python concurrent.futures
python因为其全局解释器锁GIL而无法通过线程实现真正的平行计算。
IO密集型: 读取文件,读取网络套接字频繁。
计算密集型: 大量消耗cpu的数据与逻辑计算,即平行计算。
concurrent.futures模块, 可以利用multiprocessing实现真正的平行计算。
【核心原理】 concurrent.futures会以子进程的形式,平行运行多个pyhton解释器,
可使python程序利用多核cpu来提升执行速度。由于子进程与主解释器相分离,故进程
间解释锁也是相互独立的,子进程都能够完成使用一个cpu内核。
eg:
def gcd(pair):
a, b = pair
low = min(a, b)
for i in range(low, 0, -1):
if a % i == 0 and b % i == 0:
return i
numbers = [
(1963309, 2265973), (1879675, 2493670), (2030677, 3814172),
(1551645, 2229620), (1988912, 4736670), (2198964, 7876293)
]
- 不使用多线程,多进程
import time
start = time.time()
results = list(map(gcd, numbers))
end = time.time()
print 'Took %.3f seconds.' % (end - start)
Took 2.507 seconds.
- 多线程
import time
from concurrent.futures import ThreadPoolExecutor, ProcessPoolExecutor, Executor
start = time.time()
pool = ThreadPoolExecutor(max_workers=2)
results = list(pool.map(gcd, numbers))
end = time.time()
print 'Took %.3f seconds.' % (end - start)
Took 2.840 seconds.
gcd是一个计算密集型函数,因为GIL原因,多线程无法提升效率,同时,线程启动、通信,有一定开销
所以耗时更长。
- 多进程
import time
from concurrent.futures import ThreadPoolExecutor, ProcessPoolExecutor, Executor
start = time.time()
pool = ProcessPoolExecutor(max_workers=2)
results = list(pool.map(gcd, numbers))
end = time.time()
print 'Took %.3f seconds.' % (end - start)
Took 1.861 seconds.
在双核cpu上运行多进程,比其他两个版本更快,这是因为,ProcessPoolExecutor类会
利用multiprocessing模块所提供的底层机制,完成下列操作:
1、把numbers列表中的每一项输入数据都传给map;
2、使用pickle模块对数据进行序列化,将其变成二进制形式;
3、通过本地套接字,将序列化之后的数据从主解释器所在进程, 发送到字解释器所在进程;
4、子进程中,使用pickle对二进制数据进行反序列化,将其还原成python对象;
5、引入包含gcd函数的python模块;
6、各个子进程并行对各自数据进行计算;
7、对运算结果进行反序列操作,将其转变成字节;
8、将这些字节通过socket复制到主进程中;
9、主进程对这些字节执行反序列化操作 ,将其还原成python对象;
10、将所有子进程执行结果,合并到列表之中,并返回给调用者。
concurrent.futures源码分析
- Executor
Executor是一个抽象类,提供了如下抽象方法submit,map(上面已经使用过),shutdown。
值得一提的是Executor实现了__enter__和__exit__使得其对象可以使用with操作符。
关于上下文管理和with操作符详细请参看这篇博客
http://www.cnblogs.com/kangoroo/p/7627167.html
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):
"""Submits a callable to be executed with the given arguments.
Schedules the callable to be executed as fn(*args, **kwargs) and returns
a Future instance representing the execution of the callable.
Returns:
A Future representing the given call.
"""
raise NotImplementedError()
def map(self, fn, *iterables, **kwargs):
"""Returns a iterator equivalent to map(fn, iter).
Args:
fn: A callable that will take as many arguments as there are
passed iterables.
timeout: The maximum number of seconds to wait. If None, then there
is no limit on the wait time.
Returns:
An iterator equivalent to: map(func, *iterables) but the calls may
be evaluated out-of-order.
Raises:
TimeoutError: If the entire result iterator could not be generated
before the given timeout.
Exception: If fn(*args) raises for any values.
"""
timeout = kwargs.get('timeout')
if timeout is not None:
end_time = timeout + time.time()
fs = [self.submit(fn, *args) for args in itertools.izip(*iterables)]
# Yield must be hidden in closure so that the futures are submitted
# before the first iterator value is required.
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):
"""Clean-up the resources associated with the Executor.
It is safe to call this method several times. Otherwise, no other
methods can be called after this one.
Args:
wait: If True then shutdown will not return until all running
futures have finished executing and the resources used by the
executor have been reclaimed.
