php与python 线程池多线程爬虫的例子
发布:smiling 来源: PHP粉丝网 添加日期:2018-10-31 22:56:37 浏览: 评论:0
php例子:
- <?php
- class Connect extends Worker //worker模式
- {
- public function __construct()
- {
- }
- public function getConnection()
- {
- if (!self::$ch)
- {
- self::$ch = curl_init();
- curl_setopt(self::$ch, CURLOPT_TIMEOUT, 2);
- curl_setopt(self::$ch, CURLOPT_RETURNTRANSFER, 1);
- curl_setopt(self::$ch, CURLOPT_HEADER, 0);
- curl_setopt(self::$ch, CURLOPT_NOSIGNAL, true);
- curl_setopt(self::$ch, CURLOPT_USERAGENT, "Firefox");
- curl_setopt(self::$ch, CURLOPT_FOLLOWLOCATION, 1);
- }
- /* do some exception/error stuff here maybe */
- return self::$ch;
- }
- public function closeConnection()
- {
- curl_close(self::$ch);
- }
- /**
- * Note that the link is stored statically, which for pthreads, means thread local
- * */
- protected static $ch;
- }
- class Query extends Threaded
- {
- public function __construct($url)
- {
- $this->url = $url;
- }
- public function run()
- {
- $ch = $this->worker->getConnection();
- curl_setopt($ch, CURLOPT_URL, $this->url);
- $page = curl_exec($ch);
- $info = curl_getinfo($ch);
- $error = curl_error($ch);
- $this->deal_data($this->url, $page, $info, $error);
- $this->result = $page;
- }
- function deal_data($url, $page, $info, $error)
- {
- $parts = explode(".", $url);
- $id = $parts[1];
- if ($info['http_code'] != 200)
- {
- $this->show_msg($id, $error);
- } else
- {
- $this->show_msg($id, "OK");
- }
- }
- function show_msg($id, $msg)
- {
- echo $id."\t$msg\n";
- }
- public function getResult()
- {
- return $this->result;
- }
- protected $url;
- protected $result;
- }
- function check_urls_multi_pthreads()
- {
- global $check_urls; //定义抓取的连接
- $check_urls = array( 'http://xxx.com' => "xx网",);
- $pool = new Pool(10, "Connect", array()); //建立10个线程池
- foreach ($check_urls as $url => $name)
- {
- $pool->submit(new Query($url));
- }
- $pool->shutdown();
- }
- check_urls_multi_pthreads();
python 多线程
- def handle(sid)://这个方法内执行爬虫数据处理
- pass
- class MyThread(Thread):
- """docstring for ClassName"""
- def __init__(self, sid):
- Thread.__init__(self)
- self.sid = sid
- def run():
- handle(self.sid)
- threads = []
- for i in xrange(1,11):
- t = MyThread(i)
- threads.append(t)
- t.start()
- for t in threads:
- t.join()
python 线程池爬虫
- from queue import Queue
- from threading import Thread, Lock
- import urllib.parse
- import socket
- import re
- import time
- seen_urls = set(['/'])
- lock = Lock()
- class Fetcher(Thread):
- def __init__(self, tasks):
- Thread.__init__(self)
- self.tasks = tasks
- self.daemon = True
- self.start()
- def run(self):
- while True:
- url = self.tasks.get()
- print(url)
- sock = socket.socket()
- sock.connect(('localhost', 3000))
- get = 'GET {} HTTP/1.0\r\nHost: localhost\r\n\r\n'.format(url)
- sock.send(get.encode('ascii'))
- response = b''
- chunk = sock.recv(4096)
- while chunk:
- response += chunk
- chunk = sock.recv(4096)
- links = self.parse_links(url, response)
- lock.acquire()
- for link in links.difference(seen_urls):
- self.tasks.put(link)
- seen_urls.update(links)
- lock.release()
- self.tasks.task_done()
- def parse_links(self, fetched_url, response):
- if not response:
- print('error: {}'.format(fetched_url))
- return set()
- if not self._is_html(response):
- return set()
- urls = set(re.findall(r'''(?i)href=["']?([^\s"'<>]+)''',
- self.body(response)))
- links = set()
- for url in urls:
- normalized = urllib.parse.urljoin(fetched_url, url)
- parts = urllib.parse.urlparse(normalized)
- if parts.scheme not in ('', 'http', 'https'):
- continue
- host, port = urllib.parse.splitport(parts.netloc)
- if host and host.lower() not in ('localhost'):
- continue
- defragmented, frag = urllib.parse.urldefrag(parts.path)
- links.add(defragmented)
- return links
- def body(self, response):
- body = response.split(b'\r\n\r\n', 1)[1]
- return body.decode('utf-8')
- def _is_html(self, response):
- head, body = response.split(b'\r\n\r\n', 1)
- headers = dict(h.split(': ') for h in head.decode().split('\r\n')[1:])
- return headers.get('Content-Type', '').startswith('text/html')
- class ThreadPool:
- def __init__(self, num_threads):
- self.tasks = Queue()
- for _ in range(num_threads):
- Fetcher(self.tasks)
- def add_task(self, url):
- self.tasks.put(url)
- def wait_completion(self):
- self.tasks.join()
- if __name__ == '__main__':
- start = time.time()
- pool = ThreadPool(4)
- pool.add_task("/")
- pool.wait_completion()
- print('{} URLs fetched in {:.1f} seconds'.format(len(seen_urls),time.time() - start))
Tags: python php线程池
相关文章
- ·python根据pid杀死相应进程的教程(2018-06-22)
- ·Python time时间模块用法详解(2018-09-18)
- ·使用Pdb调试Python方法总结(2018-09-18)
- ·Python守护进程和脚本单例运行详解(2018-11-01)
- ·python使用正则表达式替换匹配成功的组(2018-11-04)
- ·python+selenium实现京东自动登录及秒杀功能(2018-11-04)
- ·python进程与线程小结实例分析(2021-11-01)
推荐文章
热门文章
最新评论文章
- 写给考虑创业的年轻程序员(10)
- PHP新手上路(一)(7)
- 惹恼程序员的十件事(5)
- PHP邮件发送例子,已测试成功(5)
- 致初学者:PHP比ASP优秀的七个理由(4)
- PHP会被淘汰吗?(4)
- PHP新手上路(四)(4)
- 如何去学习PHP?(2)
- 简单入门级php分页代码(2)
- php中邮箱email 电话等格式的验证(2)