The GitHub Actions job "Label-when-reviewed" on flink-cdc.git/FLINK-40647 has succeeded. Run started by GitHub user lvyanquan (triggered by lvyanquan).
Head commit for run: 4ae9e39c5ed980d63cb1fed45cfcbcdcf0b03608 / lvyanquan <[email protected]> [FLINK-40647][runtime] Compare table keys and normalize added column types during expansion Address the review feedback on the existing-table expansion analysis. * Compare the pipeline schema's primary key with the existing target table's. A target table with matching columns and types but a different or absent primary key reported no differences, so CHECK passed and suppressed the CreateTableEvent, which in turn bypassed the connector's own key validation (FlussMetaDataApplier.sanityCheck) and let the job write under the wrong upsert semantics. Partition keys are only compared when the pipeline declares them, because most sources do not report partitioning and a two-sided comparison would flag externally partitioned target tables. Keys are compared as case-normalized sets, matching how the connectors' own checks compare them, so an ordering-only difference is not reported. * TRY_EXPAND now delegates without issuing any DDL when keys differ, instead of mutating a table whose keys can never match the pipeline. * Normalize the type of a column that is about to be added, so a type the target system cannot express is reported while probing rather than failing later when the derived DDL runs, which contradicted the documented capability-gap-is-phase-1 rule. The gate keeps the pipeline type in the emitted event. * Render every unresolved difference after an expansion, not only the incompatibilities, which printed an empty list when the applied DDL silently had no effect on the target table. The option documentation states the key comparison rule and the resulting per-mode behavior in both languages. AI-Contributed/Feature: 0/114 AI-Contributed/UT: 0/241 Report URL: https://github.com/apache/flink-cdc/actions/runs/35966610515 With regards, GitHub Actions via GitBox
