Description
When a memtable was flush, there is a small delay before the commit log replay position gets updated. If the node fails during this delay, all the updates of this memtable will be replay during commit log recovery and will end-up being over-counts.