SEZ9 commented on PR #12081: URL: https://github.com/apache/seatunnel/pull/12081#issuecomment-6029328731
@Rangsh Answering the option question first: let's go with **(a)**. The fail-loud guarantee is the whole point of this PR, the WAL worker stall on the `file:///` shared-path setup is pre-existing, and a cancellation-path carve-out in the vertex state machine deserves its own review cycle — for exactly the too-broad-catch reason you called out. Please do not implement (b) here. Two concrete asks so (a) is closed out cleanly: 1. In the worker-hardening follow-up you referenced, please record the user-visible change you described (cancel-path bookkeeping write timeout → job ends `FAILED`/`UNKNOWABLE` instead of `CANCELED`) so it is tracked explicitly, and add a one-line pointer to it in this PR's description. 2. Leave `SplitClusterFaultToleranceIT#testStreamJobRestoreInAllNodeDown` as-is — no retry wrapper or disable. If it goes red on that path again I'd rather see it than mask it. On the earlier review points: your one-line recap on `ab04db33a` got cut off right after "1a. Untimed `RequestFuture.get()` unbounded blocking — fixed". Could you re-post the rest? What I still need confirmed, one line each: - the single disruptor worker surviving an `executeResponse()` throw (given the caller-side timeout+remove race); - the writer being reused after an arbitrary write failure in `WALWorkHandler` (torn-record risk); - the timed `get(timeout, unit)` TimeoutException contract documented on the method itself, not just the class; - the Mockito test dependency for `HdfsWriterFlushSyncPathTest` in the imap-storage-file pom; - `batchQueryExecuteFailsStatus` waiting the full `writDataTimeoutMilliseconds` per entry sequentially; - the full stack trace at ERROR on every timed-out wait in `queryExecuteStatus`. Once that recap is in, I'll do a final pass and we can wrap this up. Thanks for the clear write-up of the trade-offs. <!-- streview-comment:1561 --> -- 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]
