Yes, it is a standalone application. I use multithreading to speed up data loading.

My schema is heirarchical. I have a few parent tables, and lots of child tables that refer to the parent tables via PK/FK constraints. My objects basically represent a row in a given table. For a child table, the constructors all take the parent object they are associated with, and then get their Connection object from that parent.

When I created separate threads (all in a ThreadGroup), each with their own Connection, I shared a set of parent objects loaded in the beginning of the application run from a Connection that was not in the new ThreadGroup. Hence, all child objects were referencing the parent objects Connection.

My data loader would then call commit() on the Connection belonging to an individual thread, which was not the correct one. I fixed the problem by adding a check in a central Store object, which holds a reference to the Connection, to make sure all objects in that Store have the right Connection. I'm using the mysql query "select connection_id()", which will return a unique connection id.