> They maintained a replication process across both tables as they updated the read processes before updating the write process. Say for whatever reason their offline replication process broke for 2 hours.

From the article I got the impression that both tables were being written to in the same database transaction, so this is not a possible failure scenario at all.