python支持断点续传的多线程下载示例
2019/7/13 21:57:18
本文主要是介绍python支持断点续传的多线程下载示例,对大家解决编程问题具有一定的参考价值,需要的程序猿们随着小编来一起学习吧!
#! /usr/bin/env python
#coding=utf-8
from __future__ import unicode_literals
from multiprocessing.dummy import Pool as ThreadPool
import threading
import os
import sys
import cPickle
from collections import namedtuple
import urllib2
from urlparse import urlsplit
import time
# global lock
lock = threading.Lock()
# default parameters
defaults = dict(thread_count=10,
buffer_size=10*1024,
block_size=1000*1024)
def progress(percent, width=50):
print "%s %d%%\r" % (('%%-%ds' % width) % (width * percent / 100 * '='), percent),
if percent >= 100:
print
sys.stdout.flush()
def write_data(filepath, data):
with open(filepath, 'wb') as output:
cPickle.dump(data, output)
def read_data(filepath):
with open(filepath, 'rb') as output:
return cPickle.load(output)
FileInfo = namedtuple('FileInfo', 'url name size lastmodified')
def get_file_info(url):
class HeadRequest(urllib2.Request):
def get_method(self):
return "HEAD"
res = urllib2.urlopen(HeadRequest(url))
res.read()
headers = dict(res.headers)
size = int(headers.get('content-length', 0))
lastmodified = headers.get('last-modified', '')
name = None
if headers.has_key('content-disposition'):
name = headers['content-disposition'].split('filename=')[1]
if name[0] == '"' or name[0] == "'":
name = name[1:-1]
else:
name = os.path.basename(urlsplit(url)[2])
return FileInfo(url, name, size, lastmodified)
def download(url, output,
thread_count = defaults['thread_count'],
buffer_size = defaults['buffer_size'],
block_size = defaults['block_size']):
# get latest file info
file_info = get_file_info(url)
# init path
if output is None:
output = file_info.name
workpath = '%s.ing' % output
infopath = '%s.inf' % output
# split file to blocks. every block is a array [start, offset, end],
# then each greenlet download filepart according to a block, and
# update the block' offset.
blocks = []
if os.path.exists(infopath):
# load blocks
_x, blocks = read_data(infopath)
if (_x.url != url or
_x.name != file_info.name or
_x.lastmodified != file_info.lastmodified):
blocks = []
if len(blocks) == 0:
# set blocks
if block_size > file_info.size:
blocks = [[0, 0, file_info.size]]
else:
block_count, remain = divmod(file_info.size, block_size)
blocks = [[i*block_size, i*block_size, (i+1)*block_size-1] for i in range(block_count)]
blocks[-1][-1] += remain
# create new blank workpath
with open(workpath, 'wb') as fobj:
fobj.write('')
print 'Downloading %s' % url
# start monitor
threading.Thread(target=_monitor, args=(infopath, file_info, blocks)).start()
# start downloading
with open(workpath, 'rb+') as fobj:
args = [(url, blocks[i], fobj, buffer_size) for i in range(len(blocks)) if blocks[i][1] < blocks[i][2]]
if thread_count > len(args):
thread_count = len(args)
pool = ThreadPool(thread_count)
pool.map(_worker, args)
pool.close()
pool.join()
# rename workpath to output
if os.path.exists(output):
os.remove(output)
os.rename(workpath, output)
# delete infopath
if os.path.exists(infopath):
os.remove(infopath)
assert all([block[1]>=block[2] for block in blocks]) is True
def _worker((url, block, fobj, buffer_size)):
req = urllib2.Request(url)
req.headers['Range'] = 'bytes=%s-%s' % (block[1], block[2])
res = urllib2.urlopen(req)
while 1:
chunk = res.read(buffer_size)
if not chunk:
break
with lock:
fobj.seek(block[1])
fobj.write(chunk)
block[1] += len(chunk)
def _monitor(infopath, file_info, blocks):
while 1:
with lock:
percent = sum([block[1] - block[0] for block in blocks]) * 100 / file_info.size
progress(percent)
if percent >= 100:
break
write_data(infopath, (file_info, blocks))
time.sleep(2)
if __name__ == '__main__':
import argparse
parser = argparse.ArgumentParser(description='Download file by multi-threads.')
parser.add_argument('url', type=str, help='url of the download file')
parser.add_argument('-o', type=str, default=None, dest="output", help='output file')
parser.add_argument('-t', type=int, default=defaults['thread_count'], dest="thread_count", help='thread counts to downloading')
parser.add_argument('-b', type=int, default=defaults['buffer_size'], dest="buffer_size", help='buffer size')
parser.add_argument('-s', type=int, default=defaults['block_size'], dest="block_size", help='block size')
argv = sys.argv[1:]
if len(argv) == 0:
argv = ['https://eyes.nasa.gov/eyesproduct/EYES/os/win']
args = parser.parse_args(argv)
start_time = time.time()
download(args.url, args.output, args.thread_count, args.buffer_size, args.block_size)
print 'times: %ds' % int(time.time()-start_time)
这篇关于python支持断点续传的多线程下载示例的文章就介绍到这儿,希望我们推荐的文章对大家有所帮助,也希望大家多多支持为之网!
- 2024-11-21Python编程基础教程
- 2024-11-20Python编程基础与实践
- 2024-11-20Python编程基础与高级应用
- 2024-11-19Python 基础编程教程
- 2024-11-19Python基础入门教程
- 2024-11-17在FastAPI项目中添加一个生产级别的数据库——本地环境搭建指南
- 2024-11-16`PyMuPDF4LLM`:提取PDF数据的神器
- 2024-11-16四种数据科学Web界面框架快速对比:Rio、Reflex、Streamlit和Plotly Dash
- 2024-11-14获取参数学习:Python编程入门教程
- 2024-11-14Python编程基础入门