gevent-twisted-多线程谁更快?

前言

标题有点唬人，以前了解过研究 gevent，twisted，scrapy（基于 twisted）。最近有个想法：这些东西比如做爬虫，谁的效率更好呢？我就写了以下程序（附件）测试然后用 timeit（跑 3 次，每次 10 遍，时间有限）看效果

原理：

为了防止远程网络的问题，从一个网站爬下网页代码（html），页面下载本地放在了我的本机（gentoo+apache）
然后爬虫去分析这些页面上面的链接（开始是主页），再挖掘其他页面，抓取页面关键字（我这里就是个‘py’）程序打包 Crawler.tar.bz2

先看代码树：

dongwm@localhost ~ $ tree Crawler/
Crawler/
├── common_Crawler.py  #标准爬虫，里面只是多线程编程，抓取分析类在common.py
├── common.py  #共用函数，里面只是抓取页面分析页面关键字
├── common.pyc #你懂得
├── Crawler #scrapy和django框架差不多的用法
│   ├── __init__.py
│   ├── __init__.pyc
│   ├── items.py #不需要利用，默认
│   ├── pipelines.py
│   ├── settings.py
│   ├── settings.pyc
│   └── spiders #抓取脚本文件夹
│       ├── __init__.py
│       ├── __init__.pyc
│       ├── spiders.py #我做的分析页面，这个和多线程/gevent调用的抓取分析类不同，我使用了内置方法（大家可以修改共用函数改成scrapy的方式，这样三种效果就更准确了）
│       └── spiders.pyc
├── gevent_Crawler.py #gevent版本爬虫，效果和标准版一样，抓取分析类也是common.py 保证其他环节相同，只是一个多线程，一个用协程
├── scrapy.cfg
└── scrapy_Crawler.py #因为scrapy使用是命令行，我用subproess封装了命令，然后使用timeit计算效果

2 directories, 16 files

实验前准备：

停掉我本机使用的耗费资源的进程 firefox，vmware，compiz 等，直到负载保持一个相对拨波动平衡

测试程序：

common.py

#!/usr/bin/python
#coding=utf-8

# Version 1 by Dongwm 2013/01/10
# 脚本作用：多线程抓取
# 方式： lxml + xpath + requests

import requests
from  cStringIO import StringIO
from lxml import etree

class Crawler(object):

    def __init__(self, app):
        self.deep = 2  #指定网页的抓取深度        
        self.url = '' #指定网站地址
        self.key = 'by' #搜索这个词
        self.tp = app #连接池回调实例
        self.visitedUrl = [] #抓取的网页放入列表,防止重复抓取

    def _hasCrawler(self, url): 
        '''判断是否已经抓取过这个页面'''
        return (True if url in self.visitedUrl else False)

    def getPageSource(self, url, key, deep): 
        ''' 抓取页面,分析,入库.
        '''
        if self._hasCrawler(url): #发现重复直接return
            return 
        else:
            self.visitedUrl.append(url) #发现新地址假如到这个列
        r = requests.get('http://localhost/%s' % url)
        encoding = r.encoding #判断页面的编码
        result = r.text.encode('utf-8').decode(encoding)
        #f = StringIO(r.text.encode('utf-8'))
        try:  
            self._xpath(url, result, ['a'], unicode(key, 'utf8'), deep) #分析页面中的连接地址,以及它的内容
            self._xpath(url, result, ['title', 'p', 'li', 'div'], unicode(key, "utf8"), deep) #分析这几个标签的内容
        except TypeError: #对编码类型异常处理,有些深度页面和主页的编码不同
            self._xpath(url, result, ['a'], key, deep)
            self._xpath(url, result, ['title', 'p', 'li', 'div'], key, deep)
        return True

    def _xpath(self, weburl, data, xpath, key, deep):
        page = etree.HTML(data)
        for i in xpath:
            hrefs = page.xpath(u"//%s" % i) #根据xpath标签
            if deep >1:
                for href in hrefs:
                    url = href.attrib.get('href','')
                    if not url.startswith('java') and not url.startswith('#') and not \
                        url.startswith('mailto') and url.endswith('html'):  #过滤javascript和发送邮件的链接
                            self.tp.add_job(self.getPageSource,url, key, deep-1) #递归调用,直到符合的深
            for href in hrefs:
                value = href.text  #抓取相应标签的内容
                if value:
                    m = re.compile(r'.*%s.*' % key).match(value) #根据key匹配相应内容

