Lokesh Khurana created PHOENIX-7911:
---------------------------------------
Summary: Eliminate post-abort TableNotFoundException flood by
deferring new-physical drop
Key: PHOENIX-7911
URL: https://issues.apache.org/jira/browse/PHOENIX-7911
Project: Phoenix
Issue Type: Sub-task
Reporter: Lokesh Khurana
{{TransformTool.abortTransform}} today killJobs + removeTransformRecord, but
leaves {{TRANSFORMING_NEW_TABLE}} link in place. Stale-cache Phoenix clients
keep dual-writing to a NEW physical that the abort tried to drop, producing
{{TableNotFoundException}} flood for the entire UCF tail.
*Decision:* Shape B (deferred new-physical drop):
killJob → {{Transform.disableDualWrite}} (DELETE link, shared with Gap B) →
removeTransformRecord → {{Transform.scheduleNewPhysicalDrop}} (deferred drop
after UCF window) → return SUCCESS instantly to operator. During the UCF tail,
stale-cache dual-writes route to a NEW physical that still exists → writes
succeed silently. After UCF expires, deferred-drop fires safely.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
