Scrapy框架将快速和专业地解决您的问题。它还将缓存所有请求，以便稍后可以重新运行失败的请求。

将该脚本保存为quotes_spider.py。

# quote_spiders.py
import json
import string
import scrapy
from scrapy.crawler import CrawlerProcess
from scrapy.item import Item, Field

class TextCleaningPipeline(object):
    def _clean_text(self, text):
        text = text.replace('“', '').replace('”', '')
        table = str.maketrans({key: None for key in string.punctuation})
        clean_text = text.translate(table)
        return clean_text.lower()

    def process_item(self, item, spider):
        item['text'] = self._clean_text(item['text'])
        return item

class JsonWriterPipeline(object):
    def open_spider(self, spider):
        self.file = open(spider.settings['JSON_FILE'], 'a')

    def close_spider(self, spider):
        self.file.close()

    def process_item(self, item, spider):
        line = json.dumps(dict(item)) + "\n"
        self.file.write(line)
        return item

class QuoteItem(Item):
    text = Field()
    author = Field()
    tags = Field()
    spider = Field()

class QuoteSpider(scrapy.Spider):
    name = "quotes"

    def start_requests(self):
        urls = [
            'http://quotes.toscrape.com/page/1/',
            'http://quotes.toscrape.com/page/2/',
            # ...
        ]
        for url in urls:
            yield scrapy.Request(url=url, callback=self.parse)

    def parse(self, response):
        for quote in response.css('div.quote'):
            item = QuoteItem()
            item['text'] = quote.css('span.text::text').get()
            item['author'] = quote.css('small.author::text').get()
            item['tags'] = quote.css('div.tags a.tag::text').getall()
            item['spider'] = self.name
            yield item

if __name__ == '__main__':
    settings = dict()
    settings['USER_AGENT'] = 'Mozilla/4.0 (compatible; MSIE 7.0; Windows NT 5.1)'
    settings['HTTPCACHE_ENABLED'] = True
    settings['CONCURRENT_REQUESTS'] = 20
    settings['CONCURRENT_REQUESTS_PER_DOMAIN'] = 20
    settings['JSON_FILE'] = 'items.jl'
    settings['ITEM_PIPELINES'] = dict()
    settings['ITEM_PIPELINES']['__main__.TextCleaningPipeline'] = 800
    settings['ITEM_PIPELINES']['__main__.JsonWriterPipeline'] = 801

    process = CrawlerProcess(settings=settings)
    process.crawl(QuoteSpider)
    process.start()

紧随其后的是

$ pip install Scrapy
$ python quote_spiders.py 

为了微调scraper，相应地调整CONCURRENT_REQUESTS和CONCURRENT_REQUESTS_PER_DOMAIN设置。

对于您的情况，线程可能会做的技巧，因为您可能会花费大部分时间等待响应。标准库中有一些有用的模块，如Queue，可能会有所帮助。

我以前做过类似的并行下载文件的事情，对我来说已经足够好了，但它不是你所说的那种规模。

如果您的任务对cpu的限制更大，您可能需要考虑multiprocessing模块，它将允许您利用更多的cpu /内核/线程(更多的进程不会相互阻塞，因为锁定是每个进程)

如果您希望获得尽可能好的性能，您可能会考虑使用异步I/O而不是线程。与成千上万个操作系统线程相关的开销是不小的，Python解释器内的上下文切换甚至增加了更多的开销。线程当然可以完成工作，但我怀疑异步路由将提供更好的整体性能。

具体来说，我建议使用Twisted库中的异步web客户端(http://www.twistedmatrix.com)。它有一个公认的陡峭的学习曲线，但一旦你很好地掌握了Twisted的异步编程风格，它就很容易使用。

Twisted的异步web客户端API的HowTo可以在以下地址找到:

http://twistedmatrix.com/documents/current/web/howto/client.html

我知道这是一个老问题，但在Python 3.7中，您可以使用asyncio和aiohttp来做到这一点。

import asyncio
import aiohttp
from aiohttp import ClientSession, ClientConnectorError

async def fetch_html(url: str, session: ClientSession, **kwargs) -> tuple:
    try:
        resp = await session.request(method="GET", url=url, **kwargs)
    except ClientConnectorError:
        return (url, 404)
    return (url, resp.status)

async def make_requests(urls: set, **kwargs) -> None:
    async with ClientSession() as session:
        tasks = []
        for url in urls:
            tasks.append(
                fetch_html(url=url, session=session, **kwargs)
            )
        results = await asyncio.gather(*tasks)

    for result in results:
        print(f'{result[1]} - {str(result[0])}')

if __name__ == "__main__":
    import pathlib
    import sys

    assert sys.version_info >= (3, 7), "Script requires Python 3.7+."
    here = pathlib.Path(__file__).parent

    with open(here.joinpath("urls.txt")) as infile:
        urls = set(map(str.strip, infile))

    asyncio.run(make_requests(urls=urls))

