刚刚接到报警短信，从库宕机，马上通知机房重启，在检查MySQL时，发现同步挂了，报主键冲突，询问开发是不是有往里面写数据，回答没有。


这就奇怪了，怎么会无缘无故报错呢？在检查了my.cnf配置文件，发现有个参数没有配置：

innodb_overwrite_relay_log_info = 1


当从库宕机后，重新开启主从复制同步，它可以重新执行已提交事务，这样就会造成同步失败，而这个参数就会避免这个问题的出现。


当开启了这个参数后

....
+ InnoDB: Warning: innodb_overwrite_relay_log_info is enabled. Updates of other storage engines may have problem of consistency.
+ InnoDB: relay-log.info is detected.
+ InnoDB: relay log: position 429, file name ./gauntlet3-relay-bin.000111
+ InnoDB: master log: position 280, file name gauntlet3-bin.000015
....
  InnoDB: Starting crash recovery.
....
  InnoDB: Apply batch completed
+ InnoDB: In a MySQL replication slave the last master binlog file
+ InnoDB: position 0 468, file name gauntlet3-bin.000015
+ InnoDB: and relay log file
+ InnoDB: position 0 617, file name ./gauntlet3-relay-bin.000111
  090205 17:41:31 InnoDB Plugin 1.0.2-3 started; log sequence number 57933
+ InnoDB: relay-log.info have been overwritten.
....
  090205 17:41:31 [Note] Slave SQL thread initialized, starting replication in log ``gauntlet3-bin.000015`` at position 468, relay log ``./gauntlet3-relay-bin.000111`` position: 617


已经执行完的Position点：

master log: position 280, file name gauntlet3-bin.000015

在恢复时它内部会检测到280这个点已经执行完毕，从下一个点468开始同步，并且重写relay.info文件，确保了主从同步正确。


建议在从库上添加，如果是官方MySQL，参数是relay_log_recovery=1


具体请参考：http://www.percona.com/doc/percona-server/5.5/reliability/crash_resistant_replication.html