大型系统、海量数据肯定涉及到分库分表这些提高效率的手段。由于sqlalchemy的orm思想是一张表对应一个对象，那么当我们有N张相同结构只是表名有区别的分表，sqlalchemy orm怎样处理呢。比如有如下表：

CREATE TABLE `goods_desc_0` ( `goods_id` bigint(20) unsigned NOT NULL DEFAULT '0' COMMENT '商品id', `goods_desc` text NOT NULL COMMENT '商品详细描述', PRIMARY KEY (`goods_id`) ) ENGINE=InnoDB DEFAULT CHARSET=utf8 ROW_FORMAT=COMPACT COMMENT='商品信息详情' CREATE TABLE `goods_desc_1` ( `goods_id` bigint(20) unsigned NOT NULL DEFAULT '0' COMMENT '商品id', `goods_desc` text NOT NULL COMMENT '商品详细描述', PRIMARY KEY (`goods_id`) ) ENGINE=InnoDB DEFAULT CHARSET=utf8 ROW_FORMAT=COMPACT COMMENT='商品信息详情'

class GoodsDesc(db.Model): __tablename__ = 'goods_desc' goods_id = db.Column(db.Integer, primary_key=True) goods_desc = db.Column(db.Text, default=None) def __str__(self): return "GoodsDesc => { \ goods_id:%d, goods_desc:'%s'}" % ( self.goods_id, self.goods_desc) __repr__ = __str__ # 代码示例 goods_id = 101 table_index = goods_id%2 table_name = 'goods_desc_%d' % table_index GoodsDesc.__table__.name = table_name gd = GoodsDesc.query.filter(GoodsDesc.goods_id == goods_id).first() # 这样写虽然也能工作，但是是非常危险的，因为GoodsDesc.__table__是静态全局变量， # 而且不是web程序request级别的，是app context的，非常不安全。

经过N次Google，终于找到解决方案，而且代码写起来也很优雅。代码如下：

class GoodsDesc(object): _mapper = {} @staticmethod def model(goods_id): table_index = goods_id%100 class_name = 'GoodsDesc_%d' % table_index ModelClass = GoodsDesc._mapper.get(class_name, None) if ModelClass is None: ModelClass = type(class_name, (db.Model,), { '__module__' : __name__, '__name__' : class_name, '__tablename__' : 'goods_desc_%d' % table_index, 'goods_id' : db.Column(db.Integer, primary_key=True), 'goods_desc' : db.Column(db.Text, default=None), }) GoodsDesc._mapper[class_name] = ModelClass cls = ModelClass() cls.goods_id = goods_id return cls # 外部代码调用如例如下： # ----------------------- # 新增插入 gdm = GoodsDesc.model(goods_id) gdm.goods_desc = 'desc' db.session.add(gd) # 查询 gdm = GoodsDesc.model(goods_id) gd = gdm.query.filter_by(goods_id=goods_id).first()