赞
踩
负载均衡:Load Balance,简称LB,是一种服务或基于硬件设备等实现的高可用反向代理技术,负载均 衡将特定的业务(web服务、网络流量等)分担给指定的一个或多个后端特定的服务器或设备,从而提高了 公司业务的并发处理能力、保证了业务的高可用性、方便了业务后期的水平动态扩展
1.通过ip+port决定负载均衡的去向。
2.对流量请求进行NAT处理,转发至后台服务器。
3.记录tcp、udp流量分别是由哪台服务器处理,后续该请求连接的流量都通过该服务器处理。
4.支持四层的软件 lvs:重量级四层负载均衡器。 Nginx:轻量级四层负载均衡器,可缓存。(nginx四层是通过upstream模块) Haproxy:模拟四层转发。
1.通过虚拟ur|或主机ip进行流量识别,根据应用层信息进行解析,决定是否需要进行负载均衡。
2.代理后台服务器与客户端建立连接,如nginx可代理前后端,与前端客户端tcp连接,与后端服务器建立 tcp连接, 3.支持7层代理的软件: Nginx:基于http协议(nginx七层是通过proxy_pass) Haproxy:七层代理,会话保持、标记、路径转移等。
3.支持7层代理的软件: Nginx:基于http协议(nginx七层是通过proxy_pass) Haproxy:七层代理,会话保持、标记、路径转移等
所谓的四到七层负载均衡,就是在对后台的服务器进行负载均衡时,依据四层的信息或七层的信息来决 定怎么样转发流量 四层的负载均衡,就是通过发布三层的IP地址(VIP),然后加四层的端口号,来决定哪些流量需要做负 载均衡,对需要处理的流量进行NAT处理,转发至后台服务器,并记录下这个TCP或者UDP的流量是由哪 台服务器处理的,后续这个连接的所有流量都同样转发到同一台服务器处理 七层的负载均衡,就是在四层的基础上(没有四层是绝对不可能有七层的),再考虑应用层的特征,比 如同一个Web服务器的负载均衡,除了根据VIP加80端口辨别是否需要处理的流量,还可根据七层的 URL、浏览器类别、语言来决定是否要进行负载均衡。
1.分层位置:四层负载均衡在传输层及以下,七层负载均衡在应用层及以下
2.性能 :四层负载均衡架构无需解析报文消息内容,在网络吞吐量与处理能力上较高:七层可支持解析应用 层报文消息内容,识别URL、Cookie、HTTP header等信息。
3.原理 :四层负载均衡是基于ip+port;七层是基于虚拟的URL或主机IP等。
4.功能类比:四层负载均衡类似于路由器;七层类似于代理服务器。
5.安全性:四层负载均衡无法识别DDoS攻击;七层可防御SYN Cookie/Flood攻击
我们的lvs是集成在内核中的,响应速度快,但是当客户访问的一台rs挂了之后,lvs默认并没有能力进行后端检测,而是会直接报一个connection refused的错误
这个时候就需要用到我们的haproxy了,当一台rs崩溃了,客户也不会察觉到,而且lvs只能做四层,而haproxy可以做七层负载。
在linux系统中,直接用以下命令安装haproxy即可
dnf install haproxy -y
实验环境:克隆三台主机,设置在同一个网段
haproxy:ip为172.25.254.100 主机名haproxy.karl.org
web1:ip为172.25.254.10 主机名webserver1.karl.org
web2:ip为172.25.254.20 主机名webserver2.karl.org
在haproxy中:
dnf install haproxy -y
分别在web1和web2中安装nginx服务,写入东西并启动:
- dnf install nginx -y
- echo webserver1 - 172.25.254.10 > /usr/share/nginx/html/index.html
- systemctl enable --now nginx.service
最后在haproxy上测试:
- [root@haproxy ~]# curl 172.25.254.20
- webserver2 - 172.25.254.20
- [root@haproxy ~]# curl 172.25.254.10
- webserver1 - 172.25.254.10
如果你不知道haproxy的配置文件在哪里,可以用rpm -qc命令查找:
- [root@haproxy ~]# rpm -qc haproxy
- /etc/haproxy/haproxy.cfg
- /etc/logrotate.d/haproxy
- /etc/sysconfig/haproxy
如果你在编写haproxy的配置文件的时候感觉tab不补齐不方便,你可以编写~/.vimrc
- [root@haproxy ~]# cat ~/.vimrc
- set ts=4 ai sw=4
其中ts是一个tab缩进的长度是4,ai代表自动缩进,sw代表在缩进代码时的缩进量(shiftwidth)为 4
打开/etc/haproxy/haproxy.cfg这个配置文件,找到fronted板块,添入fronted webcluster及以下的内容:
- #---------------------------------------------------------------------
- # main frontend which proxys to the backends
- #---------------------------------------------------------------------
- frontend webcluster
- bind *:80
- mode http
- use_backend webcluster-host
-
- backend webcluster-host
- balance roundrobin
- server web1 172.25.254.10:80
