python 学习笔记
来源:互联网 发布:网络兼职广告语 编辑:程序博客网 时间:2024/06/05 06:25
#!/usr/bin/python#-*-coding:utf-8-*-#db.py''' 设计数据库接口 以方便调用者使用 希望调用者可以通过:from transwarp import dbdb.create_engine(user='root',password='123456',database='test',host='127.0.0.1',port=3306)然后直接操作sql语句 users=db.select('select * from user')返回一个list 其中包含了所有的user信息。其中每一个select和update等 都隐含了自动打开和关闭数据库 这样上层调用就完全不需要关心数据库底层链接在一个数据库中执行多条sql语句 可以用with语句实现with db.connection():db.select('....')db.update('....')db.select('....')同样如果在一个数据库事务中执行多个SQL语句 也可以用with实现with db.transactions():db.select('....')db.update('....')db.select('....')'''import time,uuid,functools,threading,logging#Dict object : 重写dict 让其可以通过访问属性的方式访问对应的value'''---------------------------------------------以下是Dict类的定义-----------------------------------------------------'''class Dict(dict):'''一下是docttest.testmod()会调用作为测试的内容 也就是简单的unittest 单元测试simple dict but spport access as x.y style>>> d1 = Dict() >>> d1['x'] = 100 >>> d1.x 100 >>> d1.y = 200 >>> d1['y'] 200 >>> d2 = Dict(a=1, b=2, c='3') >>> d2.c '3' >>> d2['empty'] Traceback (most recent call last): ... KeyError: 'empty' >>> d2.empty Traceback (most recent call last): ... AttributeError: 'Dict' object has no attribute 'empty' >>> d3 = Dict(('a', 'b', 'c'), (1, 2, 3)) >>> d3.a 1 >>> d3.b 2 >>> d3.c 3''''''@time 2015-02-10@method __init__ 相当于其他语言中的构造函数zip()将两个list糅合在一起 例如:x=[1,2,3,4,5]y=[6,7,8,9,10]zip(x,y)-->就得到了[(1,6),(2,7),(3,8),(4,9),(5,10)]'''def __init__(self,names=(),values=(),**kw):super(Dict,self).__init__(**kw) #调用父类的构造方法for k,v in zip(names,values):self[k]=v'''@time 2015-02-10@method __getattr__ 相当于新增加的get方法如果对象调用的属性不存在的时候 解释器就会尝试从__getattr__()方法获得属性的值。'''def __getattr__(self,key):try:return self[key]except KeyError: raise AttributeError(r"'Dict' object has no attribute '%s'" % key)'''@time 2015-02-10@method __setattr__ 相当于新增加set方法'''def __setattr__(self,key,values):self[key]=values'''---------------------------------------------以上是Dict类的定义-----------------------------------------------------''''''@time 2015-02-10@method next_id() uuid4() make a random UUID 得到一个随机的UUID如果没有传入参数根据系统当前时间15位和一个随机得到的UUID 填充3个0 组成一个长度为50的字符串'''def next_id(t=None):'''Return next id as 50-char string args: t:unix timestamp,default to None and using time.time().'''if t is None:t=time.time()return '%015d%s000' %(int(t*1000),uuid.uuid4.hex)'''@time 2015-02-10@method _profiling 记录sql 的运行状态'''def _profiling(start,sql=''):'''单下划线开头的方法名或者属性 不会再 from moduleName import * 中被导入 也就是说只有本模块中可以访问'''t=time.time()-startif t>0.1:logging.warning('[PROFILING] [DB] %s:%s' %(t,sql))else:logging.info('[PROFILING] [DB] %s:%s' %(t,sql))class DBError(Exception):passclass MultiColumnsError(DBError):pass#global engine object 保存着mysql数据库的连接engine=Noneclass _Engine(object):def __init__(self,connect):self._connect=connectdef connect(self):return self._connect() #这里传入的参数是一个函数 将函数调用后 将函数的结果返回def create_engine(user,password,database,host='127.0.0.1',port=3306,**kw):import mysql.connector #导入mysql模块global engine #global关键字 说明这个变量在外部定义了 这是一个全局变量if engine is not None:raise DBError('Engine is already initialized') #如果连接已经存在表示连接重复 则抛出一个数据库异常params=dict(user=user,password=password,database=database,host=host,port=port) #保存了数据库的链接信息 defualts=dict(use_unicode=True,charset='utf8',collation='utf8_general_ci',autocommit=False)#保存了链接的设置 编码 等等for k,v in defualts.iteritems():#将defaults和kw中的键值对保存到params中 如果有一个key两边都存在那么保存kw的params[k]=kw.pop(k,v) #pop函数会将key为k的键值对删除并且返回k对应的value 如果k在kw中不存在 那么将会返回vparams.update(kw)params['buffered']=Trueengine=_Engine(lambda:mysql.connector.connect(**params))#在这里(lambda:mysql.connector.connect(**params))返回的是一个函数而不是一个connection对象#test connection....logging.info('Init mysql engine <%s> ok' % hex(id(engine)))'''===================以上通过engine这个全局变量就可以获得一个数据库链接,重复链接抛异常=============================''''''对数据库连接以及最基本的操作进行了封装'''class _LasyConnection(object):def __init__(self):self.connection=Nonedef cursor(self):if self.connection