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
            processes = 1
75
            Pool_ = SyncPool
76
            Queue_ = Queue.Queue
77
        else:
78
            if processes == None: processes = multiprocessing.cpu_count()
79
            Pool_ = multiprocessing.pool.Pool
80
            Queue_ = multiprocessing.Queue
81
        
82
        self.process_ct = processes
83
        self.pool = Pool_(processes)
84
        self.queue = Queue_()
85
        self.active_tasks = 0
86
        
87
        # Store a reference to the manager in self, because it will otherwise be
88
        # shutdown right away when it goes out of scope
89
        #self.manager = processing.Manager()
90
        #self.shared_rw = self.manager.Namespace()
91
        
92
        # Values that may be pickled by id()
93
        self.vars_id_dict = dicts.IdDict()
94
        self.share(self, *shared).share_vars(locals_).share_vars(globals())
95
    
96
    def share(self, *values):
97
        '''Call this on all values that should be shared writably between all
98
        processes (and be pickled by id())'''
99
        self.vars_id_dict.add(*values)
100
        return self
101
    
102
    def share_vars(self, vars_):
103
        '''Call this on all vars that should be pickled by id().
104
        Usage: self.share_vars(locals())
105
        @param vars_ {var_name: value}
106
        '''
107
        self.vars_id_dict.add_vars(vars_)
108
        return self
109
    
110
    def main_loop(self):
111
        '''Prime the task queue with at least one task before calling this''' 
112
        while True:
113
            try: call = self.queue.get(timeout=0.1) # sec
114
            except Queue.Empty:
115
                if self.active_tasks == 0: break # program done
116
                else: continue
117
            
118
            def handle_result(*args, **kw_args):
119
                self.active_tasks -= 1
120
                if call.callback != None: call.callback(*args, **kw_args)
121
            
122
            self.active_tasks += 1
123
            self.pool.apply_async(call.func, self.post_unpickle(call.args),
124
                self.post_unpickle(call.kw_args), handle_result)
125
    
126
    class Result:
127
        def get(timeout=None): raise NotImplementedError()
128
        
129
        def wait(timeout=None): raise NotImplementedError()
130
        
131
        def ready(): raise NotImplementedError()
132
        
133
        def successful(): raise NotImplementedError()
134
    
135
    def apply_async(self, func, args=(), kw_args={}, callback=None):
136
        assert callback == None, 'Callbacks not supported'
137
        
138
        call = Runnable(func, *self.prepickle(args), **self.prepickle(kw_args))
139
        call.callback = callback # store this inside the Runnable
140
        
141
        self.queue.put_nowait(call)
142
        return self.Result()
143
    
144
    def prepickle(self, value): return prepickle(value, self.vars_id_dict)
145
    
146
    def post_unpickle(self, value):
147
        return post_unpickle(value, self.vars_id_dict)
(23-23/39)