curcur commented on pull request #16714:
URL: https://github.com/apache/flink/pull/16714#issuecomment-895725445


   @gaoyunhaii Thanks for the detailed description of the "purpose of the 
change".
   
   I am overall fine with this change.
   
   Two questions:
   1. The current logic changes "broadcast restore" to "broadcast rescaling", I 
have a concern about this:
   This will somehow disable the usages of "local recovery", which will 
introduce performance issues.
   
   If it is only for broadcast operator state restore, can we pull from any 
non-empty task state and restore?
   
   2. What would be the problem if a broadcast task with an empty state? Would 
it be possible that the task get more inputs as its upstream task has already 
sent `finish` without changing parallelism?
      


-- 
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: issues-unsubscr...@flink.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


Reply via email to