The GitHub Actions job "Flink CDC CI" on flink-cdc.git/fix_flaky_tests has 
failed.
Run started by GitHub user leonardBang (triggered by leonardBang).

Head commit for run:
a8db17b0f30fd14a5bfb4abdf912a4dbeab55776 / Leonard Xu <[email protected]>
[test][connector/mysql] Fix upsert convergence race and TM remove-table 
failover hang

Two follow-up fixes to the snapshot-failover stabilization:

1. testJobManagerFailoverForNewlyAddedTableWithAheadBinlog asserted the
   binlog result right after a size-based wait plus a fixed 1s sleep. The
   binlog phase updates a row in place (PK unchanged, row count unchanged)
   and inserts another row (row count +1), so the size target is reached
   the instant the insert lands, before the in-place China->CHINA update
   necessarily arrives. Under failover load 1s was not always enough,
   producing a flaky China/CHINA assertion. Replace the size wait + sleep
   with waitForUpsertSinkContent, which polls until the sink content
   converges to the expected rows.

2. testRemoveTablesOneByOne triggered a snapshot-phase failover only after
   waiting for one sink row. For the TaskManager case
   (testTaskManagerFailoverForRemoveTable) restarting the TM mid-snapshot
   left the snapshot unable to complete, hanging until the 300s timeout.
   Restore the original immediate failover for the remove-table path; the
   size wait was only needed for the JobManager newly-added-table case.

Co-Authored-By: Claude Opus 4.7 <[email protected]>

Report URL: https://github.com/apache/flink-cdc/actions/runs/27893513268

With regards,
GitHub Actions via GitBox

Reply via email to