To provide read-only incremental snapshot we need a totally ordered identifier that could be queried before and after chunk query to denote window boundaries.

For MySQL this is the known GTID set.

Let's explore for other databases

• SQL Server - fn_dblog() might provide tx id or commit lsn, HADR metadata might provide the same
• MongoDB - correlate clusterTime from change message with oplog tail
• PostgreSQL - consider using pg_current_snapshot ()