@hupan56/wlkj 2.7.12 → 3.1.2

package/templates/qoder/scripts/{kg_build_db.py → domain/kg/build/kg_build_db.py}

@@ -1,327 +1,338 @@
-#!/usr/bin/env python
-# -*- coding: utf-8 -*-
-"""kg_build_db.py - 把 MySQL 数据库 schema 导入 kg.duckdb (数据层图谱)。
-
-和 kg_build.py 对称:
-  kg_build.py:   扫 data/code/*  → kg.duckdb (代码图谱: entities/fields/...)
-  kg_build_db.py: 扫 MySQL schema → kg.duckdb (数据图谱: DB_TABLE/db_columns/db_table_fields)
-
-产出 3 张新表 (复用现有 kg.duckdb, 不另建文件):
-  - db_tables:    表级实体 (db, table, table_comment, entity_count)
-  - db_columns:   列级真相 (db, table, column_name, data_type, is_nullable, column_comment)
-                  ← eval_prd.py 的 A1 第二层锚定读这张表
-  - db_table_fields: 表↔字段边 (table_id, column_name) 给图谱遍历用
-
-同时在 entities 表插入 DB_TABLE 类型实体 (让图谱认识"数据库表"这个概念),
-并在 edges 表连 DB_TABLE → 同名 ENDPOINT (若有), 做"改表影响哪些接口"的影响分析。
-
-安全: 只读 MySQL (INFORMATION_SCHEMA), 不碰业务数据行。
-
-增量策略 (库级签名对比, 不再全量重扫):
-  - 每个库算签名 (表名列表 + 列总数), 存 build_meta。
-  - 下次跑: 签名没变 = 该库 schema 没变 = 跳过。变了才重建该库。
-  - 比 MySQL update_time 准 (后者数据变更也会刷新, 不代表 schema 变)。
-
-用法:
-  python kg_build_db.py                    # 增量: 只重建有变化的库
-  python kg_build_db.py --db fywl_ics_quality  # 只扫指定库 (强制重建它)
-  python kg_build_db.py --rebuild          # 全量重建 (清所有签名 + 数据)
-  python kg_build_db.py --stats            # 看统计
-"""
-import os
-import sys
-import csv
-import socket
-import tempfile
-import time
-
-# 顶部 import (避免函数内延迟 import 触发 GIL 冲突: pymysql 的 C 扩展
-# 和 duckdb 在同进程内延迟加载时线程状态会打架)
-import duckdb
-import pymysql
-
-SCRIPTS = os.path.dirname(os.path.abspath(__file__))
-sys.path.insert(0, SCRIPTS)
-
-from common.paths import DATA_INDEX_DIR
-
-# MySQL 配置 (复用 mysql_mcp_server 的环境变量)
-MYSQL_HOST = os.environ.get('MYSQL_HOST', '10.54.6.9')
-MYSQL_PORT = int(os.environ.get('MYSQL_PORT', '3306'))
-MYSQL_USER = os.environ.get('MYSQL_USER', 'fywl_ics_test')
-MYSQL_PASSWORD = os.environ.get('MYSQL_PASSWORD', 'Fywl_ics_test56476')
-DB_PATH = str(DATA_INDEX_DIR / 'kg_db.duckdb')  # 数据层图谱独立文件 (不和 kg.duckdb 抢写锁)
-
-# 只扫业务库 (跳过 information_schema/mysql/sys/performance_schema)
-DB_NAME_PREFIX = 'fywl_ics_'
-
-
-def check_intranet():
-    """TCP 连通性校验 (同 mysql_mcp_server)。"""
-    try:
-        with socket.create_connection((MYSQL_HOST, MYSQL_PORT), timeout=3):
-            return True
-    except (socket.timeout, OSError):
-        return False
-
-
-def _connect_mysql():
-    return pymysql.connect(
-        host=MYSQL_HOST, port=MYSQL_PORT,
-        user=MYSQL_USER, password=MYSQL_PASSWORD,
-        charset='utf8mb4',
-        cursorclass=pymysql.cursors.DictCursor,
-        connect_timeout=5, read_timeout=30,
-    )
-
-
-def _discover_databases(cur):
-    """列出所有 fywl_ics_* 业务库。"""
-    cur.execute(
-        "SELECT schema_name FROM information_schema.schemata "
-        "WHERE schema_name LIKE %s ORDER BY schema_name",
-        (DB_NAME_PREFIX + '%',)
-    )
-    return [r['schema_name'] for r in cur.fetchall()]
-
-
-def _fetch_tables(cur, db):
-    """某库的所有表 + 注释。"""
-    cur.execute(
-        "SELECT table_name, table_comment "
-        "FROM information_schema.tables "
-        "WHERE table_schema = %s AND table_type = 'BASE TABLE' "
-        "ORDER BY table_name",
-        (db,)
-    )
-    return cur.fetchall()
-
-
-def _fetch_columns(cur, db):
-    """某库所有表的所有列。"""
-    cur.execute(
-        "SELECT table_name, column_name, data_type, column_type, "
