抱歉，您的浏览器无法访问本站
本页面需要浏览器支持（启用）JavaScript
了解详情 >

工具

chat

badge

友链

RHCA436-基于CentOS8-4-pacemaker+corosync 仲裁机制

RHCA学习笔记RH436-pacemaker

发布于：2022年5月2日

字数：2.3k 字

时长：10 分钟

RHCA436-基于CentOS8pacemaker+corosync 仲裁机制

一、仲裁原理

仲裁投票（Quorum Voting）决定集群的健康状况，或使故障自动转移，或使集群离线。当集群中的结点发生故障时，会由其他结点接手继续提供服务，不过，当结点之间通信出现问题，或大多数结点发生故障时，集群就会停止服务，可是集群可以容忍多少个结点发生故障呢？这要由仲裁配置（Quorum Configuration）决定，仲裁配置使用多数（Majority）原则，只要集群中健康运行的结点数量达到仲裁规定的数量（多数结点投赞成票），集群就会继续提供服务，否则集群就停止提供服务。在停止提供服务期间，正常结点持续监控故障结点是否恢复正常，一旦正常结点的数量恢复到仲裁规定的数量，集群就恢复正常，继续提供服务。集群的仲裁投票默认是启用的。

二、计算票数

floor是向下取整，2.5取2，2.9也取2

三、4节点集群查看仲裁情况

[root@nodea ~]# corosync-quorumtool 
Quorum information
------------------
Date:             Thu Feb 10 16:25:53 2022
Quorum provider:  corosync_votequorum
Nodes:            4        #集群内的所有存活节点
Node ID:          1        #在哪个节点上查看的
Ring ID:          1.57
Quorate:          Yes      #代表仲裁生效

Votequorum information
----------------------
Expected votes:   4        **#期待票数，集群所有参与投票的节点数量，不会变**
Highest expected: 4        **#最高票数**
Total votes:      4       ** ****#总票数，是指所有存活节点的数量**
Quorum:           3        **#集群存活，最少需要三个节点在线，=（期待票数/2+1)=4/2+1=3**
                          ** #3节点时，这里是2**
Flags:            Quorate   **#该状态表示集群存活**

Membership information
----------------------
    Nodeid      Votes Name
         1          1 nodea.private.example.com (local)
         2          1 nodeb.private.example.com
         3          1 nodec.private.example.com
         4          1 noded.private.example.com

[root@nodea ~]# pcs stonith 
  * fence_nodea  (stonith:fence_ipmilan):   Started nodea.private.example.com
  * fence_nodeb  (stonith:fence_ipmilan):   Started nodeb.private.example.com
  * fence_nodec  (stonith:fence_ipmilan):   Started nodec.private.example.com
  * fence_noded  (stonith:fence_ipmilan):   Started nodea.private.example.com


[root@nodea ~]# pcs quorum status

4节点集群，半数节点故障，集群处于不可用状态

[root@nodea ~]# corosync-quorumtool 
Quorum information
------------------
Date:             Thu Feb 10 16:30:28 2022
Quorum provider:  corosync_votequorum
Nodes:            2
Node ID:          1
Ring ID:          1.68
Quorate:          No      #2节点，仲裁不生效了

Votequorum information
----------------------
Expected votes:   4    #期待票数不会变，还是4
Highest expected: 4
Total votes:      2
Quorum:           3 Activity blocked 
Flags:            #集群不可用

Membership information
----------------------
    Nodeid      Votes Name
         1          1 nodea.private.example.com (local)
         2          1 nodeb.private.example.com

[root@nodea ~]# pcs status
Cluster name: mucluster
Cluster Summary:
  * Stack: corosync
  * Current DC: nodeb.private.example.com (version 2.0.4-6.el8-2deceaa3ae) - partition WITHOUT quorum
  * Last updated: Thu Feb 10 16:31:31 2022
  * Last change:  Thu Feb 10 15:55:21 2022 by root via cibadmin on nodea.private.example.com
  * 4 nodes configured
  * 4 resource instances configured

Node List:
  * Online: [ nodea.private.example.com nodeb.private.example.com ]
  * OFFLINE: [ nodec.private.example.com noded.private.example.com ]

Full List of Resources:
  * fence_nodea  (stonith:fence_ipmilan):   Stopped  fence也挂了**？**
  * fence_nodeb  (stonith:fence_ipmilan):   Stopped
  * fence_nodec  (stonith:fence_ipmilan):   Stopped
  * fence_noded  (stonith:fence_ipmilan):   Stopped

Daemon Status:
  corosync: active/enabled
  pacemaker: active/enabled
  pcsd: active/enabled

3节点时仲裁状态

1.先移除节点
[root@nodea ~]# pcs cluster stop noded.private.example.com
[root@nodea ~]# pcs cluster node remove noded.private.example.com

