[文章作者:张宴 本文版本:v1.0 最后修改:2008.07.27 转载请注明原文链接:http://blog.zyan.cc/post/360/]

  前言:本文阐述的是一款经过生产环境检验的千万级数据全文检索(搜索引擎)架构。本文只列出前几章的内容节选,不提供全文内容。

  在DELL PowerEdge 6850服务器(四颗64 位Inter Xeon MP 7110N处理器 / 8GB内存)、RedHat AS4 Linux操作系统、MySQL 5.1.26、MyISAM存储引擎、key_buffer=1024M环境下实测,单表1000万条记录的数据量(这张MySQL表拥有int、datetime、varchar、text等类型的10多个字段,只有主键,无其它索引),用主键(PRIMARY KEY)作为WHERE条件进行SQL查询,速度非常之快,只耗费0.01秒。

  出自俄罗斯的开源全文搜索引擎软件Sphinx,单一索引最大可包含1亿条记录,在1千万条记录情况下的查询速度为0.x秒(毫秒级)。Sphinx创建索引的速度为:创建100万条记录的索引只需3~4分钟,创建1000万条记录的索引可以在50分钟内完成,而只包含最新10万条记录的增量索引,重建一次只需几十秒。

  基于以上几点,我设计出了这套搜索引擎架构。在生产环境运行了一周,效果非常不错。有时间我会专为配合Sphinx搜索引擎,开发一个逻辑简单、速度快、占用内存低、非表锁的MySQL存储引擎插件,用来代替MyISAM引擎,以解决MyISAM存储引擎在频繁更新操作时的锁表延迟问题。另外,分布式搜索技术上已无任何问题。



  一、搜索引擎架构设计:
  1、搜索引擎架构图:
  点击在新窗口中浏览此图片

  2、搜索引擎架构设计思路:
  (1)、调用方式最简化:
  尽量方便前端Web工程师,只需要一条简单的SQL语句“SELECT ... FROM myisam_table JOIN sphinx_table ON (sphinx_table.sphinx_id=myisam_table.id) WHERE query='...';”即可实现高效搜索。

  (2)、创建索引、查询速度快:
  ①、Sphinx Search 是由俄罗斯人Andrew Aksyonoff 开发的高性能全文搜索软件包,在GPL与商业协议双许可协议下发行。
  Sphinx的特征:
  •Sphinx支持高速建立索引(可达10MB/秒,而Lucene建立索引的速度是1.8MB/秒)
  •高性能搜索(在2-4 GB的文本上搜索,平均0.1秒内获得结果)
  •高扩展性(实测最高可对100GB的文本建立索引,单一索引可包含1亿条记录)
  •支持分布式检索
  •支持基于短语和基于统计的复合结果排序机制
  •支持任意数量的文件字段(数值属性或全文检索属性)
  •支持不同的搜索模式(“完全匹配”,“短语匹配”和“任一匹配”)
  •支持作为Mysql的存储引擎

  ②、通过国外《High Performance MySQL》专家组的测试可以看出,根据主键进行查询的类似“SELECT ... FROM ... WHERE id = ...”的SQL语句(其中id为PRIMARY KEY),每秒钟能够处理10000次以上的查询,而普通的SELECT查询每秒只能处理几十次到几百次:
  点击在新窗口中浏览此图片

  ③、Sphinx不负责文本字段的存储。假设将数据库的id、date、title、body字段,用sphinx建立搜索索引。根据关键字、时间、类别、范围等信息查询一下sphinx,sphinx只会将查询结果的ID号等非文本信息告诉我们。要显示title、body等信息,还需要根据此ID号去查询MySQL数据库,或者从Memcachedb等其他的存储中取得。安装SphinxSE作为MySQL的存储引擎,将MySQL与Sphinx结合起来,是一种便捷的方法。
  创建一张Sphinx类型表,将MyISAM表的主键ID和Sphinx表的ID作一个JOIN联合查询。这样,对于MyISAM表来所,只相当于一个WHERE id=...的主键查询,WHERE后的条件都交给Sphinx去处理,可以充分发挥两者的优势,实现高速搜索查询。

  (3)、按服务类型进行分离:
  为了保证数据的一致性,我在配置Sphinx读取索引源的MySQL数据库时,进行了锁表。Sphinx读取索引源的过程会耗费一定时间,由于MyISAM存储引擎的读锁和写锁是互斥的,为了避免写操作被长时间阻塞,导致数据库同步落后跟不上,我将提供“搜索查询服务”的和提供“索引源服务”的MySQL数据库进行了分开。监听3306端口的MySQL提供“搜索查询服务”,监听3406端口的MySQL提供“索引源服务”。

