keepalive高可用haproxy实现URL资源的动静分离

实现要点：

(1) 动静分离discuzx，动静都要基于负载均衡实现； (2) 进一步测试在haproxy和后端主机之间添加varnish缓存； (3) 给出拓扑设计； (4) haproxy的设定要求： (a) 启动stats； (b) 自定义403、502和503的错误页； (c) 各组后端主机选择合适的调度方法； (d) 记录好日志； (e) 使用keepalived高可用haproxy；

拓扑结构

两台keepalived的双主模型对两台haproxy主机做高可用，两个VIP分别为10.1.253.11与10.1.253.12

haproxy主机负责接收请求、动静分离请求的图片资源、调度单台varnish缓存主机及两台httpd主机

varnish缓存主机负责缓存后端nginx服务器响应的用户上传的静态图片资源，并调度两台nginx主机

nginx主机负责响应图片资源，并为websrv主机提供nfs服务，映射为discuzx程序attachment目录

websrv安装httpd、mysql、php程序，处理discuzx程序的动态资源及未分离的静态资源，如css等

nginx服务器配置nfs服务

安装nfs服务

1.yum install nfs-utils

配置nfs共享

/etc/exports

1./data/discuz 10.1.253.66(rw,no_root_squash) 10.1.253.67(rw,no_root_squash)

创建apache用户，并授权

1.useradd -s -u 48 -g 48 apache 2.setfacl -m u:apache:rwx /data/discuz

启动nfs服务

1.systemctl start nfs.service

websrvs主机配置

安装amp程序和discuzx

关键步骤：

1.yum install httpd mysql php php-mysql php-xcache 2.mysql -uroot -p -e 'CREATE DATABASE ultrax;GRANT ALL ON ultrax.* TO ultraxuser@10.1.%.% IDENTIFIED BY "ultraxpass";FLUSH PRIVILEGES'

挂载nfs到用户上传附件路径

启动mysql、httpd并访问测试

nginx主机配置

nginx负责响应用户上传的静态图片资源，nginx的虚拟主机root路径指向nfs共享的目录即可。

为了能够将URL的资源路径映射为虚拟主机的root路径下所对应的资源，应使用nginx对请求的URL重写或重定向，显然可在最前端的haproxy主机或varnish服务或nginx服务都能够实现URL的重写，只要确保新的URL能够映射到nginx主机下该资源的路径即可。没有必要同时在haproxy、varnish、nginx都重写同一URL，考虑到后端主机的数量，我觉得可以在haproxy或varnish中重写URL。

安装nginx

配置虚拟主机

1.server { 2. listen 82; 3. server_name localhost; 4. location / { 5. root /data/discuz; 6. index index.html index.htm; 7. } 8. location ~* \.(jpg|jpeg|gif|png)$ { 9. root /data/discuz; 10. rewrite ^/.*forum/(.*)$ /$1 break; 11. } 12.}

启动nginx服务并访问测试

某资源的URL源路径

替换该URL的host为nginx主机，直接访问该URL

nginx服务器输出日志

1.10.1.250.19 - - [13/Nov/2016:9:01:53 +0800] "GET /data/attachment/forum/201611/12/174905kkys2e2wgmv25ywe.jpg HTTP/1.1" 200 126931 "-" "Mozilla/5.0 (Windows NT 6.1; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/52.0.2743.82 Safari/537.36" "-"

varnish缓存服务器

varnish服务器负责缓存响应资源，并调度nginx服务器，以及检测nginx服务的健康状态等

安装varnish

安装前需配置好epel的yum源

1.yum install varnish

配置缓存服务

配置varnish运行时参数

/etc/varnish/varnish.params

1.VARNISH_LISTEN_PORT=80 2.…… 3.VARNISH_STORAGE="malloc,128M"

配置varnish缓存服务

前面说过，对URL的重写可在varnish服务器中实现，对于有众多后端nginx主机的情况，在varnish重写URL更加方便；

在varnish中重写URL使用regsub函数实现； 为了避免与nginx服务器端的URL重写混淆，应把nginx虚拟主机配置中的URL重写注释；

/etc/varnish/default.vcl

1.vcl 4.0; 2.import directors; …… 13.backend nginx1 { 14. .host = "10.1.253.29"; 15. .port = "81"; 16. .probe = ok; 17.} 18.backend nginx2 { 19. .host = "10.1.253.29"; 20. .port = "82"; 22.} 23.sub vcl_init { 24. new RR = directors.round_robin(); 25. RR.add_backend(nginx1); 26. RR.add_backend(nginx2); 27.} 28. 29.sub vcl_recv { 30. set req.backend_hint = RR.backend(); 31. if (req.url ~ "(?i)\.(jpg|jpeg|gif|png)$") { 32. set req.url = regsub(req.url, "/.*attachment/(.*)", "/\1"); 33. } 43.} 44. 48.sub vcl_backend_response { 49. # Happens after we have read the response headers from the backend. 50. # 51. # Here you clean the response headers, removing silly Set-Cookie headers 52. # and other mistakes your backend does. 53.} 54. 55.sub vcl_deliver { 61.}

