I'm trying to get a sort of object database in memory which can support an arbitrary number of instances/connections. Lets start with the library example:
Lets say I have a Library class. I can create as many instances of this class as I want.
Each instance of Library has a collection of Library::Book objects (accessible as library.books). This collection of books is specific to that library instance.
I also have Library::Magazine which behaves the same.
And with this I can do things like library.books['Nineteen Eighty Four'].author, library.magazines.each {...}, library.magazines.delete(magazine).
Now up to this point it isn't hard. Any database model library can handle it. However I want to be able to clone the "library". Meaning I can take an instance of Library, and do library.clone, and get a brand new library which is an exact duplicate of the origin. But the new copy will have it's own collection of Books and Magazines (which was created from the origin).
This means it should create a duplicate database, and both databases can be accessed simultaneously.
It's this arbitrary number of databases with simultaneous access that seems to be very difficult. Ideally I would like it to be a relational database with foreign key constraints (so I can do something like library.categories['fiction'].books.each {...} and have a foreign key on the book referencing the category it's in).
I've been hacking away with Sequel for a bit, and it comes close in that it allows multiple sqlite in-memory databases. But it's proving extremely difficult to get it to support an arbitrary number of them (it wants a constant class representing each table in each database).
Assuming all databases have the same schema, you may be able to use Sequel's sharding support (http://sequel.jeremyevans.net/rdoc/files/doc/sharding_rdoc.html) and a single model class per table (instead of a model class per table per database). You can use Database#add_servers and Database#remove_servers to adjust the available shards at runtime.