- server web2 172.25.254.20:80
-
更改之后不要忘记了重启,不过我们是第一次配置,所以是启动,但是我查看状态发现没有启动,所以用两个命令把haproxy启动了
- [root@haproxy ~]# systemctl enable haproxy.service
- Created symlink /etc/systemd/system/multi-user.target.wants/haproxy.service → /usr/lib/systemd/system/haproxy.service.
- [root@haproxy ~]# systemctl status haproxy
- ○ haproxy.service - HAProxy Load Balancer
- Loaded: loaded (/usr/lib/systemd/system/haproxy.service; enabled; vendo>
- Active: inactive (dead)
- [root@haproxy ~]# systemctl start haproxy
- [root@haproxy ~]# systemctl status haproxy
- ● haproxy.service - HAProxy Load Balancer
- Loaded: loaded (/usr/lib/systemd/system/haproxy.service; enabled; vendo>
- Active: active (running) since Sun 2024-08-11 17:59:31 CST; 3s ago
- Process: 31019 ExecStartPre=/usr/sbin/haproxy -f $CONFIG -f $CFGDIR -c ->
- Main PID: 31021 (haproxy)
现在我们可以来测试一下haproxy有没有后端检测当rs掉线了,客户这边会不显示掉线的rs的情况,同时也是把负载转移到没有宕机的real server上
rs1宕机前:
- C:\~]$ curl 172.25.254.100
- % Total % Received % Xferd Average Speed Time Time Time Current
- Dload Upload Total Spent Left Speed
- 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 100 27 100 27 0 0 1086 0 --:--:-- --:--:-- --:--:-- 1125
- webserver1 - 172.25.254.10
-
- [C:\~]$ curl 172.25.254.100
- % Total % Received % Xferd Average Speed Time Time Time Current
- Dload Upload Total Spent Left Speed
- 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 100 27 100 27 0 0 7812 0 --:--:-- --:--:-- --:--:-- 9000
- webserver2 - 172.25.254.20
现在把rs1关闭nginx服务:
systemctl stop nginx.service
无论客户机访问多少次,已经停止nginx服务的rs1都不会再提供给客户端:
- [C:\~]$ curl 172.25.254.100
- % Total % Received % Xferd Average Speed Time Time Time Current
- Dload Upload Total Spent Left Speed
- 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 100 27 100 27 0 0 9066 0 --:--:-- --:--:-- --:--:-- 9000
- webserver2 - 172.25.254.20
-
- [C:\~]$ curl 172.25.254.100
- % Total % Received % Xferd Average Speed Time Time Time Current
- Dload Upload Total Spent Left Speed
- 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 100 27 100 27 0 0 6991 0 --:--:-- --:--:-- --:--:-- 9000
- webserver2 - 172.25.254.20
-
刚刚的haproxy配置文件写法是前后端分开写的,这里提供一种一起写的方法:
- #---------------------------------------------------------------------
- # main frontend which proxys to the backends
- #---------------------------------------------------------------------
- #frontend webcluster
- # bind *:80
- # mode http
- # use_backend webcluster-host
- #
- #backend webcluster-host
- # balance roundrobin
- # server web1 172.25.254.10:80
- # server web2 172.25.254.20:80
