On Wed, Apr 14, 2004 at 14:44:57 +0100,
Paul Breen <pbreen@computerpark.co.uk> wrote:
>
> The way we approached it was to do the replication ourselves at the
> application level. This isn't as hard as it sounds: you just open 2 db
> connections (1 to the master, 1 to the slave), start transactions on both
> connections, then write your data. If an error occurs on either
> connection, you rollback on both connections. This means you can still
> have writes that fail (for whatever reason) but both dbs are _consistent_.
This is guarenteed to work in general. When there are concurrent transactions,
one transaction may succeed on commit and the other fail. If you are using
sequences you could also end up with different values in each database.
That may or may not be a problem depending on how you use them.
