首先,过程,游泳池和队列都具有不同的使用情况。
过程用于通过创建Process对象来产生一个过程。
from multiprocessing import Process
def method1():
print "in method1"
print "in method1"
def method2():
print "in method2"
print "in method2"
p1 = Process(target=method1) # create a process object p1
p1.start() # starts the process p1
p2 = Process(target=method2)
p2.start()
池用于并行跨多个 输入值的功能执行。
from multiprocessing import Pool
def method1(x):
print x
print x**2
return x**2
p = Pool(3)
result = p.map(method1, [1,4,9])
print result # prints [1, 16, 81]
队列被用于进程间通信。现在
from multiprocessing import Process, Queue
def method1(x, l1):
print "in method1"
print "in method1"
l1.put(x**2)
return x
def method2(x, l2):
print "in method2"
print "in method2"
l2.put(x**3)
return x
l1 = Queue()
p1 = Process(target=method1, args=(4, l1,))
l2 = Queue()
p2 = Process(target=method2, args=(2, l2,))
p1.start()
p2.start()
print l1.get() # prints 16
print l2.get() # prints 8
,为你的情况下,你可以使用过程&队列(第3方法),或者你可以操纵池方法的工作(见下文)
import itertools
from multiprocessing import Pool
import sys
def method1(x):
print x
print x**2
return x**2
def method2(x):
print x
print x**3
return x**3
def unzip_func(a, b):
return a, b
def distributor(option_args):
option, args = unzip_func(*option_args) # unzip option and args
attr_name = "method" + str(option)
# creating attr_name depending on option argument
value = getattr(sys.modules[__name__], attr_name)(args)
# call the function with name 'attr_name' with argument args
return value
option_list = [1,2] # for selecting the method number
args_list = [4,2]
# list of arg for the corresponding method, (argument 4 is for method1)
p = Pool(3) # creating pool of 3 processes
result = p.map(distributor, itertools.izip(option_list, args_list))
# calling the distributor function with args zipped as (option1, arg1), (option2, arg2) by itertools package
print result # prints [16,8]
希望这有助于。
吉尔与CPU绑定操作干扰,但并没有真正影响到IO绑定操作。你的功能在做什么类型的东西? – Blender
这些函数执行相同的通用类型的东西:通过http请求获取数据,将它们存储在内存中,执行一些处理并将它们转换为numpy数组。 – PsychicLocust
“一般”不是很具体。试试线程和多处理,看看是否有区别,使用这两个模块的API是相似的。 – Blender