你可以阅读更多关于它的内容，并在这里看到一个例子。

线程绝对不是这里的答案。它们将提供进程和内核瓶颈，以及吞吐量限制，如果总体目标是“最快的方式”，这些限制是不可接受的。

稍微扭曲一点，它的异步HTTP客户端会给你更好的结果。

下面是一个“异步”解决方案，它不使用asyncio，而是使用asyncio使用的低级机制(在Linux上):select()。(或者asyncio可能使用poll或epoll，但这是类似的原理。)

它是对PyCurl示例的稍微修改版本。

(为了简单起见，它多次请求相同的URL，但您可以轻松地修改它以检索一系列不同的URL。)

(另一个轻微的修改可以使这个检索相同的URL作为一个无限循环。提示:将while url和句柄更改为while句柄，将while nprocessed<nurls更改为while 1。)

import pycurl,io,gzip,signal, time, random
signal.signal(signal.SIGPIPE, signal.SIG_IGN)  # NOTE! We should ignore SIGPIPE when using pycurl.NOSIGNAL - see the libcurl tutorial for more info

NCONNS = 2  # Number of concurrent GET requests
url    = 'example.com'
urls   = [url for i in range(0x7*NCONNS)]  # Copy the same URL over and over

# Check args
nurls  = len(urls)
NCONNS = min(NCONNS, nurls)
print("\x1b[32m%s \x1b[0m(compiled against 0x%x)" % (pycurl.version, pycurl.COMPILE_LIBCURL_VERSION_NUM))
print(f'\x1b[37m{nurls} \x1b[91m@ \x1b[92m{NCONNS}\x1b[0m')

# Pre-allocate a list of curl objects
m         = pycurl.CurlMulti()
m.handles = []
for i in range(NCONNS):
  c = pycurl.Curl()
  c.setopt(pycurl.FOLLOWLOCATION,  1)
  c.setopt(pycurl.MAXREDIRS,       5)
  c.setopt(pycurl.CONNECTTIMEOUT,  30)
  c.setopt(pycurl.TIMEOUT,         300)
  c.setopt(pycurl.NOSIGNAL,        1)
  m.handles.append(c)

handles    = m.handles  # MUST make a copy?!
nprocessed = 0
while nprocessed<nurls:

  while urls and handles:  # If there is an url to process and a free curl object, add to multi stack
    url   = urls.pop(0)
    c     = handles.pop()
    c.buf = io.BytesIO()
    c.url = url  # store some info
    c.t0  = time.perf_counter()
    c.setopt(pycurl.URL,        c.url)
    c.setopt(pycurl.WRITEDATA,  c.buf)
    c.setopt(pycurl.HTTPHEADER, [f'user-agent: {random.randint(0,(1<<256)-1):x}', 'accept-encoding: gzip, deflate', 'connection: keep-alive', 'keep-alive: timeout=10, max=1000'])
    m.add_handle(c)

  while 1:  # Run the internal curl state machine for the multi stack
    ret, num_handles = m.perform()
    if ret!=pycurl.E_CALL_MULTI_PERFORM:  break

  while 1:  # Check for curl objects which have terminated, and add them to the handles
    nq, ok_list, ko_list = m.info_read()
    for c in ok_list:
      m.remove_handle(c)
      t1 = time.perf_counter()
      reply = gzip.decompress(c.buf.getvalue())
      print(f'\x1b[33mGET  \x1b[32m{t1-c.t0:.3f}  \x1b[37m{len(reply):9,}  \x1b[0m{reply[:32]}...')  # \x1b[35m{psutil.Process(os.getpid()).memory_info().rss:,} \x1b[0mbytes')
      handles.append(c)
    for c, errno, errmsg in ko_list:
      m.remove_handle(c)
      print('\x1b[31mFAIL {c.url} {errno} {errmsg}')
      handles.append(c)
    nprocessed = nprocessed + len(ok_list) + len(ko_list)
    if nq==0: break

  m.select(1.0)  # Currently no more I/O is pending, could do something in the meantime (display a progress bar, etc.). We just call select() to sleep until some more data is available.

for c in m.handles:
  c.close()
m.close()