hiboyang commented on pull request #31715:
URL: https://github.com/apache/spark/pull/31715#issuecomment-790184675


   > I'd tend to more agree with @attilapiros on the point. It sounds not very 
proper and error-prone to let users control the behavior. I think more ideal 
approach is to let Spark decide to if the shuffle output should be unregistered 
when the executor is going to be removed.
   
   The usage here is user decides to use Remote Shuffle Service (e.g. 
Facebook/Uber's) and sets spark.shuffle.manager with a customized class which 
supports that Remote Shuffle Service. Then they will set 
spark.shuffle.markFileLostOnExecutorLost to avoid marking shuffle file lost 
when executor is lost. In other scenarios, user will not set 
spark.shuffle.markFileLostOnExecutorLost. Would you clarify the "error-prone" 
concern here?
   
   


----------------------------------------------------------------
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:
[email protected]



---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to