Was können Multiprozessing und Dill zusammen tun?

multiprocessing macht einige schlechte Entscheidungen zu Beizen. Versteh mich nicht falsch, es macht einige gute Entscheidungen, die es Ihr ermöglichen, Gurke, bestimmte Typen, also Sie können verwendet werden, in einem pool, die map-Funktion. Aber da haben wir dill tun können die Beizen -, multiprocessing-eigene Beizen wird ein wenig zu begrenzen. Eigentlich, wenn multiprocessing waren zu verwenden pickle statt cPickle... und auch fallen einige von seinen eigenen Beizen überschreibt, dann dill übernehmen könnte und geben einem viel mehr vollständige Serialisierung für multiprocessing.

Bis das passiert, gibt es eine Gabel von multiprocessing genannt pathos (die release-version ist ein wenig abgestanden, leider) , entfernt die oben genannten Einschränkungen. Pathos fügt auch einige nette features, die multiprocessing nicht, wie multi-args in der map-Funktion. Pathos ist fällig für eine Veröffentlichung, nach einigen milden Aktualisierung-vor allem Umstellung auf python 3.x.

Python 2.7.5 (default, Sep 30 2013, 20:15:49) 
[GCC 4.2.1 (Apple Inc. build 5566)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import dill
>>> from pathos.multiprocessing import ProcessingPool    
>>> pool = ProcessingPool(nodes=4)
>>> result = pool.map(lambda x: x**2, range(10))
>>> result
[0, 1, 4, 9, 16, 25, 36, 49, 64, 81]

und einfach zu zeigen, ein wenig von dem, was pathos.multiprocessing tun können...

>>> def busy_add(x,y, delay=0.01):
...     for n in range(x):
...        x += n
...     for n in range(y):
...        y -= n
...     import time
...     time.sleep(delay)
...     return x + y
... 
>>> def busy_squared(x):
...     import time, random
...     time.sleep(2*random.random())
...     return x*x
... 
>>> def squared(x):
...     return x*x
... 
>>> def quad_factory(a=1, b=1, c=0):
...     def quad(x):
...         return a*x**2 + b*x + c
...     return quad
... 
>>> square_plus_one = quad_factory(2,0,1)
>>> 
>>> def test1(pool):
...     print pool
...     print "x: %s\n" % str(x)
...     print pool.map.__name__
...     start = time.time()
...     res = pool.map(squared, x)
...     print "time to results:", time.time() - start
...     print "y: %s\n" % str(res)
...     print pool.imap.__name__
...     start = time.time()
...     res = pool.imap(squared, x)
...     print "time to queue:", time.time() - start
...     start = time.time()
...     res = list(res)
...     print "time to results:", time.time() - start
...     print "y: %s\n" % str(res)
...     print pool.amap.__name__
...     start = time.time()
...     res = pool.amap(squared, x)
...     print "time to queue:", time.time() - start
...     start = time.time()
...     res = res.get()
...     print "time to results:", time.time() - start
...     print "y: %s\n" % str(res)
... 
>>> def test2(pool, items=4, delay=0):
...     _x = range(-items/2,items/2,2)
...     _y = range(len(_x))
...     _d = [delay]*len(_x)
...     print map
...     res1 = map(busy_squared, _x)
...     res2 = map(busy_add, _x, _y, _d)
...     print pool.map
...     _res1 = pool.map(busy_squared, _x)
...     _res2 = pool.map(busy_add, _x, _y, _d)
...     assert _res1 == res1
...     assert _res2 == res2
...     print pool.imap
...     _res1 = pool.imap(busy_squared, _x)
...     _res2 = pool.imap(busy_add, _x, _y, _d)
...     assert list(_res1) == res1
...     assert list(_res2) == res2
...     print pool.amap
...     _res1 = pool.amap(busy_squared, _x)
...     _res2 = pool.amap(busy_add, _x, _y, _d)
...     assert _res1.get() == res1
...     assert _res2.get() == res2
...     print ""
... 
>>> def test3(pool): # test against a function that should fail in pickle
...     print pool
...     print "x: %s\n" % str(x)
...     print pool.map.__name__
...     start = time.time()
...     res = pool.map(square_plus_one, x)
...     print "time to results:", time.time() - start
...     print "y: %s\n" % str(res)
... 
>>> def test4(pool, maxtries, delay):
...     print pool
...     m = pool.amap(busy_add, x, x)
...     tries = 0
...     while not m.ready():
...         time.sleep(delay)
...         tries += 1
...         print "TRY: %s" % tries
...         if tries >= maxtries:
...             print "TIMEOUT"
...             break
...     print m.get()
... 
>>> import time
>>> x = range(18)
>>> delay = 0.01
>>> items = 20
>>> maxtries = 20
>>> from pathos.multiprocessing import ProcessingPool as Pool
>>> pool = Pool(nodes=4)
>>> test1(pool)
<pool ProcessingPool(ncpus=4)>
x: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17]

map
time to results: 0.0553691387177
y: [0, 1, 4, 9, 16, 25, 36, 49, 64, 81, 100, 121, 144, 169, 196, 225, 256, 289]

imap
time to queue: 7.91549682617e-05
time to results: 0.102381229401
y: [0, 1, 4, 9, 16, 25, 36, 49, 64, 81, 100, 121, 144, 169, 196, 225, 256, 289]

amap
time to queue: 7.08103179932e-05
time to results: 0.0489699840546
y: [0, 1, 4, 9, 16, 25, 36, 49, 64, 81, 100, 121, 144, 169, 196, 225, 256, 289]

>>> test2(pool, items, delay)
<built-in function map>
<bound method ProcessingPool.map of <pool ProcessingPool(ncpus=4)>>
<bound method ProcessingPool.imap of <pool ProcessingPool(ncpus=4)>>
<bound method ProcessingPool.amap of <pool ProcessingPool(ncpus=4)>>

>>> test3(pool)
<pool ProcessingPool(ncpus=4)>
x: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17]

map
time to results: 0.0523059368134
y: [1, 3, 9, 19, 33, 51, 73, 99, 129, 163, 201, 243, 289, 339, 393, 451, 513, 579]

>>> test4(pool, maxtries, delay)
<pool ProcessingPool(ncpus=4)>
TRY: 1
TRY: 2
TRY: 3
TRY: 4
TRY: 5
TRY: 6
TRY: 7
[0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30, 32, 34]

InformationsquelleAutor der Antwort Mike McKerns