Maison >base de données >tutoriel mysql >揭秘企业级web负载均衡完美架构
【51CTO.com独家特稿】相信很多朋友对 企业 级的 负载 均衡 高可用实例非常感兴趣,此篇 文章 根据成熟的线上环境而写,旨在帮助大家迅速 架构 一个 企业 级的 负载 均衡 高可用的web环境。 此系统 架构 仅映射内网VIP的80及443端口于外网的Juniper防火墙下
【51CTO.com独家特稿】相信很多朋友对企业级的负载均衡高可用实例非常感兴趣,此篇文章根据成熟的线上环境而写,旨在帮助大家迅速架构一个企业级的负载均衡高可用的web环境。
此系统架构仅映射内网VIP的80及443端口于外网的Juniper防火墙下,其它端口均关闭,内网所有机器均关闭iptables及ipfw防火墙;外网DNS指向即通过Juniper映射出来的外网地址,而此映射的地址对映的其实是内网VIP地址。这里说下端口的问题,有的朋友可能会很疑惑,这样映射端口行不?通过项目实践得知,这样完全是可行的,php-cgi需要的9000端口及MySQL的3306端口均可走内网,完全不影响业务系统的运行。
另外,我维护的电子商务网站并发大约在1000左右,此时,Nginx+Apache集群运行得非常稳定,尤其是apache,并没有想象中那般弱;其实,在内存足够(>=8G)的情况,测试时不连数据库的话,单台apache+php5能顶得住6000并发,而且相当稳定。在网站升级架构方面,我不赞成全面淘汰生级,锦上添花式的升级会更好。
第一部分:Nginx+Keepalived的说明及环境说明
喜欢看我博客或文章的朋友都知道,我一直主力推崇Nginx+Keepalived作web的负载均衡高可用架构,并积极将其用于项目方案中;Nginx负载均衡作服务器遇到的故障一般有①服务器网线松动等网络故障;②服务器硬件故障从而crash;③nginx服务死掉;遇到前二者情况,keeaplived是能起到HA的作用的;然而遇到③种情况就没有办法了,但可以通过shell监控解决这问题,从而实现真正意义上的负载均衡高可用。此篇的最新更新时间为2010年6月25号,下面将其安装步骤详细说明下:
环境:
<ol class="dp-xml"> <li class="alt">centos5.3(64位)、nginx-0.7.51、keepalived-1.1.15 </li> <li>主nginx<strong>负载</strong><strong>均衡</strong>器:192.168.0.154 </li> <li class="alt">辅nginx<strong>负载</strong><strong>均衡</strong>器:192.168.9.155 </li> <li>vip:192.168.0.188 </li> </ol>
第二部分:分别安装Nginx负载均衡器及相关配置脚本
先安装Nginx负载均衡器,nginx负载的配置就用一般的模板来配置了
<ol class="dp-xml"> <li class="alt">#添加运行nginx的用户和组www </li> <li>groupadd www </li> <li class="alt">useradd -g www www </li> <li>wget ftp://ftp.csx.cam.ac.uk/pub/software/programming/pcre/pcre-7.8.tar.gz </li> <li class="alt">tar zxvf pcre-7.8.tar.gz </li> <li>cd pcre-7.8/ </li> <li class="alt">./configure </li> <li>make && make install </li> <li class="alt">wget http://sysoev.ru/nginx/nginx-0.7.51.tar.gz </li> <li>tar zxvf nginx-0.7.51.tar.gz </li> <li class="alt">cd nginx-0.7.51/ </li> <li>./configure <span class="attribute">--user</span>=<span class="attribute-value">www</span> <span class="attribute">--group</span>=<span class="attribute-value">www</span> <span class="attribute">--prefix</span>=/usr/local/webserver/nginx --with-http_stub_status_module --with-http_ssl_module </li> <li class="alt">make && make install </li> </ol>
配置nginx负载均衡器的配置文件vim /usr/local/nginx/conf/nginx.conf,此篇文章仅仅只是我的某项目的配置文档,纯80转发;如果对nginx配置有https要求的可参考张宴的相关文章。
