Savonitar commented on PR #295:
URL: 
https://github.com/apache/flink-connector-kafka/pull/295#issuecomment-5530546385

   Hi @bowenli86 . 
   
   > Could we also clarify the reader-only restart case: after a reader is 
unregistered but before its restored-split report arrives, does Flink’s runtime 
ordering prevent a checkpoint from completing? The new state covers reports 
already received. A test or documented runtime guarantee for the 
pre-registration window would help establish the boundary of this fix.
   
   Good question, FLIP-537 explicitly guarantees that the SourceCoordinator 
receives restored-split reports before any subsequent checkpoint, and I 
verified the path in Flink 2.2.1: SourceOperator submits the report during 
restore before the task reports RUNNING, checkpoint planning rejects 
non-running tasks, and registration and snapshot actions are serialized by the 
coordinator.
   
   Since a connector unit test would only "model" this runtime ordering, I 
documented that in 
https://github.com/apache/flink-connector-kafka/pull/295/changes/bd222e5299310ec1b37ea5db877ae04f80762609
  a proper ordering test belongs in Flink’s runtime layer.
   
   Please tell me wdyt?


-- 
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