-        "is_nullable, column_key, column_comment "
-        "FROM information_schema.columns "
-        "WHERE table_schema = %s ORDER BY table_name, ordinal_position",
-        (db,)
-    )
-    return cur.fetchall()
-
-
-def _init_db_tables(con):
-    """建数据层表 (IF NOT EXISTS)。独立文件, 自带 build_meta。"""
-    con.execute("""CREATE TABLE IF NOT EXISTS db_tables (
-        db TEXT, table_name TEXT, table_comment TEXT, column_count INTEGER,
-        PRIMARY KEY (db, table_name)
-    )""")
-    con.execute("""CREATE TABLE IF NOT EXISTS db_columns (
-        db TEXT, table_name TEXT, column_name TEXT, data_type TEXT,
-        column_type TEXT, is_nullable TEXT, column_key TEXT, column_comment TEXT
-    )""")
-    con.execute("""CREATE TABLE IF NOT EXISTS db_table_fields (
-        table_id TEXT, column_name TEXT
-    )""")
-    con.execute("CREATE TABLE IF NOT EXISTS build_meta(key TEXT PRIMARY KEY, value TEXT)")
-    con.execute("CREATE INDEX IF NOT EXISTS idx_dbcol_name ON db_columns(column_name)")
-    con.execute("CREATE INDEX IF NOT EXISTS idx_dbcol_table ON db_columns(table_name)")
-
-
-def _bulk_insert(con, table, columns, rows):
-    """批量插入: JSONL → duckdb COPY (万行级 0.01s)。
-    executemany 在 DuckDB 1.4 有性能 bug (每行 ~22ms), JSONL COPY 快 5000 倍。
-    JSON 天然处理逗号/引号/换行符, 无 CSV 转义问题。
-    rows: list of tuples, 顺序与 columns 对应。"""
-    if not rows:
-        return
-    import json as _json
-    tmp = tempfile.NamedTemporaryFile(suffix='.jsonl', delete=False,
-                                      mode='w', encoding='utf-8')
-    for r in rows:
-        obj = {columns[i]: ('' if r[i] is None else str(r[i]))
-               for i in range(len(columns))}
-        tmp.write(_json.dumps(obj, ensure_ascii=False) + '\n')
-    tmp.close()
-    try:
-        sql = "COPY %s FROM '%s' (FORMAT JSON)" % (
-            table, tmp.name.replace('\\', '/'))
-        con.execute(sql)
-    finally:
-        os.unlink(tmp.name)
-
-
-def _db_signature(mcur, db):
-    """算某库的 schema 签名 (表名列表 + 列总数)。
-    用于增量: 签名没变 = 该库 schema 没变, 跳过。变了才重建该库。
-    比 update_time 准 (后者数据变更也会刷新)。"""
-    mcur.execute(
-        "SELECT table_name FROM information_schema.tables "
-        "WHERE table_schema=%s AND table_type='BASE TABLE' ORDER BY table_name",
-        (db,)
-    )
-    tables = sorted(r['table_name'] for r in mcur.fetchall())
-    mcur.execute(
-        "SELECT COUNT(*) c FROM information_schema.columns WHERE table_schema=%s",
-        (db,)
-    )
-    col_cnt = mcur.fetchone()['c']
-    return '%d:%s' % (col_cnt, ','.join(tables))
-
-
-def _load_sig(con, db):
-    """读上次该库的签名。无记录返回 None。"""
-    row = con.execute(
-        "SELECT value FROM build_meta WHERE key=?", ['db_sig:' + db]
-    ).fetchone()
-    return row[0] if row else None
-
-
-def build(dbs=None, rebuild=False):
-    """扫 MySQL schema → 写 kg_db.duckdb。
-    增量策略: 库级签名对比, 只重建 schema 有变化的库。
-    rebuild=True: 忽略签名, 全量重建。
-    """
-    if not check_intranet():
-        print('⚠️ 无法连接内网 MySQL (%s:%s)。请连接公司内网后重试。' % (MYSQL_HOST, MYSQL_PORT))
-        return 0, 0
-
-    mcon = _connect_mysql()
-    try:
-        mcur = mcon.cursor()
-        all_dbs = dbs or _discover_databases(mcur)
-
-        # 预读 duckdb 现有签名 (增量对比用)
-        con = duckdb.connect(DB_PATH)
-        _init_db_tables(con)
-        if rebuild:
-            con.execute("DELETE FROM db_tables")
