海运的博客

Pycurl底层使用libcurl，请先确定Libcurl是否已支持异步DNS解析c-ares，如不支持可升级libcurl支持异步DNS解析c-ares。
其实Libcurl更新支持为异步DNS如果已安装pycurl不用重新安装Pycurl，见https://www.haiyun.me/archives/1070.html
通过pip安装：

export PATH=/usr/local/curl/bin/:$PATH
export LD_LIBRARY_PATH="/usr/local/curl/lib/"
export PYCURL_SSL_LIBRARY=openssl
pip install pycurl

下载pycurl源码包安装：

wget http://pycurl.sourceforge.net/download/pycurl-7.19.5.1.tar.gz
tar zxvf pycurl-7.19.5.1.tar.gz 
cd pycurl-7.19.5.1
export LD_LIBRARY_PATH="/usr/local/curl/lib/"
python setup.py install --curl-config=/usr/local/curl/bin/curl-config --with-ssl

检查pycurl是否已支持异步DNS解析c-ares：

>>> import pycurl
>>> pycurl.version
'PycURL/7.19.5.1 libcurl/7.40.0 OpenSSL/1.0.1e zlib/1.2.7 c-ares/1.10.0'

安装pycurl后使用时遇到的一些错误：

pycurl.so: undefined symbol: CRYPTO_num_locks

原因：libcurl安装时--with-ssl支持

libcurl link-time ssl backend (openssl) is different from compile-time ssl backend (none/other)

原因：libcurl和pycurl编译时ssl后端不一致，调整见上和libcurl安装

pycurl: libcurl link-time version (7.19.7) is older than compile-time version (7.4.0)

原因：编译pycurl时使用的编译的libcurl动态库，不过现在pycurl现在加载的是系统自带的版本较旧的动态库，解决将编译的libcurl动态库添加到系统动态库，见ldconfig

Select模式，类似于php multi curl异步并发，连接数不能太多：

#!/usr/bin/python
# -*- coding: utf-8 -*-

import sys
import pycurl
import cStringIO

#最大连接数
num_conn = 20

queue = []
urls = ['https://www.haiyun.me/'] * 10000
for url in urls:
  queue.append(url)

num_urls = len(queue)
num_conn = min(num_conn, num_urls)
print ('----- Getting', num_urls, 'Max conn', num_conn,
       'connections -----')

m = pycurl.CurlMulti()
#初始化handle，可复用
m.handles = []
for i in range(num_conn):
  c = pycurl.Curl()
  c.body = cStringIO.StringIO()
  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)


freelist = m.handles[:]
num_processed = 0
#主循环开始
while num_processed < num_urls:

    #添加请求URL
    while queue and freelist:
      url = queue.pop()
      c = freelist.pop()
      c.setopt(pycurl.URL, url)
      c.setopt(pycurl.WRITEFUNCTION, c.body.write)
      m.add_handle(c)
      c.url = url
      #print url

    #执行请求
    while 1:
      (ret, num_handles) = m.perform()
      if ret != pycurl.E_CALL_MULTI_PERFORM:
        break

    #阻塞一会直到有连接完成
    m.select(1.0)

    #读取完成的连接
    while 1:
      (num_q, ok_list, err_list) = m.info_read()
      for c in ok_list:
        m.remove_handle(c)
        #print c.body.getvalue()
        freelist.append(c)

      for (c, errno, errmsg) in err_list:
        m.remove_handle(c)
        print ('Failed: ', c.url, errno, errmsg)
        freelist.append(c)
      num_processed = num_processed + len(ok_list) + len(err_list)
      if num_q == 0:
        break

for c in m.handles:
  c.fp = None
  c.close()
m.close()

epoll模式，php mult curl不支持此模式，tornado基于pycurl multi_socket_action封装的异步http client，每个client实例维护一个ioloop：

from tornado.httpclient import AsyncHTTPClient
from tornado.ioloop import IOLoop
count = 10000
done = 0
def handle_request(response):
  global done
  done += 1
  if (done == count):
    #结束循环
    IOLoop.instance().stop()

  if response.error:
    print "Error:", response.error
  #else:
    #print response.body
#默认client是基于ioloop实现的，配置使用Pycurl
AsyncHTTPClient.configure("tornado.curl_httpclient.CurlAsyncHTTPClient",max_clients=20)
http_client = AsyncHTTPClient()
for i in range(count):
  http_client.fetch("https://www.haiyun.me/", handle_request)
#死循环
IOLoop.instance().start()      

基于epoll的multi curl在lan环境下效果不如select，因为所有Socket都在活跃状态，所有的callback都被唤醒，会导致资源的竞争。既然都是要处理所有的Socket，直接遍历是最简单最有效的方式.
为更好的性能建议libcurl/pycurl开启异步DNS解析。