squito commented on issue #24462: [SPARK-26268][CORE] Do not resubmit tasks 
when executors are lost
URL: https://github.com/apache/spark/pull/24462#issuecomment-516149439
 
 
   why do you want to store the data files on hdfs, but the index files on the 
executors?  This seems to have the worst of both worlds -- the (bad) resiliency 
of local storage, and the (bad) performance of remote reads.  Or is the index 
file backed up somewhere as well?
   
   I definitely understand the general problem with knowing what to do about 
shuffle data when an executor is lost.  So I understand why you want to do 
something *like* what this PR does.  But it probably makes more sense to 
address this as part of the other shuffle api changes, if possible, not as 
another config.
   
   You might not be able to do everything you want -- in particular, the new 
api does not support multiple locations for shuffle data.  We decided that was 
out-of-scope, for now (but maybe a future enhancement).  Is that what you're 
looking for -- one copy on the executors local disk, and another copy on hdfs?

----------------------------------------------------------------
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]


With regards,
Apache Git Services

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

Reply via email to