During migrations we wait for writes to be replicated to a majority of nodes. To figure out how many nodes are in a majority we call getSlaveCount() (and divide it by 2 and add 1). The number reported by getSlaveCount(), however, may include old nodes that are no longer part of the set (or are now known by a different identifier). This could potentially cause us to wait for more nodes than exist - meaning that we will never acknowledge the writes as replicated.