问题
I'm in the process of improving an open-source music database, which reads songs in from my collection and stores them to an SQLite database. In turn, I'm able to leverage the database to find duplicates, run queries on my collection, and (if I so desired) find duplicate songs in the collection. To read the metadata from the music files, I'm leveraging the Mutagen library, and to store the metadata, I'm using SQLite3.
I wanted to test the code I had authored on a sizable collection, so I contacted fellow students and family, and came across a total test size of about 90,000. This also consists of mixed information - songs are of either .mp3, .ogg, or .flac format.
My main issue is with speed - the code that I have works, but it is unsuitably slow. In its current state, it runs through the test size in about 35:45. My main question: what can I do to improve the performance of this block of code? I think that it is related directly to something within either Mutagen directly or SQLite3, although I'm open to suggestions on what would be an ideal approach to improve efficiency from here.
I had gone through two iterations of improving this critical part of code. The first improvement led to a reduced runtime of 21:30, but that's still horrible. I decided to refactor the code to reduce the number of function calls made, and an attempt to improve performance. The result, however, is a regression of performance, but a huge reduction in the number of function calls made - the second batch runs for close to 51:51, which is simply unacceptable.
What follows in terms of code is for both the "improved" runtime and refactored set. Also attached are separate profiles for each piece of code.
Block 1: "Improved" Runtime
def walk(self, d):
'''Walk down the file structure iteratively, gathering file names to be read in.'''
d = os.path.abspath(d)
dirpath = os.walk(d)
for folder in dirpath:
for f in folder[2]: # for each file in the folder...
supported = 'mp3', 'ogg', 'flac'
if f.split('.')[-1] in supported:
try:
self.parse(os.path.join(folder[0], f))
if self.filecount == 2000 or self.leftover:
self.filecount = 0
try:
self.db.execute_batch_insert_statement(u"INSERT INTO song VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)", self.buf)
except Exception, e:
print e.__unicode__()
finally:
del self.buf
self.buf = [] # wipe the buffers clean so we can repeat a batch parse again.
except Exception, e:
print e.__unicode__()
try:
self.db.execute_batch_insert_statement(u"INSERT INTO song VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)", self.buf)
except Exception, e:
print e.__unicode__()
finally:
del self.buf
self.buf = [] # wipe the buffers clean so we can repeat a batch parse again.
def parse(self, filename):
'''Process and parse the music file to extract desired information.
It may be the case that, in the future, we require more information from a song than is provided at this time.
Examine all tags that can be retrieved from a mutagen.File object, and adjust the database's schema accordingly.'''
if ".ogg" in filename:
song = OggVorbis(filename)
elif ".mp3" in filename:
song = MP3(filename)
elif ".flac" in filename:
song = FLAC(filename)
else:
raise InvalidSongException(u"Song is not supported by K'atun at this time.")
filename = u'filename'
#song = mutagen.File(filename, easy=True)
artist, title, genre, track, album, bitrate, year, month = '', '', '', '', '', '', '', ''
try:
artist = song['artist'][0]
title = song['title'][0]
except Exception:
raise InvalidSongException(u"Cannot read " + filename + ": missing critical song information.")
if 'genre' in song:
genre = song['genre'][0]
else:
genre = u'Unknown'
if 'tracknumber' in song:
track = song['tracknumber'][0]
else:
track = 0
if 'album' in song:
album = song['album'][0]
else:
album = u'Unknown'
if 'date' in song:
year = song['date'][0]
else:
year = 'Unknown'
try:
bitrate = int(song.info.bitrate)
except AttributeError: # Likely due to us messing with FLAC
bitrate = 999999 # Set to a special flag value, to indicate that this is a lossless file.
