Timo Theusner created FLINK-41236:
-------------------------------------
Summary: Recover materialized tables that have no controllable
refresh job
Key: FLINK-41236
URL: https://issues.apache.org/jira/browse/FLINK-41236
Project: Flink
Issue Type: Sub-task
Components: Table SQL / Gateway, Table SQL / Planner
Reporter: Timo Theusner
{{INITIALIZING}} tables, and {{ACTIVATED}} tables whose cluster reports the
refresh job as not found or whose YARN application ended, cannot be dropped,
because {{DROP}} must cancel a job first. Let {{DROP MATERIALIZED TABLE}}
remove them from the catalog without cancelling a job or deleting a workflow,
and let {{RESUME}} start a new refresh job without state for the {{ACTIVATED}}
ones, keeping their definition and data. Otherwise the Catalog API is the only
way to recover them.
Proposed Changes
* Remove an {{INITIALIZING}} table on {{DROP}} in
{{{}DefaultMaterializedTableExecutor{}}}, in {{CONTINUOUS}} and {{FULL}} mode,
without cancelling a job or deleting a workflow
* Treat a job reported as not found, or an ended YARN application, as finished
for {{DROP}} and {{{}RESUME{}}}, which starts a new refresh job without state
* Reject {{CREATE OR ALTER}} and all other operations on such tables, with an
error naming {{DROP MATERIALIZED TABLE}}
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
