Hi community,

Thanks for the feedback on the initial proposal. I've now completed the
detailed design for STIP-23 and updated it in the GitHub issue:

https://github.com/apache/seatunnel/issues/10717

Here is a summary of the key design:

1. Signal Hierarchy
- Checkpoint barrier (Ck) remains first-class citizen
- Schema change (Sc) and flush signal (F) are second-class citizens
- F is an engine-synthesized control Record (TimerFlushSignal),
flowing in the same data plane as R / Sc / Ck

2. Configuration Model (two levels)
- env level: sink.flush.interval (milliseconds, default 0 = disabled)
- Connector level: enable_timer_flush (boolean, default false)
- Both must be set for timer flush to activate

3. Opt-in Registration
- Engine exposes TimerFlushService via SinkWriter.Context (default method)
- Connectors call register() during Writer initialization
- Transactional connectors (e.g. JDBC XA, MongoDB transactions)
skip registration and log a WARN

4. Thread Safety
- Timer callback only enqueues F into the intermediate queue
(non-blocking tryNext/offer)
- Actual flush executes on the task consume thread (never on timer thread)
- checkpointLock ensures mutual exclusion with triggerBarrier()
and sendRecordToNext()

5. Backpressure
- Queue full: drop F, retry on next timer tick
- No blocking, no retry on timer thread

6. Backward Compatibility
- New default method on SinkWriter.Context (returns NOOP)
- No changes to existing APIs or config options
- Zero impact on existing jobs with no configuration changes

The full design (including sequence diagrams, code examples,
compatibility matrix, and implementation plan) is in the issue above.

Looking forward to your review and feedback.

Best regards,
zhiwei.niu
在 2026年4月10日 +0800 21:10,M <[email protected]>,写道:
>
> Hi SeaTunnel Community,
>
> According to the latest STIP standards,I'd like to formally propose STIP-23: 
> adding engine-level timer-based flush support for Sink connectors in the Zeta 
> engine.
>
> The full proposal is tracked at: github.com/apache/seatunnel/issues/10717
>
> Looking forward to your feedback and suggestions. If there are no major 
> objections, I will proceed with the detailed design document.
>
>
> Best Regards,
> zhiwei.niu

Reply via email to