In our testing, we start up replica sets initially with one node and then add more nodes to the replica set later (comments in the code say this is for faster elections). When a config server is using auto bootstrap, this means we can end up with a single shard sharded cluster with one node for a while before we add more nodes to it.

This can end up with a scenario where we begin creating the sessions collection when there is one node in the replica set and so use w:1 for that operation despite a second node being added during the shard collection command. As a temporary solution to prevent this, we should just create embedded config server replica sets all at once, rather than adding nodes to them later.