SEZ9 commented on PR #12626:
URL: https://github.com/apache/seatunnel/pull/12626#issuecomment-6937343762

   Thanks for the detailed update on 
`c43710878bae79fc2949fa0a58cb82b1432c1858`. The notes below go only by the 
description in this thread, so I'm not marking any point resolved until it is 
confirmed against the diff.
   
   **Described as addressed (pending confirmation in the diff)**
   - **F2:** As described, 
`DefaultClassLoaderService.cleanReleasedClassLoaderReferences` catches 
`Throwable` around the cleaner invocation after the bookkeeping. If so, that 
would cover the non-`LinkageError` `Error` case, and best-effort is the right 
bar here.
   - **F4:** As described, there is a 
`CLASSLOADER_SERVICE_SKIP_RELEASED_REFERENCE_CLEANUP` switch, English and 
Chinese deployment docs, and a maintenance-contract Javadoc naming 
mongodb-driver-core 4.7.1 and Hadoop 3.1.4/3.3.6. If so, that would address the 
maintainer-pointer concern.
   
   **Remaining asks**
   - **F1:** You describe a `findLoadedClass` guard before reading `DEFAULT`, 
and you note two paths where the pool class can still be initialized.
     - When `ClassLoader.findLoadedClass` is inaccessible, the fallback 
resolves the class anyway. Could that fallback skip the Mongo step instead, 
keeping the warning about `--add-opens java.base/java.lang=ALL-UNNAMED`? If 
not, please explain why resolution is preferred.
     - Reading `DEFAULT` can still initialize a class that is loaded but not 
yet initialized. Please either avoid that or note it as a known limitation in 
the Javadoc.
   - **F3:** You mention `MongoBufferPoolPrunerContractTest` and 
`HadoopClassLoaderRootsContractTest` resolving the actual dependency 
class/field/method shapes. Which Hadoop version does the Hadoop contract test 
resolve against, and does it cover the shipped uber jar layout or only the 
plain dependency? If one of the 3.1.4/3.3.6 layouts is not covered, please say 
so in the test Javadoc.
   - **F6:** You say the de-duplication in `logStepFailure` does not apply to 
the outer release-boundary catch. Can that catch log a full stack trace at WARN 
on every release for the same repeated failure? If so, please apply the same 
de-duplication there.
   - **F5:** The explanation is cut off after "narrows the race but cannot". 
Please restate it, covering the two points below.
     - Is the `WritableComparator` registry now iterated and mutated under the 
same lock Hadoop's writers use?
     - What risk remains that the `getConf`/`setConf` pair nulls a running 
job's `Configuration`?
   - **F7:** The visible text ends partway through F5, before F7 could appear. 
Could you share the status of the `entrie(s)` wording in the CodecPool INFO 
line?
   
   <!-- streview-comment:2400 -->


-- 
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]

Reply via email to