Again, essentially what we do in concept. The dev side is a separate db (git could represent this), and once something is published past dev the record is copied to the production db (Riak in your case). The dbs have in-memory local machine cached versions, for speed, but also for reliability. You don't want your N machines depending on one db point of failure, so the local cached copies exist to mitigate that. (They receive updates via a multicast mechanism.)