-            con.execute("DELETE FROM db_columns")
-            con.execute("DELETE FROM db_table_fields")
-            con.execute("DELETE FROM build_meta WHERE key LIKE 'db_sig:%'")
-
-        changed_dbs = []
-        skipped_dbs = []
-        for db in all_dbs:
-            sig = _db_signature(mcur, db)
-            old_sig = _load_sig(con, db)
-            if not rebuild and old_sig == sig:
-                skipped_dbs.append(db)
-                continue
-            changed_dbs.append((db, sig))
-
-        con.close()  # 释放, 后面重新开写
-
-        if skipped_dbs:
-            print('跳过 %d 个未变库: %s' % (len(skipped_dbs), ', '.join(skipped_dbs)))
-        if not changed_dbs:
-            print('✓ 全部库 schema 未变化, 无需重建。')
-            return 0, 0
-        print('重建 %d 个有变化的库: %s'
-              % (len(changed_dbs), ', '.join(d for d, _ in changed_dbs)))
-
-        all_tables = []
-        all_columns = []
-        for db, sig in changed_dbs:
-            tables = _fetch_tables(mcur, db)
-            cols = _fetch_columns(mcur, db)
-            col_count = {}
-            for c in cols:
-                t = c['table_name']
-                col_count[t] = col_count.get(t, 0) + 1
-            for t in tables:
-                tn = t['table_name']
-                all_tables.append((db, tn, t.get('table_comment', '') or '',
-                                   col_count.get(tn, 0)))
-            for c in cols:
-                all_columns.append((
-                    db, c['table_name'], c['column_name'], c['data_type'],
-                    c['column_type'], c['is_nullable'], c['column_key'],
-                    c.get('column_comment', '') or '',
-                ))
-            print('  %s: %d 表, %d 列' % (db, len(tables), len(cols)))
-    finally:
-        mcon.close()
-
-    # 写 kg_db.duckdb (只写变化的库的数据)
-    con = duckdb.connect(DB_PATH)
-    try:
-        _init_db_tables(con)
-        # 删掉变化库的旧数据 (库级增量: 只删该库的, 不动其它库)
-        for db, _ in changed_dbs:
-            con.execute("DELETE FROM db_tables WHERE db=?", [db])
-            con.execute("DELETE FROM db_columns WHERE db=?", [db])
-            con.execute("DELETE FROM db_table_fields WHERE table_id LIKE ?", ['dbt:%s:%%' % db])
-
-        if all_tables:
-            _bulk_insert(con, 'db_tables',
-                         ['db', 'table_name', 'table_comment', 'column_count'],
-                         all_tables)
-        if all_columns:
-            _bulk_insert(con, 'db_columns',
-                         ['db', 'table_name', 'column_name', 'data_type',
-                          'column_type', 'is_nullable', 'column_key', 'column_comment'],
-                         all_columns)
-            tf_rows = [('dbt:%s:%s' % (row[0], row[1]), row[2])
-                       for row in all_columns]
-            _bulk_insert(con, 'db_table_fields', ['table_id', 'column_name'], tf_rows)
-
-        # 记录每个变化库的新签名 + 总构建时间
-        for db, sig in changed_dbs:
-            con.execute(
-                "INSERT OR REPLACE INTO build_meta(key, value) VALUES (?,?)",
-                ['db_sig:' + db, sig]
-            )
-        con.execute(
-            "INSERT OR REPLACE INTO build_meta(key, value) VALUES (?,?)",
-            ['db_schema_last_build', str(time.time())]
-        )
-        print('\n增量导入完成: 重建 %d 库, %d 表, %d 列 → %s'
