Saved offset is before replication slot's confirmed_flush_lsn; streaming will start from the confirmed_flush_lsn

Root cause

The LSN stored in the Kafka Connect offset is older than the slot's confirmed_flush_lsn. PostgreSQL JDBC driver ≥ 42.7.0 aggressively flushes keepalive LSNs, advancing confirmed_flush_lsn beyond the position Debezium last committed. Events between the two positions are lost.

How to fix

1. Upgrade to Debezium ≥ 2.7.4 which includes the fix for pgjdbc 42.7.x keepalive flush behavior.
2. Keep the default lsn.flush.mode=connector (Debezium only flushes on commit, not on keepalive). If you need connector_and_driver, also set offset.mismatch.strategy=trust_slot to tolerate the discrepancy safely.
3. If events were already lost: audit downstream for gaps and trigger an incremental snapshot on affected tables.