Thanks for the welcome and reply!

Unfortunately, InactiveDestroy doesn't fix the problem, and AutoInactiveDestroy seems only to be applicable to situations where the process is forking (and, in any case, doesn't fix the problem).

The test code I posted was the shortest I could come up with that generated the double error, so it doesn't really capture the use case. My actual code runs in the context of a Catalyst web server, and I'm throwing the exception to get the error message and the location in the code up to the request handler code so that it can be prettied up and sent back to the browser. So while I could give up on that and put a die every call, that handling would end up being much more explicit, not less (especially because of the added code to dig out the last SQL statement).