Sitting here at an architectural summit, there is one point of
clarification
-- the clients reside on a separate network from the data centers.
So it is
possible to lose a data center without losing the clients; they will
simply
route to another data center.
In that case, there will be a notion of promotion of a data center as
a
primary key owner, which has its own set of implications (i.e. needs
to be
more local copies, etc). And that means some sort of "fail back"
once the
original data center comes back online after any versioning merge
occurs.
(Does this make sense?)
+1. the clients in this approach are not embedded. With HR
it should be quite simple to get this functionality, by writing a chained-client. Not sure
how the REST clients would work exactly, but I think a similar approach can be used. What
kind of client access do you have BTW?