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:
- centos5.3(64位)、nginx-0.7.51、keepalived-1.1.15
- 主nginx负载均衡器:192.168.0.154
- 辅nginx负载均衡器:192.168.9.155
- vip:192.168.0.188
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
- #添加运行nginx的用户和组www
- groupaddwww
- useradd-gwwwwww
- wgetftp://ftp.csx.cam.ac.uk/pub/software/programming/pcre/pcre-7.8.tar.gz
- tarzxvfpcre-7.8.tar.gz
- cdpcre-7.8/
- ./configure
- make&&makeinstall
- wgethttp://sysoev.ru/nginx/nginx-0.7.51.tar.gz
- tarzxvfnginx-0.7.51.tar.gz
- cdnginx-0.7.51/
- ./configure--user=www--group=www--prefix=/usr/local/webserver/nginx--with-http_stub_status_module--with-http_ssl_module
- make&&makeinstall
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。
- userwwwwww;
- worker_processes8;
- pid/usr/local/nginx/logs/nginx.pid;
- worker_rlimit_nofile65535;
- events
- {
- useepoll;
- worker_connections65535;
- }
- http{
- includemime.types;
- default_typeapplication/octet-stream;
- server_names_hash_bucket_size128;
- client_header_buffer_size32k;
- large_client_header_buffers432k;
- client_max_body_size8m;
- sendfileon;
- tcp_nopushon;
- keepalive_timeout60;
- tcp_nodelayon;
- fastcgi_connect_timeout300;
- fastcgi_send_timeout300;
- fastcgi_read_timeout300;
- fastcgi_buffer_size64k;
- fastcgi_buffers464k;
- fastcgi_busy_buffers_size128k;
- fastcgi_temp_file_write_size128k;
- gzipon;
- gzip_min_length1k;
- gzip_buffers416k;
- gzip_http_version1.0;
- gzip_comp_level2;
- gzip_typestext/plainapplication/x-javascripttext/cssapplication/xml;
- gzip_varyon;
- upstreambackend
- {
- server192.168.1.102:80;
- server192.168.1.103:80;
- server192.168.1.105:80;
- }
- server{
- listen80;
- server_namewww.yuhongchun027.com;
- location/{
- root/var/www;
- indexindex.jspindex.htmindex.html;
- proxy_redirectoff;
- proxy_set_headerHost$host;
- proxy_set_headerX-Real-IP$remote_addr;
- proxy_set_headerX-Forwarded-For$proxy_add_x_forwarded_for;
- proxy_passhttp://backend;
- }
- location/nginx{
- access_logon;
- auth_basic"NginxStatus";
- auth_basic_user_file/usr/local/nginx/htpasswd;
- }
- log_formataccess'$remote_addr-$remote_user[$time_local]"$request"'
- '$status$body_bytes_sent"$http_referer"'
- '"$http_user_agent"$http_x_forwarded_for';
- access_log/var/log/access.logaccess;
- }
- }
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。
- wgethttp://www.keepalived.org/software/keepalived-1.1.15.tar.gz
- #tarzxvfkeepalived-1.1.15.tar.gz
- #cdkeepalived-1.1.15
- #./configure--prefix=/usr/local/keepalived
- #make
- #makeinstall
- #cp/usr/local/keepalived/sbin/keepalived/usr/sbin/
- #cp/usr/local/keepalived/etc/sysconfig/keepalived/etc/sysconfig/
- #cp/usr/local/keepalived/etc/rc.d/init.d/keepalived/etc/init.d/
- #mkdir/etc/keepalived
- #cd/etc/keepalived/
- vimkeepalived.conf
- !ConfigurationFileforkeepalived
- global_defs{
- notification_email{
- yuhongchun027@163.com
- }
- notification_email_fromkeepalived@chtopnet.com
- smtp_server127.0.0.1
- smtp_connect_timeout30
- router_idLVS_DEVEL
- }
- vrrp_instanceVI_1{
- stateMASTER
- interfaceeth0
- virtual_router_id51
- mcast_src_ip192.168.0.154<==主nginx的IP地址
- priority100
- advert_int1
- authentication{
- auth_typePASS
- auth_passchtopnet
- }
- virtual_ipaddress{
- 192.168.0.188<==vip地址
- }
- }
- #servicekeepalivedstart
我们来看一下日志:
- [root@ltos~]#tail/var/log/messages
- Oct603:25:03ltosavahi-daemon[2306]:Registeringnewaddressrecordfor192.168.0.188oneth0.
- Oct603:25:03ltosavahi-daemon[2306]:Registeringnewaddressrecordfor192.168.0.154oneth0.
- Oct603:25:03ltosavahi-daemon[2306]:RegisteringHINFOrecordwithvalues'I686'/'LINUX'.
- Oct603:25:23ltosavahi-daemon[2306]:Withdrawingaddressrecordforfe80::20c:29ff:feb9:eeaboneth0.
- Oct603:25:23ltosavahi-daemon[2306]:Withdrawingaddressrecordfor192.168.0.154oneth0.
- Oct603:25:23ltosavahi-daemon[2306]:Hostnameconflict,retryingwith<ltos-31>
很显然vrrp已经启动,我们还可以通过命令来检查
- [root@ltoshtml]#ipaddr
- 1:lo:<LOOPBACK,UP,LOWER_UP>mtu16436qdiscnoqueue
- link/loopback00:00:00:00:00:00brd00:00:00:00:00:00
- inet127.0.0.1/8scopehostlo
- inet6::1/128scopehost
- valid_lftforeverpreferred_lftforever
- 2:eth0:<BROADCAST,MULTICAST,UP,LOWER_UP>mtu1500qdiscpfifo_fastqlen1000
- link/ether00:0c:29:ba:9b:e7brdff:ff:ff:ff:ff:ff
- inet192.168.0.154/24brd192.168.0.255scopeglobaleth0
- inet192.168.0.188/32scopeglobaleth0
- inet6fe80::20c:29ff:feba:9be7/64scopelink
- valid_lftforeverpreferred_lftforever
- 3:sit0:<NOARP>mtu1480qdiscnoop
- link/sit0.0.0.0brd0.0.0.0
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:
- !ConfigurationFileforkeepalived
- global_defs{
- notification_email{
- yuhongchun027@163.com
- }
- notification_email_fromkeepalived@chtopnet.com
- smtp_server127.0.0.1
- smtp_connect_timeout30
- router_idLVS_DEVEL
- }
- vrrp_instanceVI_1{
- stateBACKUP
- interfaceeth0
- virtual_router_id51
- mcast_src_ip192.168.0.155<==辅nginx的IP的地址
- priority100
- advert_int1
- authentication{
- auth_typePASS
- auth_passchtopnet
- }
- virtual_ipaddress{
- 192.168.0.188
- }
- }
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:
- vim/root/nginx_pid.sh
- #!/bin/bash
- while:
- do
- nginxpid=`ps-Cnginx--no-header|wc-l`
- if[$nginxpid-eq0];then
- /usr/local/nginx/sbin/nginx
- sleep5
- nginxpid=`ps -C nginx --no-header | wc -l`
- if[$nginxpid-eq0];then
- /etc/init.d/keepalivedstop
- fi
- fi
- sleep5
- done
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。