self.buf.append((filename, artist, filename.split('.')[-1], title, genre, track, album, bitrate, year, time.time()))
self.filecount += 1
Sat Dec 24 21:24:23 2011 modified.dat
70626027 function calls (70576436 primitive calls) in 1290.127 CPU seconds
Ordered by: cumulative time
List reduced from 666 to 28 due to restriction <28>
ncalls tottime percall cumtime percall filename:lineno(function)
1 0.033 0.033 1290.127 1290.127 parser.py:6(<module>)
1 0.000 0.000 1290.090 1290.090 parser.py:90(main)
1 0.000 0.000 1286.493 1286.493 parser.py:24(__init__)
1 1.826 1.826 1286.335 1286.335 parser.py:35(walk)
90744 2.376 0.000 1264.788 0.014 parser.py:55(parse)
90744 11.840 0.000 1250.401 0.014 lib/mutagen/__init__.py:158(File)
376019 613.881 0.002 613.881 0.002 {method 'seek' of 'file' objects}
90744 1.231 0.000 580.143 0.006 /usr/lib/pymodules/python2.7/mutagen/apev2.py:458(score)
671848 530.346 0.001 530.346 0.001 {method 'read' of 'file' objects}
90742 0.846 0.000 242.337 0.003 /usr/lib/pymodules/python2.7/mutagen/__init__.py:68(__init__)
63944 2.471 0.000 177.050 0.003 /usr/lib/pymodules/python2.7/mutagen/id3.py:1973(load)
63944 0.526 0.000 119.326 0.002 /usr/lib/pymodules/python2.7/mutagen/easyid3.py:161(__init__)
63944 4.649 0.000 118.077 0.002 /usr/lib/pymodules/python2.7/mutagen/id3.py:89(load)
26782 1.073 0.000 64.435 0.002 /usr/lib/pymodules/python2.7/mutagen/ogg.py:434(load)
127531 0.464 0.000 59.314 0.000 /usr/lib/pymodules/python2.7/mutagen/id3.py:76(__fullread)
63944 1.078 0.000 54.060 0.001 /usr/lib/pymodules/python2.7/mutagen/mp3.py:68(__init__)
26782 0.638 0.000 53.613 0.002 /usr/lib/pymodules/python2.7/mutagen/ogg.py:379(find_last)
66487 3.167 0.000 50.136 0.001 /usr/lib/pymodules/python2.7/mutagen/mp3.py:106(__try)
855079 6.415 0.000 33.237 0.000 /usr/lib/pymodules/python2.7/mutagen/id3.py:279(__read_frames)
816987 0.904 0.000 24.491 0.000 /usr/lib/pymodules/python2.7/mutagen/id3.py:321(__load_framedata)
816987 2.805 0.000 23.587 0.000 /usr/lib/pymodules/python2.7/mutagen/id3.py:1023(fromData)
60803/11257 0.370 0.000 19.036 0.002 /usr/lib/python2.7/os.py:209(walk)
11256 14.651 0.001 14.651 0.001 {posix.listdir}
816973 3.265 0.000 13.140 0.000 /usr/lib/pymodules/python2.7/mutagen/id3.py:996(_readData)
879103 4.936 0.000 11.473 0.000 /usr/lib/pymodules/python2.7/mutagen/id3.py:964(__init__)
872462 0.967 0.000 11.336 0.000 /usr/lib/pymodules/python2.7/mutagen/__init__.py:78(__getitem__)
63944 1.969 0.000 10.871 0.000 /usr/lib/pymodules/python2.7/mutagen/id3.py:443(update_to_v24)
619380 1.396 0.000 8.521 0.000 /usr/lib/pymodules/python2.7/mutagen/easyid3.py:175(__getitem__)
Block 2: Refactored Code
def walk(self, d):
'''Walk down the file structure iteratively, gathering file names to be read in.'''
d = os.path.abspath(d)
dirpath = os.walk(d)
parsecount = 0
start = time.time()
for folder in dirpath:
for f in folder[2]: # for each file in the folder...
filetype = f.split('.')[-1].lower()
if filetype == 'mp3':
try:
self.read_mp3(os.path.join(folder[0], f).decode('utf_8'))
except Exception, e:
print e.__unicode__()
elif filetype == 'ogg':
try:
self.read_vorbis(os.path.join(folder[0], f).decode('utf_8'))
except Exception, e:
print e.__unicode__()
elif filetype == 'flac':
try:
self.read_flac(os.path.join(folder[0], f).decode('utf_8'))
except Exception, e:
print e.__unicode__()
else:
continue
if self.filecount == 2000 or self.leftover:
self.filecount = 0
print "Time differential: %1.4f s" % (time.time() - start)
self.batch_commit()
try:
print "Wrapping up"
self.batch_commit()
except Exception, e:
print e.__unicode__()
finally:
print "Elapsed time: " + str(time.time()-start)
def batch_commit(self):
'''Insert new values into the database in large quantities.'''
self.db.execute_batch_insert_statement(u"INSERT INTO song VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)", self.buf)
self.buf = []
def read_mp3(self, filename):
'''Read and extract an MP3 file's tags. This makes use of the ID3 standard, not the easy ID3 tag system.'''
artist, title, genre, track, album, bitrate, year = '', '', '', '', '', 0, ''
song = MP3(filename)
keys = song.keys()
try:
artist = song['TPE1'].__unicode__()
title = song['TIT2'].__unicode__()
except KeyError, e:
raise InvalidSongException(u"Cannot read " + filename + ": missing critical song information.")
genre = song['TCON'].__unicode__() if "TCON" in keys else u'Unknown'
track = song['TRCK'].__unicode__() if "TRCK" in keys else u'0'
album = song['TALB'].__unicode__() if "TALB" in keys else u'Unknown'
bitrate = int(song.info.bitrate)
year = song['TDRC'].__unicode__() if "TDRC" in keys else u'Unknown'
self.buf.append((filename, artist, filename.split('.')[-1], title, genre, track, album, bitrate, year, time.time()))
self.filecount += 1
def read_vorbis(self, filename):
'''Read and extract an Ogg Vorbis file's tags.'''
song = OggVorbis(filename)
artist, title, genre, track, album, bitrate, year = '', '', '', '', '', 0, ''
try:
artist = song['artist'][0]
title = song['title'][0]
except KeyError, e:
raise InvalidSongException(u"Cannot read " + filename + ": missing critical song information.")
