Multiprocessing.Array bug / shared numpy array
 

Hi,

The documentation for the Multiprocessing.Array says:

multiprocessing.Array(typecode_or_type, size_or_initializer, *,
lock=True)¶

....
If lock is False then access to the returned object will not be
automatically protected by a lock, so it will not necessarily be
“process-safe”.
....

However:
In [48]: mp.Array('i',1,lock=False)
---------------------------------------------------------------------------
AssertionError Traceback (most recent call
last)

/Library/Frameworks/Python.framework/Versions/2.6/lib/python2.6/
multiprocessing/__init__.pyc in Array(typecode_or_type,
size_or_initializer, **kwds)
252 '''
253 from multiprocessing.sharedctypes import Array
--> 254 return Array(typecode_or_type, size_or_initializer,
**kwds)
255
256 #


/Library/Frameworks/Python.framework/Versions/2.6/lib/python2.6/
multiprocessing/sharedctypes.pyc in Array(typecode_or_type,
size_or_initializer, **kwds)
85 if lock is None:
86 lock = RLock()
---> 87 assert hasattr(lock, 'acquire')
88 return synchronized(obj, lock)
89

AssertionError:

-------
I.e. it looks like lock=false is not actually supported. Or am I
reading this wrong? If not, I can submit a bug report.


I am trying to create a shared, read-only numpy.ndarray between
several processes. After some googling the basic idea is:

sarr = mp.Array('i',1000)
ndarr = scipy.frombuffer(sarr._obj,dtype='int32')

Since it will be read only (after being filled once in a single
process) I don't think I need any locking mechanism. However is this
really true given garbage collection, reference counts and other
implicit things going on?

Or is there a recommended better way to do this?

Thanks

Re: Multiprocessing.Array bug / shared numpy array
 

On 2009-10-08 15:14 PM, Felix wrote:

> I am trying to create a shared, read-only numpy.ndarray between
> several processes. After some googling the basic idea is:
>
> sarr = mp.Array('i',1000)
> ndarr = scipy.frombuffer(sarr._obj,dtype='int32')
>
> Since it will be read only (after being filled once in a single
> process) I don't think I need any locking mechanism. However is this
> really true given garbage collection, reference counts and other
> implicit things going on?
>
> Or is there a recommended better way to do this?

I recommend using memory-mapped arrays for such a purpose.

You will want to ask further numpy questions on the numpy mailing list:

http://www.scipy.org/Mailing_Lists

--
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless enigma
that is made terrible by our own mad attempt to interpret it as though it had
an underlying truth."
-- Umberto Eco