Routing queries to correct shard in cluster (no replies)

Hi,

I am learning about mysql cluster and wasn't able to find an answer to this question from whitepapers.

How does a query request from application get routed to the correct shard in a MySQL cluster?

The path that the query can take could be one of these:
1. Application -> MySQL Server -> Data Nodes (NDB)
2. Application (using CLusterJPA) -> Data Nodes (NDB)

I am more interested in the second path.
My intuition is that query routing happens at the NDB API layer, but could be wrong.

Any pointers would be really appreciated.

Thanks,