keepalived：
HTTP_GET //使用keepalived获取后端real server健康状态检测
SSL_GET(https) //这里以为这后端使用的是http协议
TCP_CHECK
下面演示基于TCP_CHECK做检测
# man keepalived //查看TCP_CHECK配置段
# TCP healthchecker
TCP_CHECK
{
# ======== generic connection options
# Optional IP address to connect to.
# The default is the realserver IP //默认使用real server的IP
connect_ip <IP ADDRESS> //可省略
# Optional port to connect to
# The default is the realserver port
connect_port <PORT> //可省略
# Optional interface to use to
# originate the connection
bindto <IP ADDRESS>
# Optional source port to
# originate the connection from
bind_port <PORT>
# Optional connection timeout in seconds.
# The default is 5 seconds
connect_timeout <INTEGER>
# Optional fwmark to mark all outgoing
# checker packets with
fwmark <INTEGER>
# Optional random delay to start the initial check
# for maximum N seconds.
# Useful to scatter multiple simultaneous
# checks to the same RS. Enabled by default, with
# the maximum at delay_loop. Specify 0 to disable
warmup <INT>
# Retry count to make additional checks if check
# of an alive server fails. Default: 1
retry <INT>
# Delay in seconds before retrying. Default: 1
delay_before_retry <INT>
} #TCP_CHECK
# cd /etc/keepalived
# vim keepalived.conf //两台keepalived都要设置
1 virtual_server 192.168.184.150 80 { //这里可以合并
2 delay_loop 6
3 lb_algo wrr
4 lb_kind DR
5 net_mask 255.255.0.0
6 protocol TCP
7 sorry_server 127.0.0.1 80
8
9 real_server 192.168.184.143 80 {
10 weight 1
11 TCP_CHECK {
12 connect_timeout 3
13 }
14 }
15
16 real_server 192.168.184.144 80 {
17 weight 2
18 TCP_CHECK {
19 connect_timeout 3
20 }
21 }
22 }
systemctl restart keepalived
# systemctl status keepalived
1 ● keepalived.service - LVS and VRRP High Availability Monitor 2 Loaded: loaded (/usr/lib/systemd/system/keepalived.service; disabled; vendor preset: disabled) 3 Active: active (running) since Thu 2018-12-13 23:11:06 CST; 1min 32s ago 4 Process: 6233 ExecStart=/usr/sbin/keepalived $KEEPALIVED_OPTIONS (code=exited, status=0/SUCCESS) 5 Main PID: 6234 (keepalived) 6 CGroup: /system.slice/keepalived.service 7 ├─6234 /usr/sbin/keepalived -D 8 ├─6235 /usr/sbin/keepalived -D 9 └─6236 /usr/sbin/keepalived -D 10 11 Dec 13 23:11:11 node1 Keepalived_healthcheckers[6235]: Check on service [192.168.184.144]:80 failed after 1 retry. 12 Dec 13 23:11:11 node1 Keepalived_healthcheckers[6235]: Removing service [192.168.184.144]:80 from VS [192.168.184.150]:80 13 Dec 13 23:11:11 node1 Keepalived_healthcheckers[6235]: Remote SMTP server [127.0.0.1]:25 connected. 14 Dec 13 23:11:11 node1 Keepalived_healthcheckers[6235]: SMTP alert successfully sent. 15 Dec 13 23:11:14 node1 Keepalived_vrrp[6236]: Sending gratuitous ARP on eth0 for 192.168.184.150 16 Dec 13 23:11:14 node1 Keepalived_vrrp[6236]: VRRP_Instance(VI_1) Sending/queueing gratuitous ARPs on eth0 for 192.168.184.150 17 Dec 13 23:11:14 node1 Keepalived_vrrp[6236]: Sending gratuitous ARP on eth0 for 192.168.184.150 18 Dec 13 23:11:14 node1 Keepalived_vrrp[6236]: Sending gratuitous ARP on eth0 for 192.168.184.150 19 Dec 13 23:11:14 node1 Keepalived_vrrp[6236]: Sending gratuitous ARP on eth0 for 192.168.184.150 20 Dec 13 23:11:14 node1 Keepalived_vrrp[6236]: Sending gratuitous ARP on eth0 for 192.168.184.150 //发送广播地址已经添加 21 You have new mail in /var/spool/mail/root