  (4)、“主索引+增量索引”更新方式:
  一般网站的特征:信息发布较为频繁;刚发布完的信息被编辑、修改的可能性大;两天以前的老帖变动性较小。
  基于这个特征,我设计了Sphinx主索引和增量索引。对于前天17:00之前的记录建立主索引,每天凌晨自动重建一次主索引;对于前天17:00之后到当前最新的记录,间隔3分钟自动重建一次增量索引。

  (5)、“Ext3文件系统+tmpfs内存文件系统”相结合:
  为了避免每3分钟重建增量索引导致磁盘IO较重,从而引起系统负载上升,我将主索引文件创建在磁盘,增量索引文件创建在tmpfs内存文件系统“/dev/shm/”内。“/dev/shm/”内的文件全部驻留在内存中,读写速度非常快。但是,重启服务器会导致“/dev/shm/”内的文件丢失,针对这个问题,我会在服务器开机时自动创建“/dev/shm/”内目录结构和Sphinx增量索引。

  (6)、中文分词词库:
  我根据“自整理的中文分词库”+“搜狗拼音输入法细胞词库”+“LibMMSeg高频字库”+... 综合整理成一份中文分词词库,出于某些考虑暂不提供。你可以使用LibMMSeg自带的中文分词词库。



  二、MySQL+Sphinx+SphinxSE安装步骤:
  1、安装python支持(以下针对CentOS系统,其他Linux系统请使用相应的方法安装)
yum install -y python python-devel


  2、编译安装LibMMSeg(LibMMSeg是为Sphinx全文搜索引擎设计的中文分词软件包,其在GPL协议下发行的中文分词法,采用Chih-Hao Tsai的MMSEG算法。LibMMSeg在本文中用来生成中文分词词库。)

  以下压缩包“sphinx-0.9.8-rc2-chinese.zip”中包含mmseg-0.7.3.tar.gz、sphinx-0.9.8-rc2.tar.gz以及中文分词补丁。


unzip sphinx-0.9.8-rc2-chinese.zip
tar zxvf mmseg-0.7.3.tar.gz
cd mmseg-0.7.3/
./configure
make
make install
cd ../


  3、编译安装MySQL 5.1.26-rc、Sphinx、SphinxSE存储引擎
wget http://dev.mysql.com/get/Downloads/MySQL-5.1/mysql-5.1.26-rc.tar.gz/from/http://mirror.x10.com/mirror/mysql/
tar zxvf mysql-5.1.26-rc.tar.gz

tar zxvf sphinx-0.9.8-rc2.tar.gz
cd sphinx-0.9.8-rc2/
patch -p1 < ../sphinx-0.98rc2.zhcn-support.patch
patch -p1 < ../fix-crash-in-excerpts.patch
cp -rf mysqlse ../mysql-5.1.26-rc/storage/sphinx
cd ../

cd mysql-5.1.26-rc/
sh BUILD/autorun.sh
./configure --with-plugins=sphinx --prefix=/usr/local/mysql-search/ --enable-assembler --with-extra-charsets=complex --enable-thread-safe-client --with-big-tables --with-readline --with-ssl --with-embedded-server --enable-local-infile
make && make install
cd ../

cd sphinx-0.9.8-rc2/
CPPFLAGS=-I/usr/include/python2.4
LDFLAGS=-lpython2.4
./configure --prefix=/usr/local/sphinx --with-mysql=/usr/local/mysql-search
make
make install
cd ../

mv /usr/local/sphinx/etc/sphinx.conf /usr/local/sphinx/etc/sphinx.conf.old




  第二章第3节之后的正文内容不予公布,全文的目录如下(共24页):

  点击在新窗口中浏览此图片

  点击在新窗口中浏览此图片

  点击在新窗口中浏览此图片



  2010年2月5日增加:

  文档全文请访问: http://blog.zyan.cc/sphinx_search/




技术大类 » 搜索引擎技术 | 评论(132) | 引用(0) | 阅读(207248)
louis vuitton uk Email Homepage
2011-11-23 09:04
This louis vuitton uk for sale belongs to the sounding just what are termed as Louis Vuitton vintage best sellers, many other products and services for the reason that range appearing companies.You will easily notice the unfold zippers of this coach outlet store online. That is the decoration. There are some inside pockets for you as well. They are easy to match your clothes and to carry.Let us inspire your inner beauty with fine christian louboutin sale. Purse the elegance in bridal wedding. Enjoy the fashion.
sunairmax Email Homepage
2012-4-19 10:39
Now we give large discount to customers,some could save up to 40% or even 50%.We promise we would offer you the best service as possible as we can.We are expecting your visiting http://www.sunairmax.org
zra0g617 Email Homepage
2012-4-19 23:09
Nike air griffey max sportswear from the latest version of design studio looks like a straight line down from the Nike ACG mountain terrain park here. Off-set with pastel neon accents fall, Nike Sportswear Air Max a forest may be a prudent and ice throw, Nike Air Max 95 till the fall of raising its charming head. On the other hand, you could just continue to move forward, these rocks into the air and pull off Max's now a khaki on the appearance. Air Max 95 Air Max less costly 1 using the traditional design of all components, this name is air max 95's color is set separately. I absolutely love these Nike Air Max 2012. I needed a new style for work, they look great and even feel better. Very soft sole nike air max 90 New Arrival. It is important for my work shoes to have ventilation. I&rsquo;ll be back before Christmas.
lfq618
2012-4-26 20:19
张老师,我按照你的步骤在linux上配置时,编译sphinx时总是出错:tokenizer_zhcn.cpp:42:   instantiated from here/usr/local/include/mmseg/Singleton.h:95: error: ‘atexit’ was not declared in this scopemake[2]: *** [tokenizer_zhcn.o] Error 1make[2]: Leaving directory `/opt/sphinx-0.9.8-rc2/src'make[1]: *** [all] Error 2make[1]: Leaving directory `/opt/sphinx-0.9.8-rc2/src'make: *** [all-recursive] Error 1希望帮解决下,小弟不胜感激, 嘿嘿
coach factory outlet Email Homepage
2012-5-17 11:40
Today, following half a century, mentor leather-bases coach factory outlet continues to be the delicate craft of leather-based master is accountable for,Would you like to meet more friends, or go with the times? If yes, coach factory online is opening welcome doors to you.in the market you definitely can find various colorways that are designed in as well as the high quality that applied in. For most of you would like to come. So just come to our coach factory outlet online store to choose one.
louis vuitton sale Email Homepage
2012-5-17 11:40
Louis Vuitton belt at louis vuitton sale is one kind of fashion accessory with high cost performance among the Louis Vuitton accessories.Offering quality LV products with favorable prices, louis vuitton outlet store is at your service. Hurry up, or you can not seize the chance.in fact, louis vuitton is one of the most famous fashion design master.he opened the fist suitcase shop called after his name.
coach outlet online Email Homepage
2012-5-17 11:40
I heard of coach outlet online through the advertisement when I was shopping. And now I often brow the webpage and buy Coach bags online.It is a symbol regarding position not to mention nature.Here I would like to launch a excellent bags pertaining to business men.Which may be coach outlet store.coach outlet has always been simple,durable style features to win consumers.The products are more flexible,with easy bleaching,wear characteristics,and simply use a damp cloth.
coach outlet, Email Homepage
2012-5-17 11:40
Remember the coach outlet provide coach bags which won't be deteriorated into its overall styles by any means. It will maintain its looks, colors, and uniqueness for long time.coach outlet store online has been voted by Hour Detroit magazine readers as the Best of Detroit in their 12th annual readers'poll.Lots of women like which usually amount normally include a coach outlet online ,it provides coziness to many girls that don't even think it is a great bushel of great interest directly to them.
xujie777 Email
2012-5-18 17:06
We aim to make all our customers satisfy with our products. You will find a variety of Men's fashion louis vuitton uk, fashion Women's cheap Louis Vuitton bags in our store at affordable price.Thinking of interesting ways to cost a milestone birthday? louis vuitton online shop had one of the most distinctive distinctive celebrations.bakery along with living room operated by means of about three moment louis vuitton online Most effective Pastry Chef’s of the year Rammy Nominee Chef’s.
xujie777 Email
2012-5-18 17:06
The choices are likely to be basically countless seeing that louis vuitton outlet occurs with the help of completely new and also incredible concepts once in a while.Louis vuitton Wholesale Monogram Canvas HandbagsLouis Vuitton Collection Beach Handbags louis vuitton bags outlet Damier Canvas HandbagsLouis vuitton Mahina HandbagsLouis Vuitton Monogram Mini Lin HandbagsLouis Vuitton Monogram Multicolore HandbagsLouis vuitton Monogram Vernis HandbagsLouis Vuitton Wholesale Epi Leather HandbagsLouis Vuitton For Men HandbagsLouis Vuitton Damier Canvas WalletsLouis Vuitton Epi Leather WalletsLouis Vuitton Monogram Canvas WalletsLouis Vuitton Monogram Vernis WalleLouis Vuitton ShoesLouis Vuitton Men wallets.As the Authentic Louis Vuitton are so high-priced, so came the louis vuitton handbags outlet.