<ol class="dp-xml"> <li class="alt">user www www; </li> <li>worker_processes 8; </li> <li class="alt"> </li> <li>pid /usr/local/nginx/logs/nginx.pid; </li> <li class="alt">worker_rlimit_nofile 65535; </li> <li> </li> <li class="alt">events </li> <li>{ </li> <li class="alt">use epoll; </li> <li>worker_connections 65535; </li> <li class="alt">} </li> <li>http{ </li> <li class="alt">include mime.types; </li> <li>default_type application/octet-stream; </li> <li class="alt">server_names_hash_bucket_size 128; </li> <li>client_header_buffer_size 32k; </li> <li class="alt">large_client_header_buffers 4 32k; </li> <li>client_max_body_size 8m; </li> <li class="alt">sendfile on; </li> <li>tcp_nopush on; </li> <li class="alt">keepalive_timeout 60; </li> <li>tcp_nodelay on; </li> <li class="alt">fastcgi_connect_timeout 300; </li> <li>fastcgi_send_timeout 300; </li> <li class="alt">fastcgi_read_timeout 300; </li> <li>fastcgi_buffer_size 64k; </li> <li class="alt">fastcgi_buffers 4 64k; </li> <li>fastcgi_busy_buffers_size 128k; </li> <li class="alt">fastcgi_temp_file_write_size 128k; </li> <li>gzip on; </li> <li class="alt">gzip_min_length 1k; </li> <li>gzip_buffers 4 16k; </li> <li class="alt">gzip_http_version 1.0; </li> <li>gzip_comp_level 2; </li> <li class="alt">gzip_types text/plain application/x-javascript text/css application/xml; </li> <li>gzip_vary on; </li> <li class="alt"> </li> <li>upstream backend </li> <li class="alt">{ </li> <li>server 192.168.1.102:80; </li> <li class="alt">server 192.168.1.103:80; </li> <li>server 192.168.1.105:80; </li> <li class="alt">} </li> <li>server { </li> <li class="alt">listen 80; </li> <li>server_name www.yuhongchun027.com; </li> <li class="alt">location / { </li> <li>root /var/www ; </li> <li class="alt">index index.jsp index.htm index.html; </li> <li>proxy_redirect off; </li> <li class="alt">proxy_set_header Host $host; </li> <li>proxy_set_header X-Real-IP $remote_addr; </li> <li class="alt">proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; </li> <li>proxy_pass http://backend; </li> <li class="alt">} </li> <li> </li> <li class="alt">location /nginx { </li> <li>access_log on; </li> <li class="alt">auth_basic "NginxStatus"; </li> <li>auth_basic_user_file /usr/local/nginx/htpasswd; </li> <li class="alt">} </li> <li> </li> <li class="alt">log_format access '$remote_addr - $remote_user [$time_local] "$request" ' </li> <li>'$status $body_bytes_sent "$http_referer" ' </li> <li class="alt">'"$http_user_agent" $http_x_forwarded_for'; </li> <li>access_log /var/log/access.log access; </li> <li class="alt"> </li> <li>} </li> <li class="alt">} </li> </ol>
小节:
第一部分和第二部分讲的是如何通过安装Nginx来达到负载均衡后端web集群的过程,Nginx能实现自动切换后端有故障的web服务器;但Nginx负载均衡器出了问题怎么办呢,它们之间是如何实现无故障转移的呢?
