I created a scheduler job on 2am repairing cass cluster. It runs repair -pr on each node. And after repair, it will use nodetool tablestats checking each table's 'Percent repaired' if it's 100%. Then at 6am, it will again nodetool tablestats checking 'Percent repaired'.
So I found after repair at 2am, all tables will show 'Percent repair:100%'. Then at 6am, the metric drop under 100% on one or two tables.
Between 2am and 6am, there were no data write or update at all.
Could someone explain this phenomenon?