-              % (len(changed_dbs), len(all_tables), len(all_columns), os.path.basename(DB_PATH)))
-        return len(all_tables), len(all_columns)
-    finally:
-        con.close()
-
-
-def query_table_columns(table):
-    """查询某表的所有列 (给 eval_prd 等用)。返回 {column_name_lower}。"""
-    try:
-        con = duckdb.connect(DB_PATH, read_only=True)
-        rows = con.execute(
-            "SELECT DISTINCT column_name FROM db_columns WHERE table_name = ?",
-            [table]
-        ).fetchall()
-        con.close()
-        return {r[0].lower() for r in rows if r[0]}
-    except Exception:
-        return set()
-
-
-def stats():
-    """数据层图谱统计。"""
-    try:
-        con = duckdb.connect(DB_PATH, read_only=True)
-        t = con.execute("SELECT COUNT(*) FROM db_tables").fetchone()[0]
-        c = con.execute("SELECT COUNT(*) FROM db_columns").fetchone()[0]
-        by_db = con.execute(
-            "SELECT db, COUNT(*) FROM db_tables GROUP BY db ORDER BY COUNT(*) DESC"
-        ).fetchall()
-        con.close()
-        print('数据层图谱 (%s): %d 表 / %d 列' % (os.path.basename(DB_PATH), t, c))
-        for db, cnt in by_db:
-            print('  %s: %d 表' % (db, cnt))
-    except Exception as e:
-        print('统计失败: %s' % e)
-
-
-if __name__ == '__main__':
-    args = sys.argv[1:]
-    rebuild = '--rebuild' in args
-    db_filter = None
-    for i, a in enumerate(args):
-        if a == '--db' and i + 1 < len(args):
-            db_filter = [args[i + 1]]
-    if '--stats' in args:
-        stats()
-    else:
-        build(dbs=db_filter, rebuild=rebuild)
-        stats()
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+# v3.0 路径自举: 引导到 common/bootstrap, 统一 sys.path 逻辑
+import os as _o, sys as _s
+_f = _o.path.abspath(__file__)
+for _ in range(10):
+    _f = _o.path.dirname(_f)
+    _cp = _o.path.join(_f, 'foundation')
+    if _o.path.isfile(_o.path.join(_cp, 'bootstrap.py')):
+        break
+if _cp not in _s.path: _s.path.insert(0, _cp)
+from bootstrap import setup; setup()
+
+"""kg_build_db.py - 把 MySQL 数据库 schema 导入 kg.duckdb (数据层图谱)。
+
+和 kg_build.py 对称:
+  kg_build.py:   扫 data/code/*  → kg.duckdb (代码图谱: entities/fields/...)
+  kg_build_db.py: 扫 MySQL schema → kg.duckdb (数据图谱: DB_TABLE/db_columns/db_table_fields)
+
+产出 3 张新表 (复用现有 kg.duckdb, 不另建文件):
+  - db_tables:    表级实体 (db, table, table_comment, entity_count)
+  - db_columns:   列级真相 (db, table, column_name, data_type, is_nullable, column_comment)
+                  ← eval_prd.py 的 A1 第二层锚定读这张表
+  - db_table_fields: 表↔字段边 (table_id, column_name) 给图谱遍历用
+
+同时在 entities 表插入 DB_TABLE 类型实体 (让图谱认识"数据库表"这个概念),
+并在 edges 表连 DB_TABLE → 同名 ENDPOINT (若有), 做"改表影响哪些接口"的影响分析。
+
+安全: 只读 MySQL (INFORMATION_SCHEMA), 不碰业务数据行。
+
+增量策略 (库级签名对比, 不再全量重扫):
+  - 每个库算签名 (表名列表 + 列总数), 存 build_meta。
+  - 下次跑: 签名没变 = 该库 schema 没变 = 跳过。变了才重建该库。
+  - 比 MySQL update_time 准 (后者数据变更也会刷新, 不代表 schema 变)。
+
+用法:
+  python kg_build_db.py                    # 增量: 只重建有变化的库
+  python kg_build_db.py --db fywl_ics_quality  # 只扫指定库 (强制重建它)
+  python kg_build_db.py --rebuild          # 全量重建 (清所有签名 + 数据)
+  python kg_build_db.py --stats            # 看统计
+"""
+import os
+import sys
+import csv
+import socket
+import tempfile
+import time
+
+# 顶部 import (避免函数内延迟 import 触发 GIL 冲突: pymysql 的 C 扩展
+# 和 duckdb 在同进程内延迟加载时线程状态会打架)
+import duckdb
+import pymysql
+
+SCRIPTS = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))  # 子包→scripts/根
