To clarify the requirement: much like you, we're not looking for synchronous replication, which would be too slow. The branches and central server can be different for about 5 to 10 minutes. But the branches need to be able to function independently if the network or central goes down. Londiste looks interesting, though the documentation seems a bit sparse. Is it really that simple to set up, or is there a fair amount of trial and error in the setup process?
> There is one central server with 19 branches. Some tables need to replicate > from the central server to the branches. Other tables are centralised from > the branches into one totalling table at the centre. A few tables need to > replicate in both directions.
I'm working on some projects here with this very need (and same scale), and I plan to use londiste (master/slaves asynchronous solution) replication solution, which I already use in production on some other project.
The fact is that we want the "branches" to still be fully available in case of network or central server failure, so we don't buy into synchronous replication; which is not available yet into PostgreSQL as far as I know, even if one of the basics building-block alternatives is provided into -core, namely Two Phase Commit. http://www.postgresql.org/docs/8.3/static/sql-prepare-transaction.html
