Comment 4 for bug 1499669

I suspect the solution is to explicitly use transactions in all of the database updates. That way, if an exception occurs due to the thread being killed in the middle of a transaction then the transaction will be rolled back, and there's no risk of attempting to do something else to the DB in the middle of an implicit transaction.

An alternative would be to try to catch exceptions and roll back at a higher level, but that makes it much easier to miss some cases.