即使在删除对象之后,Python struct.pack也会导致内存泄漏

Python struct.pack causing a memory leak, even after deleting object

我正在使用pyaudio检测音频源的活动。我正在一个经常死掉的流中创建每个音频事件的wav文件。使用memory_profiler,我注意到record_to_file()方法中的pack方法通常使用的内存量是删除数据对象所能恢复的4倍。此代码取自检测和录制python中的音频

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
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
from sys import byteorder
import sys
from array import array
import struct

import gc

import pyaudio
import wave
import subprocess
import objgraph
from memory_profiler import profile
from guppy import hpy

THRESHOLD = 5000
CHUNK_SIZE = 1024
FORMAT = pyaudio.paInt16
RATE = 44100

def is_silent(snd_data):
   "Returns 'True' if below the 'silent' threshold"
    return max(snd_data) < THRESHOLD


def normalize(snd_data):
   "Average the volume out"
    MAXIMUM = 16384
    times = float(MAXIMUM)/max(abs(i) for i in snd_data)

    r = array('h')
    for i in snd_data:
        r.append(int(i*times))
    return r


def trim(snd_data):
   "Trim the blank spots at the start and end"
    def _trim(snd_data):
        snd_started = False
        r = array('h')

        for i in snd_data:
            if not snd_started and abs(i)>THRESHOLD:
                snd_started = True
                r.append(i)

            elif snd_started:
                r.append(i)
        return r

    # Trim to the left
    snd_data = _trim(snd_data)

    # Trim to the right
    snd_data.reverse()
    snd_data = _trim(snd_data)
    snd_data.reverse()
    return snd_data


def add_silence(snd_data, seconds):
   "Add silence to the start and end of 'snd_data' of length 'seconds' (float)"
    r = array('h', [0 for i in xrange(int(seconds*RATE))])
    r.extend(snd_data)
    r.extend([0 for i in xrange(int(seconds*RATE))])
    return r


def record():
   """g
    Record a word or words from the microphone and
    return the data as an array of signed shorts.

    Normalizes the audio, trims silence from the
    start and end, and pads with 0.5 seconds of
    blank sound to make sure VLC et al can play
    it without getting chopped off.
   """

    p = pyaudio.PyAudio()
    stream = p.open(format=FORMAT, channels=1, rate=RATE,
        input=True, output=True,
        frames_per_buffer=CHUNK_SIZE)

    num_silent = 0
    snd_started = False

    r = array('h')

    while 1:
        # little endian, signed short
        snd_data = array('h', stream.read(CHUNK_SIZE))
        if byteorder == 'big':
            snd_data.byteswap()
        r.extend(snd_data)

        silent = is_silent(snd_data)

        if silent and snd_started:
            num_silent += 1
        elif not silent and not snd_started:
            snd_started = True

        if snd_started and num_silent > 400:
            break

    sample_width = p.get_sample_size(FORMAT)
    stream.stop_stream()
    stream.close()
    p.terminate()

    #r = normalize(r)
    r = trim(r)
    r = add_silence(r, 0.5)

    return sample_width, r

@profile
def record_to_file(path):
   "Records from the microphone and outputs the resulting data to 'path'"
    sample_width, data = record()

    data = struct.pack('<' + ('h'*len(data)), *data)
    print(sys.getsizeof(data))
    wf = wave.open(path, 'wb')
    wf.setnchannels(1)
    wf.setsampwidth(sample_width)
    wf.setframerate(RATE)
    wf.writeframes(data)
    wf.close()
    del data
    gc.collect()


if __name__ == '__main__':
    count = 1
    h = hpy()
    f = open('heap.txt','w')
    objgraph.show_growth(limit=3)
    while(1):
        filename = 'demo' + str(count) + '.wav'
        print("please speak a word into the microphone")
        record_to_file(filename)
        print("done - result written to {0}".format(filename))
        cmd = 'cd"C:\\Users\\user\\Desktop\
audio" & ffmpeg\\bin\\ffmpeg -i {0} -acodec libmp3lame {1}.mp3'
.format(filename, filename)
       "subprocess.call(cmd, shell=True)"
        count += 1
        objgraph.show_growth()
        print h.heap()

下面是内存分析器模块输出的一个迭代:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Line #    Mem usage    Increment   Line Contents
================================================
   117     19.6 MiB      0.0 MiB   @profile
   118                             def record_to_file(path):
   119                                "Records from the microphone and outputs the resulting data to 'path'"
   120     22.4 MiB      2.8 MiB       sample_width, data = record()
   125     31.3 MiB      8.9 MiB       data = struct.pack('<' + ('h'*len(data)), *data)
   126     31.3 MiB      0.0 MiB       print(sys.getsizeof(data))
   127     31.3 MiB      0.0 MiB       wf = wave.open(path, 'wb')
   128     31.3 MiB      0.0 MiB       wf.setnchannels(1)
   129     31.3 MiB      0.0 MiB       wf.setsampwidth(sample_width)
   130     31.4 MiB      0.0 MiB       wf.setframerate(RATE)
   131     31.4 MiB      0.0 MiB       wf.writeframes(data)
   132     31.4 MiB      0.0 MiB       wf.close()
   133     30.4 MiB     -0.9 MiB       del data
   134     27.9 MiB     -2.5 MiB       gc.collect()

使用vs调试进程时,我看到系统内存中有大量"h"字符,这可能是发生泄漏的原因。任何帮助都将不胜感激


struct类保留项目缓存,以便更快地访问。清除结构缓存的唯一方法是调用struct._clearcache()方法。这里有一个正在使用的例子。

警告!这是一种_方法,可以随时更改。有关这些方法类型的说明,请参见此处。

这里和这里的python论坛上都有关于这个"内存泄漏"的讨论。