Is it possible to configure, or use a design pattern or approach, hibernate to support the DB faulting, continue to transact to L2 cache, then on the DB being restarted having the pending changes write back out to the DB?

In principle the idea would be to pre-fetch everything into the L2 cache of a single App Server and during the DB failure still support transactions with the caveat that obviously the changes being made are at risk should the app server fail before the DB restarts and the changes are written back to the DB.

There are lots of issues with this type of operation but for our situation is possible.

However, from my investigations I can't find out whether this can be done so it may be a non-starter.

What I'm looking for is transactional, concurrent multi-threaded, in-memory object graph with asynchronous write to RDBMS with a rich OR mapping facility.

cheers,
- Dean