    def work(self):
        self.tp.add_job(self.getPageSource, self.url, self.key, self.deep)
        self.tp.wait_for_complete() #等待线程池完成

common_Crawler.py

#!/usr/bin/python
#coding=utf-8

# Version 1 by Dongwm 2013/01/10
# 脚本作用：多线程



import time
import threading
import Queue
from common import Crawler

#lock = threading.Lock()   #设置线程锁


class MyThread(threading.Thread):

    def __init__(self, workQueue, timeout=1, **kwargs):
        threading.Thread.__init__(self, kwargs=kwargs)
        self.timeout = timeout #线程在结束前等待任务队列多长时间
        self.setDaemon(True)  #设置deamon,表示主线程死掉,子线程不跟随死掉
        self.workQueue = workQueue
        self.start() #初始化直接启动线程

    def run(self):
        '''重载run方法'''
        while True:
            try:
                #lock.acquire() #线程安全上锁 PS:queue 实现就是线程安全的，没有必要上锁 ,否者可以put/get_nowait
                callable, args = self.workQueue.get(timeout=self.timeout) #从工作队列中获取一个任务
                res = callable(*args)  #执行的任务
                #lock.release()  #执行完,释放锁 
            except Queue.Empty: #任务队列空的时候结束此线程
                break
            except Exception, e:
                return -1


class ThreadPool(object):

    def __init__(self, num_of_threads):
         self.workQueue = Queue.Queue()
         self.threads = []
         self.__createThreadPool(num_of_threads)

    def __createThreadPool(self, num_of_threads):
        for i in range(num_of_threads):
             thread = MyThread(self.workQueue)
             self.threads.append(thread)

    def wait_for_complete(self):
        '''等待所有线程完成'''
        while len(self.threads):
            thread = self.threads.pop()
            if thread.isAlive():  #判断线程是否还存活来决定是否调用join
                thread.join()

    def add_job( self, callable, *args):
        '''增加任务,放到队列里面'''
        self.workQueue.put((callable, args))
def main():

    tp = ThreadPool(10) 
    crawler = Crawler(tp)
    crawler.work()

if __name__ == '__main__':

    import timeit
    t = timeit.Timer("main()") 
    t.repeat(3, 10)

gevent_Crawler.py

#!/usr/bin/python
#coding=utf-8

# Version 1 by Dongwm 2013/01/10
# 脚本作用：gevent

import gevent.monkey
gevent.monkey.patch_all()
from gevent.queue import Empty, Queue
import gevent
from common import Crawler

class GeventLine(object):

    def __init__(self, workQueue, timeout=1, **kwargs):
        self.timeout = timeout #线程在结束前等待任务队列多长时间
        self.workQueue = workQueue

    def run(self):
        '''重载run方法'''
        while True:
            try:
                callable, args = self.workQueue.get(timeout=self.timeout) #从工作队列中获取一个任务
                res = callable(*args)  #执行的任务
                print res
            except Empty:
                break
            except Exception, e:
                print e
                return -1

class GeventPool(object):

    def __init__(self, num_of_threads):
             self.workQueue = Queue()
             self.threads = []
             self.__createThreadPool(num_of_threads)

    def __createThreadPool(self, num_of_threads):
        for i in range(num_of_threads):
             thread = GeventLine(self.workQueue)
             self.threads.append(gevent.spawn(thread.run))


    def wait_for_complete(self):
        '''等待所有线程完成'''

        while len(self.threads):
            thread = self.threads.pop()
            thread.join()
        gevent.shutdown()

    def add_job( self, callable, *args):
        '''增加任务,放到队列里面'''
        self.workQueue.put((callable, args))

def main():
    tp = GeventPool(10) 
    crawler = Crawler(tp)
    crawler.work()

if __name__ == '__main__':

    import timeit
    t = timeit.Timer("main()") 
    t.repeat(3, 10)