启动varnish并访问测试

1.systemctl start varnish

—访问varnish服务器下该资源的URL

—nginx服务器端的访问日志

1.10.1.253.29 - - [13/Nov/2016:22:21:43 +0800] "GET /forum/201611/12/174905kkys2e2wgmv25ywe.jpg HTTP/1.1" 304 0 "-" "Mozilla/5.0 (Windows NT 6.1; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/52.0.2743.82 Safari/537.36" "10.1.250.19"

效果就是：只要重写的路径下存在该资源，无论URL中该资源的前的路径是什么，都统统能够重写为自定义路径下的相同资源

haproxy主机配置

安装

1.yum install haproxy

配置文件

配置文件路径：/etc/haproxy/haproxy.cfg

主要是定义前端和后端的配置，其中前端基于acl对URI进行匹配控制：

url_static_geg条件为URI的起始路径，url_static_end条件为URI的后缀名

只有同时满足以上两个条件才调用static主机组，其余的URL使用默认的dynamic主机组

此外，还定义了错误响应码的重定向到另一主机的

以及开启了haproxy的stats页面

配置frontend前端

1.frontend main *:80 2. acl url_static_beg path_beg -i /data/attachment 3. acl url_static_end path_end -i .jpg .gif .png .css .js 4. 5. use_backend static if url_static_beg url_static_end 6. 7. default_backend dynamic 8. 9. errorloc 503 http://10.1.253.29:82/errorpage/503sorry.html 10. errorloc 403 http://10.1.253.29:82/errorpage/403sorry.html 11. errorloc 502 http://10.1.253.29:82/errorpage/502sorry.html

配置backend后端

1.backend dynamic 2. balance roundrobin 4. server web1 10.1.253.66:81 check cookie amp1 5. server web2 10.1.253.66:82 check cookie amp2 6. 7.backend static 8. balance roundrobin 9. server ngx1 10.1.253.29:81 check 10. server ngx2 10.1.253.29:82 check

配置stats页面

1.listen stats 2. bind : 3. stats enable 4. stats uri /admin?stats …… 7. stats refresh 10s 8. stats admin if TRUE 9. stats hide-version

测试结果

keepalived高可用haproxy服务

安装

1.yum install keepalived

配置双主模型的keepalived主机

/etc/keepalived/keepalived.conf

1.global_defs { 2. notification_email { 3. root@localhost 4. } 5. notification_email_from keepalived@jasonmc.com 6. smtp_server localhost 7. smtp_connect_timeout 30 8. router_id node1 9. vrrp_mcast_group4 224.22.29.1 10.} 11.vrrp_script chk_down { 12. script "[[ -f /etc/keepalived/down ]]&& exit 1 || exit 0" 13. interval 1 14. weight -5 15.} 16.vrrp_script chk_haproxy { 17. script "killall -0 haproxy && exit 0 || exit 1" 18. interval 1 19. weight -5 20.} 21.vrrp_instance VI_1 { 22. state MASTER 23. interface eno16777736 24. virtual_router_id 10 25. priority 96 26. advert_int 10 27. authentication { 28. auth_type PASS 29. auth_pass 1a7b2ce6 30. } 31. virtual_ipaddress { 32. 10.1.253.11 dev eno16777736 33. } 34. track_script { 35. chk_down 36. chk_haproxy 37. } 38.} 39.vrrp_instance VI_2 { 40. state BACKUP 41. interface eno16777736 42. virtual_router_id 11 43. priority 100 44. advert_int 11 45. authentication { 46. …… 48. }

启动keepalived服务并测试

1.systemctl start keepalived

haproxy1与haproxy2同时上线时

—haproxy1拥有VIP1 10.1.253.11

—haproxy2拥有VIP2 10.1.253.12

触发haproxy1下线操作

VI_1(即haproxy1)上在/etc/keepalived/目录下建立down文件，让keepalived的track_script功能检测到此文件并实现下线功能。

