Savonitar opened a new pull request, #331: URL: https://github.com/apache/flink-connector-kafka/pull/331
## What is the purpose of the change Reduce weekly CI work by skipping each released-version row unchanged since its last successful Java and Python builds, while always testing snapshots and refreshing released rows after 28 days. In the [September 27 weekly run](https://github.com/apache/flink-connector-kafka/actions/runs/36283436510), only the Java 1.20.5 / v3.4 job failed. Both v5.0 rows and the v4.0 row passed. With this per-row baseline in place and unchanged connector, shared-CI, and weekly workflow revisions, the following scheduled run would retry v3.4 and test the snapshot while skipping those three green released rows. ## Brief change log - Store a baseline per released row, keyed by its Flink version, connector branch, and JDK configuration. Validate it against the preceding run's latest Java and Python job results, regardless of the overall conclusion. - Use the run's resolved reusable-workflow revisions and pin both connector checkouts to the recorded SHA. Keep explicit caller names for matching jobs to rows. - Include the `weekly.yml` file blob SHA at the run's `context.sha` in each baseline, so changed workflow parameters rebuild released rows while unrelated commits do not. - Preserve the original successful-build timestamp while a row is skipped. Rebuild changed, unverified, or 28-day-old rows; manual dispatch runs the full matrix. - Explain the model above `prepare_matrix` and execute regression tests against the inline scripts in PR CI, failing on unexpected warnings or swallowed mock assertions. ## Verifying this change Run `node --test .github/scripts/weekly-matrix.test.js`. Coverage includes partial failures, snapshot failures, expiry across skipped weeks, referenced-CI and weekly workflow changes, unchanged workflow content across commits, legacy baselines, manual dispatch, unavailable metadata, and strict mock validation. All 65 tests pass. Red-green checks confirmed that the regression tests fail without their corresponding fixes. A separate replay of the per-row selection using the September 27 run's public job results and unchanged branch/shared-CI fixtures selected only the snapshot and v3.4. `./mvnw clean verify` passed on a clean checkout of the pushed commit with Java 17, including the connector integration tests. YAML parsing and `git diff --check` also passed. A live run of the updated weekly workflow has not been performed. If resolved reusable-workflow metadata or the weekly workflow blob SHA is unavailable during preparation, the workflow tests all rows and records no reusable baseline. ## Does this pull request potentially affect one of the following parts: - Dependencies: no - Public API, Table options, or PyFlink wrappers: no - Checkpointed connector state, serializers, or exactly-once delivery: no - Per-record code paths: no ## Documentation - New connector feature: no - Feature documentation: not applicable - Changes to both documentation languages: not applicable --- ##### Was generative AI tooling used to co-author this PR? - [X] Yes Generated-by: Codex (GPT-6) -- 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]
