On Feb 13, 2008, at 9:14 AM, Nigel Thomas wrote:
MarcusMulti-master replication is a bitch (logically as well as technically). Ideally you want to know where / when data is updated; and if possible have a single master for each row, or at least a well understood conflict resolution policy.<SNIP>
I worked once on a multimaster replication implementation where we associated individual rows with the owning master site by assigning a series of numbers to each master site ... thus, EUROPE primary key might be between 1000000000 to 1999999999; PACIFIC primary key ranged between 2000000000 to 299999999, etc.
It works great as long as you can be assured you're never going to run out of numbers.