Zerolzj opened a new pull request, #10714: URL: https://github.com/apache/paimon/pull/10714
### Purpose Bound nested test concurrency in the Flink 1 connector lane without changing test coverage or timeouts. The lane currently runs the Maven reactor with `-T 2C` while Surefire inherits `flink.forkCount=1C`. These independent CPU-scaled settings multiply across simultaneously active connector modules. On a four-CPU runner, the six selected modules permit an upper bound of 24 test JVMs; many test classes start their own MiniCluster. This is an execution-budget problem, not a change to Flink runtime parallelism. Observed while investigating the failed connector check in #10662: - The upstream job used eight Maven worker threads. - `RemoveOrphanBlobsActionITCase.testDistributedCleanupDoesNotHangWithPipelinedShuffle` exceeded its existing 60-second timeout in Flink 1.19; the same case took about 46 seconds in Flink 1.18. - With clean JDK 8 builds locally, the Flink 1.19 test class passed in about 14 seconds on the PR head and 17 seconds on its unchanged base. The complete 21-case Blob cleanup class also passed on the PR head. These observations motivate bounding CI resource contention, but do not establish the precise cause of the upstream timeout. Linux validation is still pending. This PR is separate from the bucket-function proposal and changes no Java production code. ### Changes - Retain `-T 2C` for dependency compilation with tests skipped. - Use `-T 1C` and `-Dflink.forkCount=1` only during Flink 1 connector tests. Reactor scheduling remains parallel, but test JVM counts no longer multiply by CPU count at both layers. - Add a dry-run command regression which checks the aggregate upper bound on 1, 2, 4 and 8 CPU runners. The previous commands fail this budget; the new commands pass. - Preserve CDC, all five Flink 1 connector versions, verification goals, existing test timeouts and the other CI lanes. ### Validation - `python3 -m unittest discover -s tools/ci -p 'test_*.py'`: 31 passed. The new budget test first failed on the old script for all four runner sizes. - `bash -n tools/ci/run-java-tests.sh`: passed. - Dry-run output confirms dependency build remains `-T 2C install -DskipTests -am`; test execution is `-T 1C verify -Dflink.forkCount=1` without `-am`. - Local Java/Blob cleanup results above are diagnostic controls on the existing code, not a claim that the modified CI lane has already passed on Linux. One commit, two files, 17 added lines. No tests disabled, no retries added, no timeout increase. -- 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]
