DanielLeens commented on PR #12626: URL: https://github.com/apache/seatunnel/pull/12626#issuecomment-6905412781
Thanks for the detailed follow-up. I checked the current diff at `c43710878bae79fc2949fa0a58cb82b1432c1858` against these points. - **Scope:** the diff now contains nine files: four deployment documentation files, the MongoDB and Hadoop contract tests, `DefaultClassLoaderService`, `ReleasedClassLoaderReferenceCleaner`, and its engine test. The previously integrated source/checkpoint files are absent. - **F1:** `stopMongoBufferPoolPruner` calls `findLoadedClass` before reading `DEFAULT` and returns when the class was not loaded or belongs to another loader. This avoids defining the pool class merely because the driver is on the job jar path when reflective access is available. There is still an explicit fallback to class resolution if `ClassLoader.findLoadedClass` is inaccessible; the warning names `--add-opens java.base/java.lang=ALL-UNNAMED`. Also, loaded does not imply initialized: reading `DEFAULT` can still initialize an already-loaded class. I am not claiming an unconditional no-initialization guarantee. - **F2:** `DefaultClassLoaderService.cleanReleasedClassLoaderReferences` catches `Throwable` around the cleaner invocation after the bookkeeping. A non-`LinkageError` `Error` from the cleaner is therefore caught at this release boundary rather than escaping directly. Individual cleanup operations also have their own guards. This is best-effort handling, not a guarantee that logging itself can survive a terminal VM resource failure. - **F3:** the diff includes `MongoBufferPoolPrunerContractTest` in connector-mongodb and `HadoopClassLoaderRootsContractTest` in connector-file-base. They resolve the actual dependency class/field/method shapes, rather than only engine fixtures. This source check is not a claim that every dependency version or runtime cleanup behavior has been validated. - **F4/F6:** `CLASSLOADER_SERVICE_SKIP_RELEASED_REFERENCE_CLEANUP` is defined and read in `DefaultClassLoaderService`; the English and Chinese hybrid/separated deployment documents describe it. The cleaner's maintenance-contract Javadoc records mongodb-driver-core 4.7.1 and Hadoop 3.1.4/3.3.6 as the referenced layouts. `logStepFailure` de-duplicates identical step/exception/message signatures: the first occurrence is WARN with a stack trace, subsequent occurrences DEBUG. This applies to that path, not every warning or the outer release-boundary catch. - **F5:** the comment immediately before the comparator configuration reread explicitly says it narrows the race but cannot close it from outside Hadoop. - **F7:** the CodecPool INFO text now says `entries`. The CI information has also changed: [Code style on attempt 2](https://github.com/DanielLeens/seatunnel/actions/runs/43552064252/job/132483854127) passed on this same head, so another style retry is unnecessary. The overall Build remains failed: all-connectors-it-5 JDK 11 and jdbc-connectors-it-part-7 JDK 11/17 are still red. I have not classified those remaining failures as unrelated or claimed merge readiness. No local functional execution was performed for this follow-up. -- 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]
