一、前言
2013年下半年大部时候都用在LVS实施上面了,一路摸爬滚打走来,遇到各种问题,感谢同事们对我的帮助和指导,感谢领导对我的信任,本文总结一下lvs集群(ospf+fullnat)的详细部署情况以及需要注意的问题点,先大概说一下LVS在我们公司的应用情况吧,LVS在我们公司走过了以下三个阶段:
阶段一,一个业务一套LVS调度(主备模式),优缺点如下:
优点:业务和业务之间隔离,A业务有问题不会影响B业务
缺点:1、管理不方便,2、LB多了虚拟路由ID冲突导致业务异常,3、业务量足够大LB成为瓶颈
阶段二,一个IDC一套LVS调度(主备模式),优缺点如下:
优点、业务统一集中管理
缺点:1、A业务突发上流(超过LB的承受能力)会影响整个集群上的业务,2、LB很容易成为瓶颈
阶段三,一个IDC一套调度(ospf+fullnat集群模式),优缺点如下:
优点:1、LB调度机自由伸缩,横向线性扩展(最多机器数受限于三层设备允许的等价路由数目 ),2、业务统一集中管理,3、LB资源全利用,All Active。不存在备份机
缺点:部署相对比较复杂
二、环境说明
1、架构图
2、fullnat是淘宝开源的一种lvs转发模式,主要思想:引入local address(内网ip地址),cip-vip转换为lip->rip,而 lip和rip均为IDC内网ip,可以跨vlan通讯,这刚好符合我们的需求,因为我们的内网是划分了vlan的。
3、环境说明
软件环境: 系统版本:centos6.4 keepalived版本:v1.2.2 ospfd版本:version 0.99.20 zebra版本: version 0.99.20 交换机: 外网核心交换IP:114.11x.9x.1 内网核心交换IP:10.10.2.1 LB1外网机柜交换IP:114.11x.9x.122 LB2外网机柜交换IP:114.11x.9x.160 LB1内网机柜交换IP:10.10.15.254 LB2内网机柜交换IP:10.10.11.254 LB1: 调度机IP(外网bond1):10.10.254.18/30 ##外网需要配置一个与核心交换机联通的私有地址 调度机IP(内网bond0):10.10.15.77 内网分发私有网段:10.10.251.0/24 ##local address 外网ospf转发网段网关:10.10.254.17 LB2: 调度机IP(外网bond1):10.10.254.22/30 #外网需要配置一个与核心交换机联通的私有地址 调度机IP(内网bond0):10.10.11.77 内网分发私有网段:10.10.250.0/24 ##local address 外网ospf转发网段网关:10.10.254.21
为了提升网络吞吐量,网络冗余,我们LB上网卡是做bond的,详细说明如下图:
三、具体部署
部署分三大部分,网卡绑定、ospf配置和lvs配置,下面依次介绍:
A、网卡绑定部分
1、服务器(LB1)上配置如下:
[root@lvs_cluster_C1 ~]# vi /etc/sysconfig/network-scripts/ifcfg-bond0 DEVICE=bond0 ONBOOT=yes BOOTPROTO=static IPADDR=10.10.11.77 ##LB2为10.10.15.77 NETMASK=255.255.255.0 USERCTL=no TYPE=Ethernet [root@lvs_cluster_C1 ~]# vi /etc/sysconfig/network-scripts/ifcfg-bond1 DEVICE=bond1 ONBOOT=yes BOOTPROTO=static IPADDR=10.10.254.22 ##LB2为10.10.254.18 NETMASK=255.255.255.252 USERCTL=no TYPE=Ethernet [root@lvs_cluster_C1 ~]# vi /etc/sysconfig/network-scripts/ifcfg-eth0 DEVICE=eth0 BOOTPROTO=none ONBOOT=yes USERCTL=no MASTER=bond0 SLAVE=yes [root@lvs_cluster_C1 ~]# vi /etc/sysconfig/network-scripts/ifcfg-eth1 DEVICE=eth1 BOOTPROTO=none ONBOOT=yes USERCTL=no MASTER=bond0 SLAVE=yes [root@lvs_cluster_C1 ~]# vi /etc/sysconfig/network-scripts/ifcfg-eth2 DEVICE=eth2 BOOTPROTO=none ONBOOT=yes USERCTL=no MASTER=bond1 SLAVE=yes [root@lvs_cluster_C1 ~]# vi /etc/sysconfig/network-scripts/ifcfg-eth3 DEVICE=eth3 BOOTPROTO=none ONBOOT=yes USERCTL=no MASTER=bond1 SLAVE=yes [root@lvs_cluster_C1 ~]# vi /etc/modprobe.d/openfwwf.conf options b43 nohwcrypt=1 qos=0 alias bond0 bonding options bond0 miimon=100 mode=0 #bond的几种模式的详细说明我之前写的博文中有介绍或者去百度、谷歌吧 alias bond1 bonding options bond1 miimon=100 mode=0 alias net-pf-10 off
