zephyr45 commented on PR #13060: URL: https://github.com/apache/seatunnel/pull/13060#issuecomment-6967720978
@SEZ9 Per-finding note for head `cf52689e`. That commit only deletes the `SinkContextProxy` override and its `MultiTableSinkWriterTest` case; it does not touch Kafka or Spark code, so everything below still applies. - **F1:** `validateRestoredStateOwner` and the modulo check were removed in `585f69b7`; restore consumes whatever states the engine assigns, so a Flink contiguous scale-down no longer throws. Tests: `KafkaSinkWriterTest#shouldFenceContiguouslyAssignedFlinkNamespacesAfterScaleDown`, `#shouldFenceEveryAssignedWriterNamespaceAfterScaleDown`. - **F2:** With `transaction_prefix` unset, each restored state is parsed against its own saved generated prefix and every assigned namespace is fenced (`585f69b7`). Test: `#shouldRestoreMixedGeneratedPrefixesAfterScaleDown`. - **F3:** A writer restored without state (scale-up) starts at sequence 1 in its current-index base namespace; `initTransactions` on that ID fences an earlier attempt before the first send, without walking the namespace history (`d8143b58`, `b9ed7c12`). Tests: `#shouldReuseGeneratedPrefixWhenStatelessWriterRetries`, `#shouldNotWalkStatelessWriterTransactionHistory`. The documented `transaction.timeout.ms` boundary for transactions outside checkpoint state is unchanged. - **F4:** Fixed in this PR, not deferred. By "unchanged" I meant that `cf52689e` didn't touch it. `SparkSinkWriterContext#getStableSubtaskIndex()` returns the partition ID (`9e4aad72`), so a task retry derives the same namespace and fences the previous attempt's open transaction. Tests: `KafkaIT#shouldFenceAbandonedAttemptWithStableIndex` (task IDs 99 → 100, same partition, real broker), `SparkDataWriterFactoryTest` / `SparkSinkTest#shouldSeparateTaskIdentityFromStablePartitionAcrossRetriesAndEpochs`. - **F7:** Yes. In both EN and ZH `Kafka.md`, the `transaction_prefix` row describes `<prefix>-<writer>-<sequence>` and links to Semantics; the rescaling paragraph is scoped to Zeta and Flink (`7887e909`, updated in `b9ed7c12` and `9e4aad72`). - **F5/F6:** Deferred to #13298, as agreed. CI on `cf52689e` ([fork Build 44125840341](https://github.com/zephyr45/seatunnel/actions/runs/44125840341)): unit tests and both `kafka-connector-it` jobs passed. Six jobs failed for infrastructure reasons unrelated to this PR: a missing Maven wrapper class (2), a 404 on the `atmoz/sftp:alpine-3.7` image (2), and `JdbcIrisIT` container start-up timeouts (2). -- 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]
