Python标准库为我们提供了threading和multiprocessing模块编写相应的多线程/多进程代码。从Python3.2开始,标准库为我们提供了concurrent.futures模块,它提供了ThreadPoolExecutor和ProcessPoolExecutor两个类,实现了对threading和multiprocessing的更高级的抽象,对编写线程池/进程池提供了直接的支持。
concurrent.futures基础模块是executor和future。
Executor
Executor是一个抽象类,它不能被直接使用。它为具体的异步执行定义了一些基本的方法。
ThreadPoolExecutor和ProcessPoolExecutor继承了Executor,分别被用来创建线程池和进程池的代码。
class Executor(object):
"""This is an abstract base class for concrete asynchronous executors."""
def submit(self, fn, *args, **kwargs):
raise NotImplementedError()
def map(self, fn, *iterables, timeout=None):
if timeout is not None:
end_time = timeout + time.time()
fs = [self.submit(fn, *args) for args in zip(*iterables)]
def result_iterator():
try:
for future in fs:
if timeout is None:
yield future.result()
else:
yield future.result(end_time - time.time())
finally:
for future in fs:
future.cancel()
return result_iterator()
def shutdown(self, wait=True):
pass
def __enter__(self):
return self
def __exit__(self, exc_type, exc_val, exc_tb):
self.shutdown(wait=True)
return False
submit()方法
Executor中定义了submit()
方法,这个方法的作用是提交一个可执行的回调task
,并返回一个future实例。future对象代表的就是给定的调用。
通过下面的例子来理解submit对线程池/进程池的操作。
改写为进程池形式很简单,把ThreadPoolExecutor替换为ProcessPoolExecutor即可。如果需要提交多个task,可以通过循环多次submit()。
map()方法
除了submit,Exectuor还为我们提供了map方法,这个方法返回一个map(func, *iterables)迭代器,迭代器中的回调执行返回的结果有序的。可以通过下面的例子来理解:
# coding: utf-8
from concurrent.futures import ThreadPoolExecutor as Pool
import requests
URLS = ['http://www.baidu.com', 'http://qq.com', 'http://sina.com']
def task(url, timeout=10):
return requests.get(url, timeout=timeout)
pool = Pool(max_workers=3)
results = pool.map(task, URLS)
for ret in results:
print('%s, %s' % (ret.url, len(ret.content)))
执行结果
http://www.baidu.com/, 2381
http://www.qq.com/, 252160
http://www.sina.com.cn/, 607265
Future
Future
可以理解为一个在未来完成的操作,这是异步编程的基础。通常情况下,我们执行io操作,访问url时(如下)在等待结果返回之前会产生阻塞,cpu不能做其他事情,而Future的引入帮助我们在等待的这段时间可以完成其他的操作。
import requests
data = requests.get('http://www.baidu.com').content
print len(data)
Future实例是由Executor.submit()
创建的。Future提供了丰富的方法来处理调用。
# coding: utf-8
from concurrent.futures import ThreadPoolExecutor as Pool
from concurrent.futures import as_completed
import requests
URLS = ['http://qq.com', 'http://sina.com', 'http://www.baidu.com', ]
def task(url, timeout=10):
return requests.get(url, timeout=timeout)
with Pool(max_workers=3) as executor:
future_tasks = [executor.submit(task, url) for url in URLS]
for f in future_tasks:
if f.running():
print('%s is running' % str(f))
for f in as_completed(future_tasks):
try:
ret = f.done()
if ret:
f_ret = f.result()
print('%s, done, result: %s, %s' % (str(f), f_ret.url, len(f_ret.content)))
except Exception as e:
f.cancel()
print(str(e))
运行结果:
<Future at 0x7fc2716e1f60 state=running> is running
<Future at 0x7fc27136d4e0 state=running> is running
<Future at 0x7fc27136d710 state=running> is running
<Future at 0x7fc27136d710 state=finished returned Response>, done, result: http://www.baidu.com/, 2381
<Future at 0x7fc2716e1f60 state=finished returned Response>, done, result: http://www.qq.com/, 252343
<Future at 0x7fc27136d4e0 state=finished returned Response>, done, result: http://www.sina.com.cn/, 602366
从运行结果可以看出,as_completed不是按照URLS列表元素的顺序返回的。这也表明,并发访问不通的url时,没有阻塞。
wait
wait方法接会返回一个tuple(元组),tuple中包含两个set(集合),一个是completed(已完成的)另外一个是uncompleted(未完成的)。使用wait方法的一个优势就是获得更大的自由度,它接收三个参数FIRST_COMPLETED, FIRST_EXCEPTION和ALL_COMPLETE,默认设置为ALL_COMPLETED。
# coding: utf-8
from concurrent.futures import ThreadPoolExecutor as Pool
from concurrent.futures import wait
import requests
URLS = ['http://qq.com', 'http://sina.com', 'http://www.baidu.com', ]
def task(url, timeout=10):
return requests.get(url, timeout=timeout)
with Pool(max_workers=3) as executor:
future_tasks = [executor.submit(task, url) for url in URLS]
for f in future_tasks:
if f.running():
print('%s is running' % str(f))
results = wait(future_tasks)
done = results[0]
for x in done:
print(x)
wait有timeout和return_when两个参数可以设置。
timeout控制wait()方法返回前等待的时间。
return_when决定方法什么时间点返回:如果采用默认的ALL_COMPLETED,程序会阻塞直到线程池里面的所有任务都完成;如果采用FIRST_COMPLETED参数,程序并不会等到线程池里面所有的任务都完成。