+sys.path.insert(0, SCRIPTS)
+
+from foundation.core.paths import DATA_INDEX_DIR
+
+# MySQL 配置 (复用 mysql_mcp_server 的环境变量)
+MYSQL_HOST = os.environ.get('MYSQL_HOST', '10.54.6.9')
+MYSQL_PORT = int(os.environ.get('MYSQL_PORT', '3306'))
+MYSQL_USER = os.environ.get('MYSQL_USER', 'fywl_ics_test')
+MYSQL_PASSWORD = os.environ.get('MYSQL_PASSWORD', 'Fywl_ics_test56476')
+DB_PATH = str(DATA_INDEX_DIR / 'kg_db.duckdb')  # 数据层图谱独立文件 (不和 kg.duckdb 抢写锁)
+
+# 只扫业务库 (跳过 information_schema/mysql/sys/performance_schema)
+DB_NAME_PREFIX = 'fywl_ics_'
+
+
+def check_intranet():
+    """TCP 连通性校验 (同 mysql_mcp_server)。"""
+    try:
+        with socket.create_connection((MYSQL_HOST, MYSQL_PORT), timeout=3):
+            return True
+    except (socket.timeout, OSError):
+        return False
+
+
+def _connect_mysql():
+    return pymysql.connect(
+        host=MYSQL_HOST, port=MYSQL_PORT,
+        user=MYSQL_USER, password=MYSQL_PASSWORD,
+        charset='utf8mb4',
+        cursorclass=pymysql.cursors.DictCursor,
+        connect_timeout=5, read_timeout=30,
+    )
+
+
+def _discover_databases(cur):
+    """列出所有 fywl_ics_* 业务库。"""
+    cur.execute(
+        "SELECT schema_name FROM information_schema.schemata "
+        "WHERE schema_name LIKE %s ORDER BY schema_name",
+        (DB_NAME_PREFIX + '%',)
+    )
+    return [r['schema_name'] for r in cur.fetchall()]
+
+
+def _fetch_tables(cur, db):
+    """某库的所有表 + 注释。"""
+    cur.execute(
+        "SELECT table_name, table_comment "
+        "FROM information_schema.tables "
+        "WHERE table_schema = %s AND table_type = 'BASE TABLE' "
+        "ORDER BY table_name",
+        (db,)
+    )
+    return cur.fetchall()
+
+
+def _fetch_columns(cur, db):
+    """某库所有表的所有列。"""
+    cur.execute(
+        "SELECT table_name, column_name, data_type, column_type, "
+        "is_nullable, column_key, column_comment "
+        "FROM information_schema.columns "
+        "WHERE table_schema = %s ORDER BY table_name, ordinal_position",
+        (db,)
+    )
+    return cur.fetchall()
+
+
+def _init_db_tables(con):
+    """建数据层表 (IF NOT EXISTS)。独立文件, 自带 build_meta。"""
+    con.execute("""CREATE TABLE IF NOT EXISTS db_tables (
+        db TEXT, table_name TEXT, table_comment TEXT, column_count INTEGER,
+        PRIMARY KEY (db, table_name)
+    )""")
+    con.execute("""CREATE TABLE IF NOT EXISTS db_columns (
+        db TEXT, table_name TEXT, column_name TEXT, data_type TEXT,
+        column_type TEXT, is_nullable TEXT, column_key TEXT, column_comment TEXT
+    )""")
+    con.execute("""CREATE TABLE IF NOT EXISTS db_table_fields (
+        table_id TEXT, column_name TEXT
+    )""")
+    con.execute("CREATE TABLE IF NOT EXISTS build_meta(key TEXT PRIMARY KEY, value TEXT)")
+    con.execute("CREATE INDEX IF NOT EXISTS idx_dbcol_name ON db_columns(column_name)")
+    con.execute("CREATE INDEX IF NOT EXISTS idx_dbcol_table ON db_columns(table_name)")
+
+
+def _bulk_insert(con, table, columns, rows):
+    """批量插入: JSONL → duckdb COPY (万行级 0.01s)。
+    executemany 在 DuckDB 1.4 有性能 bug (每行 ~22ms), JSONL COPY 快 5000 倍。
+    JSON 天然处理逗号/引号/换行符, 无 CSV 转义问题。
+    rows: list of tuples, 顺序与 columns 对应。"""
+    if not rows:
+        return
+    import json as _json
+    tmp = tempfile.NamedTemporaryFile(suffix='.jsonl', delete=False,
+                                      mode='w', encoding='utf-8')
