There has been a lot of discussion around binlog servers recently for MySQL. This seems to be spurred by the release of MaxScale binlog server. Some recent posts regarding this: https://mariadb/blog/mariadb-replication-maxscale-and-need-binlog-server/ blog.booking/mysql_slave_scaling_and_more.html We actually have been using binlog tailers for a long time in production to serve as an off-master DR type solution. In the event of a master host failure, generally the in-rack binlog tailer will have the majority of events that never made it to the slave and then we can replay those missing events. We do this without using anything special, just a small python wrapper around mysqlbinlog. The python script is just to do management of the mysqlbinlog process to ensure it is always running and tailing the log files. We are now moving to a model where those off-host binlog readers can actually run in semi-sync mode. This allows us to guarantee that the tailer has all of the events that executed on the master. I suspect that MaxScale will get this capability at sometime in the near future. For more background on our setup, Yoshinori wrote a detailed blog post about our architecture at: yoshinorimatsunobu.blogspot/2014/04/semi-synchronous-replication-at-facebook.html
Posted on: Wed, 16 Jul 2014 16:31:25 +0000
Trending Topics
Recently Viewed Topics
© 2015