-
- listen webcluster
- bind *:80
- mode http
- balance roundrobin
- server web1 172.25.254.10:80
- server web2 172.25.254.20:80

重启haproxy
systemctl restart haproxy.service
可以发现测试结果和刚刚的是一致的
- [C:\~]$ curl 172.25.254.100
- % Total % Received % Xferd Average Speed Time Time Time Current
- Dload Upload Total Spent Left Speed
- 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 100 27 100 27 0 0 1034 0 --:--:-- --:--:-- --:--:-- 1038
- webserver2 - 172.25.254.20
-
- [C:\~]$ curl 172.25.254.100
- % Total % Received % Xferd Average Speed Time Time Time Current
- Dload Upload Total Spent Left Speed
- 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 100 27 100 27 0 0 8287 0 --:--:-- --:--:-- --:--:-- 9000
- webserver2 - 172.25.254.20
-
在默认情况下,haproxy只有一个进程:
- [root@haproxy ~]# pstree -p | grep haproxy
- |-haproxy(31115)---haproxy(31117)---{haproxy}(31118)
设置haproxy worker的进程数为2
vim /etc/haproxy/haproxy.cfg
在global板块的最后一行添加nbproc 2
- # utilize system-wide crypto-policies
- ssl-default-bind-ciphers PROFILE=SYSTEM
- ssl-default-server-ciphers PROFILE=SYSTEM
- nbproc 2
- #---------------------------------------------------------------------
- # common defaults that all the 'listen' and 'backend' sections will
- # use if not designated in their block
- #---------------------------------------------------------------------
重启haproxy
systemctl restart haproxy.service
这个时候再过滤一下haproxy的进程数
- [root@haproxy ~]# pstree -p | grep haproxy
- |-haproxy(31168)-+-haproxy(31171)
- | `-haproxy(31172)
我们就使用nbproc 2参数把haproxy的进程数更改为了2
这里科普两个参数:
cpu-map 1 0 全 局 绑定haproxy worker 进程至指定CPU,将第1个work进程绑 定至0号CPU
cpu-map 2 1 全 局 绑定haproxy worker 进程至指定CPU,将第2个work进程绑 定至1号CPU
添加参数到haproxy配置文件中,重启查看效果
- nbproc 2
- cpu-map 1 0
- cpu-map 2 1
- #---------------------------------------------------------------------
- # common defaults that all the 'listen' and 'backend' sections will
- # use if not designated in their block
- #---------------------------------------------------------------------
查看效果
- [root@haproxy ~]# pstree -p | grep haproxy
- |-haproxy(31184)-+-haproxy(31187)
- | `-haproxy(31188)
定向haproxy的日志:
查看global中的log那一行的参数可知,haproxy的日志属于local2
- log 127.0.0.1 local2
-
- chroot /var/lib/haproxy
- pidfile /var/run/haproxy.pid
- maxconn 10000
- user haproxy
- group haproxy
- daemon
进入rsyslog的日志采集文件
vim /etc/rsyslog.conf
开启udp采集,在rhel9中需要开启,在rhel7版本和欧拉操作系统不需要
- # Provides UDP syslog reception
- # for parameters see http://www.rsyslog.com/doc/imudp.html
- module(load="imudp") # needs to be done just once
- input(type="imudp" port="514")
