I'd like to thank everyone for the ideas.
Glassfish with toplink and postgres (with another database in it) are
already present, so I don't consider them as overhead. The project
already uses JPA, so I'd prefer not to have two kinds of persistence
architectures.
The problem is that I need the data replicated between two nodes.
h2 clustering looks interesting, if I can automate the recovery.
ha-jdbc seems to have that aspect covered too, so that's my prime
candidate right now.
I tried to figure out which postgresql multimaster replication
projects are still alive but already stable, but an RDBMS-agnostic
solution is always preferred anyway.
Thanks again everyone!
Gabor