第三部分:安装Keepalived,让其分别作web及Nginx的HA
安装keepalived,并将其做成服务模式,方便以后调试。
<ol class="dp-xml"> <li class="alt">wget http://www.keepalived.org/software/keepalived-1.1.15.tar.gz </li> <li>#tar zxvf keepalived-1.1.15.tar.gz </li> <li class="alt">#cd keepalived-1.1.15 </li> <li>#./configure <span class="attribute">--prefix</span>=/usr/local/keepalived </li> <li class="alt">#make </li> <li>#make install </li> <li class="alt">#cp /usr/local/keepalived/sbin/keepalived /usr/sbin/ </li> <li>#cp /usr/local/keepalived/etc/sysconfig/keepalived /etc/sysconfig/ </li> <li class="alt">#cp /usr/local/keepalived/etc/rc.d/init.d/keepalived /etc/init.d/ </li> <li>#mkdir /etc/keepalived </li> <li class="alt">#cd /etc/keepalived/ </li> <li> </li> <li class="alt">vim keepalived.conf </li> <li>! Configuration File for keepalived </li> <li class="alt">global_defs { </li> <li> notification_email { </li> <li class="alt"> yuhongchun027@163.com </li> <li> } </li> <li class="alt"> notification_email_from keepalived@chtopnet.com </li> <li> smtp_server 127.0.0.1 </li> <li class="alt"> smtp_connect_timeout 30 </li> <li> router_id LVS_DEVEL </li> <li class="alt">} </li> <li>vrrp_instance VI_1 { </li> <li class="alt"> state MASTER </li> <li> interface eth0 </li> <li class="alt"> virtual_router_id 51 </li> <li> mcast_src_ip 192.168.0.154 <span class="tag">==主nginx的IP地址 </span> </li> <li class="alt"> priority 100 </li> <li> advert_int 1 </li> <li class="alt"> authentication { </li> <li> auth_type PASS </li> <li class="alt"> auth_pass chtopnet </li> <li> } </li> <li class="alt"> virtual_ipaddress { </li> <li> 192.168.0.188 <span class="tag">==vip地址 </span> </li> <li class="alt"> } </li> <li>} </li> <li class="alt">#service keepalived start </li> </ol>
我们来看一下日志:
<ol class="dp-xml"> <li class="alt">[root@ltos ~]# tail /var/log/messages </li> <li>Oct 6 03:25:03 ltos avahi-daemon[2306]: Registering new address record for 192.168.0.188 on eth0. </li> <li class="alt">Oct 6 03:25:03 ltos avahi-daemon[2306]: Registering new address record for 192.168.0.154 on eth0. </li> <li>Oct 6 03:25:03 ltos avahi-daemon[2306]: Registering HINFO record with values 'I686'/'LINUX'. </li> <li class="alt">Oct 6 03:25:23 ltos avahi-daemon[2306]: Withdrawing address record for fe80::20c:29ff:feb9:eeab on eth0. </li> <li>Oct 6 03:25:23 ltos avahi-daemon[2306]: Withdrawing address record for 192.168.0.154 on eth0. </li> <li class="alt">Oct 6 03:25:23 ltos avahi-daemon[2306]: Host name conflict, retrying with <span class="tag"><span class="tag-name">ltos-31</span><span class="tag">></span> </span> </li> </ol>
很显然vrrp已经启动,我们还可以通过命令来检查
