On Wed, Aug 7, 2019 at 3:34 PM Daniel Fink (PDF) <daniel.fink@xxxxxxx> wrote: > My current idea is to lock both tables completely from access (the queried and the updated one) so that postgresql does not have to ensure isolation for concurrent queries by keeping a copy of each row. I'm not sure that locking will prevent the snapshotting and the WAL machinery, but someone more expert on the are could clarify this. Since the column is nullable, I would apply it outside of the transaction, and then do the update. If that still fails, I would try to split the update on small chunks (after all, it's an update, so it is smething you can line up data). Luca
