SEZ9 commented on PR #12626:
URL: https://github.com/apache/seatunnel/pull/12626#issuecomment-6888184811

   Thanks for the update at `c43710878bae79fc2949fa0a58cb82b1432c1858`.
   
   On scope: the description says the previously integrated source/checkpoint 
files are no longer in the diff. I don't have that visible in the thread yet, 
so could you confirm the current file list is the nine-file classloader patch 
only? Once I can see that in the diff I'll consider the scope concern closed.
   
   Going through my earlier findings:
   
   - **F1 (forced `PowerOfTwoBufferPool` init in the Mongo cleanup)** – The 
description says the cleanup "checks MongoDB's pruning root without loading it 
unnecessarily", which is the behaviour I was after. Could you point me to how 
the check avoids triggering `<clinit>` when the driver is merely on the jar 
path but was never used? The same note observes that "reflective access can 
initialize a loaded class", so I want to be sure the no-load path is the common 
case rather than a best-effort fallback.
   - **F2 (`clean()` not fully best-effort)** – Still open from what I can see. 
Please confirm that a non-`LinkageError` `Error` thrown inside the cleaner is 
caught and logged rather than escaping `releaseClassLoader` after the loader 
has been removed from the maps; otherwise one bad loader aborts the caller's 
loop over the remaining jars.
   - **F3 (reflective contract tested only against fixtures)** – Still open. A 
test that resolves the class/field/method names against the real 
mongodb-driver-core and Hadoop artifacts on the classpath (even a simple 
"layout still matches" assertion) would keep a dependency bump from silently 
turning this back into a leak. If that is not feasible in this PR, please say 
so and we can track it separately.
   - **F4 (engine hard-coding connector internals)** – You mention a 
"documented process switch"; please point me to where that documentation lives, 
and add a short code-level note near the reflective constants stating which 
library versions they were verified against.
   - **F5 (unlocked comparator registry / `getConf`/`setConf` reread)** – 
Acknowledged as a known, non-atomic limit. I'm fine accepting that for the 
release-time path as long as it is stated in a comment at the call site.
   - **F6 (WARN stack trace per release, no operator switch)** – Could you 
point me to the switch in the diff so I can confirm it? Also, is the 
layout-drift warning now de-duplicated, or still logged with a full stack trace 
on every release?
   - **F7 (`entrie(s)` grammar)** – Please confirm the CodecPool INFO message 
was fixed.
   
   On CI: the Code style job (run `43552064252`, job `132316548455`) failed 
before Spotless ran because of the Maven Wrapper download, so it tells us 
nothing about formatting. Please re-run it once the wrapper download is healthy 
so we have a green style check on the current head.
   
   Once F2, F3, and the pointers for F4–F6 are addressed (or explicitly 
deferred), I'm happy to take another pass.
   
   <!-- streview-comment:2358 -->


-- 
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