shudder We have a business relationship with a financial company based out of Germany, and their system (which sits on top of replicated Oracle) defines "idempotent" to mean "our database already has a cache of the answer we already gave you". Oh, and they gave us a JAR file of client code that does client-side failover, with a fixed 8-second delay to account for the Oracle replication lag between sites. No coordination between different instances of our client. So now, when one client fails over due to a connectivity blip, the other clients get cached error messages (logically equivalent to 404 Not Found). "Idempotent", indeed.