right, the case I'm trying to account for is re-starting the backfill process that might have failed without needing to track the status of the relations elsewhere, so I can just re-run the full syncing job and upsert any existing relations. Specifically so that a batch of 1k relation inserts don't fail because one of the batch already existed. Easy to work around, just +1ing a use case for a

REPLACE_UNSAFE

type of operation