On 28.1.2012, at 2.23, Stan Hoeppner wrote:
With dsync-based replication it's possible to avoid both of these problems, because application-level replication can intelligently handle situations where asynchronous replication results in data conflicts. (This kind of conflict resolution is also what I hope to do with some nosql database in future when Dovecot supports them.) I've been working on dsync-based easy-to-use replication recently, and it's almost in a condition where I'm going to start using it myself (maybe this weekend).
Can you provide a basic diagram/high level description of how this dsync replication would be configured to work over a 2 node wide area network?
I'll write a description at some point.. It's anyway meant to be more scalable than just 2 nodes, so the idea is to have userdb lookup return the 2 (or more) replicas.
Are we looking at something like period scripts, something more automatic, a replication daemon?
It's a replication daemon that basically calls "doveadm sync" when needed (via doveadm server connection). Initially it's not as optimal from performance point of view as it could, but should get better. :)