示例：
HTTP_GET {
url {
path /
status_code 200
}
connect_timeout 3
nb_get_retry 3
delay_before_retry 3
}
TCP_CHECK {
connect_timeout 3
}
HA Services:
nginx
100: -25
96: -20 79 --> 99 --> 79
博客作业：
keepalived 高可用 ipvs
nginx
active/active
Linux HA Cluster
LB, HA, HP, hadoop
LB:
传输层：lvs
应用层：nginx, haproxy, httpd, perlbal, ats, varnish
HA:
vrrp: keepalived
AIS: heartbeat, OpenAIS, corosync/pacemaker, cman/rgmanager(conga) RHCS
HA:
故障场景：
硬件故障：
设计缺陷
使用过久自然损坏
人为故障
…… ……
软件故障
设计缺陷
bug
人为误操作
……
A=MTBF/(MTBF+MTTR)
MTBF: Mean Time Between Failure
MTTR: Mean Time To Repair
0<A<1: 百分比
90%, 95%, 99%
99.9%, 99.99%, 99.999%
提供冗余：
network partition： vote system
隔离：
STONITH：shoot the other node on the head 节点级别隔离
Fence: 资源级别的隔离
failover domain：
fda: node1, node5
fdb: node2, node5
fdc: node3, node5
fdd: node4, node5
资源的约束性：
位置约束：资源对节点的倾向性；
排列约束：资源彼此间是否能运行于同一节点的倾向性；
顺序约束：多个资源启动顺序依赖关系；
vote system:
少数服从多数：quorum
> total/2
with quorum: 拥有法定票数
without quorum: 不拥有法定票数
两个节点(偶数个节点)：
Ping node
qdisk
failover
failback
Messaging Layer:
heartbeat
v1
v2
v3
corosync
cman
Cluster Resource Manager(CRM):
heartbeat v1 haresources (配置接口：配置文件haresources)
heartbeat v2 crm (在每个节点运行一个crmd(5560/tcp)守护进程，有命令行接口crmsh; GUI: hb_gui)
heartbeat v3, pacemaker (配置接口：crmsh, pcs; GUI: hawk(suse), LCMC, pacemaker-gui)
rgmanager (配置接口：cluster.conf, system-config-cluster, conga(webgui), cman_tool, clustat)
组合方式：
heartbeat v1 (haresources)
heartbeat v2 (crm)
heartbeat v3 + pacemaker
corosync + pacemaker
corosync v1 + pacemaker (plugin)
corosync v2 + pacemaker (standalone service)
cman + rgmanager
corosync v1 + cman + pacemaker
RHCS: Red Hat Cluster Suite
RHEL5: cman + rgmanager + conga (ricci/luci)
RHEL6: cman + rgmanager + conga (ricci/luci)
corosync + pacemaker
corosync + cman + pacemaker
RHEL7: corosync + pacemaker
Resource Agent：
service: /etc/ha.d/haresources.d/目录下的脚本；
LSB: /etc/rc.d/init.d/目录下的脚本；
OCF：Open Cluster Framework
provider:
STONITH:
Systemd:
资源类型：
primitive：主资源，原始资源；在集群中只能运行一个实例；
clone：克隆资源，在集群中可运行多个实例；
匿名克隆、全局惟一克隆、状态克隆(主动、被动)
multi-state(master/slave)：克隆资源的特殊实现；多状态资源；
group: 组资源；
启动或停止；
资源监视
相关性：
资源属性：
priority: 优先级；
target-role：started, stopped, master;
is-managed: 是否允许集群管理此资源；
resource-stickiness: 资源粘性；
allow-migrate: 是否允许迁移；
约束：score
位置约束：资源对节点的倾向性；
(-oo, +oo):
任何值+无穷大=无穷大
任何值+负无穷大=负无穷大
无穷大+负无穷大=负无穷大
排列约束：资源彼此间是否能运行于同一节点的倾向性；
(-oo, +oo)
顺序约束：多个资源启动顺序依赖关系；
(-oo, +oo)
Mandatory
安装配置：
CentOS 7： corosync v2 + pacemaker
corosync v2: vote system
pacemaker: 独立服务
集群的全生命周期管理工具：
pcs: agent(pcsd)
crmsh: agentless (pssh)
配置集群的前提：
(1) 时间同步；
(2) 基于当前正在使用的主机名互相访问；
(3) 是否会用到仲裁设备；
web serivce：
vip: 172.16.100.91
httpd
回顾：AIS HA
