I understand that a single DML statement is better than using bulk collect for all having intermediate commits. My only concern is if I'm loading a large amount of data like 100 million records into a 800 million record table with foreign keys and indexes and the session gets killed, the rollback might take a long time which is not acceptable. Using bulk collect forall with interval commits is slower than a single straight merge statement, but in case of dead session, the rollback time won't be as bad and a reload of the not yet committed data will not be as bad. To design a recoverable data load that may not be affected as badly, is bulk collect + for all the right approach?