The log mining query may unintentionally skip records while processing large batches of rows due to how the SCN range predicate is applied:
DBZ-2875, we allowed the startScn to be inclusive but this lead to repeated events. The fix there should have been to move the inclusiveness such that the query read:
Additionally, the query also does not scope COMMIT and ROLLBACK rows to be within the SCN boundary, which should not be the case.
This was sourced from the discussion started https://issues.redhat.com/browse/DBZ-2875?focusedCommentId=15954247&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15954247