from threading import Thread, current_thread
num = 0
def calc():
global num
print 'thread %s is running...' % current_thread().name
for _ in xrange(10000):
num += 1
print 'thread %s ended.' % current_thread().name
if __name__ == '__main__':
print 'thread %s is running...' % current_thread().name
threads = []
for i in range(5):
threads.append(Thread(target=calc))
threads[i].start()
for i in range(5):
threads[i].join()
print 'global num: %d' % num
print 'thread %s ended.' % current_thread().name
thread MainThread is running...
thread Thread-34 is running...
thread Thread-34 ended.
thread Thread-35 is running...
thread Thread-36 is running...
thread Thread-37 is running...
thread Thread-38 is running...
thread Thread-35 ended.
thread Thread-38 ended.
thread Thread-36 ended.
thread Thread-37 ended.
global num: 30668
thread MainThread ended.
我们发现 num 的值是 30668，事实上，num 的值是不确定的，你再运行一遍，会发现结果变了。
原因是因为 num += 1 不是一个原子操作，也就是说它在执行时被分成若干步：
计算 num + 1，存入临时变量 tmp 中；
将 tmp 的值赋给 num.
由于线程是交替运行的，线程在执行时可能中断，就会导致其他线程读到一个脏值。
为了保证计算的准确性，我们就需要给 num += 1 这个操作加上锁。当某个线程开始执行这个操作时，由于该线程获得了锁，因此其他线程不能同时执行该操作，只能等待，直到锁被释放，这样就可以避免修改的冲突。创建一个锁可以通过 threading.Lock() 来实现，代码如下：
from threading import Thread, current_thread, Lock
num = 0
lock = Lock()
def calc():
global num
print 'thread %s is running...' % current_thread().name
for _ in xrange(10000):
lock.acquire() # 获取锁
num += 1
lock.release() # 释放锁
print 'thread %s ended.' % current_thread().name
if __name__ == '__main__':
print 'thread %s is running...' % current_thread().name
threads = []
for i in range(5):
threads.append(Thread(target=calc))
threads[i].start()
for i in range(5):
threads[i].join()
print 'global num: %d' % num
print 'thread %s ended.' % current_thread().name
让我们看下执行结果：
thread MainThread is running...
thread Thread-44 is running...
thread Thread-45 is running...
thread Thread-46 is running...
thread Thread-47 is running...
thread Thread-48 is running...
thread Thread-45 ended.
thread Thread-47 ended.
thread Thread-48 ended.
thread Thread-46 ended.
thread Thread-44 ended.
global num: 50000
thread MainThread ended.
GIL 锁
讲到 Python 中的多线程，就不得不面对 GIL 锁，GIL 锁的存在导致 Python 不能有效地使用多线程实现多核任务，因为在同一时间，只能有一个线程在运行。
GIL 全称是 Global Interpreter Lock，译为全局解释锁。早期的 Python 为了支持多线程，引入了 GIL 锁，用于解决多线程之间数据共享和同步的问题。但这种实现方式后来被发现是非常低效的，当大家试图去除 GIL 的时候，却发现大量库代码已重度依赖 GIL，由于各种各样的历史原因，GIL 锁就一直保留到现在。