is None:connection=engine.connect()logging.info('open connection <%s>...' % hex(id(connection)))self.connection = connectionreturn self.connection.cursor()def commit(self):self.connection.commit()def rollback(self):#print '================='#print self.connectionself.connection.rollback()def cleanup(self):if self.connection:connection = self.connectionself.connection=Nonelogging.info('colse connection <%s>...' %hex(id(connection)))connection.close()'''接下来解决对于不同的线程数据库链接应该是不一样的 于是创建一个变量 是一个threadlocal 对象'''class _DbCtx(threading.local):'''Thread local object that holds connection info'''def __init__(self):self.connection=Noneself.transactions = 0def is_init(self):return not self.connection is None #判断是否已经进行了初始化def init(self):logging.info('open lazy connection...')self.connection=_LasyConnection()#打开一个数据库链接#print threading.current_thread().name#print id(self.connection)self.transactions=0def cleanup(self):self.connection.cleanup()self.connection=Nonedef cursor(self):'''return cursor'''return self.connection.cursor()#由于它继承threading.local 是一个threadlocal对象 所以它对于每一个线程都是不一样的。#所以当需要数据库连接的时候就使用它来创建 _db_ctx=_DbCtx()'''=====================================以上通过_db_ctx就可以打开和关闭链接=============================================='''#通过with语句让数据库链接可以自动创建和关闭'''with 语句: with 后面的语句会返回 _ConnectionCtx 对象 然后调用这个对象的 __enter__方法得到返回值 返回值赋值给as后面的变量 然后执行 with下面的语句 执行完毕后 调用那个对象的 __exit__()方法'''class _ConnectionCtx(object):'''_ConnectionCtx object that can open and close connection context._ConnectionCtx object can nested and only the most outer connection has effectwith connection():passwith connectino():pass'''def __enter__(self):global _db_ctxself.should_cleanup=Falseif not _db_ctx.is_init():_db_ctx.init()self.should_cleanup=Truereturn selfdef __exit__(self,type,value,trace):global _db_ctxif self.should_cleanup:_db_ctx.cleanup()def connection():'''return _ConnectionCtx object that can be used by 'with' statement :with connection:pass'''return _ConnectionCtx()#采用装饰器的方法 让其能够进行共用同一个数据库连接def with_connection(func):'''Decorater for reuse connection@with_connectiondef foo(*args,**kw):f1()f2()f3()'''#@functools.wraps(func)def wrapper(*args,**kw):with connection():return func(*args,**kw)return wrapper#================================================以下是事务处理======================================================='''知识点与数据库连接知识点相同'''class _TransactionCtx(object):'''_transactionCtx object that can handle transactionswith _transactionCtx():pass'''def __enter__(self):global _db_ctx#print '++++++++++++++++++++++++++++++++++++++++++++++++++'self.should_close_conn = Falseif not _db_ctx.is_init():#needs open a connection first:_db_ctx.init()self.should_close_conn=True_db_ctx.transactions=_db_ctx.transactions+1logging.info('begin transactions....' if _db_ctx.transactions==1 else 'join current transactions')#print '===========',_db_ctx.is_init()return selfdef __exit__(self,type,value,trace):global _db_ctx_db_ctx.transactions=_db_ctx.transactions-1try:if _db_ctx.transactions==0:#print '----------------type:',type#print '----------------value:',value#print '----------------trace:',traceif type is None:self.commit()else:self.rollback()finally:if self.should_close_conn:_db_ctx.cleanup()def commit(self):global _db_ctxlogging.info('commit transaction....')try:_db_ctx.connection.commit()logging.info('commit ok.')except:logging.warning('commit failed.try rollback...')#print dir(_db_ctx.connection)_db_ctx.connection.rollback()logging.warning('rollback ok')raisedef rollback(self):global _db_ctxlogging.warning('rollback transaction....')