We discovered that mongo writes aren't atomic. The technical
term I think is "durable" but I think of this as related to
atomicity because executing a write command and storing the
written data to disk should be a single action. In short, mongo
will respond with a return code stating that an update was
successfully stored *before* the data is actually persisted to
disk!! In practice this hasn't (yet) bit us since DB writes are
fairly uncommon in our application; most users are reading, and
the writes which do occur are typically for different documents
based on how they're structured, which limits the possibility of
clashing. Even so - from a theoretical standpoint this scares
us. MongoDB claims to be "webscale" (e.g. it can handle thousands
of writes per second as it would only trigger a single write to
disk), however, at what cost?