rdblue edited a comment on issue #23401: [SPARK-26513][Core] : Trigger GC on 
executor node idle
URL: https://github.com/apache/spark/pull/23401#issuecomment-450701012
 
 
   > Dynamic scale down is often done fairly conservatively when combined with 
cached blocks
   
   I agree. We (Netflix) actually don't internally recommend caching when using 
dynamic allocation for most ETL workloads, and we recommend careful settings 
with ML workloads. So when I'm talking about dynamic allocation, I mean a case 
where executors time out fairly quickly as a stage enters its long tail, which 
is when this would primarily take effect. I don't think it would hurt, but I 
don't think it would help much either.
   
   I also don't mean to say that I think this is a bad idea. It may be worth 
putting in as an option for certain workloads. I'm just skeptical that it is a 
good practice for a default and I didn't see much validation across workloads 
in the paper.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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