_db_ctx.connection.rollback()logging.info('rollback ok...')def transaction(): ''' Create a transaction object so can use with statement: with transaction(): pass >>> def update_profile(id, name, rollback): ... u = dict(id=id, name=name, email='%s@test.org' % name, passwd=name, last_modified=time.time()) ... insert('user', **u) ... r = update('update user set passwd=? where id=?', name.upper(), id) ... if rollback: ... raise StandardError('will cause rollback...') >>> with transaction(): ... update_profile(900301, 'Python', False) >>> select_one('select * from user where id=?', 900301).name u'Python' >>> with transaction(): ... update_profile(900302, 'Ruby', True) Traceback (most recent call last): ... StandardError: will cause rollback... >>> select('select * from user where id=?', 900302) [] ''' return _TransactionCtx()def with_transaction(func):''' A decorator that makes function around transaction. >>> @with_transaction ... def update_profile(id, name, rollback): ... u = dict(id=id, name=name, email='%s@test.org' % name, passwd=name, last_modified=time.time()) ... insert('user', **u) ... r = update('update user set passwd=? where id=?', name.upper(), id) ... if rollback: ... raise StandardError('will cause rollback...') >>> update_profile(8080, 'Julia', False) >>> select_one('select * from user where id=?', 8080).passwd u'JULIA' >>> update_profile(9090, 'Robert', True) Traceback (most recent call last): ... StandardError: will cause rollback... >>> select('select * from user where id=?', 9090) [] '''@functools.wraps(func)def wrapper(*args,**kw):_start=time.time()with transaction():return func(*args,**kw)_profiling(_start)return wrapper#===============================以上是事务处理======================================================================def _select(sql,first,*args):'execute select SQL and return unique result or list results'global _db_ctxcursor = Nonesql = sql.replace('?','%s')logging.info('SQL:%s,ARGS:%s' %(sql,args))try:cursor = _db_ctx.connection.cursor()cursor.execute(sql,args)if cursor.description:names=[x[0] for x in cursor.description] #返回结果集的描述 x[0] 是描述结果集的列名if first:values=cursor.fetchone()if not values:return Nonereturn Dict(names,values)return [Dict(names,x) for x in cursor.fetchall()]finally:if cursor:cursor.close()@with_connectiondef select_one(sql,*args):''' Execute select SQL and expected one result. If no result found, return None. If multiple results found, the first one returned. >>> u1 = dict(id=100, name='Alice', email='alice@test.org', passwd='ABC-12345', last_modified=time.time()) >>> u2 = dict(id=101, name='Sarah', email='sarah@test.org', passwd='ABC-12345', last_modified=time.time()) >>> insert('user', **u1) 1 >>> insert('user', **u2) 1 >>> u = select_one('select * from user where id=?', 100) >>> u.name u'Alice' >>> select_one('select * from user where email=?', 'abc@email.com') >>> u2 = select_one('select * from user where passwd=? order by email', 'ABC-12345') >>> u2.name u'Alice' '''return _select(sql,True,*args)@with_connectiondef select_int(sql, *args):''' Execute select SQL and expected one int and only one int result. >>> n = update('delete from user') >>> u1 = dict(id=96900, name='Ada', email='ada@test.org', passwd='A-12345', last_modified=time.time()) >>> u2 = dict(id=96901, name='Adam', email='adam@test.org', passwd='A-12345', last_modified=time.time()) >>> insert('user', **u1) 1 >>> insert('user', **u2) 1 >>> select_int('select count(*) from user') 2 >>> select_int('select count(*) from user where email=?', 'ada@test.org') 1 >>> select_int('select count(*) from user where email=?', 'notexist@test.org') 0 >>> select_int('select id from user where email=?', 'ada@test.org') 96900 >>> select_int('select id, name from user where email=?', 'ada@test.org') Traceback (most recent call last): ... MultiColumnsError: Expect only one column. '''d = _select(sql, True, *args)if len(d)!