最近PXC 5.7出现脑裂，前端Navicate连接到MySQL时，提示WSREP has not yet prepared node for application use错误。脑裂可是个不容易忽视的问题啊，严重的导致数据错乱。你懂的，就好比酒驾，迷糊啊，都乱了。下文是本次故障的具体描述。
一、环境说明
当前的集群采用了2个数据节点+1个仲裁节点，两节点均有写入数据
重启了仲裁节点后，数据节点发生脑裂
操作系统CentOS 7
数据库版本Percona XtraDB Cluster 5.7.20-18-57-log
故障现象
Navicate客户端可以连接，但是查询数据报错：WSREP has not yet prepared node for application use
2节点状态查看都显示为non-Primary
mysql> show status like '%wsrep_cluster%';
+--------------------------+--------------------------------------+
| Variable_name | Value |
+--------------------------+--------------------------------------+
| wsrep_cluster_conf_id | 6 |
| wsrep_cluster_size | 3 |
| wsrep_cluster_state_uuid | cd96b06a-0a1d-11e8-99d2-837e6f3b95a9 |
| wsrep_cluster_status | non-Primary |
+--------------------------+--------------------------------------+二、故障分析
节点1日志
2018-05-29T11:36:56.595913+08:00 0 [Note] WSREP: (45125952, 'tcp://0.0.0.0:4567') connection to peer 17b576f8 with addr tcp://192.168.1.253:4567 timed out, no messages seen in PT3S (gmcast.peer_timeout)
2018-05-29T11:36:56.602911+08:00 0 [Note] WSREP: (45125952, 'tcp://0.0.0.0:4567') turning message relay requesting on, nonlive peers: tcp://192.168.1.253:4567
2018-05-29T11:37:05.600073+08:00 0 [Note] WSREP: Current view of cluster as seen by this node
view (view_id(NON_PRIM,029e6b40,17) ##此处已提示为非主节点
memb {
45125952,0
}
joined {
}
left {
}
partitioned {
029e6b40,0
17b576f8,0
}
)
2018-05-29T11:37:05.631952+08:00 0 [Note] WSREP: Current view of cluster as seen by this node
view (view_id(NON_PRIM,45125952,18)
memb {
45125952,0
}
joined {
}
left {
}
partitioned {
029e6b40,0
17b576f8,0
}
)
2018-05-29T11:37:05.634073+08:00 0 [Note] WSREP: declaring 029e6b40 at tcp://192.168.1.249:4567 stable
2018-05-29T11:37:05.634674+08:00 0 [Note] WSREP: Current view of cluster as seen by this node
view (view_id(NON_PRIM,029e6b40,19)
memb {
029e6b40,0
45125952,0
}
joined {
}
left {
}
partitioned {
17b576f8,0
}
)
2018-05-29T11:37:05.663943+08:00 0 [Note] WSREP: New COMPONENT: primary = no, bootstrap = no, my_idx = 0, memb_num = 1
2018-05-29T11:37:05.670701+08:00 151042 [Warning] WSREP: Send action {(nil), 563, TORDERED} returned -107 (Transport endpoint is not connected)
2018-05-29T11:37:05.670763+08:00 151042 [Note] WSREP: --------- CONFLICT DETECTED -------- ###检查到冲突
2018-05-29T11:37:05.670778+08:00 151042 [Note] WSREP: cluster conflict due to certification failure for threads:节点2日志
2018-05-28T23:59:07.757523+08:00 0 [Note] WSREP: Member 2.0 (pxc-cluster-node-1) synced with group.
2018-05-29T09:47:36.679991+08:00 6095593 [Note] Aborted connection 6095593 to db: 'DB71' user: 'user71' host: '192.168.1.253' (Got an error reading communication packets)
2018-05-29T09:47:36.680062+08:00 6095607 [Note] Aborted connection 6095607 to db: 'DB71' user: 'user71' host: '192.168.1.253' (Got an error reading communication packets)
.............................................#Author : Leshami Blog : http://blog.csdn.net/leshami
2018-05-29T11:34:11.981171+08:00 6104858 [Note] Aborted connection 6104858 to db: 'DB71' user: 'user71' host: '192.168.1.253' (Got an error reading communication packets)
2018-05-29T11:36:02.203516+08:00 0 [Note] WSREP: (029e6b40, 'tcp://0.0.0.0:4567') connection to peer 17b576f8 with addr tcp://192.168.1.253:4567 timed out, no messages seen in PT3S (gmcast.peer_timeout)
2018-05-29T11:36:02.203708+08:00 0 [Note] WSREP: (029e6b40, 'tcp://0.0.0.0:4567') turning message relay requesting on, nonlive peers: tcp://192.168.1.253:4567
2018-05-29T11:36:03.203906+08:00 0 [Note] WSREP: (029e6b40, 'tcp://0.0.0.0:4567') reconnecting to 17b576f8 (tcp://192.168.1.253:4567), attempt 0
2018-05-29T11:36:03.704413+08:00 0 [Note] WSREP: declaring node with index 1 suspected, timeout PT5S (evs.suspect_timeout)
2018-05-29T11:36:03.704482+08:00 0 [Note] WSREP: evs::proto(029e6b40, OPERATIONAL, view_id(REG,029e6b40,17)) suspecting node: 17b576f8
