The last job was the scheme of one master vs slave in different regions. The task of balancing at the database level is not solved, the access was predominantly local. You can live, but as You can see, the replication can lag, and it makes it uneven. Another mistake — channels, bitch, still not as reliable as might be desired. Missing link between the servers is a replica of rose. For this reason, it is sufficient periodically lost the ability to record something in the master "afar". So in any case suggest to enter at least a simple monitoring and try to understand how much it will rebate specifically in your case, and assess whether it is suitable. If You have graphics lag time replication, control access to the master from each point, where will it write to live will be, maybe not easier, but more predictable :)
\r
Can also appear floating problems with code that expects no delays. For example, it registers a new user (you start in master database), but to do really can not do anything, because its data is not reached to the slave. This problem looks pretty stupid, but there may be more cunning manifestation.
