One of the biggest problems with VDiff was always that it was fragile — any ephemeral error along the way, such as a mysql connection getting timed out, would cause it to terminate — and it was not resumable. This was very problematic for larger tables which could take hours or days to complete:

• Any little problem would cause the entire job to fail
• You would have to start over from scratch
• The cost was high for each run as you have to diff the entire table again each time

We should address all of these issues in VDiff2 by making it resumable based on the last row (LastPK) that we had previously in the VDiff.