2018-05-29T11:36:03.704507+08:00 0 [Note] WSREP: evs::proto(029e6b40, OPERATIONAL, view_id(REG,029e6b40,17)) suspected node without join message, declaring inactive
2018-05-29T11:36:04.204630+08:00 0 [Note] WSREP: declaring node with index 1 inactive (evs.inactive_timeout)
2018-05-29T11:36:06.204943+08:00 0 [Note] WSREP: (029e6b40, 'tcp://0.0.0.0:4567') connection to peer 00000000 with addr tcp://192.168.1.253:4567 timed out, no messages seen in PT3S (gmcast.peer_timeout)
2018-05-29T11:36:09.206069+08:00 0 [Note] WSREP: declaring node with index 2 suspected, timeout PT5S (evs.suspect_timeout)
2018-05-29T11:36:09.206168+08:00 0 [Note] WSREP: evs::proto(029e6b40, GATHER, view_id(REG,029e6b40,17)) suspecting node: 45125952
2018-05-29T11:36:09.706352+08:00 0 [Note] WSREP: evs::proto(029e6b40, GATHER, view_id(REG,029e6b40,17)) suspecting node: 45125952
2018-05-29T11:36:10.206318+08:00 0 [Note] WSREP: (029e6b40, 'tcp://0.0.0.0:4567') connection to peer 00000000 with addr tcp://192.168.1.253:4567 timed out, no messages seen in PT3S (gmcast.peer_timeout)
2018-05-29T11:36:10.206514+08:00 0 [Note] WSREP: evs::proto(029e6b40, GATHER, view_id(REG,029e6b40,17)) suspecting node: 45125952
2018-05-29T11:36:10.706763+08:00 0 [Note] WSREP: evs::proto(029e6b40, GATHER, view_id(REG,029e6b40,17)) suspecting node: 45125952
2018-05-29T11:36:11.204659+08:00 0 [Warning] WSREP: evs::proto(029e6b40, GATHER, view_id(REG,029e6b40,17)) install timer expired
evs::proto(evs::proto(029e6b40, GATHER, view_id(REG,029e6b40,17)), GATHER) {
current_view=Current view of cluster as seen by this node
view (view_id(REG,029e6b40,17)
memb {
029e6b40,0
17b576f8,0
45125952,0
}
joined {
}
left {
}
partitioned {
}
),
input_map=evs::input_map: {aru_seq=5278704,safe_seq=5278704,node_index=node: {idx=0,range=[5278710,5278709],safe_seq=5278704} node: {idx=1,range=[5278705,5278704],safe_seq=5278704} node: {idx=2,range=[5278710,5278709],safe_seq=5278704} },
fifo_seq=10557624,
last_sent=5278709,
known:三、故障解决
对于当前的脑裂，由于在测试环境，因此直接停止其中一个节点(1.248)，然后将剩余的节点强制提升为主节点
首先停止仲裁节点，然后停止1.249节点
如下在1.249节点上做如下操作
mysql> SET GLOBAL wsrep_provider_options='pc.bootstrap=YES';
Query OK, 0 rows affected (0.00 sec)
--再次查看此时wsrep_cluster_status为Primary
mysql> show status like 'wsrep_cluster%';
+--------------------------+--------------------------------------+
| Variable_name | Value |
+--------------------------+--------------------------------------+
| wsrep_cluster_conf_id | 4 |
| wsrep_cluster_size | 1 |
| wsrep_cluster_state_uuid | cd96b06a-0a1d-11e8-99d2-837e6f3b95a9 |
| wsrep_cluster_status | Primary |
+--------------------------+--------------------------------------+
4 rows in set (0.01 sec)
在第一个节点248上启动
[root@node248 ~]# systemctl start mysql
--集群状态查看，此时也已正常
mysql> show status like 'wsrep_cluster%';
+--------------------------+--------------------------------------+
| Variable_name | Value |
+--------------------------+--------------------------------------+
| wsrep_cluster_conf_id | 5 |
| wsrep_cluster_size | 2 |
| wsrep_cluster_state_uuid | cd96b06a-0a1d-11e8-99d2-837e6f3b95a9 |
| wsrep_cluster_status | Primary |
+--------------------------+--------------------------------------+
在第三个节点启动仲裁节点
[root@ydq4 ~]# /etc/init.d/garb start
正在启动 /usr/bin/garbd： [确定]
mysql> show status like 'wsrep_cluster%';
+--------------------------+--------------------------------------+
| Variable_name | Value |
+--------------------------+--------------------------------------+
| wsrep_cluster_conf_id | 6 |
| wsrep_cluster_size | 3 |
| wsrep_cluster_state_uuid | cd96b06a-0a1d-11e8-99d2-837e6f3b95a9 |
| wsrep_cluster_status | Primary |
+--------------------------+--------------------------------------+四、后记
1、此次脑裂时应用程序端已出现节点无法连接，并且有日志记录到mysql error log
2、在数据节点已出现异常时，冒然重启仲裁节点不可取
3、建议使用3个数据节点最佳
4、多节点数据写入，最好按模块将写划分到不同节点
5、节点修复参考：http://galeracluster.com/documentation-webpages/twonode.html
上传日志：