To scale up our MongoDB cluster, we were forced to split a running replicaset into two. The problem was that we couldn't afford any downtime, so we had to migrate while still serving many hundred queries per second. This blog post describes the whole process.