postgresql 克隆不带pg_dump的Postgres模式

nbewdwxp  于 2023-04-20  发布在  PostgreSQL
关注(0)|答案(1)|浏览(110)

我需要一种方法来系统地克隆Postgres模式到另一个数据库,没有数据,通过SQL查询。我不能使用pg_dump(因为它很复杂)。解决方案如下:
1.连接到要克隆的数据库。
1.获取模式
1.连接到新数据库(我们要克隆到的位置)
1.执行架构脚本
是否可以通过sql查询或类似的方式获取pg_dump用于转储模式的查询?例如,通过sql编辑器运行pg_dump?

更新:

对于克隆模式,我指的是特定的schema(例如public)及其内容。

jk9hmnmh

jk9hmnmh1#

这段Python 3代码将使用psycopg包将源数据库的模式和表复制到目标数据库。
如果你想复制空值、键或索引,你可能需要进一步细化由这段代码生成的CREATE TABLE查询。你可以很容易地将这些CREATE TABLE脚本保存到文件中,以供单独使用或执行。
底部还有一个小块,如果您希望这样做,它将复制表数据。

import psycopg

source = {
    'host': 'source_host', 
    'port': '5432', 
    'dbname': 'source_database',
    'user': 'postgres', 
    'password': 'postgres'
    }

target = {
    'host': 'target_host', 
    'port': '5432', 
    'dbname': 'target_database',
    'user': 'postgres', 
    'password': 'postgres'
    }

sourcedb = psycopg.connect(**source)
targetdb = psycopg.connect(**target)

with sourcedb:
    with targetdb:
        targetdb.autocommit = False
        with sourcedb.cursor() as sourceCursor:
            with targetdb.cursor() as targetCursor:
                # get list of schemas in database
                query = f"""SELECT      schema_name 
                            FROM        information_schema.schemata
                            WHERE           schema_name not in ('pg_catalog',
                                                                'pg_toast', 
                                                                'information_schema')
                            ORDER BY    schema_name ASC;"""
                sourceCursor.execute(query)
                schemas = [tupleRow[0] for tupleRow in sourceCursor.fetchall()]

                for schema in schemas:
                    # get list of tables in schema
                    query = f"""SELECT      table_name 
                                FROM        information_schema.tables 
                                WHERE           table_schema = '{schema}' 
                                            AND table_type = 'BASE TABLE'
                                ORDER BY    table_name ASC;"""
                    sourceCursor.execute(query)
                    schemaTables = [tupleRow[0] for tupleRow in sourceCursor.fetchall()]

                    # create schema on target_database if not exists
                    query = f'CREATE SCHEMA IF NOT EXISTS "{schema}";'
                    targetCursor.execute(query)

                    for table in schemaTables:
                        # generate a 'CREATE TABLE' query from source 
                        # table's column names and data types
                        query = f"""SELECT       column_name
                                                ,data_type
                                    FROM        INFORMATION_SCHEMA.COLUMNS 
                                    WHERE           table_name = '{table}' 
                                                AND table_schema = '{schema}'
                                    ORDER BY    ordinal_position ASC;"""
                        sourceCursor.execute(query)
                        columns = ['"{}" {}\n'.format(i[0], i[1]) for i in sourceCursor.fetchall()]
                        columnsString = ' ,'.join(columns)
                        createTableQuery = f'CREATE TABLE IF NOT EXISTS "{schema}"."{table}"\n( {columnsString});\n\n'

                        # run the CREATE TABLE on the target_database
                        targetCursor.execute(createTableQuery)

                        # uncomment this block to copy the data inside the tables
                        # from source_database to target_database
                        
                        # with sourceCursor.copy(f"COPY {schema}.{table} TO STDOUT;") as sourceCopy:
                        #     with targetCursor.copy(f"COPY {schema}.{table} FROM STDIN;") as targetWrite:
                        #         for data in sourceCopy:
                        #             targetWrite.write(data)

        targetdb.commit()

相关问题