xujie777 Email
2012-5-18 17:06
coach outlet has become a popular shopping experience for consumers around the world, and a desirable distribution channel for manufacturer's and retailers.Getting your hands on coach outlet store online can be a hefty investment of hundreds of dollars.But do not despair,the Coach Outlet store could be the answer to your prayers.coach outlet online is your smart choice when you want to get the discount Coach accessory. You can find the exact Coach Bags and other accessory you want at a low price that's right for you. My dear friends, let the coach outlet online pave your way into the world of high fashion with their ultimate fashion factory.
xujie123
2012-5-21 11:22
Our online store offers you discounted Designer louis vuitton replica wallet at present. You could find them in desirable quality and price. If you don't mind high class louis vuitton uk, have a good time here.louis vuitton Store Online Handbags can also bring great accuracy as well as practical applicability and fashionable.Have you ever dreamed of being as charming as Madonna? Have you ever thought of becoming an envy of all your friends? If so, come to louis vuitton outlet.<br/>
xujie555
2012-5-21 13:46
coach factory outlet uses graceful accessories to match the classical logo of coach, which is the best combination of coach. The handmade coach products?can make you more charming and graceful.coach factory online provides people many coach goods. If you wish to snatch the coach handbag, then this best method is made for that you like for coach discount.coach factory outlet online is a fashion brand to ensure its quality. With designer coach shoulder bags, you will always attract people's attention. The bags will emphasize your personal style and taste.<br/>
xujie333
2012-5-21 15:12
coach outlet online Store guarantee that all the coach handbags offered are own high quality. In addition , all of them are sold at an unexpected low price.If you want to purchase, just visit their website.coach factory outlet is really sizzling kinds of shopping way for you. With the usage of the replica designer coach bags, you can surely be able to change your individual looks in a stunning manner.Coach bags on sale from the coach outlet are cheap or discount prices that you certainly will stand out from the crowd on your next camping trip!<br/>
cheap oakley sunglasses Email Homepage
2012-7-18 13:31
Various designer Oakley Sunglasses Outlet for Sale in our store. Oakley Sunglasses at possible lowest price and high quality, fast delivery.Hot summer has already come, when you go out the most imporant things you need take is a pair of sunglasses. These fashion and stylish Cheap Oakley Sunglasses can protect your eyes well in the strong sunlight. We do not simply need a pair of sunglasses, we need protection from UV rays, we need to be cool in summer, show our understanding of fashion. In fact, all we need is a pair of Discount Oakley Sunglasses.
coach88888888 Email
2012-8-13 17:51
If you desire to go to Coach Factory Outlet, but have no idea in which to go, you can research online. It is no doubt that there is drastically information and details about it for the reference.Coach Factory Online are loved by many people, when you walk in the street, you could see many people take coach styles.hat experts claim Coach Outlet Online shopping is in the changes they are available in.Coach Outlet will send you a coupon in the post to use in their upon only after you type a achieve. The Coach bags are customarily somewhat many than the ones in the stores.
bj198993 Email
2012-8-24 19:14
正在学习中
HGJH Email Homepage
2012-9-27 08:53
You can pamper your self by buying a pair of Karen Millen dresses artist shoe occasionally. Get fantastic superior men's put on that offers you comfort and style. put on good men's artist footwear and make an impression among your friends. As?these times folks are acknowledged by their style, attire and appears therefore it is necessary to consider fantastic treatment of your complete attire. A guy appears handsome when he clubs up the artist Karen Millen dress shoe having a artist outfit.There are various shoes specifications that the manufacturer must consider treatment of and therefore there are distinct manufacturers within the industry that caters towards demands of this type of customers. footwear are an crucial style accessory, which includes a practical objective too. There are a amount of pairs of footwear that you just Karen Millen Ireland should absolutely maintain within your closet. A formal pair along having a casual plus a walking shoes.
ariadne Email Homepage
2012-9-27 16:56
对这块没有什么研究,学习。。。。
领悟
2012-12-5 11:59
不错,学习了,zan
分页: 6/7 第一页 上页 1 2 3 4 5 6 7 下页 最后页
发表评论
表情
emotemotemotemotemot
emotemotemotemotemot
emotemotemotemotemot
emotemotemotemotemot
emotemotemotemotemot
打开HTML
打开UBB
打开表情
隐藏
记住我
昵称   密码   游客无需密码
网址   电邮   [注册]