跨数据中心主控复制的一个好的解决方案是什么?

Lets say we have a distributed system where data is sharded by user ids. In most cases each shard is changed by user who owns this shard. There are more then one datacenter and users are balanced between these datacenters using DNS, cookies or somehow else. I.e. in most cases every user is served by one datacenter. This means we can replicate data between datacenters in master-master manner. Of course conflicting writes are possible, but they are rare and could be resolved using, say, CRDTs or vector clocks. And if some datacenter fails users could be just redirected to other datacenters!Lets say we have a distributed system where dat