You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Weiqing Yang (Jira)" <ji...@apache.org> on 2026/08/03 00:20:00 UTC
[jira] [Created] (FLINK-40299) Add end-to-end state migration tests for RowData schema evolution
Weiqing Yang created FLINK-40299:
------------------------------------
Summary: Add end-to-end state migration tests for RowData schema evolution
Key: FLINK-40299
URL: https://issues.apache.org/jira/browse/FLINK-40299
Project: Flink
Issue Type: Sub-task
Components: Table SQL / Runtime
Reporter: Weiqing Yang
Adds coverage that exercises the full restore path on RocksDB rather than the serializer snapshot in isolation, in the module that sees both the state backend and RowData.
Cases:
* A nullable field added to ValueState<RowData>, option enabled: state restores, existing values intact, the new field null.
* The same change with the option disabled: restore is rejected.
* The same change with state TTL enabled: values migrate correctly and TTL timestamps are preserved.
* State shaped like an interval join buffer (RowData below List and Tuple serializers): restore is rejected rather than migrated.
The last two matter most. TTL is the one case where a RowData serializer is both opted in and wrapped in a composite, so its correctness rests on the TTL-aware migration path rather than on structure. The interval-join case is the fail-closed guarantee: migration is not propagated into nested serializers by composite snapshots, so that shape must be rejected rather than silently mis-migrated.
Depends on FLINK-40298.
FLIP-527: https://cwiki.apache.org/confluence/spaces/FLINK/pages/353601981/FLIP-527+State+Schema+Evolution+for+RowData
--
This message was sent by Atlassian Jira
(v8.20.10#820010)