综述
多线程是程序设计中的一个重要方面,尤其是在服务器Deamon程序方面。无论何种系统,线程调度的开销都比传统的进程要快得多。
Python可以方便地支持多线程。可以快速创建线程、互斥锁、信号量等等元素,支持线程读写同步互斥。美中不足的是,Python的运行在Python虚拟机上,创建的多线程可能是虚拟的线程,需要由Python虚拟机来轮询调度,这大大降低了Python多线程的可用性。希望高版本的Python可以解决这个问题,发挥多CPU的最大效率。
网上有些朋友说要获得真正多CPU的好处,有两种方法:
1.可以创建多个进程而不是线程,进程数和cpu一样多。 2.使用Jython 或 IronPython,可以得到真正的多线程。
闲话少说,下面看看Python如何建立线程
Python线程创建
使用threading模块的 Thread类
类接口如下
class Thread( group=None, target=None, name=None, args=(), kwargs={})
需要关注的参数是target和args. target 是需要子线程运行的目标函数,args是函数的参数,以tuple的形式传递。
以下代码创建一个指向函数worker 的子线程
def worker(a_tid,a_account): ... th = threading.Thread(target=worker,args=(i,acc) ) ;
启动这个线程
th.start()
等待线程返回
threading.Thread.join(th) 或者th.join()
如果你可以对要处理的数据进行很好的划分,而且线程之间无须通信,那么你可以使用:创建=》运行=》回收的方式编写你的多线程程序。但是如果线程之间需要访问共同的对象,则需要引入互斥锁或者信号量对资源进行互斥访问。
下面讲讲如何创建互斥锁
创建锁 g_mutex = threading.Lock() .... 使用锁 for ... : #锁定,从下一句代码到释放前互斥访问 g_mutex.acquire() a_account.deposite(1) #释放 g_mutex.release()
最后,模拟一个公交地铁IC卡缴车费的多线程程序
有10个读卡器,每个读卡器收费器每次扣除用户一块钱进入总账中,每读卡器每天一共被刷10000000次。账户原有100块。所以最后的总账应该为10000100。先不使用互斥锁来进行锁定(注释掉了锁定代码),看看后果如何。
import time,datetime import threading def worker(a_tid,a_account): global g_mutex print "Str " , a_tid, datetime.datetime.now() for i in range(1000000): #g_mutex.acquire() a_account.deposite(1) #g_mutex.release() print "End " , a_tid , datetime.datetime.now() class Account: def __init__ (self, a_base ): self.m_amount=a_base def deposite(self,a_amount): self.m_amount+=a_amount def withdraw(self,a_amount): self.m_amount-=a_amount if __name__ == "__main__": global g_mutex count = 0 dstart = datetime.datetime.now() print "Main Thread Start At: " , dstart #init thread_pool thread_pool = [] #init mutex g_mutex = threading.Lock() # init thread items acc = Account(100) for i in range(10): th = threading.Thread(target=worker,args=(i,acc) ) ; thread_pool.append(th) # start threads one by one for i in range(10): thread_pool[i].start() #collect all threads for i in range(10): threading.Thread.join(thread_pool[i]) dend = datetime.datetime.now() print "count=",acc.m_amount print "Main Thread End at: " ,dend , " time span " , dend-dstart;
注意,先不用互斥锁进行临界段访问控制,运行结果如下:
Main Thread Start At: 2009-01-13 00:17:55.296000 Str 0 2009-01-13 00:17:55.312000 Str 1 2009-01-13 00:17:55.453000 Str 2 2009-01-13 00:17:55.484000 Str 3 2009-01-13 00:17:55.531000 Str 4 2009-01-13 00:17:55.562000 Str 5 2009-01-13 00:17:55.609000 Str 6 2009-01-13 00:17:55.640000 Str 7 2009-01-13 00:17:55.687000 Str 8 2009-01-13 00:17:55.718000 Str 9 2009-01-13 00:17:55.781000 End 0 2009-01-13 00:18:06.250000 End 1 2009-01-13 00:18:07.500000 End 4 2009-01-13 00:18:07.531000 End 2 2009-01-13 00:18:07.562000 End 3 2009-01-13 00:18:07.593000 End 9 2009-01-13 00:18:07.609000 End 7 2009-01-13 00:18:07.640000 End 8 2009-01-13 00:18:07.671000 End 5 2009-01-13 00:18:07.687000 End 6 2009-01-13 00:18:07.718000 count= 3434612 Main Thread End at: 2009-01-13 00:18:07.718000 time span 0:00:12.422000
从结果看到,程序确实是多线程运行的。但是由于没有对对象Account进行互斥访问,所以结果是错误的,只有3434612,比原预计少了很多。
把上面阴影部分代码的注释打开,运行结果如下
Main Thread Start At: 2009-01-13 00:26:12.156000 Str 0 2009-01-13 00:26:12.156000 Str 1 2009-01-13 00:26:12.390000 Str 2 2009-01-13 00:26:12.437000 Str 3 2009-01-13 00:26:12.468000 Str 4 2009-01-13 00:26:12.515000 Str 5 2009-01-13 00:26:12.562000 Str 6 2009-01-13 00:26:12.593000 Str 7 2009-01-13 00:26:12.640000 Str 8 2009-01-13 00:26:12.671000 Str 9 2009-01-13 00:26:12.718000 End 0 2009-01-13 00:27:01.781000 End 1 2009-01-13 00:27:05.890000 End 5 2009-01-13 00:27:06.046000 End 7 2009-01-13 00:27:06.078000 End 4 2009-01-13 00:27:06.109000 End 2 2009-01-13 00:27:06.140000 End 6 2009-01-13 00:27:06.156000 End 8 2009-01-13 00:27:06.187000 End 3 2009-01-13 00:27:06.203000 End 9 2009-01-13 00:27:06.234000 count= 10000100 Main Thread End at: 2009-01-13 00:27:06.234000 time span 0:00:54.078000
这次可以看到,结果正确了。运行时间比不进行互斥多了很多,需要花54秒才能运行(我机器烂,没钱更新,呵呵),不过这也是同步的代价,没办法。