cma-management/backend/scripts/migrate_subjects_entity_id_20260923.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""`subjects` 账套分层 DDL + 存量回填 — 2026-09-23

任务：`0-P1-project-20260923-entity-dimension-convergence-fullstack` §三③
（裁定件 `project-entity-dimension-ruling-20260923` §3.1「方案 A：分层」）

背景
----
`subjects` 207 行混装且 `subject_code` **全局唯一**（`uk_subject_code`）：
  · 43 行标准准则科目（1001 库存现金 …，`remark IS NULL`）= 天然全局
  · 164 行 `remark='zxt-sync'` 酣客（智享通）企业明细   = 账套私有
双 token 实测两账套拿到**同一份 207 行** ⇒ 博海页面出现酣客的
`100106 酒窖备用金` / `100107 现金备用卡（唐）` / `100108 友赞商户`（结构性串配置）。

分层口径（`knowledge_articles` 同款已验模式）
------------------------------------------
  `entity_id IS NULL` = 全局标准准则科目（两账套共享）
  `entity_id = N`     = 账套 N 私有明细
  唯一键 `(subject_code)` → `(entity_id, subject_code)`
  ⚠️ MySQL 把 NULL 视为互不相同的索引值 ⇒ 「全局行同码重复」索引拦不住，
     由写入口（`POST /api/cma/bot/subjects` UPSERT 判重）兜住，不可依赖索引。

动作（四段可独立跑，互不隐含）
----------------------------
  --check     只读预检（默认）：列/索引现状 + 行数分布 + 唯一键重复检查，零写入
  --apply     DDL：ADD COLUMN entity_id INT NULL + INDEX + UNIQUE(entity_id,subject_code)
              − DROP INDEX uk_subject_code（幂等：已存在的步骤自动跳过）
  --backfill  存量回填（**须 --evidence <json>**，与 DDL 分开放行）：
              `remark='zxt-sync'` 的 164 行 → `--target-entity`（默认 1）；其余保持 NULL
  --down      回滚：重建 `uk_subject_code`（先断言无同码重复，fail-closed）→ 删新唯一键 → 删列
  --show-create  打印 SHOW CREATE TABLE subjects 原文（验收证据）

生产保护（生产资产派单纪律）
--------------------------
  目标库名 == 'cma' 时 `--apply/--backfill/--down` 一律拒绝；须 `--allow-prod` **且**
  环境变量 `CMA_DDL_APPROVED=1`（= 已拿到 ops 审批单）才放行。`--check/--show-create` 只读不受限。

用法
----
  venv/bin/python3 scripts/migrate_subjects_entity_id_20260923.py --check
  CMA_DDL_APPROVED=1 venv/bin/python3 scripts/migrate_subjects_entity_id_20260923.py \\
      --apply --allow-prod
  CMA_DDL_APPROVED=1 venv/bin/python3 scripts/migrate_subjects_entity_id_20260923.py \\
      --backfill --allow-prod --evidence /path/to/evidence.json --target-entity 1
"""
from __future__ import annotations

import argparse
import json
import os
import sys

sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))

from sqlalchemy import create_engine, inspect, text  # noqa: E402

TABLE = "subjects"
COLUMN = "entity_id"
NEW_UNIQUE = "uk_entity_subject"
OLD_UNIQUE = "uk_subject_code"
NEW_INDEX = "ix_subjects_entity_id"
PROD_DB_NAMES = ("cma",)
DDL_ADD_COLUMN = (
    f"ALTER TABLE {TABLE} ADD COLUMN {COLUMN} INT NULL "
    "COMMENT '企业ID（NULL=全局标准准则科目 / N=账套私有明细）' AFTER id"
)
DDL_ADD_UNIQUE = f"ALTER TABLE {TABLE} ADD UNIQUE INDEX {NEW_UNIQUE} ({COLUMN}, subject_code)"
DDL_DROP_OLD_UNIQUE = f"ALTER TABLE {TABLE} DROP INDEX {OLD_UNIQUE}"
DDL_ADD_OLD_UNIQUE = f"ALTER TABLE {TABLE} ADD UNIQUE INDEX {OLD_UNIQUE} (subject_code)"
DDL_DROP_COLUMN = f"ALTER TABLE {TABLE} DROP COLUMN {COLUMN}"