=1:raise MultiColumnsError('Expect only one column.')return d.values()[0]@with_connectiondef select(sql, *args):''' Execute select SQL and return list or empty list if no result. >>> u1 = dict(id=200, name='Wall.E', email='wall.e@test.org', passwd='back-to-earth', last_modified=time.time()) >>> u2 = dict(id=201, name='Eva', email='eva@test.org', passwd='back-to-earth', last_modified=time.time()) >>> insert('user', **u1) 1 >>> insert('user', **u2) 1 >>> L = select('select * from user where id=?', 900900900) >>> L [] >>> L = select('select * from user where id=?', 200) >>> L[0].email u'wall.e@test.org' >>> L = select('select * from user where passwd=? order by id desc', 'back-to-earth') >>> L[0].name u'Eva' >>> L[1].name u'Wall.E' '''return _select(sql, False, *args)@with_connectiondef _update(sql, *args):#print 'a++++++++++++++++++++++++++++'global _db_ctxcursor = Nonesql = sql.replace('?', '%s')logging.info('SQL: %s, ARGS: %s' % (sql, args))try:cursor = _db_ctx.connection.cursor()cursor.execute(sql, args)r = cursor.rowcountif _db_ctx.transactions==0:# no transaction enviroment:logging.info('auto commit')_db_ctx.connection.commit()return rfinally:if cursor:cursor.close()def insert(table, **kw):''' Execute insert SQL. >>> u1 = dict(id=2000, name='Bob', email='bob@test.org', passwd='bobobob', last_modified=time.time()) >>> insert('user', **u1) 1 >>> u2 = select_one('select * from user where id=?', 2000) >>> u2.name u'Bob' >>> insert('user', **u2) Traceback (most recent call last): ... IntegrityError: 1062 (23000): Duplicate entry '2000' for key 'PRIMARY' '''#print '======================='cols, args = zip(*kw.iteritems())sql = 'insert into `%s` (%s) values (%s)' % (table, ','.join(['`%s`' % col for col in cols]), ','.join(['?' for i in range(len(cols))]))return _update(sql, *args)def update(sql, *args):r''' Execute update SQL. >>> u1 = dict(id=1000, name='Michael', email='michael@test.org', passwd='123456', last_modified=time.time()) >>> insert('user', **u1) 1 >>> u2 = select_one('select * from user where id=?', 1000) >>> u2.email u'michael@test.org' >>> u2.passwd u'123456' >>> update('update user set email=?, passwd=? where id=?', 'michael@example.org', '654321', 1000) 1 >>> u3 = select_one('select * from user where id=?', 1000) >>> u3.email u'michael@example.org' >>> u3.passwd u'654321' >>> update('update user set passwd=? where id=?', '***', '123\' or id=\'456') 0 '''#print sql,argsreturn _update(sql, *args)'''@with_connectiondef a():global _db_ctxprint _db_ctx.is_init()'''if __name__ == '__main__':logging.basicConfig(level=logging.DEBUG)#Dict()#create_engine('root','123456','pythonstudy')#print engine.connect()#create_engine('root','123456','pythonstudy')''''import threadingd1=threading.Thread(target=_db_ctx.init)d2=threading.Thread(target=_db_ctx.init)d1.start()d2.start()d1.join()d2.join()这样测试可以看到每一个线程的数据库连接的id都不是一样的 可以知道每一个线程拥有不同的数据库链接'''create_engine('root', '123456', 'pythonstudy')'''a()print _db_ctx''''''with transaction():print 'dd'u1=dict(id=900301,name='python',email='python@test.org' ,passwd='python',last_modified=time.time())insert('user',**u1)print 'hellp''''update('drop table if exists user')update('create table user (id int primary key, name text, email text, passwd text, last_modified real)')import doctestdoctest.testmod()
初始源代码来自:<a target=_blank href="https://github.com/michaelliao/awesome-python-webapp/tree/day-01" target="_blank" style="color: rgb(0, 136, 204); text-decoration: none; font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif; font-size: 14px; line-height: 20px;">https://github.com/michaelliao/awesome-python-webapp/</a>
1 0
- 【python】:python学习笔记
- Python学习笔记-Python起步
- Python学习笔记-Python基础
- Python学习笔记-Python对象
- Python基本概念--Python学习笔记
- Python学习笔记:Python函数
- python学习笔记-python安装
- Python学习笔记--Python基础
- 【Python】python基础学习笔记
- python学习笔记3
- Python学习笔记
- Boost.Python学习笔记
- python学习笔记1
- python学习笔记2
- python学习笔记(2)
- Python学习笔记 Module
- Python学习笔记.
- python学习笔记(1)
- JNI
- 菜单项MenuListener。。其实就是这种方法替换了OnOpetionsSelectedItem方法
- AppCan获得B轮1亿元融资并宣布移动引擎开源
- php输出格式化的那点事
- CSS 文字跳动
- python 学习笔记
- <script>标签的问题
- 【设计模式】Java设计模式第二课之单例模式
- JDBC:元数据 && 获取插入记录的主键值 && _JDBC_处理 Blob
- 在Android上怎样实现JAVA和JS交互呢
- 【AppCan年会纪实】妹子奖金海岛游、奖奖奖拿拿拿
- (原创)尽管少写那么多代码,但省下来的时间又在哪里呢
- jBPM工作流应用
- CocurrentHashMap和Hashtable的区别