Messaging Layer:
heartbeat v1, v2, v3
corosync v1, v2(votequorum)
OpenAIS
CRM:
pacemaker
配置接口：crmsh (agentless), pssh
pcs (agent), pcsd
conga(ricci/luci)
group, constraint
rgmanager(cman)
resource group:
failover domain
配置：
全局属性：property, stonith-enable等等；
高可用服务：资源，通过RA
RA:
LSB: /etc/rc.d/init.d/
systemd：/etc/systemd/system/multi-user.wants
处于enable状态的服务；
OCF: [provider]
heartbeat
pacemaker
linbit
service
stonith
高可用集群的可用方案：
heartbeat v1
heartbeat v2
heartbeat v3 + pacemaker X
corosync + pacemaker
cman + rgmanager
corosync + cman + pacemaker
corosync + pacemaker
keepalived
HA Cluster(2)
Heartbeat信息传递：
Unicast, udpu
Mutlicast, udp
Broadcast
组播地址：用于标识一个IP组播域；IANA把D类地址留给组播使用：224.0.0.0-239.255.255.255
永久组播地址：224.0.0.0-224.0.0.255
临时组播地址：224.0.1.0-238.255.255.255
本地组播地址：239.0.0.0-239.255.255.255
示例配置文件：
totem {
version: 2
crypto_cipher: aes128
crypto_hash: sha1
secauth: on
interface {
ringnumber: 0
bindnetaddr: 172.16.0.0
mcastaddr: 239.185.1.31
mcastport: 5405
ttl: 1
}
}
nodelist {
node {
ring0_addr: 172.16.100.67
nodeid: 1
}
node {
ring0_addr: 172.16.100.68
nodeid: 2
}
node {
ring0_addr: 172.16.100.69
nodeid: 3
}
}
logging {
fileline: off
to_stderr: no
to_logfile: yes
logfile: /var/log/cluster/corosync.log
to_syslog: no
debug: off
timestamp: on
logger_subsys {
subsys: QUORUM
debug: off
}
}
quorum {
provider: corosync_votequorum
}
HA Web Service:
vip: 172.16.100.92, ocf:heartbeat:IPaddr
httpd: systemd
nfs shared storage: ocf:heartbeat:Filesystem
HA Cluster工作模型:
A/P：两节点集群; active/passive;
without-quorum-policy={stop|ignore|suicide|freeze}
A/A：双主模型
N-M: N个节点，M个服务，N>M;
N-N: N个节点，N个服务；
network partition:
brain-split：块级别的共享存储时，非常危险；
vote quorum:
with quorum > total/2
without quorum <= total/2
stop
ignore
suicide
freeze
CAP:
C: consistency
A: availiability
P: partition tolerance
webip, webstore, webserver
node1: 100 + 0 + 0
node2: 0 + 0 + 0
node3: 0 + 0 + 0
node2: 50+50+50
A --> B --> C
C --> B --> A
pcs：
cluster
auth
setup
resource
describe
list
create
delete
constraint
colocation
order
location
property
list
set
status
config
博客作业：
(1) 手动配置，多播：corosync+pacemaker+crmsh, 配置高可用的mysql集群，datadir指向的路径为nfs导出路径；
(2) pcs/pcsd，单播：corosync+pacemaker， 配置高可用的web集群；
单播配置示例：
某些环境中可能不支持组播。这时应该配置 Corosync 使用单播，下面是使用单播的 Corosync 配置文件的一部分：
totem {
#...
interface {
ringnumber: 0
bindnetaddr: 192.168.42.0
broadcast: yes
mcastport: 5405
}
interface {
ringnumber: 1
bindnetaddr: 10.0.42.0
broadcast: yes
mcastport: 5405
}
transport: udpu
}
nodelist {
node {
ring0_addr: 192.168.42.1
ring1_addr: 10.0.42.1
nodeid: 1
}
node {
ring0_addr: 192.168.42.2
ring1_addr: 10.0.42.2
nodeid: 2
}
}
如果将 broadcast 设置为 yes ，集群心跳将通过广播实现。设置该参数时，不能设置 mcastaddr 。
transport 配置项决定集群通信方式。要完全禁用组播，应该配置单播传输参数 udpu 。这要求将所有的节点服务器信息写入 nodelist ，也就是需要在配署 HA 集群之前确定节点组成。配认配置是 udp 。通信方式类型还支持 udpu 和 iba 。
在 nodelist 之下可以为某一节点设置只与该节点相关的信息，这些设置项只能包含在 node 之中，即只能对属于集群的节点服务器进行设置，而且只应包括那些与默认设置不同的参数。每台服务器都必须配置 ring0_addr 。
原文地址：https://www.cnblogs.com/hanshanxiaoheshang/p/10117152.html