By doing the work with secondary connections to the underlying databases,
asynchronously and in a separate process, this should have minimal impact on
the performance of the storage node. Extra complexity comes from backends that
may lose connection to the database (here MySQL): this commit fully implements
reconnection.