1参数就算只有一个也要写成（arg,）的形式
2在类里面必须声明成静态方法
3不能将要执行子进程的方法写到于Pool或者Process同级的方法内部
4执行子进程的方法如果是使用map进程池的时候必须只能接受一个参数，所以多参数的时候要使用zip包裹起来，使用apply参数必须分开，不能只写为一个
5有的对象不可以被pickle则不能当作参数传递
6多进程中可以传递lock，但不能直接传递queue对象，只能使用manager的list等对像，或者必须在实例化queue对象的时候初始化queue才行，举例：
import multiprocessing as MP
def f(msg):
f.q.put(msg)
def q_init(q):
f.q = q
if __name__ == "__main__":
q = MP.Queue()
p = MP.Pool(None, q_init, (q,))
for i in range(0,8):
p.apply_async(f, (i,))
p.close()
p.join()
while not q.empty():
print(q.get())
。。。。未完待续
7 python2的multiprocessing不可以把方法写到类里面，必须写出来才能执行
8 不能在多个子进程中写同一个h5py文件然后再join（）之后再关闭文件，这样会导致写入的数据再关闭时被瞬间清空，不知道原因，只可以再单个子进程中写了数据就关闭才行 。
9 须在close（）方法前加上result.get()方法才能捕捉子进程的错误，否则有可能子进程出错了就直接跳过了而不报错。该方法对map和apply都实用，已经测试过了
10 The multiprocessing module has a major limitation: it only accepts certain functions, and in certain situations. For instance any class methods, lambdas, or functions defined in __main__ wont’ work. This is due to the way Python “pickles” (read: serializes) data and sends it to the worker processes. “Pickling” simply can’t handle a lot of different types of Python objects.
Fortunately, there is a fork of the multiprocessing module called multiprocess that works just fine (pip install --user multiprocess). multiprocess uses dillinstead of pickle to serialize Python objects (read: send your data and functions to the Python workers), and does not suffer the same issues. Usage is identical:
# shut down the old workers
pool.close()
from multiprocess import Pool
pool = Pool(8)
%timeit pool.map(lambda x: time.sleep(0.1), range(24))
pool.close()