genre = song['genre'][0] if genre else u'Unknown'
track = song['tracknumber'][0] if 'tracknumber' in song else u'0'
album = song['album'][0] if 'album' in song else u'Unknown'
bitrate = int(song.info.bitrate)
year = song['date'][0] if 'date' in song else 'Unknown'
self.buf.append((filename, artist, filename.split('.')[-1], title, genre, track, album, bitrate, year, time.time()))
self.filecount += 1
def read_flac(self, filename):
'''Read and extract a FLAC file's tags.'''
song = FLAC(filename)
artist, title, genre, track, album, bitrate, year = '', '', '', '', '', 0, ''
try:
artist = song['artist'][0]
title = song['title'][0]
except KeyError, e:
raise InvalidSongException(u"Cannot read " + filename + ": missing critical song information.")
genre = song['genre'][0] if genre else u'Unknown'
track = song['tracknumber'][0] if 'tracknumber' in song else u'0'
album = song['album'][0] if 'album' in song else u'Unknown'
bitrate = 999999 # Special flag for K'atun; will know that this is a lossless file
year = song['date'][0] if 'date' in song else 'Unknown'
self.buf.append((filename, artist, filename.split('.')[-1], title, genre, track, album, bitrate, year, time.time()))
self.filecount += 1
Mon Dec 26 03:22:34 2011 refactored.dat
59939763 function calls (59890172 primitive calls) in 3111.490 CPU seconds
Ordered by: cumulative time
List reduced from 559 to 28 due to restriction <28>
ncalls tottime percall cumtime percall filename:lineno(function)
1 0.001 0.001 3111.490 3111.490 parser.py:6(<module>)
1 0.000 0.000 3111.477 3111.477 parser.py:138(main)
1 0.000 0.000 3108.242 3108.242 parser.py:27(__init__)
1 1.760 1.760 3108.062 3108.062 parser.py:40(walk)
46 0.103 0.002 2220.618 48.274 parser.py:78(batch_commit)
46 0.002 0.000 2220.515 48.272 db_backend.py:127(execute_batch_insert_statement)
46 2184.900 47.498 2184.900 47.498 {method 'executemany' of 'sqlite3.Connection' objects}
90747 0.515 0.000 845.343 0.009 /usr/lib/pymodules/python2.7/mutagen/__init__.py:68(__init__)
426651 640.459 0.002 640.459 0.002 {method 'read' of 'file' objects}
63945 1.847 0.000 582.267 0.009 parser.py:83(read_mp3)
63945 2.372 0.000 577.245 0.009 /usr/lib/pymodules/python2.7/mutagen/id3.py:1973(load)
63945 0.307 0.000 514.927 0.008 /usr/lib/pymodules/python2.7/mutagen/id3.py:72(__init__)
63945 0.256 0.000 514.620 0.008 /usr/lib/pymodules/python2.7/mutagen/_util.py:103(__init__)
63945 0.225 0.000 514.363 0.008 /usr/lib/pymodules/python2.7/mutagen/__init__.py:35(__init__)
63945 4.188 0.000 514.139 0.008 /usr/lib/pymodules/python2.7/mutagen/id3.py:89(load)
127533 0.802 0.000 455.713 0.004 /usr/lib/pymodules/python2.7/mutagen/id3.py:76(__fullread)
63945 1.029 0.000 432.574 0.007 /usr/lib/pymodules/python2.7/mutagen/id3.py:202(__load_header)
26786 0.504 0.000 270.216 0.010 parser.py:102(read_vorbis)
26786 1.095 0.000 267.578 0.010 /usr/lib/pymodules/python2.7/mutagen/ogg.py:434(load)
26782 0.627 0.000 143.492 0.005 /usr/lib/pymodules/python2.7/mutagen/ogg.py:379(find_last)
221337 121.448 0.001 121.448 0.001 {method 'seek' of 'file' objects}
97603 1.797 0.000 118.799 0.001 /usr/lib/pymodules/python2.7/mutagen/ogg.py:66(__init__)
26786 0.342 0.000 114.656 0.004 lib/mutagen/oggvorbis.py:40(__init__)
63945 0.646 0.000 58.809 0.001 lib/mutagen/mp3.py:68(__init__)
66480 3.377 0.000 57.489 0.001 lib/mutagen/mp3.py:106(__try)
47 35.609 0.758 35.609 0.758 {method 'commit' of 'sqlite3.Connection' objects}
855108 6.184 0.000 32.181 0.000 /usr/lib/pymodules/python2.7/mutagen/id3.py:279(__read_frames)
60803/11257 0.385 0.000 31.885 0.003 /usr/lib/python2.7/os.py:209(walk)
回答1:
If you feel up to it, you could dive into threading :) - It is quite easy in python.
Create two Queue objects, one for IN and one for OUT and in between multiple threads scan the files. (Worker threads, and Queue's are threadsafe)
It should be only ~80loc more code, and you probably can keep the current functions, just wrap em in appropriate classes. (to make "Thread" take it)
But 20min for 90k songs does not seem to be completely out of line. It requires a lot of random access on the disk, and seeking is slow (10ms). so 90,000files, one seek per file is 15min already.
来源：https://stackoverflow.com/questions/8640701/optimizing-python-code-using-sqlite3-mutagen