The only way I've found to accomplishing this is to do the full refresh in stages (which I know sounds like a contradiction). In our case we're replicating from MySQL to Postgres, so for a full refresh I'll drop all the tables in Postgres, set the select list to A*, then run the pipeline; add B*, run the pipeline; add C*, run the pipeline; etc. That way when the error occurs only 1/26th of the tables are affected, I can fix that, then move on.