<ol class="dp-xml"> <li class="alt">[root@ltos html]# ip addr </li> <li>1: lo: <span class="tag"><span class="tag-name">LOOPBACK</span>,UP,LOWER_UP<span class="tag">></span> mtu 16436 qdisc noqueue </span> </li> <li class="alt"> link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 </li> <li> inet 127.0.0.1/8 scope host lo </li> <li class="alt"> inet6 ::1/128 scope host </li> <li> valid_lft forever preferred_lft forever </li> <li class="alt">2: eth0: <span class="tag"><span class="tag-name">BROADCAST</span>,MULTICAST,UP,LOWER_UP<span class="tag">></span> mtu 1500 qdisc pfifo_fast qlen 1000 </span> </li> <li> link/ether 00:0c:29:ba:9b:e7 brd ff:ff:ff:ff:ff:ff </li> <li class="alt"> inet 192.168.0.154/24 brd 192.168.0.255 scope global eth0 </li> <li> inet 192.168.0.188/32 scope global eth0 </li> <li class="alt"> inet6 fe80::20c:29ff:feba:9be7/64 scope link </li> <li> valid_lft forever preferred_lft forever </li> <li class="alt">3: sit0: <span class="tag"><span class="tag-name">NOARP</span><span class="tag">></span> mtu 1480 qdisc noop </span> </li> <li> link/sit 0.0.0.0 brd 0.0.0.0 </li> </ol>
说明vip已经启动,这样主服务器就配置好了,辅机的配置大致一样,除了配置文件有少部分的变化,下面贴出辅机的配置文件:
<ol class="dp-xml"> <li class="alt">! Configuration File for keepalived </li> <li>global_defs { </li> <li class="alt"> notification_email { </li> <li> yuhongchun027@163.com </li> <li class="alt"> } </li> <li> notification_email_from keepalived@chtopnet.com </li> <li class="alt"> smtp_server 127.0.0.1 </li> <li> smtp_connect_timeout 30 </li> <li class="alt"> router_id LVS_DEVEL </li> <li>} </li> <li class="alt">vrrp_instance VI_1 { </li> <li> state BACKUP </li> <li class="alt"> interface eth0 </li> <li> virtual_router_id 51 </li> <li class="alt"> mcast_src_ip 192.168.0.155 <span class="tag">==辅nginx的IP的地址 </span> </li> <li> priority 100 </li> <li class="alt"> advert_int 1 </li> <li> authentication { </li> <li class="alt"> auth_type PASS </li> <li> auth_pass chtopnet </li> <li class="alt"> } </li> <li> virtual_ipaddress { </li> <li class="alt"> 192.168.0.188 </li> <li> } </li> <li class="alt">} </li> </ol>
第四部分:针对Keepalived的不足,用Nginx_pid.sh来监控nginx进程,实现真正意义上的负载均衡高可用。
针对Nginx+Keepalived,编写nginx监控脚本nginx_pid.sh,此脚本思路其实也很简单,即放置在后台一直监控nginx进程;如进程消失,尝试重启nginx,如是失败则立即停掉本机的keepalived服务,让另一台负载均衡器接手,此脚本直接从生产环境下载:
<ol class="dp-xml"> <li class="alt">vim /root/nginx_pid.sh </li> <li>#!/bin/bash </li> <li class="alt">while : </li> <li>do </li> <li class="alt"> <span class="attribute">nginxpid</span>=`ps -C nginx --no-header | wc -l` </li> <li> if [ $nginxpid -eq 0 ];then </li> <li class="alt"> /usr/local/nginx/sbin/nginx </li> <li> sleep 5 </li> <li>nginxpid=`ps -C nginx --no-header | wc -l`</li> <li class="alt"> if [ $nginxpid -eq 0 ];then </li> <li> /etc/init.d/keepalived stop </li> <li class="alt"> fi </li> <li> fi </li> <li class="alt"> sleep 5 </li> <li>done </li> </ol>
然后置于后台运行 sh /root/nginx_pid.sh &,这种写法是错误的,这样你用root用户logout后,此进程会消失;正确写法为nohup/bin/bash /root/nginx_pid.sh &,附带下注释:如果你正在运行一个进程,而且你觉得在退出帐户时该进程还不会结束,那么可以使用nohup命令。该命令可以在你退出root帐户之后继续运行相应的进程。nohup就是不挂起的意思( no hang up),哈哈,差点老马失蹄了。
后记:
我的线上环境网络非常复杂,这也是LVS+Keepalived失败的原因。目前此套架构在1000并发的电子商务网站非常稳定,带来的直接影响就是nginx_backup一直处于闲置状态。相对于张宴的双机轮询而言,我感觉他的可能更加完美,因为目前我的Nginx仅仅只做了负载均衡器,如果以后有机会我会尝试做负载均衡器/反向代理加速。