[GitHub] [kafka] vvcephei commented on pull request #9515: KAFKA-10561: read offsets directly from checkpoint for uninitialized tasks

[GitBox <gi...@apache.org>]

vvcephei commented on pull request #9515:
URL: https://github.com/apache/kafka/pull/9515#issuecomment-717683830


   Thanks for this quick fix, @ableegoldman !
   
   It looks like the bug before was that we would skip to encode these uninitialized tasks since they'd look "apparently stateless", and now we'll just try to read the checkpoint instead. If the checkpoint file is empty, though, it seems like the outcome is the same, though, right? We would not encode anything, just like a stateless task?
   
   Although, as I'm typing this, I'm realizing the bug is probably that when the task _is_ initialized, we'd report some high offsetSum, which the assignor interprets as a low lag, and when it's _not_ initialized, then we report nothing, which the assignor interprets as a high lag (since the assignor will independently verify if it's a stateless task or not). In that case, when we legitimately have no checkpoint file, then it's ok to report nothing, because we legitimately have a high lag. And it won't flip-flop in any case, because once the task gets initialized, its lag will still be about the same.
   
   Did I get that right?


----------------------------------------------------------------
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.

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