—haproxy1上关于keepalived的日志输出：

1.Nov 14 13:18:55 h1 Keepalived_vrrp[54901]: VRRP_Script(chk_down) failed 2.Nov 14 13:19:01 h1 Keepalived_vrrp[54901]: VRRP_Instance(VI_1) Received higher prio advert 3.Nov 14 13:19:01 h1 Keepalived_vrrp[54901]: VRRP_Instance(VI_1) Entering BACKUP STATE 4.Nov 14 13:19:01 h1 Keepalived_vrrp[54901]: VRRP_Instance(VI_1) removing protocol VIPs. 5.Nov 14 13:19:01 h1 Keepalived_healthcheckers[54900]: Netlink reflector reports IP10.1.253.11 removed

haproxy2主机无法收到haproxy1多播发送的HEARTBEAT信息，将成为VI_1的MASTER主机。

—haproxy2上关于keepalived的日志输出：

1.Nov 14 13:19:01 h1 Keepalived_vrrp[58092]: VRRP_Instance(VI_1) forcing a new MASTER election 2.Nov 14 13:19:01 h1 Keepalived_vrrp[58092]: VRRP_Instance(VI_1) forcing a new MASTER election 3.Nov 14 13:19:11 h1 Keepalived_vrrp[58092]: VRRP_Instance(VI_1) Transition to MASTER STATE 4.Nov 14 13:19:21 h1 Keepalived_vrrp[58092]: VRRP_Instance(VI_1) Entering MASTER STATE 5.Nov 14 13:19:21 h1 Keepalived_vrrp[58092]: VRRP_Instance(VI_1) setting protocol VIPs. 6.Nov 14 13:19:21 h1 Keepalived_vrrp[58092]: VRRP_Instance(VI_1) Sending gratuitousARPs on eno16777736 for 10.1.253.11 7.Nov 14 13:19:21 h1 Keepalived_healthcheckers[58091]: Netlink reflector reports IP10.1.253.11 added 8.Nov 14 13:19:26 h1 Keepalived_vrrp[58092]: VRRP_Instance(VI_1) Sending gratuitousARPs on eno16777736 for 10.1.253.11

—haproxy2上将同时拥有 VIP1 与VIP2

触发haproxy1重新上线操作

把VI_1(即haproxy1)中/etc/keepalived/目录下down文件移除，让keepalived的track_script功能检测不到此文件实现重新上线的功能。

—haproxy1上关于keepalived的日志输出：

1.Nov 14 13:58:02 h1 Keepalived_vrrp[67748]: VRRP_Script(chk_down) succeeded 2.Nov 14 13:58:12 h1 Keepalived_vrrp[67748]: VRRP_Instance(VI_1) Transition to MASTER STATE 3.Nov 14 13:58:22 h1 Keepalived_vrrp[67748]: VRRP_Instance(VI_1) Entering MASTER STATE 4.Nov 14 13:58:22 h1 Keepalived_vrrp[67748]: VRRP_Instance(VI_1) setting protocol VIPs. 5.Nov 14 13:58:22 h1 Keepalived_vrrp[67748]: VRRP_Instance(VI_1) Sending gratuitousARPs on eno16777736 for 10.1.253.11 6.Nov 14 13:58:22 h1 Keepalived_healthcheckers[67747]: Netlink reflector reports IP10.1.253.11 added 7.Nov 14 13:58:27 h1 Keepalived_vrrp[67748]: VRRP_Instance(VI_1) Sending gratuitousARPs on eno16777736 for 10.1.253.11

—haproxy2上关于keepalived的日志输出：

1.Nov 14 13:58:12 h1 Keepalived_vrrp[58092]: VRRP_Instance(VI_1) Received higher prio advert 2.Nov 14 13:58:12 h1 Keepalived_vrrp[58092]: VRRP_Instance(VI_1) Entering BACKUP STATE 3.Nov 14 13:58:12 h1 Keepalived_vrrp[58092]: VRRP_Instance(VI_1) removing protocol VIPs. 4.Nov 14 13:58:12 h1 Keepalived_healthcheckers[58091]: Netlink reflector reports IP10.1.253.11 removed

—再次恢复为haproxy1拥有VIP1，haproxy2拥有VIP2

测试结果

—同时访问 VIP1 或 VIP2 都能正常访问由haproxy代理的discuzx网站

—对于用户上传的附件资源，由varnish服务器或nginx服务器进行响应

总结

HAProxy是一款纯粹的高性能反向代理服务器，能够代理应用层协议，也可以定义mode tcp让代理传输层协议。HAProxy能够代理HTTP协议和TCP协议，支持代理Web Server、Dynamic Engine、DateBase，且能够检测后端主机的健康状态，实现后端主机的HA。其内建的stats管理页能够非常方便查看前端、后端主机的状态，简单的操作就能实现后端主机的上下线。

关于URL的重写，上文中已经说明可在HAProxy代理服务器、Varnish缓存服务器或Nginx主机上实现，但为了便于管理较多的后端主机，通常选择在HAProxy服务器或者Varnish缓存服务器上实现URL的重写。

HAProxy代理服务器的单进程事件驱动模型使得其能够处理大并发请求，使用弹性二叉树算法存储的连接会话能够非常灵活的进行管理，对于后端主机调度算法也能做到非常精细。