2.查看仲裁状态
[root@nodea ~]# corosync-quorumtool 
Quorum information
------------------
Date:             Thu Feb 10 16:43:29 2022
Quorum provider:  corosync_votequorum
Nodes:            3
Node ID:          1
Ring ID:          1.74
Quorate:          Yes

Votequorum information
----------------------
Expected votes:   3
Highest expected: 3
Total votes:      3
Quorum:           2  **#至少两个节点存活，集群才会存活**
Flags:            Quorate 

Membership information
----------------------
    Nodeid      Votes Name
         1          1 nodea.private.example.com (local)
         2          1 nodeb.private.example.com
         3          1 nodec.private.example.com

仲裁参数

man 5 votequorum
--auto_tie_breaker : 当出现脑裂时，票数时50%对50%，那么节点id小的一边正常工作，相当于静态优先级
演示： pcs quorum update auto_tie_breaker=1 
down两个节点，集群还是正常运行，仲裁也生效，如果只剩一个节点了，则不生效了，就需要加--two_node参数 


--wait_for_all : 集群在第一次启动时，等待所有的节点都启动后再计算仲裁，否则一旦访问不到节点就fence，那么重启的节点永远会起不来

--two_node : 针对集群只有2个节点时，如果有一个节点故障，期待票数会设置为1，也可以继续提供服务，当开启--wait_for_all 时，该参数功能自动开启，当节点数大于2时，该参数功能自动关闭

--last_man_standing : 开启该功能，集群会每10s钟计算一次期待票数，特别时针对集群降级时，一台一台进行下线时，期待票数会很快的进行计算，期待票数的随着集群节点的减少而逐渐降低，直到只剩下2台为止

说明：
不加此参数，如果集群内有节点故障，那么期待票数不会变，可能会导致超过半数节点故障时，集群将不再工作
加上此参数，如果集群内有节点故障，那么期待票数会每隔10s钟重新计算直到只剩2台，即使节点故障也不会导致集群不工作的场景，如果只剩2台节点，则需要仲裁服务器来解决集群不工作的场景，或者下面的那个参数，下面参数无法解决脑裂问题，仲裁服务器可以解决

修改仲裁参数

  1. 打印仲裁参数
[root@nodea ~]# pcs quorum config
Options:

[root@nodea ~]# vim /etc/corosync/corosync.conf
quorum {
    provider: corosync_votequorum
}
  1. 先停止集群
[root@nodea ~]# pcs cluster stop --all
  1. 命令行修改仲裁参数，所有节点都会生效
#命令行修改参数
[root@nodea ~]# pcs quorum update auto_tie_breaker=1 wait_for_all=1 last_man_standing=1
Checking corosync is not running on nodes...
nodea.private.example.com: corosync is not running
noded.private.example.com: corosync is not running
nodeb.private.example.com: corosync is not running
nodec.private.example.com: corosync is not running
Sending updated corosync.conf to nodes...
nodea.private.example.com: Succeeded
nodeb.private.example.com: Succeeded
nodec.private.example.com: Succeeded
noded.private.example.com: Succeeded

#查看仲裁参数
[root@nodea ~]# pcs quorum config
Options:
  auto_tie_breaker: 1
  last_man_standing: 1
  wait_for_all: 1

#查看配置文件已被修改
[root@nodea ~]# cat /etc/corosync/corosync.conf
quorum {
    provider: corosync_votequorum
    last_man_standing: 1
    wait_for_all: 1
    auto_tie_breaker: 1

}
  1. 配置文件修改参数
#修改配置文件
[root@nodea ~]# cat /etc/corosync/corosync.conf
quorum {
    provider: corosync_votequorum
    last_man_standing: 1
    wait_for_all: 1
    auto_tie_breaker: 1

}

#同步配置文件到其他节点
[root@nodea ~]# pcs cluster sync
nodea.private.example.com: Succeeded
nodeb.private.example.com: Succeeded
nodec.private.example.com: Succeeded
noded.private.example.com: Succeeded

#查看仲裁参数
[root@nodea ~]# pcs quorum config
Options:
    auto_tie_breaker: 1
    last_man_standing: 1
    wait_for_all: 1
  1. 重新启动集群
[root@nodea ~]# pcs cluster start --all
  1. 查看仲裁结果
[root@nodea ~]# corosync-quorumtool 
Quorum information
------------------
Date:             Fri Feb 11 06:28:19 2022
Quorum provider:  corosync_votequorum
Nodes:            4
Node ID:          1
Ring ID:          1.a8
Quorate:          Yes

