SEZ9 commented on PR #13261:
URL: https://github.com/apache/seatunnel/pull/13261#issuecomment-6906659838

   Thanks for the update — good to see the Apache Build (131695248136) and the 
fork Build (43367464099) now pass on 
`24b8e5153a604d7327a40a30202a2af595902b94`. A green build doesn't close out the 
review items, though, so here is what is still open from my side:
   
   1. **Double assignment on Zeta (F1)** — `addSplitsBack` now resets `assigned 
= false`, but `assignSplit()` is still an unsynchronized check-then-act. On 
Zeta, `addSplitsBack` and `registerReader` run under different locks, so with 
parallelism > 1 the single split can be handed out twice. Please either guard 
the `assigned` flip-and-assign in one critical section, or explain why that 
interleaving cannot occur.
   
   2. **Engine-level restart verification (F2/F5/F6)** — the new tests cover 
the Flink ordering (reader removed before the split comes back) and now have 
deterministic ordering, which is useful. They still don't cover the Zeta 
ordering: `addSplitsBack` -> `registerReader` -> `run()`, and `run()` after a 
re-assignment to confirm the guard against a second `new SingleSplit(null)`. 
Beyond the unit test, please verify against the real engine that the split is 
actually returned and reassigned on restart, not just that the enumerator 
callbacks behave — and note in the PR description how that was checked.
   
   3. **Javadoc (F4)** — the empty-list case is described, but please add a 
sentence on *why* returning early is safe (the reader restores its own split 
from checkpoint state).
   
   4. **Style (F7)** — make the test class and methods `public` to match the 
module's existing tests, and consider a debug log on the take-back path, 
consistent with sibling enumerators.
   
   For the record, I agree the fix itself is correct for Flink's 
regional-failover contract (F3): the reader is unregistered before 
`addSplitsBack`, so clearing `assigned` parks the split until the reader 
re-registers, with no state/option/serialization change. Once items 1–2 are 
addressed I think this is close.
   
   <!-- streview-comment:2381 -->


-- 
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