2、交换机上配置
bond0:10.10.15.77 eth0-GigabitEthernet1/0/29 eth1-GigabitEthernet1/0/30 LB1对应的内网机柜交换机操作: interface Bridge-Aggregation10 port access vlan 150 interface GigabitEthernet1/0/29 port link-aggregation group 10 interface GigabitEthernet1/0/30 port link-aggregation group 10 bond1:10.10.254.18 GigabitEthernet1/0/26 GigabitEthernet1/0/28 LB1对应的外网机柜交换机操作: vlan 50 interface Bridge-Aggregation10 port access vlan 50 interface GigabitEthernet1/0/26 port link-aggregation group 10 interface GigabitEthernet1/0/28 port link-aggregation group 10 bond0:10.10.11.77 eth0:GigabitEthernet1/0/14 eth1:GigabitEthernet1/0/05 LB2对应的内网机柜交换机操作: interface Bridge-Aggregation110 port access vlan 110 interface GigabitEthernet1/0/05 port link-aggregation group 110 interface GigabitEthernet1/0/14 port link-aggregation group 110 bond1:10.10.254.22 eth2:GigabitEthernet1/0/38 eth3:GigabitEthernet1/0/46 LB2对应的外网机柜交换机操作: vlan 60 interface Bridge-Aggregation110 port access vlan 60 interface GigabitEthernet1/0/38 port link-aggregation group 110 interface GigabitEthernet1/0/46 port link-aggregation group 110 display link-aggregation verbose #查看绑定状态是否ok
B、ospf配置部分
1、交换机上配置:
外网核心操作: vlan 50 vlan 60 interface Vlan-interface50 ip address 10.10.254.17 255.255.255.252 ospf timer hello 1 ospf timer dead 4 ospf dr-priority 96 interface Vlan-interface60 ip address 10.10.254.21 255.255.255.252 ospf timer hello 1 ospf timer dead 4 ospf dr-priority 95 #配置ospf的参数, timer hello是发送hello包的间隔,timer dead是存活的死亡时间。默认是10、40,hello包是ospf里面维持邻居关系的报文,这里配置是每秒发送一个,当到4秒还没有收到这个报文,就会认为这个邻居已经丢失,需要修改路由 ospf 1 area 0.0.0.0 network 10.10.254.16 0.0.0.3 network 10.10.254.20 0.0.0.3 内网核心: interface Vlan-interface110 ip address 10.10.250.1 255.255.255.0 sub interface Vlan-interface150 ip address 10.10.251.1 255.255.255.0 sub
2、服务器上配置
mkdir /etc/quagga/ mkdir -p /var/log/quagga/ chmod -R 777 /var/log/quagga/
配置文件:
cat /etc/quagga/zebra.conf hostname lvs_cluster_C2 ##LB2为:hostname lvs_cluster_C1 cat /etc/quagga/ospfd.conf log file /var/log/quagga/ospfd.log log stdout log syslog interface bond1 ip ospf hello-interval 1 ip ospf dead-interval 4 router ospf ospf router-id 10.10.254.17 ##LB2为:10.10.254.21 log-adjacency-changes auto-cost reference-bandwidth 1000 network 114.11x.9x.0/24 area 0.0.0.0 network 10.10.254.16/30 area 0.0.0.0 ##LB2为:10.10.254.20/30
ospfd的启动脚本:
