Project

General

Profile

1
# Parallel processing
2

    
3
import cPickle
4
import itertools
5
import Queue
6
import rand
7
import types
8
import warnings
9

    
10
import collection
11
import dicts
12
import exc
13
from Runnable import Runnable
14

    
15
def try_pickle(value):
16
    try: cPickle.dumps(value)
17
    except Exception, e:
18
        exc.add_msg(e, 'Tried to pickle: '+repr(value))
19
        raise
20

    
21
def prepickle(value, vars_id_dict_):
22
    def filter_(value, is_leaf):
23
        id_ = id(value)
24
        if id_ in vars_id_dict_: value = id_
25
        # Try pickling the value. If it fails, we'll get a full traceback here,
26
        # which is not provided with pickling errors in multiprocessing's Pool.
27
        elif is_leaf: try_pickle(value)
28
        return value
29
    return collection.rmap(filter_, value)
30

    
31
def post_unpickle(value, vars_id_dict_):
32
    def filter_(value, is_leaf):
33
        if type(value) == int: value = vars_id_dict_.get(value, value)
34
            # get() returns the value itself if it isn't a known id()
35
        return value
36
    return collection.rmap(filter_, value)
37

    
38
class SyncPool:
39
    '''A dummy synchronous Pool to use if multiprocessing is not available'''
40
    def __init__(self, processes=None): pass
41
    
42
    class Result:
43
        def __init__(self, value): self.value = value
44
        
45
        def get(timeout=None): return self.value
46
        
47
        def wait(timeout=None): pass
48
        
49
        def ready(): return True
50
        
51
        def successful(): return True # TODO: False if raised exception
52
    
53
    def apply_async(self, func, args=(), kw_args={}, callback=None):
54
        if callback == None: callback = lambda v: None
55
        
56
        value = func(*args, **kw_args)
57
        callback(value)
58
        return self.Result(value)
59

    
60
class MultiProducerPool:
61
    '''A multi-producer pool. You must call pool.main_loop() in the thread that
62
    created this to process new tasks.'''
63
    
64
    def __init__(self, processes=None, locals_={}, *shared):
65
        '''
66
        @param processes If 0, uses SyncPool
67
        @post The # processes actually used is made available in self.process_ct
68
        '''
69
        try:
70
            if processes == 0: raise ImportError('turned off')
71
            import multiprocessing
72
            import multiprocessing.pool
73
        except ImportError, e:
74
            warnings.warn(UserWarning('Not using parallel processing: '+str(e)))
75
            processes = 1
76
            Pool_ = SyncPool
77
            Queue_ = Queue.Queue
78
        else:
79
            if processes == None: processes = multiprocessing.cpu_count()
80
            Pool_ = multiprocessing.pool.Pool
81
            Queue_ = multiprocessing.Queue
82
        
83
        self.process_ct = processes
84
        self.pool = Pool_(processes)
85
        self.queue = Queue_()
86
        self.active_tasks = 0
87
        
88
        # Store a reference to the manager in self, because it will otherwise be
89
        # shutdown right away when it goes out of scope
90
        #self.manager = processing.Manager()
91
        #self.shared_rw = self.manager.Namespace()
92
        
93
        # Values that may be pickled by id()
94
        self.vars_id_dict = dicts.IdDict()
95
        self.share(self, *shared).share_vars(locals_).share_vars(globals())
96
    
97
    def share(self, *values):
98
        '''Call this on all values that should be shared writably between all
99
        processes (and be pickled by id())'''
100
        self.vars_id_dict.add(*values)
101
        return self
102
    
103
    def share_vars(self, vars_):
104
        '''Call this on all vars that should be pickled by id().
105
        Usage: self.share_vars(locals())
106
        @param vars_ {var_name: value}
107
        '''
108
        self.vars_id_dict.add_vars(vars_)
109
        return self
110
    
111
    def main_loop(self):
112
        '''Prime the task queue with at least one task before calling this''' 
113
        while True:
114
            try: call = self.queue.get(timeout=0.1) # sec
115
            except Queue.Empty:
116
                if self.active_tasks == 0: break # program done
117
                else: continue
118
            
119
            def handle_result(*args, **kw_args):
120
                self.active_tasks -= 1
121
                if call.callback != None: call.callback(*args, **kw_args)
122
            
123
            self.active_tasks += 1
124
            self.pool.apply_async(call.func, self.post_unpickle(call.args),
125
                self.post_unpickle(call.kw_args), handle_result)
126
    
127
    class Result:
128
        def get(timeout=None): raise NotImplementedError()
129
        
130
        def wait(timeout=None): raise NotImplementedError()
131
        
132
        def ready(): raise NotImplementedError()
133
        
134
        def successful(): raise NotImplementedError()
135
    
136
    def apply_async(self, func, args=(), kw_args={}, callback=None):
137
        assert callback == None, 'Callbacks not supported'
138
        
139
        call = Runnable(func, *self.prepickle(args), **self.prepickle(kw_args))
140
        call.callback = callback # store this inside the Runnable
141
        
142
        self.queue.put_nowait(call)
143
        return self.Result()
144
    
145
    def prepickle(self, value): return prepickle(value, self.vars_id_dict)
146
    
147
    def post_unpickle(self, value):
148
        return post_unpickle(value, self.vars_id_dict)
(20-20/35)