There's an open-source implementation of a different approach to the same problem in ToroDB (http://github.com/torodb/torodb http://github.com/torodb/torodb).

There's an open-source implementation of a different approach to the same problem in ToroDB (http://github.com/torodb/torodb http://github.com/torodb/torodb). ToroDB creates tables dynamically, which match the types (mapped to columns) of every level of the JSON documents, obtaining a fully relational database. This is effectively "partitioning" (automatically) the documents based on their effective "type". ToroDB actually uses PostgreSQL as its backend to store the data relationally. Hope this information is interesting. Disclaimer: I am one of the developers of ToroDB