SEZ9 commented on PR #12096:
URL: https://github.com/apache/seatunnel/pull/12096#issuecomment-6889190128

   Status check on unchanged head `aa30f22ef2bd316cb707d89ee233e183a152a0d5`.
   
   Agreed on the framing: the successful fork Build 41162813818 is good to see, 
but it does not address the open source-level findings, so this head is not 
approvable yet. What remains:
   
   - **F1 (HIGH)** – constructor-time retention cleanup can still delete the 
checkpoint/savepoint the job is being restored from. The restore target must be 
excluded from the deletion candidates.
   - **F2 / F4** – retention restore enumerates all durable pipeline states 
without separating savepoints from checkpoints, so user-created savepoints 
become eligible for retention deletion after restart/failover.
   - **F3** – cleanup retry assumes `deleteCheckpoint` is idempotent; a 
partially-deleted batch can wedge cleanup permanently, and the retention 
backlog is unbounded if deletion keeps failing.
   - **F5** – the recovery scan deserializes the full state payload of every 
retained checkpoint synchronously inside the coordinator constructor; this 
needs to be bounded.
   - **F7** – only `CheckpointStorageException` is caught; a `RuntimeException` 
from storage can escape the constructor and abort coordinator recreation during 
active-master failover.
   - **F6 / F8** – `checkpoint.storage.max-retained` semantics changed (exactly 
max-retained kept, one `deleteCheckpoint` call per completed checkpoint) with 
no documentation update.
   
   On the overlap raised in @davidzollo's 20261107 note: as stated there, a 
maintainer needs to reconcile the overlapping retention PRs before this branch 
can proceed. I am not in a position to make that call, so I will leave it to 
the maintainers.
   
   If this PR remains the selected implementation, suggested order for the next 
revision:
   1. F1 first (exclude the restore target), with a unit test that restores 
from the oldest retained checkpoint and asserts it survives cleanup.
   2. F5 and F3 together, as the two stability blockers: bound materialization 
during the recovery scan and bound the pending-deletion backlog.
   3. Then F2/F4, F7, and the docs update for F6/F8.
   
   Thanks for keeping the CI result and the source findings separate — it makes 
the remaining work clear.
   
   <!-- streview-comment:2370 -->


-- 
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