+    for r in rows:
+        obj = {columns[i]: ('' if r[i] is None else str(r[i]))
+               for i in range(len(columns))}
+        tmp.write(_json.dumps(obj, ensure_ascii=False) + '\n')
+    tmp.close()
+    try:
+        sql = "COPY %s FROM '%s' (FORMAT JSON)" % (
+            table, tmp.name.replace('\\', '/'))
+        con.execute(sql)
+    finally:
+        os.unlink(tmp.name)
+
+
+def _db_signature(mcur, db):
+    """算某库的 schema 签名 (表名列表 + 列总数)。
+    用于增量: 签名没变 = 该库 schema 没变, 跳过。变了才重建该库。
+    比 update_time 准 (后者数据变更也会刷新)。"""
+    mcur.execute(
+        "SELECT table_name FROM information_schema.tables "
+        "WHERE table_schema=%s AND table_type='BASE TABLE' ORDER BY table_name",
+        (db,)
+    )
+    tables = sorted(r['table_name'] for r in mcur.fetchall())
+    mcur.execute(
+        "SELECT COUNT(*) c FROM information_schema.columns WHERE table_schema=%s",
+        (db,)
+    )
+    col_cnt = mcur.fetchone()['c']
+    return '%d:%s' % (col_cnt, ','.join(tables))
+
+
+def _load_sig(con, db):
+    """读上次该库的签名。无记录返回 None。"""
+    row = con.execute(
+        "SELECT value FROM build_meta WHERE key=?", ['db_sig:' + db]
+    ).fetchone()
+    return row[0] if row else None
+
+
+def build(dbs=None, rebuild=False):
+    """扫 MySQL schema → 写 kg_db.duckdb。
+    增量策略: 库级签名对比, 只重建 schema 有变化的库。
+    rebuild=True: 忽略签名, 全量重建。
+    """
+    if not check_intranet():
+        print('⚠️ 无法连接内网 MySQL (%s:%s)。请连接公司内网后重试。' % (MYSQL_HOST, MYSQL_PORT))
+        return 0, 0
+
+    mcon = _connect_mysql()
+    try:
+        mcur = mcon.cursor()
+        all_dbs = dbs or _discover_databases(mcur)
+
+        # 预读 duckdb 现有签名 (增量对比用)
+        con = duckdb.connect(DB_PATH)
+        _init_db_tables(con)
+        if rebuild:
+            con.execute("DELETE FROM db_tables")
+            con.execute("DELETE FROM db_columns")
+            con.execute("DELETE FROM db_table_fields")
+            con.execute("DELETE FROM build_meta WHERE key LIKE 'db_sig:%'")
+
+        changed_dbs = []
+        skipped_dbs = []
+        for db in all_dbs:
+            sig = _db_signature(mcur, db)
+            old_sig = _load_sig(con, db)
+            if not rebuild and old_sig == sig:
+                skipped_dbs.append(db)
+                continue
+            changed_dbs.append((db, sig))
+
+        con.close()  # 释放, 后面重新开写
+
+        if skipped_dbs:
+            print('跳过 %d 个未变库: %s' % (len(skipped_dbs), ', '.join(skipped_dbs)))
+        if not changed_dbs:
+            print('✓ 全部库 schema 未变化, 无需重建。')
+            return 0, 0
+        print('重建 %d 个有变化的库: %s'
+              % (len(changed_dbs), ', '.join(d for d, _ in changed_dbs)))
+
+        all_tables = []
+        all_columns = []
+        for db, sig in changed_dbs:
+            tables = _fetch_tables(mcur, db)
+            cols = _fetch_columns(mcur, db)
+            col_count = {}
+            for c in cols:
+                t = c['table_name']
+                col_count[t] = col_count.get(t, 0) + 1
+            for t in tables:
+                tn = t['table_name']
+                all_tables.append((db, tn, t.get('table_comment', '') or '',
+                                   col_count.get(tn, 0)))
+            for c in cols:
+                all_columns.append((
+                    db, c['table_name'], c['column_name'], c['data_type'],