[root@lvs_cluster_C1 ~]# cat /etc/init.d/ospfd #!/bin/bash # chkconfig: - 16 84 # config: /etc/quagga/ospfd.conf ### BEGIN INIT INFO # Provides: ospfd # Short-Description: A OSPF v2 routing engine # Description: An OSPF v2 routing engine for use with Zebra ### END INIT INFO # source function library . /etc/rc.d/init.d/functions # Get network config . /etc/sysconfig/network # quagga command line options . /etc/sysconfig/quagga RETVAL=0 PROG="ospfd" cmd=ospfd LOCK_FILE=/var/lock/subsys/ospfd CONF_FILE=/etc/quagga/ospfd.conf case "$1" in start) # Check that networking is up. [ "${NETWORKING}" = "no" ] && exit 1 # The process must be configured first. [ -f $CONF_FILE ] || exit 6 if [ `id -u` -ne 0 ]; then echo $"Insufficient privilege" 1>&2 exit 4 fi echo -n $"Starting $PROG: " daemon $cmd -d $OSPFD_OPTS RETVAL=$? [ $RETVAL -eq 0 ] && touch $LOCK_FILE echo ;; stop) echo -n $"Shutting down $PROG: " killproc $cmd RETVAL=$? [ $RETVAL -eq 0 ] && rm -f $LOCK_FILE echo ;; restart|reload|force-reload) $0 stop $0 start RETVAL=$? ;; condrestart|try-restart) if [ -f $LOCK_FILE ]; then $0 stop $0 start fi RETVAL=$? ;; status) status $cmd RETVAL=$? ;; *) echo $"Usage: $PROG {start|stop|restart|reload|force-reload|try-restart|status}" exit 2 esac exit $RETVAL
zebra的启动脚本:
[root@lvs_cluster_C1 ~]# cat /etc/init.d/zebra #!/bin/bash # chkconfig: - 15 85 # config: /etc/quagga/zebra.conf ### BEGIN INIT INFO # Provides: zebra # Short-Description: GNU Zebra routing manager # Description: GNU Zebra routing manager ### END INIT INFO # source function library . /etc/rc.d/init.d/functions # quagga command line options . /etc/sysconfig/quagga RETVAL=0 PROG="zebra" cmd=zebra LOCK_FILE=/var/lock/subsys/zebra CONF_FILE=/etc/quagga/zebra.conf case "$1" in start) # Check that networking is up. [ "${NETWORKING}" = "no" ] && exit 1 # The process must be configured first. [ -f $CONF_FILE ] || exit 6 if [ `id -u` -ne 0 ]; then echo $"Insufficient privilege" 1>&2 exit 4 fi echo -n $"Starting $PROG: " /sbin/ip route flush proto zebra daemon $cmd -d $ZEBRA_OPTS RETVAL=$? [ $RETVAL -eq 0 ] && touch $LOCK_FILE echo ;; stop) echo -n $"Shutting down $PROG: " killproc $cmd RETVAL=$? [ $RETVAL -eq 0 ] && rm -f $LOCK_FILE echo ;; restart|reload|force-reload) $0 stop $0 start RETVAL=$? ;; condrestart|try-restart) if [ -f $LOCK_FILE ]; then $0 stop $0 start fi RETVAL=$? ;; status) status $cmd RETVAL=$? ;; *) echo $"Usage: $0 {start|stop|restart|reload|force-reload|try-restart|status}" exit 2 esac exit $RETVAL
quagga的配置:
