Mike Matrigali wrote:
One suggestion was using an open source product which captured the info at the jdbc level. Also others have used triggers to provide the necessary information.
One problem with statement-based replication (e.g., at JDBC level) is that statements are not idempotent. This makes it difficult to do recovery on a replicated instance and to catch up with the other database instances. After a crash one will normally have to copy the entire database instead of just doing a replay of what has happened while the instance was down. (Ref. Sequoia). This would be easier with a row-based replication scheme.
With respect to trigger-based implementations, the biggest problem seems to be performance. I do not know whether this is inherent with using triggers, but it seems like the overhead is pretty big in many database systems.
-- Øystein