def is_sqlite(engine) -> bool:
    """沙箱演练（SQLite）与生产（MySQL）的 DDL 方言差异：索引/列定义语句形态不同"""
    return engine.dialect.name == "sqlite"


def stmt_add_column(engine) -> str:
    """生产形态 = MySQL（带 COMMENT / AFTER）；沙箱 = SQLite 精简形态（语法不支持二者）"""
    if is_sqlite(engine):
        return f"ALTER TABLE {TABLE} ADD COLUMN {COLUMN} INT NULL"
    return DDL_ADD_COLUMN


def stmt_add_index(engine, name: str, cols: list, unique: bool = False) -> str:
    cols_s = ", ".join(cols)
    if is_sqlite(engine):
        return f"CREATE {'UNIQUE ' if unique else ''}INDEX {name} ON {TABLE} ({cols_s})"
    return f"ALTER TABLE {TABLE} ADD {'UNIQUE ' if unique else ''}INDEX {name} ({cols_s})"


def stmt_drop_index(engine, name: str) -> str:
    return f"DROP INDEX {name}" if is_sqlite(engine) else f"ALTER TABLE {TABLE} DROP INDEX {name}"


def db_name_of(engine) -> str:
    try:
        return engine.url.database or ""
    except Exception:
        return ""


def state(engine) -> dict:
    insp = inspect(engine)
    tables = set(insp.get_table_names())
    out = {"table_exists": TABLE in tables, "has_column": False,
           "indexes": [], "unique_keys": [], "counts": {}}
    if not out["table_exists"]:
        return out
    out["has_column"] = COLUMN in {c["name"] for c in insp.get_columns(TABLE)}
    out["indexes"] = sorted({i["name"] for i in insp.get_indexes(TABLE)})
    out["unique_keys"] = sorted({uc["name"] for uc in insp.get_unique_constraints(TABLE)})
    with engine.connect() as conn:
        out["counts"]["total"] = conn.execute(text(f"SELECT COUNT(*) FROM {TABLE}")).scalar()
        if out["has_column"]:
            rows = conn.execute(text(
                f"SELECT IFNULL({COLUMN},'NULL'), COUNT(*) FROM {TABLE} GROUP BY 1 ORDER BY 1")).fetchall()
            out["counts"]["by_entity"] = [[str(r[0]), int(r[1])] for r in rows]
        rows = conn.execute(text(
            f"SELECT IFNULL(remark,'(null)'), COUNT(*) FROM {TABLE} GROUP BY 1 ORDER BY 2 DESC")).fetchall()
        out["counts"]["by_remark"] = [[str(r[0]), int(r[1])] for r in rows]
        dup_code = conn.execute(text(
            f"SELECT COUNT(*) FROM (SELECT subject_code FROM {TABLE} GROUP BY subject_code "
            f"HAVING COUNT(*) > 1) t")).scalar()
        out["duplicate_subject_code_groups"] = int(dup_code)
        if out["has_column"]:
            dup_pair = conn.execute(text(
                f"SELECT COUNT(*) FROM (SELECT {COLUMN}, subject_code FROM {TABLE} "
                f"GROUP BY {COLUMN}, subject_code HAVING COUNT(*) > 1) t")).scalar()
            out["duplicate_entity_code_groups"] = int(dup_pair)
    return out