将采集到的日志保存在/var/log/haproxy.log中
- Save boot messages also to boot.log
- local7.* /var/log/boot.log
- local2.* /var/log/haproxy.log
重启rsyslog
systemctl restart rsyslog.service
默认这个目录是没有创建的
- [root@haproxy ~]# ll /var/log/haproxy.log
- ls: 无法访问 '/var/log/haproxy.log': 没有那个文件或目录
我们在客户端访问haproxy的时候,由于rs1的nginx此时处于报废状态,所以haproxy会让客户无法察觉,并把原来rs1的负载全分配给rs2,此时rsyslog便会采集到haproxy的日志:
- [C:\~]$ curl 172.25.254.100
- % Total % Received % Xferd Average Speed Time Time Time Current
- Dload Upload Total Spent Left Speed
- 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --100 27 100 27 0 0 7690 0 --:--:-- --:--:-- --:--:-- 9000
- webserver2 - 172.25.254.20
- [root@haproxy ~]# ll /var/log/haproxy.log
- -rw------- 1 root root 174 8月 11 22:46 /var/log/haproxy.log
查看一下日志的具体内容:
- [root@haproxy ~]# cat /var/log/haproxy.log
- Aug 11 22:46:57 localhost haproxy[31326]: 172.25.254.1:52537 [11/Aug/2024:22:46:57.640] webcluster webcluster/web2 0/0/0/1/1 200 226 - - ---- 1/1/0/0/+1 0/0 "GET / HTTP/1.1"
HAProxy通过固定参数 balance 指明对后端服务器的调度算法
balance参数可以配置在listen或backend选项中
HAProxy的调度算法分为静态和动态调度算法
有些算法可以根据参数在静态和动态算法中相互转换
按照事先定义好的规则轮询公平调度,不关心后端服务器的当前负载、连接数和响应速度 等,且无法实时修改权重(只能为0和1,不支持其它值),只能靠重启HAProxy生效。
static-rr:
基于权重的轮询调度 不支持运行时利用socat进行权重的动态调整(只支持0和1,不支持其它值)
不支持端服务器慢启动
其后端主机数量没有限制,相当于LVS中的 wrr
first
根据服务器在列表中的位置,自上而下进行调度
其只会当第一台服务器的连接数达到上限,新请求才会分配给下一台服务
其会忽略服务器的权重设置
不支持用socat进行动态修改权重,可以设置0和1,可以设置其它值但无效
基于后端服务器状态进行调度适当调整
新请求将优先调度至当前负载较低的服务器
权重可以在haproxy运行时动态调整无需重启
roundrobin
1. 基于权重的轮询动态调度算法
2. 支持权重的运行时调整,不同于lvs中的rr轮训模式
3. HAProxy中的roundrobin支持慢启动(新加的服务器会逐渐增加转发数)
4. 其每个后端backend中最多支持4095个real server
5. 支持对real server权重动态调整
6. roundrobin为默认调度算法,此算法使用广泛
leastconn leastconn
加权的最少连接的动态
支持权重的运行时调整和慢启动
即:根据当前连接最少的后端服务器而非权重进行优先调度(新客户 端连接)
比较适合长连接的场景使用,比如:MySQL等场景。
其它算法即可作为静态算法,又可以通过选项成为动态算法
source
源地址hash,基于用户源地址hash并将请求转发到后端服务器,后续同一个源地址请求将被转发至同一 个后端web服务器。此方式当后端服务器数据量发生变化时,会导致很多用户的请求转发至新的后端服 务器,默认为静态方式,但是可以通过hash-type支持的选项更改这个算法一般是在不插入Cookie的TCP 模式下使用,也可给拒绝会话cookie的客户提供最好的会话粘性,适用于session会话保持但不支持 cookie和缓存的场景源地址有两种转发客户端请求到后端服务器的服务器选取计算方式,分别是取模法 和一致性hash
map-base 取模法
map-based:取模法,对source地址进行hash计算,再基于服务器总权重的取模,最终结果决定将此请 求转发至对应的后端服务器。 此方法是静态的,即不支持在线调整权重,不支持慢启动,可实现对后端服务器均衡调度 缺点是当服务器的总权重发生变化时,即有服务器上线或下线,都会因总权重发生变化而导致调度结果 整体改变
一致性hash
一致性哈希,当服务器的总权重发生变化时,对调度结果影响是局部的,不会引起大的变动hash(o) mod n 该hash算法是动态的,支持使用 socat等工具进行在线权重调整,支持慢启动
hash环偏斜问题
增加虚拟服务器IP数量,比如:一个后端服务器根据权重为1生成1000个虚拟IP,再hash。而后端服务器权 重为2则生成2000的虚拟IP,再bash,最终在hash环上生成3000个节点,从而解决hash环偏斜问题
hash对象 Hash对象到后端服务器的映射关系:
一致性hash示意图 后端服务器在线与离线的调度方式:
uri
基于对用户请求的URI的左半部分或整个uri做hash,再将hash结果对总权重进行取模后
