一个节点无法启动 Galera Cluster 中的 mariadb

Question

我的集群中有 3 个节点，名称：ha-node1, ha-node2, ha-node3。

ha-node2现在无法启动mariadb.service，之前正常：

我使用以下命令来显示日志：

[root@ha-node2 ~]# systemctl status mariadb.service
● mariadb.service - MariaDB database server
   Loaded: loaded (/usr/lib/systemd/system/mariadb.service; enabled; vendor preset: disabled)
  Drop-In: /etc/systemd/system/mariadb.service.d
           └─migrated-from-my.cnf-settings.conf
   Active: failed (Result: timeout) since Mon 2017-07-31 12:00:33 CST; 13min ago
  Process: 59147 ExecStartPre=/bin/sh -c [ ! -e /usr/bin/galera_recovery ] && VAR= ||   VAR=`/usr/bin/galera_recovery`; [ $? -eq 0 ]   && systemctl set-environment _WSREP_START_POSITION=$VAR || exit 1 (code=killed, signal=TERM)
  Process: 59138 ExecStartPre=/bin/sh -c systemctl unset-environment _WSREP_START_POSITION (code=exited, status=0/SUCCESS)

Jul 31 11:59:02 ha-node2 systemd[1]: Starting MariaDB database server...
Jul 31 12:00:33 ha-node2 systemd[1]: mariadb.service start-pre operation timed out. Terminating.
Jul 31 12:00:33 ha-node2 systemd[1]: Failed to start MariaDB database server.
Jul 31 12:00:33 ha-node2 systemd[1]: Unit mariadb.service entered failed state.
Jul 31 12:00:33 ha-node2 systemd[1]: mariadb.service failed.

并使用journal -xe:

[root@ha-node2 ~]# journalctl -xe
Jul 31 12:14:49 ha-node2 attrd[2281]:   notice: Removing all ha-node3 attributes for attrd_peer_change_cb
Jul 31 12:14:49 ha-node2 attrd[2281]:   notice: Lost attribute writer ha-node3
Jul 31 12:14:49 ha-node2 attrd[2281]:   notice: Removing ha-node3/3 from the membership list
Jul 31 12:14:49 ha-node2 attrd[2281]:   notice: Purged 1 peers with id=3 and/or uname=ha-node3 from the membership cache
Jul 31 12:14:49 ha-node2 xinetd[1175]: START: mysqlchk pid=66355 from=::ffff:192.168.8.102
Jul 31 12:14:49 ha-node2 xinetd[1175]: EXIT: mysqlchk status=1 pid=66340 duration=0(sec)
Jul 31 12:14:49 ha-node2 xinetd[1175]: EXIT: mysqlchk status=1 pid=66341 duration=0(sec)
Jul 31 12:14:49 ha-node2 xinetd[1175]: EXIT: mysqlchk signal=13 pid=66355 duration=0(sec)
Jul 31 12:14:49 ha-node2 corosync[1444]:  [TOTEM ] A new membership (192.168.8.102:11184) was formed. Members
Jul 31 12:14:49 ha-node2 corosync[1444]:  [QUORUM] Members[1]: 2
Jul 31 12:14:49 ha-node2 corosync[1444]:  [MAIN  ] Completed service synchronization, ready to provide service.
Jul 31 12:14:49 ha-node2 crmd[2285]:   notice: State transition S_ELECTION -> S_INTEGRATION [ input=I_ELECTION_DC cause=C_TIMER_POPPED origin=election_timeout_popped ]
Jul 31 12:14:49 ha-node2 crmd[2285]:  warning: FSA: Input I_ELECTION_DC from do_election_check() received in state S_INTEGRATION
Jul 31 12:14:49 ha-node2 crmd[2285]:   notice: Notifications disabled
Jul 31 12:14:49 ha-node2 corosync[1444]:  [TOTEM ] A new membership (192.168.8.101:11188) was formed. Members joined: 1 3
Jul 31 12:14:49 ha-node2 pacemakerd[1675]:    error: Node ha-node1[1] appears to be online even though we think it is dead
Jul 31 12:14:49 ha-node2 pacemakerd[1675]:   notice: pcmk_cpg_membership: Node ha-node1[1] - state is now member (was lost)
Jul 31 12:14:49 ha-node2 pacemakerd[1675]:    error: Node ha-node3[3] appears to be online even though we think it is dead
Jul 31 12:14:49 ha-node2 pacemakerd[1675]:   notice: pcmk_cpg_membership: Node ha-node3[3] - state is now member (was lost)
Jul 31 12:14:49 ha-node2 attrd[2281]:   notice: crm_update_peer_proc: Node ha-node1[1] - state is now member (was (null))
Jul 31 12:14:49 ha-node2 cib[2277]:   notice: crm_update_peer_proc: Node ha-node1[1] - state is now member (was (null))
Jul 31 12:14:49 ha-node2 corosync[1444]:  [QUORUM] This node is within the primary component and will provide service.
Jul 31 12:14:49 ha-node2 corosync[1444]:  [QUORUM] Members[3]: 1 2 3
Jul 31 12:14:49 ha-node2 corosync[1444]:  [MAIN  ] Completed service synchronization, ready to provide service.
Jul 31 12:14:49 ha-node2 pacemakerd[1675]:   notice: Membership 11188: quorum acquired (3)
Jul 31 12:14:49 ha-node2 stonith-ng[2279]:   notice: crm_update_peer_proc: Node ha-node1[1] - state is now member (was (null))
Jul 31 12:14:49 ha-node2 cib[2277]:   notice: crm_update_peer_proc: Node ha-node3[3] - state is now member (was (null))
Jul 31 12:14:49 ha-node2 attrd[2281]:   notice: crm_update_peer_proc: Node ha-node3[3] - state is now member (was (null))
Jul 31 12:14:49 ha-node2 stonith-ng[2279]:   notice: crm_update_peer_proc: Node ha-node3[3] - state is now member (was (null))
Jul 31 12:14:49 ha-node2 crmd[2285]:   notice: Membership 11188: quorum acquired (3)
Jul 31 12:14:49 ha-node2 crmd[2285]:   notice: pcmk_quorum_notification: Node ha-node1[1] - state is now member (was lost)
Jul 31 12:14:49 ha-node2 crmd[2285]:   notice: pcmk_quorum_notification: Node ha-node3[3] - state is now member (was lost)
Jul 31 12:14:49 ha-node2 attrd[2281]:   notice: Recorded attribute writer: ha-node3
Jul 31 12:14:49 ha-node2 attrd[2281]:   notice: Processing sync-response from ha-node3

---

EDIT-1

我用clustercheck查看ha-node2中cluster的状态，发现连接已关闭，galera cluster节点没有同步：

[root@ha-node2 ~]# clustercheck 
HTTP/1.1 503 Service Unavailable
Content-Type: text/plain
Connection: close
Content-Length: 36

Galera cluster node is not synced.

并且在ha-node1和ha-node3中是连接关闭的，并且galera集群节点是同步的：

[root@ha-node1 ~]# clustercheck 
HTTP/1.1 200 OK
Content-Type: text/plain
Connection: close
Content-Length: 32

Galera cluster node is synced.

Answer 1

我个人在消息xinetd[1175]: EXIT: mysqlchk signal=13 中看到了问题。从 xinetd 手册页是

For services with type =
       INTERNAL, SIGTERM signal will be sent.  For  services  without  type  =
       INTERNAL, SIGKILL signall will be sent.

在你的情况下是SIGPIPE 13 Write on a pipe with no reader, Broken pipe (POSIX)

最可能的原因是： Jul 31 12:14:49 ha-node2 attrd[2281]: notice: Lost attribute writer ha-node3 你必须追踪这个以找出 ha-node2 和 ha-node3 之间发生的事情（为此我推荐 tcpdump） .

我必须看到更多您的 mysqlchk 或制作您的 HAproxy 的脚本。

如果没有任何帮助，您也可以尝试替代 HAClustering 解决方案https://github.com/olafz/percona-clustercheck