def print_check(st: dict) -> None:
    print("== 只读预检（inspector / SELECT，零写入）==")
    print(f"  table_exists={st['table_exists']} has_column={st['has_column']}")
    print(f"  indexes={st['indexes']}")
    print(f"  unique_keys={st['unique_keys']}")
    print(f"  counts={st['counts']}")
    print(f"  duplicate_subject_code_groups={st.get('duplicate_subject_code_groups')}")
    if "duplicate_entity_code_groups" in st:
        print(f"  duplicate_entity_code_groups={st['duplicate_entity_code_groups']}")


def do_apply(engine) -> int:
    st = state(engine)
    if not st["table_exists"]:
        print("!!! 目标库没有 subjects 表，拒绝 --apply")
        return 3
    if st.get("duplicate_subject_code_groups"):
        print("!!! subjects.subject_code 已有重复组，先处置数据再加/换唯一键")
        return 3
    done = []
    with engine.begin() as conn:
        if not st["has_column"]:
            add_col = stmt_add_column(engine)
            conn.execute(text(add_col))
            done.append(add_col)
        cur = state(engine)
        present = set(cur["indexes"]) | set(cur["unique_keys"])
        add_idx = stmt_add_index(engine, NEW_INDEX, [COLUMN])
        if NEW_INDEX not in present:
            conn.execute(text(add_idx))
            done.append(add_idx)
        add_uk = stmt_add_index(engine, NEW_UNIQUE, [COLUMN, "subject_code"], unique=True)
        if NEW_UNIQUE not in present:
            conn.execute(text(add_uk))
            done.append(add_uk)
        if OLD_UNIQUE in present:
            drop_old = stmt_drop_index(engine, OLD_UNIQUE)
            conn.execute(text(drop_old))
            done.append(drop_old)
    print("== APPLY 完成，执行语句：")
    for s in done:
        print("   " + s)
    if not done:
        print("   （无语句执行：已是目标形态，幂等）")
    print_check(state(engine))
    return 0


def do_backfill(engine, evidence_path: str, target: int) -> int:
    if not evidence_path or not os.path.exists(evidence_path):
        print("!!! --backfill 必须提供 --evidence <json>（DAMA 只读取证 + 财务确认件锚点），拒绝执行")
        return 3
    ev = json.load(open(evidence_path, encoding="utf-8"))
    need = ("dama_evidence", "finance_confirmation", "counts")
    missing = [k for k in need if not ev.get(k)]
    if missing:
        print(f"!!! 取证件缺字段 {missing}，拒绝执行（fail-closed）")
        return 3
    st = state(engine)
    if not st["has_column"]:
        print("!!! entity_id 列未落地，先跑 --apply")
        return 3
    by_remark = dict(st["counts"]["by_remark"])
    live_private = int(by_remark.get("zxt-sync", 0))
    live_global = int(by_remark.get("(null)", 0))
    exp = ev["counts"]
    if int(exp.get("e1", -1)) != live_private or int(exp.get("global_null", -1)) != live_global:
        print(f"!!! 取证件行数锚点与现网不符：evidence e1={exp.get('e1')}/global_null="
              f"{exp.get('global_null')} vs live {live_private}/{live_global}，拒绝执行")
        return 3
    with engine.begin() as conn:
        n1 = conn.execute(text(
            f"UPDATE {TABLE} SET {COLUMN} = :e WHERE remark = 'zxt-sync'"), {"e": target}).rowcount
        n0 = conn.execute(text(
            f"UPDATE {TABLE} SET {COLUMN} = NULL WHERE remark IS NULL AND {COLUMN} IS NOT NULL")).rowcount
    print(f"== BACKFILL 完成：remark='zxt-sync' → entity_id={target} 命中 {n1} 行；"
          f"标准准则科目置 NULL {n0} 行（evidence: {evidence_path}）")
    print_check(state(engine))
    return 0


