###使用SQLalchemy操作数据库
####一、导库
from sqlalchemy.ext.declarative import declarative_base
from sqlalchemy import Column, Integer, String, ForeignKey, UniqueConstraint, Index
from sqlalchemy.orm import sessionmaker, relationship
from sqlalchemy import create_engine
####二、 连接MySQL
engine = create_engine("mysql+pymysql://root:123@127.0.0.1:3306/db3?charset=utf8", max_overflow=5)
Base = declarative_base()
'''
create table users(
id int auto_increment primary key,
name varchar(32) not null default '',
extra varchar(32) not null default ''
)
'''
'''
Usertype
id title xxoo
1 普通用户
row.xxoo : 多条记录对象
'''
class UserType(Base):
__tablename__ = 'usertype' ### 表名
id = Column(Integer, autoincrement=True, primary_key=True)
title = Column(String(32), nullable=False, server_default='')
'''
users
id name extra type_id
1 zekai nb 1
usertype = releationship('Usertype')
row.usertype
'''
class Users(Base):
__tablename__ = 'users'
id = Column(Integer, autoincrement=True, primary_key=True)
name = Column(String(32), nullable=False, server_default='')
extra = Column(String(32), nullable=False, server_default='')
type_id = Column(Integer, ForeignKey(UserType.id))
usertype = relationship('UserType', backref='xxoo')
__table_args__ = (
# UniqueConstraint('id', 'name', name='uix_id_name'), ### 联合唯一索引
# Index('ix_name_extra', 'name', 'extra'), ### 组合索引
)
def drop_db():
Base.metadata.drop_all(engine)
def create_db():
## 会将当前执行文件中所有继承自Base类的类,生成表
Base.metad。ata.create_all(engine)
# drop_db()
create_db()
####操作表中的数据
Session = sessionmaker(bind=engine)
session = Session()
#### 增加一条数据
# obj = UserType(title='普通用户')
# session.add(obj)
#### 添加多条数据
# session.add_all([
# UserType(title='VIP用户'),
# UserType(title='VIP中P用户'),
# UserType(title='SVIP用户'),
# UserType(title='黑金用户')
# ])
### 查询
#### 查询全部 返回的是一个列表, 列表中套对象
# res = session.query(UserType).all()
# for row in res:
# print(row.id, row.name)
### 查询一条数据
# res = session.query(UserType).first()
# print(res)
#### where条件
# res = session.query(UserType).filter(UserType.name=='VIP用户', UserType.id==2).all()
# for row in res:
# print(row.id, row.name)
# print(res[0].name, res[0].id)
# res = session.query(UserType).filter_by(name='VIP用户').all()
# print(res)
#### 删除:
# session.query(UserType).filter(UserType.id>3).delete()
#### 修改
# session.query(UserType).filter(UserType.id == 3).update({"name" : "SVIP用户"})
### MySQL 高级查询操作
#### 通配符 分组 分页 排序 between and in not in
### between ... and ...
# res = session.query(UserType).filter(UserType.id.between(1,3)).all()
# for row in res:
# print(row.id, row.title)
### in 操作 bool_
# res = session.query(UserType).filter(UserType.id.in_([1,3,4])).all()
# print(res)
#### not in
# select * from Usertype
# res = session.query(UserType).filter(~UserType.id.in_([1,3,4])).all()
# for row in res:
# print(row.id, row.title)
#
# session.query(UserType).filter(Users.id.in_(session.query(Users.id).filter_by(name='eric'))).all()
# res = session.query(UserType.title).filter(~UserType.id.in_([1,3,4])).all()
# for row in res:
# print( row.title)
from sqlalchemy import and_, or_
#### 通配符
# ret = session.query(UserType).all()
from sqlalchemy.sql import func
# res = session.query(
# Users.type_id,
# func.max(Users.id),
# func.min(Users.id)).group_by(Users.type_id).all()
#
# print(res)
ret = session.query(
func.max(Users.id),
func.min(Users.id)).group_by(Users.type_id).having(func.min(Users.id) >2).all()
### left join
###1. 查询某一个用户的用户类型
#### 第一种方法:
# res = session.query(Users,UserType).join(UserType, isouter=True).all()
# # print(res)
# for row in res:
# print(row[0].id, row[0].name, row[1].title)
#### 第二种方法
# res = session.query(Users).all()
# for row in res:
# print(row.id, row.name, row.extra, row.usertype.title)
#### 2. 某一个类型下面的用户
### 第一种方法
# res = session.query(UserType).all()
# for row in res:
# print(row.id, row.title, session.query(Users).filter(Users.type_id == row.id).all() )
#### 第二种方法
# res = session.query(UserType).all()
#
# for row in res:
# print(row.id, row.title, row.xxoo)
# 必须提交任务后才能修改数据(重要)
session.commit()
session.close()