"""
pass
def __enter__(self):
return self
def __exit__(self, exc_type, exc_val, exc_tb):
self.shutdown(wait=True)
return False
下面我们以线程ProcessPoolExecutor的方式说明其中的各个方法。
map
map(self, fn, iterables, **kwargs)
map方法的实例我们上面已经实现过,值得注意的是,返回的results列表是有序的,顺序和iterables迭代器的顺序一致。
这里我们使用with操作符,使得当任务执行完成之后,自动执行shutdown函数,而无需编写相关释放代码。
import time
from concurrent.futures import ThreadPoolExecutor, ProcessPoolExecutor, Executor
start = time.time()
with ProcessPoolExecutor(max_workers=2) as pool:
results = list(pool.map(gcd, numbers))
print 'results: %s' % results
end = time.time()
print 'Took %.3f seconds.' % (end - start)
产出结果是:
results: [1, 5, 1, 5, 2, 3]
Took 1.617 seconds.
submit
submit(self, fn, *args, **kwargs)
submit方法用于提交一个可并行的方法,submit方法同时返回一个future实例。
future对象标识这个线程/进程异步进行,并在未来的某个时间执行完成。future实例表示线程/进程状态的回调。
import time
from concurrent.futures import ThreadPoolExecutor, ProcessPoolExecutor, Executor
start = time.time()
futures = list()
with ProcessPoolExecutor(max_workers=2) as pool:
for pair in numbers:
future = pool.submit(gcd, pair)
futures.append(future)
print 'results: %s' % [future.result() for future in futures]
end = time.time()
print 'Took %.3f seconds.' % (end - start)
产出结果是:
results: [1, 5, 1, 5, 2, 3]
Took 2.289 seconds.
future
submit函数返回future对象,future提供了跟踪任务执行状态的方法。比如判断任务是否执行中future.running(),判断任务是否执行完成future.done()等等。
as_completed方法传入futures迭代器和timeout两个参数
默认timeout=None,阻塞等待任务执行完成,并返回执行完成的future对象迭代器,迭代器是通过yield实现的。
timeout>0,等待timeout时间,如果timeout时间到仍有任务未能完成,不再执行并抛出异常TimeoutError
import time
from concurrent.futures import ThreadPoolExecutor, ProcessPoolExecutor, Executor, as_completed
start = time.time()
with ProcessPoolExecutor(max_workers=2) as pool:
futures = [ pool.submit(gcd, pair) for pair in numbers]
for future in futures:
print '执行中:%s, 已完成:%s' % (future.running(), future.done())
print '#### 分界线 ####'
for future in as_completed(futures, timeout=2):
print '执行中:%s, 已完成:%s' % (future.running(), future.done())
end = time.time()
print 'Took %.3f seconds.' % (end - start)
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。
import time
from concurrent.futures import ThreadPoolExecutor, ProcessPoolExecutor, Executor, as_completed, wait, ALL_COMPLETED, FIRST_COMPLETED, FIRST_EXCEPTION
start = time.time()
with ProcessPoolExecutor(max_workers=2) as pool:
futures = [ pool.submit(gcd, pair) for pair in numbers]
for future in futures:
print '执行中:%s, 已完成:%s' % (future.running(), future.done())
print '#### 分界线 ####'
done, unfinished = wait(futures, timeout=2, return_when=ALL_COMPLETED)
for d in done:
print '执行中:%s, 已完成:%s' % (d.running(), d.done())
print d.result()
end = time.time()
print 'Took %.3f seconds.' % (end - start)
由于设置了ALL_COMPLETED,所以wait等待所有的task执行完成,可以看到6个任务都执行完成了。
执行中:True, 已完成:False
执行中:True, 已完成:False
执行中:True, 已完成:False
执行中:True, 已完成:False
执行中:False, 已完成:False
执行中:False, 已完成:False
分界线
执行中:False, 已完成:True
执行中:False, 已完成:True
执行中:False, 已完成:True
执行中:False, 已完成:True
执行中:False, 已完成:True
执行中:False, 已完成:True
Took 1.518 seconds.
如果我们将配置改为FIRST_COMPLETED,wait会等待直到第一个任务执行完成,返回当时所有执行成功的任务。这里并没有做并发控制。
重跑,结构如下,可以看到执行了2个任务。
执行中:True, 已完成:False
执行中:True, 已完成:False
执行中:True, 已完成:False
执行中:True, 已完成:False
执行中:False, 已完成:False
执行中:False, 已完成:False
分界线
执行中:False, 已完成:True
执行中:False, 已完成:True
Took 1.517 seconds.