hongzhi-gao commented on PR #19241: URL: https://github.com/apache/iotdb/pull/19241#issuecomment-6842368478
Would it make sense to split this into two dependent PRs, with the recovery guarantees landing before the retry behavior? 1. **Recovery and idempotency foundations:** preserve acknowledged unfinished staging across shutdown and cleanup; make partial imports recoverable without duplicate installation or publication effects; coordinate terminal outcomes with replication WAL durability; and keep snapshots and lifecycle operations consistent with that protocol. The related lock-order and initialization/deletion fixes could belong here. 2. **Retry behavior:** distinguish temporarily incomplete replicated PREPARE/COMMIT commands from permanent failures, return the retryable status, and validate eventual completion through disconnect/reconnect and restart ClusterITs. The first PR would need to be independently safe and testable. If there is no safe intermediate implementation, keeping one PR with clearly separated commit groups would be preferable to forcing the split. As a follow-up, it may also be worth exploring **targeted retransmission of missing PIECEs**: let the receiver identify missing pieces and resend only those, rather than retransmitting the original consensus batch. This would require stable piece identities, retained retransmission sources, and idempotent duplicate handling. I would keep it outside this fix; it would not replace the recovery and WAL guarantees above. -- 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]