根据最终结果将请求转发到后端指定服务器
适用于后端是缓存服务器场景
默认是静态算法,也可以通过hash-type指定map-based和consistent,来定义使用取模法还是一致性 hash
web服务器中需要记录客户端的真实IP地址,用于做访问统计、安全防护、行为分析、区域排行等场景
当我们的客户端访问的是nginx发布的东西时,取消haproxy配置文件中我注释的这一行,便可看见是哪个ip访问的nginx
- defaults
- mode http
- log global
- option httplog
- option dontlognull
- option http-server-close
- #option forwardfor except 127.0.0.0/8
- option redispatch
- retries 3
- timeout http-request 10s
- timeout queue 1m
- timeout connect 10s
- timeout client 1m
- timeout server 1m
- timeout http-keep-alive 10s
- timeout check 10s
- maxconn 3000

更改后重启haproxy
systemctl restart haproxy.service
模拟客户访问rs
- [C:\~]$ curl 172.25.254.100
- % Total % Received % Xferd Average Speed Time Time Time Current
- Dload Upload Total Spent Left Speed
- 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 100 27 100 27 0 0 8348 0 --:--:-- --:--:-- --:--:-- 9000
- webserver2 - 172.25.254.20
查看我们的nginx日志文件,可以发现172.25.254.1正是客户端的ip
- [root@webserver2 ~]# > /var/log/nginx/access.log
- [root@webserver2 ~]# cat /var/log/nginx/access.log
- 172.25.254.100 - - [11/Aug/2024:23:27:26 +0800] "GET / HTTP/1.1" 200 27 "-" "curl/8.7.1" "172.25.254.1"
打开haproxy配置文件
vim /etc/haproxy/haproxy.cfg
- frontend testacl
- bind :80
- mode http
- ########### ACL settings #######################
- acl url_static path_end -i .jpg .png .css .js .html
- acl url_php path_end -i .php
- ########### host ###########################
- use_backend static_host if url_static
- use_backend php_host if url_php
- ########### default server ###################
- default_backend default_webserver
-
-
- backend static_host
- mode http
- server web2 192.168.0.101:80 check weight 1 inter 3s fall 3 rise 5
- backend php_host
- mode http
- server web1 192.168.0.102:80 check weight 1 inter 3s fall 3 rise 5
-
- backend default_webserver
- mode http
- server web1 172.25.254.10:80 check weight 1 inter 3s fall 3 rise 5

测试一下:
- [root@rs1 ~]# echo css 192.168.0.101 > /usr/share/nginx/html/index.css
- [root@rs2 ~]# echo php 192.168.0.102 > /var/www/html/index.php
- [root@node10 ~]# curl 172.25.254.100/index.php
- php 192.168.0.102
- [root@node10 ~]# curl 172.25.254.100/index.css
- css 192.168.0.101
打开haproxy配置文件
vim /etc/haproxy/haproxy.cfg
- frontend testacl
- bind :80
- mode http
- ########### ACL settings #######################
- acl url_static path_end -i .jpg .png .css .js .html
- acl url_static path_end -m sub /static /images /javascript
- acl acl_app path_beg -m sub /api
- ########### host ###########################
- use_backend static_host if url_static
- use_backend api_host if acl_app