+                    c['column_type'], c['is_nullable'], c['column_key'],
+                    c.get('column_comment', '') or '',
+                ))
+            print('  %s: %d 表, %d 列' % (db, len(tables), len(cols)))
+    finally:
+        mcon.close()
+
+    # 写 kg_db.duckdb (只写变化的库的数据)
+    con = duckdb.connect(DB_PATH)
+    try:
+        _init_db_tables(con)
+        # 删掉变化库的旧数据 (库级增量: 只删该库的, 不动其它库)
+        for db, _ in changed_dbs:
+            con.execute("DELETE FROM db_tables WHERE db=?", [db])
+            con.execute("DELETE FROM db_columns WHERE db=?", [db])
+            con.execute("DELETE FROM db_table_fields WHERE table_id LIKE ?", ['dbt:%s:%%' % db])
+
+        if all_tables:
+            _bulk_insert(con, 'db_tables',
+                         ['db', 'table_name', 'table_comment', 'column_count'],
+                         all_tables)
+        if all_columns:
+            _bulk_insert(con, 'db_columns',
+                         ['db', 'table_name', 'column_name', 'data_type',
+                          'column_type', 'is_nullable', 'column_key', 'column_comment'],
+                         all_columns)
+            tf_rows = [('dbt:%s:%s' % (row[0], row[1]), row[2])
+                       for row in all_columns]
+            _bulk_insert(con, 'db_table_fields', ['table_id', 'column_name'], tf_rows)
+
+        # 记录每个变化库的新签名 + 总构建时间
+        for db, sig in changed_dbs:
+            con.execute(
+                "INSERT OR REPLACE INTO build_meta(key, value) VALUES (?,?)",
+                ['db_sig:' + db, sig]
+            )
+        con.execute(
+            "INSERT OR REPLACE INTO build_meta(key, value) VALUES (?,?)",
+            ['db_schema_last_build', str(time.time())]
+        )
+        print('\n增量导入完成: 重建 %d 库, %d 表, %d 列 → %s'
+              % (len(changed_dbs), len(all_tables), len(all_columns), os.path.basename(DB_PATH)))
+        return len(all_tables), len(all_columns)
+    finally:
+        con.close()
+
+
+def query_table_columns(table):
+    """查询某表的所有列 (给 eval_prd 等用)。返回 {column_name_lower}。"""
+    try:
+        con = duckdb.connect(DB_PATH, read_only=True)
+        rows = con.execute(
+            "SELECT DISTINCT column_name FROM db_columns WHERE table_name = ?",
+            [table]
+        ).fetchall()
+        con.close()
+        return {r[0].lower() for r in rows if r[0]}
+    except Exception:
+        return set()
+
+
+def stats():
+    """数据层图谱统计。"""
+    try:
+        con = duckdb.connect(DB_PATH, read_only=True)
+        t = con.execute("SELECT COUNT(*) FROM db_tables").fetchone()[0]
+        c = con.execute("SELECT COUNT(*) FROM db_columns").fetchone()[0]
+        by_db = con.execute(
+            "SELECT db, COUNT(*) FROM db_tables GROUP BY db ORDER BY COUNT(*) DESC"
+        ).fetchall()
+        con.close()
+        print('数据层图谱 (%s): %d 表 / %d 列' % (os.path.basename(DB_PATH), t, c))
+        for db, cnt in by_db:
+            print('  %s: %d 表' % (db, cnt))
+    except Exception as e:
+        print('统计失败: %s' % e)
+
+
+if __name__ == '__main__':
+    args = sys.argv[1:]
+    rebuild = '--rebuild' in args
+    db_filter = None
+    for i, a in enumerate(args):
+        if a == '--db' and i + 1 < len(args):
+            db_filter = [args[i + 1]]
+    if '--stats' in args:
+        stats()
+    else:
+        build(dbs=db_filter, rebuild=rebuild)
+        stats()