Crawler/spiders/spiders.py

from scrapy.contrib.spiders import CrawlSpider, Rule
from scrapy.selector import HtmlXPathSelector
from scrapy.contrib.linkextractors.sgml import SgmlLinkExtractor
from scrapy.item import Item

class MySpider(CrawlSpider):
    name = 'localhost'
    allowed_domains = ['localhost']
    start_urls = ['http://localhost']
    rules = ( 
        Rule(SgmlLinkExtractor(allow=(r'http://localhost/.*')), callback="parse_item"),  
    )  
    def parse_item(self, response):
        hxs = HtmlXPathSelector(response)
        hxs.select('//*[@*]/text()').re(r'py')  #实现了common.py里面的抓取和分析，但是common.py是抓取五种标签，分2次抓取，这里是抓取所有标签，不够严禁

scrapy_Crawler.py #时间有限，没有研究模块调用，也不够严禁


#!/usr/bin/python
#coding=utf-8

# Version 1 by Dongwm 2013/01/10
# 脚本作用：scrapy

from subprocess import call

def main():
    call('scrapy crawl localhost --nolog', shell=True)

if __name__ == '__main__':

    import timeit
    t = timeit.Timer("main()") 
    t.repeat(3, 10)

实验过程

1. 同时启动三个终端，一起跑（手点回车，肯定有点延迟）

dongwm@localhost ~/Crawler $ python scrapy_Crawler.py
10000000 loops, best of 3: 0.024 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0223 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0223 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0223 usec per loop
10000000 loops, best of 3: 0.0222 usec per loop
10000000 loops, best of 3: 0.0223 usec per loop #他是最快跑完的，非常快～～  数据很稳定

dongwm@localhost ~/Crawler $ python gevent_Crawler.py
100000000 loops, best of 3: 0.0134 usec per loop
100000000 loops, best of 3: 0.0131 usec per loop
100000000 loops, best of 3: 0.0132 usec per loop
100000000 loops, best of 3: 0.0132 usec per loop
100000000 loops, best of 3: 0.0132 usec per loop
100000000 loops, best of 3: 0.0132 usec per loop
100000000 loops, best of 3: 0.0132 usec per loop
100000000 loops, best of 3: 0.0132 usec per loop
100000000 loops, best of 3: 0.0132 usec per loop
100000000 loops, best of 3: 0.0132 usec per loop
100000000 loops, best of 3: 0.0134 usec per loop
100000000 loops, best of 3: 0.0132 usec per loop
100000000 loops, best of 3: 0.0133 usec per loop
100000000 loops, best of 3: 0.0133 usec per loop
100000000 loops, best of 3: 0.0133 usec per loop
100000000 loops, best of 3: 0.0132 usec per loop
100000000 loops, best of 3: 0.0133 usec per loop
100000000 loops, best of 3: 0.0132 usec per loop
100000000 loops, best of 3: 0.0126 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0123 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0123 usec per loop  #跑得很慢，不知道是不是timeit的原因(或者调用的优先级太低，抢资源能力不行)，很奇怪，但是它的数据最快，数据稳定在0.0123-0.0133


dongwm@localhost ~/Crawler $ python common_Crawler.py
100000000 loops, best of 3: 0.0274 usec per loop
10000000 loops, best of 3: 0.0245 usec per loop
10000000 loops, best of 3: 0.0252 usec per loop
10000000 loops, best of 3: 0.0239 usec per loop
10000000 loops, best of 3: 0.025 usec per loop
10000000 loops, best of 3: 0.0273 usec per loop
10000000 loops, best of 3: 0.0255 usec per loop
10000000 loops, best of 3: 0.0261 usec per loop
10000000 loops, best of 3: 0.0275 usec per loop
10000000 loops, best of 3: 0.0261 usec per loop
10000000 loops, best of 3: 0.0257 usec per loop
10000000 loops, best of 3: 0.0273 usec per loop
10000000 loops, best of 3: 0.0241 usec per loop
10000000 loops, best of 3: 0.0257 usec per loop
10000000 loops, best of 3: 0.0275 usec per loop
10000000 loops, best of 3: 0.0241 usec per loop
10000000 loops, best of 3: 0.0259 usec per loop
10000000 loops, best of 3: 0.0251 usec per loop
10000000 loops, best of 3: 0.0193 usec per loop
10000000 loops, best of 3: 0.0176 usec per loop
100000000 loops, best of 3: 0.0199 usec per loop
100000000 loops, best of 3: 0.0167 usec per loop
100000000 loops, best of 3: 0.018 usec per loop
10000000 loops, best of 3: 0.0179 usec per loop
100000000 loops, best of 3: 0.0173 usec per loop
100000000 loops, best of 3: 0.0172 usec per loop
100000000 loops, best of 3: 0.018 usec per loop
100000000 loops, best of 3: 0.0162 usec per loop
100000000 loops, best of 3: 0.0179 usec per loop
100000000 loops, best of 3: 0.0171 usec per loop  #第二跑得快，但是还是数据不稳定，时间在0.017-0.026之间

