When a changefeed is resumed, TiCDC needs to scan the historical versions of data in TiKV to catch up with the incremental data logs generated during the pause. The replication process proceeds only after the scan is completed. The scan process might take several to tens of minutes.