def do_down(engine, yes: bool) -> int:
    st = state(engine)
    if not st["has_column"]:
        print("== DOWN 跳过：entity_id 列不存在（已是回滚态）")
        return 0
    if st.get("duplicate_subject_code_groups"):
        print(f"!!! 已存在同码多行（{st['duplicate_subject_code_groups']} 组）⇒ 无法重建全局唯一键 "
              f"uk_subject_code，拒绝 --down（fail-closed：先人工处置同码行）")
        return 3
    if st["counts"]["by_entity"] and len(st["counts"]["by_entity"]) > 1 and not yes:
        print(f"!!! 列内已有分账套数据 {st['counts']['by_entity']}：回滚会丢失账套归属，须显式 --yes")
        return 3
    with engine.begin() as conn:
        present = set(st["indexes"]) | set(st["unique_keys"])
        if NEW_UNIQUE in present:
            conn.execute(text(stmt_drop_index(engine, NEW_UNIQUE)))
        cur = state(engine)
        present = set(cur["indexes"]) | set(cur["unique_keys"])
        if NEW_INDEX in present:
            conn.execute(text(stmt_drop_index(engine, NEW_INDEX)))
        if OLD_UNIQUE not in present:
            conn.execute(text(stmt_add_index(engine, OLD_UNIQUE, ["subject_code"], unique=True)))
        conn.execute(text(DDL_DROP_COLUMN))
    print("== DOWN 完成：重建 uk_subject_code + 删除 entity_id 列")
    print_check(state(engine))
    return 0


def main():
    ap = argparse.ArgumentParser(description="subjects 账套分层 DDL/回填")
    g = ap.add_mutually_exclusive_group()
    g.add_argument("--check", action="store_true", help="只读预检（默认）")
    g.add_argument("--apply", action="store_true", help="执行 DDL（幂等）")
    g.add_argument("--backfill", action="store_true", help="存量回填（须 --evidence）")
    g.add_argument("--down", action="store_true", help="回滚")
    ap.add_argument("--show-create", action="store_true", help="打印 SHOW CREATE TABLE")
    ap.add_argument("--url", default=None, help="目标库 SQLAlchemy URL（默认 backend/.env）")
    ap.add_argument("--allow-prod", action="store_true", help="生产库放行开关（须 CMA_DDL_APPROVED=1）")
    ap.add_argument("--evidence", default=None, help="--backfill 的取证件 JSON")
    ap.add_argument("--target-entity", type=int, default=1, help="--backfill 目标账套（默认 1=酣客）")
    ap.add_argument("--yes", action="store_true", help="--down 有分账套数据时的确认开关")
    args = ap.parse_args()

    if args.url:
        engine = create_engine(args.url, echo=False)
    else:
        from app.database import get_engine
        engine = get_engine()

    dbname = db_name_of(engine)
    print(f"目标库: {dbname or '(未知)'} | dialect={engine.dialect.name} | host={engine.url.host or 'socket'}")
    is_prod = dbname in PROD_DB_NAMES

    if (args.apply or args.backfill or args.down) and is_prod:
        if not args.allow_prod or os.getenv("CMA_DDL_APPROVED") != "1":
            print(f"!!! 拒绝执行：目标库是生产库 {dbname}。")
            print("!!! 生产 DDL/回填须 ops 审批后另派执行（本单边界：不得直连生产写入）。")
            print("!!! 放行方式：--allow-prod 且 export CMA_DDL_APPROVED=1（持有审批单）。")
            return 3

    if args.apply:
        return do_apply(engine)
    if args.backfill:
        return do_backfill(engine, args.evidence, args.target_entity)
    if args.down:
        return do_down(engine, args.yes)

    st = state(engine)
    print_check(st)
    if is_prod:
        print("  [注意] 目标是生产库：以上仅为只读预检，未做任何写入。")
    if args.show_create:
        if engine.dialect.name != "mysql":
            print("== SHOW CREATE TABLE 跳过（非 MySQL 目标）==")
        else:
            with engine.connect() as conn:
                print(conn.execute(text(f"SHOW CREATE TABLE {TABLE}")).fetchone()[1])
    return 0


if __name__ == "__main__":
    sys.exit(main())