mariadb galera cluster 中的节点宕机日志解释以及如何避免ddl造成集群范围ha
作者：互联网
2016-07-07 11:05:49 140011799914240 [Note] WSREP: (fb55d9c9, 'tcp://0.0.0.0:4567') turning message relay requesting on, nonlive peers: tcp://197.3.xxx.152:4567
发现nonlive的节点197.3.xxx.152
2016-07-07 11:05:51 140011799914240 [Note] WSREP: (fb55d9c9, 'tcp://0.0.0.0:4567') reconnecting to 51f4c07a (tcp://197.3.xxx.152:4567), attempt 0
重新尝试发送一个keepalived的链接。
2016-07-07 11:05:55 140011799914240 [Note] WSREP: evs::proto(fb55d9c9, GATHER, view_id(REG,128a493c,122))suspecting node: 51f4c07a
经过5秒还没有收到响应，怀疑节点已经宕掉，51f4c07a即197.3.xxx.152。
2016-07-07 11:05:55 140011799914240 [Note] WSREP: evs::proto(fb55d9c9, GATHER, view_id(REG,128a493c,122))suspected node without join message, declaring inactive
宣判节点宕掉。declaring inactive
2016-07-07 11:05:55 140011799914240 [Note] WSREP: declaring 128a493c at tcp://197.3.xxx.150:4567 stable
获取存活的节点信息。
2016-07-07 11:05:55 140011799914240 [Note] WSREP: Node 128a493c state prim
获取存活的节点的状态。
2016-07-07 11:05:55 140011799914240 [Note] WSREP: view(view_id(PRIM,128a493c,123) memb {
128a493c,0
fb55d9c9,0
} joined {
} left {
} partitioned {
51f4c07a,0
})
对宕掉的节点进行移除，即partitioned
2016-07-07 11:05:55 140011799914240 [Note] WSREP: save pc into disk
2016-07-07 11:05:55 140011799914240 [Note] WSREP:forgetting 51f4c07a (tcp://197.3.xxx.152:4567)
2016-07-07 11:05:55 140011791521536 [Note] WSREP: New COMPONENT: primary = yes, bootstrap = no, my_idx = 1, memb_num = 2
打印新的primary COMPONENT 信息。
2016-07-07 11:05:55 140011791521536 [Note] WSREP: STATE EXCHANGE: Waiting for state UUID.
2016-07-07 11:05:55 140011799914240 [Note] WSREP: (fb55d9c9, 'tcp://0.0.0.0:4567') turning message relay requesting off
2016-07-07 11:05:55 140011791521536 [Note] WSREP: STATE EXCHANGE: sent state msg: b8a08882-43ef-11e6-884e-2beaea270cdb
2016-07-07 11:05:55 140011791521536 [Note] WSREP: STATE EXCHANGE: got state msg: b8a08882-43ef-11e6-884e-2beaea270cdb from 0 (node3)
2016-07-07 11:05:55 140011791521536 [Note] WSREP: STATE EXCHANGE: got state msg: b8a08882-43ef-11e6-884e-2beaea270cdb from 1 (node2)
2016-07-07 11:05:55 140011791521536 [Note] WSREP: Quorum results:
version = 4,
component = PRIMARY,
conf_id = 33,
members = 2/2 (joined/total),
act_id = 128561512,
last_appl. = 0,
protocols = 0/7/3 (gcs/repl/appl),
group UUID = afbfeed1-38e8-11e6-84b1-d20f5b484535
新的集群状态信息在新的primary component 内的节点进行交换与确认。
2016-07-07 11:05:55 140011791521536 [Note] WSREP: Flow-control interval: [23, 23]
2016-07-07 11:05:55 140011908541184 [Note] WSREP: New cluster view: global state: afbfeed1-38e8-11e6-84b1-d20f5b484535:128561512, view# 34: Primary, number of nodes: 2, my index: 1, protocol version 3
2016-07-07 11:05:55 140011908541184 [Note] WSREP: wsrep_notify_cmd is not defined, skipping notification.
2016-07-07 11:05:55 140011908541184 [Note] WSREP: REPL Protocols: 7 (3, 2)
2016-07-07 11:05:55 140011808306944 [Note] WSREP: Service thread queue flushed.
2016-07-07 11:05:55 140011908541184 [Note] WSREP: Assign initial position for certification: 128561512, protocol version: 3
2016-07-07 11:05:55 140011808306944 [Note] WSREP: Service thread queue flushed.
2016-07-07 11:05:58 140011799914240 [Note] WSREP: cleaning up 51f4c07a (tcp://197.3.xxx.152:4567)
完全剔除197.3.xxx.152。
上面的日志中，涉及到进行节点宕机检查的timeout 时间，详细信息请参考官方文档，作者做了截图，请查看！
几个相关的参数有相对大小的关系，特殊设置时请注意：
顺便提一下，在galera 集群中，执行ddl会造成整个集群hang ,如果ddl只要执行的时间较长，在交易频率的系统上，则就造成了悲剧。下面的方式可以避免这种情况。
To run an ALTER statement in this manner, on each node run the following queries:
在galera集群的每一个节点上执行下面的步骤：
1. Change the Schema Upgrade method to Rolling Schema Upgrade.
SET wsrep_OSU_method='RSU';
设置为滚动schema变更模式。
2. Run the ALTER statement.
执行alter 语句。
3. Reset the Schema Upgrade method back to Total Order Isolation.
SET wsrep_OSU_method='TOI';
将schema的变更模式重新设置为TOI。
标签：galera,11,07,05,55,WSREP,宕机,Note,ha 来源： https://blog.51cto.com/15057824/2648692