https://bugzilla.redhat.com/show_bug.cgi?id=1193177
--- Comment #12 from Stuart D Gathman stuart@gathman.org --- Short answer: I do not have a revised patch
Long answer: My reading of the docs is that because there is only one table involved, the existing patch should work. If there is a lock failure, it will fail on the first insert or update - both of which already have a ROLLBACK. Changing the transaction mode to IMMEDIATE would merely cause the failure on the BEGIN instead. So I don't think a revised patch is needed. This is not really a correct transaction anyway, because it is updating with data collected from outside the transaction. It is purely a (necessary) performance hack.