Chris Travers wrote:
> I guess I am thinking along different lines than you. I was thinking
> that the simplest solution would be to have master/slave replication for
> *approved* transactions only and no replication for initial commits
> prior to approval. This makes the assumption that a single transaction
> will be committed on a single server, and that a single transaction will
> not be split over multiple servers. In this way, you can commit a
> pending transaction to any single server, and when it is approved, it
> gets replicated via the master. See below for more.
This works if you don't care that multiple servers commit transactions
that force a budget or bank account to be exceeded.
> Thinking about this.... The big issue is that you only want to
> replicate the deltas, not the entire account. I am still thinking
> master/slave, but something where the deltas are replicated in the
> background or where the user, in checking his account, is actually
> querying the home server. This second issue could be done via dblink or
> DBI-Link and would simply require that a master table linking the
> accounts with home servers be replicated (this should, I think, be
> fairly low-overhead).
Except what you know have is your system fails if any server fail or is
inaccessible.