Abhishek Chennaka created KUDU-3800:
---------------------------------------
Summary: Diff scan drops rows whose full lifecycle is contained in
the scan window
Key: KUDU-3800
URL: https://issues.apache.org/jira/browse/KUDU-3800
Project: Kudu
Issue Type: Improvement
Reporter: Abhishek Chennaka
Assignee: Abhishek Chennaka
Kudu's diff scan does not surface a row that is deleted at both endpoints of
(snap_start, snap_end], even when it had a full INSERT → DELETE lifecycle
inside the window. include_deleted_rows doesn't help — it only surfaces rows
live at snap_start.
This breaks two-tier consistency: a row previously migrated to a second tier
and re-inserted-and-deleted inside a diff-scan window is never reported, so any
stale copy downstream is never reconciled.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
