Rangsh commented on PR #12081:
URL: https://github.com/apache/seatunnel/pull/12081#issuecomment-5627847969

   @DanielLeens thanks for the from-scratch re-review of `1de6f1b8` and for 
catching the fail-close / `FileMapStore` interaction — agreed this was a 
blocking durability regression, not a design nit.
   
   Pushed **`f17373ac5`** to address Issue 1 (and the Issue 3 Javadoc clarity 
note):
   
   ### Issue 1 (blocking) — fail-close must be visible/actionable
   
   - Exposed the sticky fail-close flag: 
`WALWorkHandler.isAppendBlockedAfterWriteFailure()` → `WALDisruptor` → 
`IMapFileStorage.isAppendPermanentlyBlocked()` (new default on `IMapStorage`, 
default `false`).
   - `FileMapStore.store()` / `storeAll()` no longer discard the storage-layer 
failure signal: they throw `IMapStorageException` when the write is not durable.
   - When fail-close has tripped, the exception message is explicit that APPEND 
is **permanently** blocked for this process and that an engine-node restart is 
required — so Hazelcast write-through / the checkpoint coordinator can fail 
loudly instead of silently advancing in-memory IMap state while WAL persistence 
has stopped.
   
   Coverage:
   - `FileMapStoreTest`: store / storeAll throw on underlying failure; 
fail-closed path asserts the permanent / restart message.
   - `WALWorkHandlerSurvivabilityTest`: asserts the sticky getter after the 
first write failure.
   
   ### Issue 3 (non-blocking) — Javadoc permanence
   
   - Class/field Javadoc on `WALWorkHandler` now states explicitly that 
fail-close is sticky for the handler/process lifetime with **no in-process 
reset**.
   
   ### Issue 2 (CI)
   
   - Will rely on the new head’s CI re-run for `doris-connector-it` / 
`mysql-cdc-connector-it`; still treating those as unrelated to this module if 
they flake again.
   
   Happy to take another pass once you’ve re-traced `f17373ac5`.


-- 
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