[root@lvs_cluster_C1 ~]# cat /etc/sysconfig/quagga # # Default: Bind all daemon vtys to the loopback(s) only # QCONFDIR="/etc/quagga" BGPD_OPTS="-A 127.0.0.1 -f ${QCONFDIR}/bgpd.conf" OSPF6D_OPTS="-A ::1 -f ${QCONFDIR}/ospf6d.conf" OSPFD_OPTS="-A 127.0.0.1 -f ${QCONFDIR}/ospfd.conf" RIPD_OPTS="-A 127.0.0.1 -f ${QCONFDIR}/ripd.conf" RIPNGD_OPTS="-A ::1 -f ${QCONFDIR}/ripngd.conf" ZEBRA_OPTS="-A 127.0.0.1 -f ${QCONFDIR}/zebra.conf" ISISD_OPTS="-A ::1 -f ${QCONFDIR}/isisd.conf" # Watchquagga configuration (please check timer values before using): WATCH_OPTS="" WATCH_DAEMONS="zebra bgpd ospfd ospf6d ripd ripngd" # To enable restarts, uncomment this line (but first be sure to edit # the WATCH_DAEMONS line to reflect the daemons you are actually using): #WATCH_OPTS="-Az -b_ -r/sbin/service_%s_restart -s/sbin/service_%s_start -k/sbin/service_%s_stop"
3、服务启动:
/etc/init.d/zebra start && chkconfig zebra on /etc/init.d/ospfd start && chkconfig ospfd on PS:先启动zebra再启动ospf,不然LB会学习不到路由信息
C、lvs部署部分
1、安装组件
#安装带fullnat功能内核(淘宝已开源) rpm -ivh kernel-2.6.32-220.23.3.el6.x86_64.rpm kernel-firmware-2.6.32-220.23.3.el6.x86_64.rpm --force #安装lvs-tools(ipvsadm,keepalived,quagga),这些工具都是依据新内核修改过的,所以不要用原生的 rpm -ivh lvs-tools-1.0.0-77.el6.x86_64.rpm
2、添加local_address网段
cat /opt/sbin/ipadd.sh #!/bin/bash arg=$1 dev=bond0 network="10.10.251" ##LB2 10.10.250 seq="2 254" function start() { for i in `seq $seq` do ip addr add $network.$i/32 dev $dev done } function stop() { for i in `seq $seq` do ip addr del $network.$i/32 dev $dev done } case "$arg" in start) start ;; stop) stop ;; restart) stop start ;; esac echo "/opt/sbin/ipadd.sh" >> /etc/rc.local ##加入开机启动
3、keepalived的配置文件
1>配置问的大概说明
. ├── gobal_module ##全局配置文件 ├── info.txt ##记录集群部署的业务信息 ├── keepalived.conf ##主配置文件 ├── kis_ops_test.conf ##业务配置文件 ├── local_address.conf ##local_address ├── lvs_module ##所有业务的include配置 └── realserver #rs目录 └── kis_ops_test_80.conf ##业务的realserver的配置文件
2>配置文件的内容说明
[root@lvs_cluster_C1 keepalived]# cat /etc/keepalived/gobal_module
! global configure file global_defs { notification_email { [email protected] } notification_email_from [email protected] smtp_server 127.0.0.1 smtp_connect_timeout 30 router_id LVS_CLUSTER }
[root@lvs_cluster_C1 keepalived]# cat /etc/keepalived/local_address.conf
local_address_group laddr_g1 { 10.10.250.2-254 ##LB1 10.10.251.2-254 }
PS:local_address每个LB不能重复
[root@lvs_cluster_C1 keepalived]# cat /etc/keepalived/lvs_module
include ./kis_ops_test.conf #测试
