The test wrongly assumes that the moveChunk command issued by a completely fresh mongos always succeeds. This is not always true: this refresh at the beginning of the moveChunk command in the mongos might not see that the collection was already sharded and throw an exception: the reason is that the CatalogCache reads from the config server with Nearest read preference so it might end up reading the majority committed information from a secondary node of the CSRS that hasn't replicated the shardCollection operation yet.