The term "rollback" refers to transactional operations, but bulk-copying is NOT done in
a transactional context (it would be orders of magnitude slower if it were). If you lose
hardware during a transfer, it would be wise to start the entire copy over again (from
the state your production database was in during the backup snapshot you did just
before the bulk load).
Manually editing rows in the metadata tables is only supported in the context of being
asked to do so by Tech Support staff. You really ought to open an Tech Support incident
after catastrophic failure like this, since there are millions of failure modes, and it is not
possible to document everything that might happen.
- V