- ########### default server ###################
- default_backend default_webserver
-
-
- backend static_host
- mode http
- server web2 192.168.0.101:80 check weight 1 inter 3s fall 3 rise 5
- backend api_host
- mode http
- server web1 192.168.0.102:80 check weight 1 inter 3s fall 3 rise 5
- backend default_webserver
- mode http
- server web1 172.25.254.10:80 check weight 1 inter 3s fall 3 rise 5
-

创建相关文件
- [root@rs1 ~]# mkdir /usr/share/nginx/html/static
- [root@rs1 ~]# echo static 192.168.0.101 > /usr/share/nginx/html/static/index.html
- [root@rs2 ~]# mkdir /var/www/html/api/
- [root@rs2 ~]# echo api 192.168.0.102 > /var/www/html/api/index.html
测试访问
- [root@node10 ~]# curl 172.25.254.100/api/
- api 192.168.0.102
- [root@node10 ~]# curl 172.25.254.100/static/
- static 192.168.0.101
haproxy可以实现https的证书安全,从用户到haproxy为https,从haproxy到后端服务器用http通信 但基于性能考虑,生产中证书都是在后端服务器比如nginx上实现
- #配置HAProxy支持https协议,支持ssl会话;
- bind *:443 ssl crt /PATH/TO/SOME_PEM_FILE
- #指令 crt 后证书文件为PEM格式,需要同时包含证书和所有私钥
- cat demo.key demo.crt > demo.pem
- #把80端口的请求重向定443
- bind *:80
- redirect scheme https if !{ ssl_fc }
- haproxy ~]# mkdir /etc/haproxy/certs/
- haproxy ~]# openssl req -newkey rsa:2048 \
- -nodes -sha256 –keyout /etc/haproxy/certs/timinglee.org.key \
- -x509 -days 365 -out /etc/haproxy/certs/timinglee.org.crt
- haproxy ~]# vim /etc/haproxy/haproxy.cfg
- frontend webserver
- bind *:80
- redirect scheme https if !{ ssl_fc }
- mode http
- use_backend webcluster
- frontend webserver-https
- bind *:443 ssl crt /etc/haproxy/timinglee.org.pem
- mode http
- use_backend webcluster
- backend webcluster
- mode http
- balance roundrobin
- server web1 172.25.254.200:80 check inter 3s fall 3 rise 5
- server web2 172.25.254.201:80 check inter 3s fall 3 rise 5
- [root@客户端 ~]#curl -IkL http://172.25.254.100
- HTTP/1.1 302 Found
- content-length: 0
- location: https://www.timinglee.org/
- cache-control: no-cache
- HTTP/1.1 200 OK
- date: Sat, 04 Apr 2020 02:31:31 GMT
- server: Apache/2.4.6 (CentOS) PHP/5.4.16
- last-modified: Thu, 02 Apr 2020 01:44:13 GMT
- etag: "a-5a244f01f8adc"
- accept-ranges: bytes
- content-length: 10
- content-type: text/html; charset=UTF-8
- [root@centos6 ~]#curl -Ik https://www.timinglee.org
- HTTP/1.1 200 OK
- date: Sat, 04 Apr 2020 02:31:50 GMT
- server: Apache/2.4.6 (CentOS) PHP/5.4.16
- last-modified: Thu, 02 Apr 2020 01:44:28 GMT
- etag: "a-5a244f0fd5175"
- accept-ranges: bytes
- content-length: 10
- content-type: text/html; charset=UTF-8

至此,haproxy的笔记大致总结完成。
Copyright © 2003-2013 www.wpsshop.cn 版权所有,并保留所有权利。