0] == 10.0
+        # 15m 面: 同窗 → 只有 06-20 14:35 行(end 当日带时间戳不被右界误伤)
+        panel_15 = unified_provider.get_closes_panel(
+            ["600600.XSHG"], "2024-06-18", "2024-06-20", interval="15m", fq="raw")
+        assert [str(d)[:16] for d in panel_15.index] == ["2024-06-20 14:35"]
+        assert panel_15.iloc[0, 0] == 11.0
+
+    def test_get_price_end_date_inclusive_bare_date(self, unified_provider):
+        """PriceFetcher 逐只路径同语义: 裸日期 end 当日含、次日排。"""
+        conn = sqlite3.connect(unified_provider.db_path)
+        conn.executemany("INSERT INTO dbbardata VALUES(?,?,?,?,?,?,?,?,?,?,?)", [
+            ("600600", "SSE", "2024-06-19", "d", 100, 1e3, 0, 1.0, 1.0, 1.0, 10.0),
+            ("600600", "SSE", "2024-06-21", "d", 100, 1e3, 0, 1.0, 1.0, 1.0, 12.0),
+        ])
+        conn.commit()
+        conn.close()
+        df = unified_provider.get_price(
+            "600600.XSHG", start_date="2024-06-18", end_date="2024-06-20")
+        assert len(df) == 1
+        assert df.iloc[-1]["close"] == 10.0