Data-split technique from Apply4: create an empty copy of the source table, insert only the records that should remain (e.g. only this region's rows), then rename the original to `<table>_old` and rename the copy to the original name. Watch out for constraints and auto-increment values; leave a ~10% ID buffer in one region to avoid ID clashes between the two regions for a while. Proved faster than deleting the cross-region rows from the live table based on the team's tests.