[GitHub] [kafka] ramesh-muthusamy commented on pull request #9319: KAFKA-10413 Allow even distribution of lost/new tasks when more than one worker j…

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

ramesh-muthusamy commented on pull request #9319:
URL: https://github.com/apache/kafka/pull/9319#issuecomment-727573081


   > After writing another test, I confirmed my impression after first reading the code. So, this fix would help if all the lost workers joined after leaving the group at about the same time. But if a completely new worker joins in addition to a returning worker, then the fact that we distribute the tasks between the new workers, will lead to an imbalance compared to any existing workers. For example,
   > 
   > With 2 workers with each having 8 tasks each, if 1 worker leaves and rejoins within the scheduled delay but also another new worker joins, then we'll end up with the worker that never left having 8 tasks and the other two workers having 4 tasks each.
   
   @kkonstantine  I think we are already capturing this scenario, since we are taking all the existing workers while trying to do the revocation even the worker which had 8 tasks in this case and was running fine would participate in the revocation process. This would result in the existing worker with 3 tasks and the two new workers with 3 and 2 tasks respectively. 


----------------------------------------------------------------
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