关于python:multiprocessing:如何在多个进程之间共享一个dict?

multiprocessing: How do I share a dict among multiple processes?

一个程序,它创建几个可在可连接队列Q上工作的进程,并最终可能操纵全局字典D来存储结果。 (因此每个子进程可以使用D来存储其结果,并查看其他子进程正在生成的结果)

如果我在子进程中打印字典D,我会看到已对其进行的修改(即在D上)。 但是在主进程加入Q之后,如果我打印D,那就是空的dict!

我知道这是一个同步/锁定问题。 有人能告诉我这里发生了什么,以及如何同步访问D?


一般答案涉及使用Manager对象。改编自文档:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
from multiprocessing import Process, Manager

def f(d):
    d[1] += '1'
    d['2'] += 2

if __name__ == '__main__':
    manager = Manager()

    d = manager.dict()
    d[1] = '1'
    d['2'] = 2

    p1 = Process(target=f, args=(d,))
    p2 = Process(target=f, args=(d,))
    p1.start()
    p2.start()
    p1.join()
    p2.join()

    print d

输出:

1
2
$ python mul.py
{1: '111', '2': 6}


多处理与线程不同。每个子进程都将获得主进程内存的副本。通常,状态通过通信(管道/插座),信号或共享存储器共享。

多处理使一些抽象可用于您的用例 - 通过使用代理或共享内存将其视为本地的共享状态:http://docs.python.org/library/multiprocessing.html#sharing-state-between-processes

相关部分:

  • http://docs.python.org/library/multiprocessing.html#shared-ctypes-objects
  • http://docs.python.org/library/multiprocessing.html#module-multiprocessing.managers


我想分享我自己的工作,这比管理器的字典更快,比使用大量内存并且不适用于Mac OS的pyshmht库更简单,更稳定。虽然我的dict只适用于普通字符串,但目前是不可变的。
我使用线性探测实现,并在表后面的单独内存块中存储键和值对。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
from mmap import mmap
import struct
from timeit import default_timer
from multiprocessing import Manager
from pyshmht import HashTable


class shared_immutable_dict:
    def __init__(self, a):
        self.hs = 1 << (len(a) * 3).bit_length()
        kvp = self.hs * 4
        ht = [0xffffffff] * self.hs
        kvl = []
        for k, v in a.iteritems():
            h = self.hash(k)
            while ht[h] != 0xffffffff:
                h = (h + 1) & (self.hs - 1)
            ht[h] = kvp
            kvp += self.kvlen(k) + self.kvlen(v)
            kvl.append(k)
            kvl.append(v)

        self.m = mmap(-1, kvp)
        for p in ht:
            self.m.write(uint_format.pack(p))
        for x in kvl:
            if len(x) <= 0x7f:
                self.m.write_byte(chr(len(x)))
            else:
                self.m.write(uint_format.pack(0x80000000 + len(x)))
            self.m.write(x)

    def hash(self, k):
        h = hash(k)
        h = (h + (h >> 3) + (h >> 13) + (h >> 23)) * 1749375391 & (self.hs - 1)
        return h

    def get(self, k, d=None):
        h = self.hash(k)
        while True:
            x = uint_format.unpack(self.m[h * 4:h * 4 + 4])[0]
            if x == 0xffffffff:
                return d
            self.m.seek(x)
            if k == self.read_kv():
                return self.read_kv()
            h = (h + 1) & (self.hs - 1)

    def read_kv(self):
        sz = ord(self.m.read_byte())
        if sz & 0x80:
            sz = uint_format.unpack(chr(sz) + self.m.read(3))[0] - 0x80000000
        return self.m.read(sz)

    def kvlen(self, k):
        return len(k) + (1 if len(k) <= 0x7f else 4)

    def __contains__(self, k):
        return self.get(k, None) is not None

    def close(self):
        self.m.close()

uint_format = struct.Struct('>I')


def uget(a, k, d=None):
    return to_unicode(a.get(to_str(k), d))


def uin(a, k):
    return to_str(k) in a


def to_unicode(s):
    return s.decode('utf-8') if isinstance(s, str) else s


def to_str(s):
    return s.encode('utf-8') if isinstance(s, unicode) else s


def mmap_test():
    n = 1000000
    d = shared_immutable_dict({str(i * 2): '1' for i in xrange(n)})
    start_time = default_timer()
    for i in xrange(n):
        if bool(d.get(str(i))) != (i % 2 == 0):
            raise Exception(i)
    print 'mmap speed: %d gets per sec' % (n / (default_timer() - start_time))


def manager_test():
    n = 100000
    d = Manager().dict({str(i * 2): '1' for i in xrange(n)})
    start_time = default_timer()
    for i in xrange(n):
        if bool(d.get(str(i))) != (i % 2 == 0):
            raise Exception(i)
    print 'manager speed: %d gets per sec' % (n / (default_timer() - start_time))


def shm_test():
    n = 1000000
    d = HashTable('tmp', n)
    d.update({str(i * 2): '1' for i in xrange(n)})
    start_time = default_timer()
    for i in xrange(n):
        if bool(d.get(str(i))) != (i % 2 == 0):
            raise Exception(i)
    print 'shm speed: %d gets per sec' % (n / (default_timer() - start_time))


if __name__ == '__main__':
    mmap_test()
    manager_test()
    shm_test()

我的笔记本电脑的性能结果是:

1
2
3
mmap speed: 247288 gets per sec
manager speed: 33792 gets per sec
shm speed: 691332 gets per sec

简单的用法示例:

1
2
ht = shared_immutable_dict({'a': '1', 'b': '2'})
print ht.get('a')


除了@ senderle之外,有些人可能也想知道如何在这里使用multiprocessing.Pool的功能。

好处是Manager实例有一个.Pool()方法,它模仿顶层`多处理的所有熟悉的API。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
from itertools import repeat
import multiprocessing as mp
import os
import pprint

def f(d):
    pid = os.getpid()
    d[pid] ="Hi, I was written by process %d" % pid

if __name__ == '__main__':
    with mp.Manager() as manager:
        d = manager.dict()
        with manager.Pool() as pool:
            pool.map(f, repeat(d, 10))
        # `d` is a DictProxy object that can be converted to dict
        pprint.pprint(dict(d))

输出:

1
2
3
4
5
6
7
8
9
10
11
$ python3 mul.py
{22562: 'Hi, I was written by process 22562',
 22563: 'Hi, I was written by process 22563',
 22564: 'Hi, I was written by process 22564',
 22565: 'Hi, I was written by process 22565',
 22566: 'Hi, I was written by process 22566',
 22567: 'Hi, I was written by process 22567',
 22568: 'Hi, I was written by process 22568',
 22569: 'Hi, I was written by process 22569',
 22570: 'Hi, I was written by process 22570',
 22571: 'Hi, I was written by process 22571'}

这是一个稍微不同的示例,其中每个进程只将其进程ID记录到全局DictProxy对象d


也许你可以试试pyshmht,为Python共享基于内存的哈希表扩展。

注意

  • 它没有经过全面测试,仅供参考。

  • 它目前缺乏用于多处理的锁/ sem机制。