#####2. 挨个启动，待负载保持一个相对拨波动平衡在换另一个

dongwm@localhost ~/Crawler $ python scrapy_Crawler.py
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0122 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0123 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0123 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0122 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0123 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0122 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0122 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0126 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop   #数据很稳定，在0.0122-0.0126之间 机器负载在1.3左右,最高超过了1.4（闲暇0.6左右）

dongwm@localhost ~/Crawler $ python gevent_Crawler.py
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0126 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0126 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0126 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0126 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0126 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0126 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0126 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop
100000000 loops, best of 3: 0.0125 usec per loop
100000000 loops, best of 3: 0.0124 usec per loop  #数据很稳定，在0.0124-0.0126之间 机器负载在1.2左右（闲暇0.6左右）

dongwm@localhost ~/Crawler $ python common_Crawler.py
10000000 loops, best of 3: 0.0135 usec per loop
100000000 loops, best of 3: 0.0185 usec per loop
10000000 loops, best of 3: 0.0174 usec per loop
100000000 loops, best of 3: 0.019 usec per loop
10000000 loops, best of 3: 0.016 usec per loop
10000000 loops, best of 3: 0.0181 usec per loop
10000000 loops, best of 3: 0.0146 usec per loop
100000000 loops, best of 3: 0.0192 usec per loop
10000000 loops, best of 3: 0.0165 usec per loop
10000000 loops, best of 3: 0.0176 usec per loop
10000000 loops, best of 3: 0.0177 usec per loop
10000000 loops, best of 3: 0.0182 usec per loop
100000000 loops, best of 3: 0.0195 usec per loop
10000000 loops, best of 3: 0.0163 usec per loop
10000000 loops, best of 3: 0.0161 usec per loop
100000000 loops, best of 3: 0.0191 usec per loop
100000000 loops, best of 3: 0.0193 usec per loop
10000000 loops, best of 3: 0.0147 usec per loop
100000000 loops, best of 3: 0.0197 usec per loop
10000000 loops, best of 3: 0.0178 usec per loop
10000000 loops, best of 3: 0.0172 usec per loop
100000000 loops, best of 3: 0.022 usec per loop
100000000 loops, best of 3: 0.0191 usec per loop
10000000 loops, best of 3: 0.0208 usec per loop
10000000 loops, best of 3: 0.0144 usec per loop
10000000 loops, best of 3: 0.0201 usec per loop
100000000 loops, best of 3: 0.0195 usec per loop
100000000 loops, best of 3: 0.0231 usec per loop
10000000 loops, best of 3: 0.0149 usec per loop
100000000 loops, best of 3: 0.0211 usec per loop #数据有点不稳定，浮动较大，但是最要在0.016-0.019  机器负载曾经长时间在1.01,最高未超过1.1 （闲暇0.6左右）

一些我的看法

虽然我的实验有不够严禁的地方，我的代码能力也有限（希望有朋友看见代码能提供修改意见或更 NB 的版本），但是效果还是比较明显的，我总结下

gevent 确实性能很好，并且很稳定，占用 io 一般 (据说长时间使用有内存泄露的问题？我不理解)
scrapy 这个框架把爬虫封装的很好，只需要最少的代码就能实现，性能也不差 gevent
多线程编程确实有瓶颈，并且不稳定