Votequorum information
----------------------
Expected votes:   4        ** ****#期待票数为4**
Highest expected: 4
Total votes:      4
Quorum:           3  
Flags:            Quorate **WaitForAll LastManStanding AutoTieBreaker  #多了这些参数**

Membership information
----------------------
    Nodeid      Votes Name
         1          1 nodea.private.example.com (local)
         2          1 nodeb.private.example.com
         3          1 nodec.private.example.com
         4          1 noded.private.example.com
  1. 测试4节点集群，故障1台节点后，期待票数变化
#noded关机
[root@noded ~]# init 0

#再查看期待票数重新计算
[root@nodea ~]# corosync-quorumtool 
Quorum information
------------------
Date:             Fri Feb 11 06:29:30 2022
Quorum provider:  corosync_votequorum
Nodes:            3
Node ID:          1
Ring ID:          1.ac
Quorate:          Yes

Votequorum information
----------------------
Expected votes:   3         ** ****#期待票数重新计算为3**
Highest expected: 3
Total votes:      3
Quorum:           2  
Flags:            Quorate WaitForAll LastManStanding AutoTieBreaker 

Membership information
----------------------
    Nodeid      Votes Name
         1          1 nodea.private.example.com (local)
         2          1 nodeb.private.example.com
         3          1 nodec.private.example.com
  1. 测试4节点集群，故障2台节点，集群仍然正常工作
[root@nodea ~]# corosync-quorumtool 
Quorum information
------------------
Date:             Fri Feb 11 06:32:07 2022
Quorum provider:  corosync_votequorum
Nodes:            2
Node ID:          1
Ring ID:          1.b0
Quorate:          Yes    ** ****#状态依然为仲裁 **

Votequorum information
----------------------
Expected votes:   2     **#期待票数变为2，集群仍然继续工作**
Highest expected: 2
Total votes:      2
Quorum:           2  
Flags:            Quorate WaitForAll LastManStanding AutoTieBreaker 

Membership information
----------------------
    Nodeid      Votes Name
         1          1 nodea.private.example.com (local)
         2          1 nodeb.private.example.com


[root@nodea ~]# pcs cluster status
Cluster Status:
 Cluster Summary:
   * Stack: corosync
   * Current DC: nodeb.private.example.com (version 2.0.4-6.el8-2deceaa3ae) - partition with quorum
   * Last updated: Fri Feb 11 06:32:33 2022
   * Last change:  Fri Feb 11 06:17:33 2022 by hacluster via crmd on nodea.private.example.com
   * 4 nodes configured
   * 4 resource instances configured
 Node List:
   * Online: [ nodea.private.example.com nodeb.private.example.com ]
   * OFFLINE: [ nodec.private.example.com noded.private.example.com ]

PCSD Status:
  nodea.private.example.com: Online
  nodeb.private.example.com: Online
  noded.private.example.com: Offline
  nodec.private.example.com: Offline

**#fence也是正常的**
[root@nodea ~]# pcs status 
Full List of Resources:
  * fence_nodea  (stonith:fence_ipmilan):   Started nodea.private.example.com
  * fence_nodeb  (stonith:fence_ipmilan):   Started nodeb.private.example.com
  * fence_nodec  (stonith:fence_ipmilan):   Started nodea.private.example.com
  * fence_noded  (stonith:fence_ipmilan):   Started nodeb.private.example.com

9.只剩下1个节点，集群仍然正常工作

[root@nodea ~]# corosync-quorumtool 
Quorum information
------------------
Date:             Fri Feb 11 06:36:45 2022
Quorum provider:  corosync_votequorum
Nodes:            1
Node ID:          1
Ring ID:          1.b4
Quorate:          Yes

Votequorum information
----------------------
Expected votes:   1
Highest expected: 1
Total votes:      1
Quorum:           1  
Flags:            Quorate WaitForAll LastManStanding AutoTieBreaker 

Membership information
----------------------
    Nodeid      Votes Name
         1          1 nodea.private.example.com (local)

**#fence也是正常的，如果fence显示stop，可以用pcs stonith refresh 刷新一下**
[root@nodea ~]# pcs status 
Full List of Resources:
  * fence_nodea  (stonith:fence_ipmilan):   Started nodea.private.example.com
  * fence_nodeb  (stonith:fence_ipmilan):   Started nodeb.private.example.com
  * fence_nodec  (stonith:fence_ipmilan):   Started nodea.private.example.com
  * fence_noded  (stonith:fence_ipmilan):   Started nodeb.private.example.com

说明：如果不加仲裁参数，超过半数节点故障，则集群不可用，fence也会故障
问题：解决2节点集群不工作（不工作是因为怕脑裂）的场景，可以用到仲裁服务器或者仲裁磁盘实现

更新于：2024年7月25日

字数：2.3k 字

时长：10 分钟

评论