fatmanverse commented on PR #11487:
URL: https://github.com/apache/seatunnel/pull/11487#issuecomment-5018333222

   @nzw921rx Thanks for the detailed review. I addressed the requested changes 
in commit e22ab12d7.\n\n1. Snapshot-only restore now treats the checkpointed 
table set as authoritative. It collects tables from processed tables, remaining 
tables, remaining snapshot splits, and assigned snapshot splits, and does not 
call catalog discovery on the snapshot-only restore path. Legacy checkpoint 
state is also marked as having checkpointed remaining tables before the 
snapshot assigner opens, so it cannot rediscover the catalog. The new unit 
coverage verifies that a newly created table is excluded after restore.\n\n2. 
The E2E now reuses MysqlCDCDriverResolver and copies the Maven-resolved 
Connector/J into the container. It no longer downloads a driver from inside the 
container, and the test uses the declared 8.0.27 driver.\n\n3. Both existing 
public HybridSplitAssigner constructors are retained and delegate to the new 
overloads with snapshotOnly = false.\n\n4. Snapshot-only validation now runs 
 before datasource, dialect, deserializer, and offset initialization. It 
rejects stop modes other than never, startup.specific-offset.*, and 
startup.timestamp. The negative test verifies that startup.timestamp fails 
before source initialization.\n\n5. The E2E seeds 103 rows and sets 
snapshot.split.size = 1, so multiple snapshot splits are exercised. It starts 
the job asynchronously, waits until rows from early splits are visible while 
later splits are still outstanding, updates and deletes rows from completed 
snapshot ranges, and verifies that the sink remains exactly the original 
snapshot result. The test passed all 5 runtime invocations: Flink 1.13.6, 
1.15.3, 1.18.0, 1.20.1, and Zeta. The Zeta run reported 103 records read, 103 
written, and 0 failed.\n\n6. exactly_once remains supported for both initial 
and snapshot startup modes. The snapshot mode documentation now describes the 
bounded low-to-high-watermark backfill semantics and the controlled 
concurrent-write E2E coverage. Snap
 shot-only still never enters continuous binlog streaming.\n\nValidation 
completed:\n- ./mvnw spotless:apply\n- CDC base targeted tests: 6 tests 
passed\n- MySQL factory tests: 2 tests passed\n- MysqlCDCSnapshotOnlyIT: 5 
tests passed\n- git diff --check\n\nThe repository-wide -DskipTests verify is 
still blocked by pre-existing origin/dev compilation errors in connector-jdbc, 
including the duplicate setNullToStatementByDataType method and missing 
generated methods. No connector-jdbc files are changed in this PR. The full 
test run is also blocked by the existing seatunnel-config-shade shaded Typesafe 
Config ClassNotFoundException.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to