Distributed databases with a near-real-time multi-master configuration – such as BigCouch, coming soon to Apache CouchDB – must deal with the potential of simultaneous modifications of a single resource. While the approach taken by multiple single-machine Apache CouchDB servers using regular HTTP replication is well understood, the situation changes a little bit when dealing with BigCouch-style internal replication inside a cluster.
I think it’s time to have a better understanding of what this means, and what impact this has on you as an application developer. Most of the time, there’s no change – to your app, a BigCouch-style cluster looks and feels like a single Apache CouchDB node. But when making near-simultaneous writes to the same document from different clients, you may experience document conflicts that you wouldn’t have with an Apache CouchDB 1.x single server.
How does this happen? Bear with me – this gets a bit complex. Hopefully this diagram will help.