nginx是著名的非专职全七层负载均衡器,在用惯了四层LVS后,终于碰上了麻烦:LVS后端的4台RS磁盘都较小(20G),跑不到一天就塞满了东西;而根据预估,实际上一天时间该节点也就只有20G的文件增长。很显然,因为lvs转发的轮询算法,导致RS重复缓存了相同的文件。 针对这个情况,可以有两个办法(我只想到两个,欢迎大家补充): 1、架构拆分,把不同的几个域名分别指向不同的server,这个在DNS上就能完成,不过就丧失了lvs的冗余;也可以用nginx的upstream+server配置,分别指向不通的RS,不过不同域名文件数量如果相差比较大的话,RS的负载就不均衡了…… 2、url_hash,采用HAproxy的loadbalance uri或者nginx的upstream_hash模块,都可以做到针对url进行哈希算法式的负载均衡转发。那么,就开始试试nginx的url_hash负载均衡吧: 1、安装部署: cd /tmp wget ftp://ftp.csx.cam.ac.uk/pub/soft ... re/pcre-8.01.tar.gz tar zxvf pcre-8.01.tar.gz wget http://wiki.nginx.org/images/7/78/Nginx_upstream_hash-0.3.tar.gz tar zxvf Nginx_upstream_hash-0.3.tar.gz wget http://sysoev.ru/nginx/nginx-0.7.65.tar.gz tar zxvf nginx-0.7.65.tar.gz vi nginx-0.7.65/src/http/ngx_http_upstream.h struct ngx_http_upstream_srv_conf_s { ngx_http_upstream_peer_t peer; void **srv_conf; ngx_array_t *servers; /* ngx_http_upstream_server_t */ ngx_array_t *values; ngx_array_t *lengths; ngx_uint_t retries; ngx_uint_t flags; ngx_str_t host; u_char *file_name; ngx_uint_t line; in_port_t port; in_port_t default_port; }; 为了安全,可以修改一下nginx的version信息:vi nginx-0.7.65/src/core/nginx.h #define NGINX_VERSION "2.6.STABLE21" #define NGINX_VER "squid/" NGINX_VERSION vi nginx-0.7.65/src/http/ngx_http_header_filter_module.c static char ngx_http_server_string[] = "Server: squid/2.6.STABLE21" CRLF; vi nginx-0.7.65/src/http/ngx_http_special_response.c static u_char ngx_http_error_tail[] = "<hr><center>squid/2.6.STABLE21</center>" CRLF "</body>" CRLF "</html>" CRLF cd nginx-0.7.65 ./configure --prefix=/home/nginx --with-pcre=/tmp/pcre-8.01 --with-http_stub_status_module --with-http_ssl_module --without-http_rewrite_module --add-module=/tmp/nginx_upstream_hash-0.3 vi auto/cc/gcc # debug #CFLAGS="$CFLAGS -g" make && make install 这样就安装完成了。上面有个地方要注意,即—with-pcre=PATH,填的是pcre的源码文件路径。 2、配置文件 user nobody nobody; worker_processes 10; #error_log logs/error.log; #error_log logs/error.log notice; #error_log logs/error.log info; #pid logs/nginx.pid; worker_rlimit_nofile 51200; events { use epoll; worker_connections 51200; } http { include mime.types; default_type application/octet-stream; keepalive_timeout 120; tcp_nodelay on; upstream images10.anjukestatic.com { server 211.151.78.50:80; server 211.151.79.219:80; server 119.167.247.139; server 113.6.255.78; } upstream images9.anjukestatic.com { server 121.14.58.54:80; server 121.14.225.184:80; server 59.34.215.146:80; } server { listen 80; server_name images10.anjukestatic.com; location / { proxy_pass http://images10.anjukestatic.com; proxy_set_header Host $host; proxy_set_header X-Real-IP $remote_addr; proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; } log_format images10.anjukestatic.com '$remote_addr - $remote_user [$time_local] $request ' '"$status" $body_bytes_sent "$http_referer" ' '"$http_user_agent" "$http_x_forwarded_for"'; access_log /data1/logs/img10.log images10.anjukestatic.com; } server { listen 80; server_name images9.anjukestatic.com; location / { proxy_pass http://images9.anjukestatic.com; proxy_set_header Host $host; proxy_set_header X-Real-IP $remote_addr; proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; } log_format images9.anjukestatic.com '$remote_addr - $remote_user [$time_local] $request ' '"$status" $body_bytes_sent "$http_referer" ' '"$http_user_agent" "$http_x_forwarded_for"'; access_log /data1/logs/img9.log images9.anjukestatic.com; } } 以上配置的问题: 1、RS中不能设置nginx本机的其他端口,我本来设定的server 11.11.11.10:3128,希望能把nginx本机也开上squid,省出一台机器来。结果在确认配置了DNS的情况下,返回状态码全是503…… 2、RS一旦有宕机的,nginx不会重算hash,导致部分url返回错误信息;而启用hash_again标签的话,其他RS就都乱了。 3、RS中默认logformat中将显示nginx的IP。 解决办法: 1、不知道 2、不采用hash_again标签而采用error_page重定向到专门的备份服务器保障访问 3、修改RS的logformat,把%>a改成%{X-Real_IP}>h即可。 最后的根本性问题: 对nginx下的RS集群进行增减操作,是否会对hash表产生影响?nginx_upstream_hash目录中的CHANGES有如下三条: Changes with upstream_hash 0.3 06 Aug 2008 *) Bugfix: infinite loop when retrying after a 404 and the "not_found" flag of *_next_upstream was set. *) Change: no more "hash_method" directive. Hash method is always CRC-32. *) Change: failover strategy is compatible with PECL Memcache. nginx的wiki上,关于hash_again的doc这么写到: Number of times to rehash the value and choose a different server if the backend connection fails. Increase this number to provide high availability. 关于PECL Memcache,请参考下列链接: http://www.surfchen.org/archives/348 http://tech.idv2.com/2008/07/24/memcached-004/ 尤其是第二个链接,其中关于rehash的解释,很好的解释了为什么大家都不推荐使用hash_again标签。 由此可知,upstream_hash模块,使用的是余数计算standard+CRC32方式,10+2的存活率是17%,3+1的存活率是23%! 而存活率最高的是consistent+CRC32方式,存活率是n/(n+m)*100%,10+2是83%,3+1是75%。 nginx的wiki中,还有另一个3rd模块upstream_consistent_hash,下回可以试试; 网上还有针对upstream_hash模块的补丁http://www.sanotes.net/wp-content/uploads/2009/06/nginx_upstream_hash.pdf,好模块就是有人研究呀~~ from:http://cdn-wangsheng.blog.sohu.com/145515265.html |
|Archiver|手机版|小黑屋|创星网络
( 苏ICP备11027519号 )
|网站地图
GMT+8, 2025-2-23 01:33 , Processed in 0.053349 second(s), 16 queries .
Powered by Discuz! X3
© 2001-2013 Comsenz Inc.