[root@lvs_cluster_C1 keepalived]# cat /etc/keepalived/kis_ops_test.conf
#把local_address include进来 include ./local_address.conf #VIP组,可以有多个vip virtual_server_group kis_ops_test_80 { 114.11x.9x.185 80 #kis_ops_test } virtual_server group kis_ops_test_80 { delay_loop 7 lb_algo wrr lb_kind FNAT protocol TCP nat_mask 255.255.255.0 persistence_timeout 0 #回话保持机制,默认为0 syn_proxy ##开启此参数可以有效防范SynFlood*** laddr_group_name laddr_g1 alpha #开启alpha模式:启动时默认rs是down的状态,健康检查通过后才会添加到vs pool omega #开启omega模式,清除rs时会执行相应的脚本(rs的notify_up,quorum_up) quorum 1 #服务是否有效的阀值(正常工作rs的wight值) hysteresis 0 #延迟系数跟quorum配合使用 #高于或低于阀值时会执行以下脚本。 quorum_up " ip addr add 114.11x.9x.185/32 dev lo ;" quorum_down " ip addr del 114.11x.9x.185/32 dev lo ;" include ./realserver/kis_ops_test_80.conf }
[root@lvs_cluster_C1 ~]# cat /etc/keepalived/realserver/kis_ops_test_80.conf
#real_server 10.10.2.240 80 { # weight 1 # inhibit_on_failure # TCP_CHECK { # connect_timeout 3 # nb_get_retry 3 ##TCP_CHECK 方式此参数不生效 # delay_before_retry 3 ##TCP_CHECK 方式此参数不生效 # connect_port 80 # } #} real_server 10.10.2.240 80 { weight 1 inhibit_on_failure HTTP_GET { url { path /abc digest 134b225d509b9c40647377063d211e75 } connect_timeout 3 nb_get_retry 3 delay_before_retry 3 connect_port 80 } }
到这里LB上的配置基本上完成了,还有一点需要配置哈,那就是要在LB上配置路由策略,不然vip是ping不通的,但是不影响http访问,具体配置如下:
echo "from 114.11x.9x.0/24 table LVS_CLUSTER" >/etc/sysconfig/network-scripts/rule-bond1 echo "default table LVS_CLUSTER via 10.10.254.21 dev bond1" >/etc/sysconfig/network-scripts/route-bond1 ##LB1 via 10.10.254.17 echo "203 LVS_CLUSTER" >> /etc/iproute2/rt_tables /etc/init.d/network restart ##重启网络
四、realserver的配置
1、realserver需要更换带toa模块的内核,如果不更改的话你的web服务(比如nginx)的日志获取不到用户的真实IP,而是记录了LB的local_address的IP
centos5系列的系统: rpm -ivh kernel-2.6.18-274.18.2.el5.kis_toa.x86_64.rpm centos6系列的系统: rpm -ivh kernel-toa-2.6.32-220.23.3.el6.kis_toa.x86_64.rpm
2、realserver要能和LB的local_address互访即可,这个要看大家的内网环境了。
五、LB的调优
1、网卡调优,这个很重要,如果不调优大流量下,cpu单核耗尽会把LB搞死的,本人亲身体验过,
具体看参考我之前的文章,高并发、大流量网卡调优
2、内核参数调优
#该参数决定了,网络设备接收数据包的速率比内核处理这些包的速率快时,允许送到队列的数据包的最大数目 net.core.netdev_max_backlog = 500000 #开启路由转发功能 net.ipv4.ip_forward = 1 #修改文件描述符 fs.nr_open = 5242880 fs.file-max = 4194304 sed -i 's/1024/4194304/g' /etc/security/limits.conf
PS:目前集群有这么个问题,realserver不能访问自己的vip,因为有些业务确实有这样的需要,最后想到以下解决办法:
1、如果realserver上有公网IP且和vip是同一个段则需要加路由: route add -net vip netmask 255.255.255.255 gw 114.11x.9x.1 2、如果realserver走的nat网关,而nat网关的出口是和vip同一个网段,那就需要在nat上加如上功能的路由即可