tgravescs edited a comment on issue #26696: [WIP][SPARK-18886][CORE] Make 
locality wait time be the time since a TSM's available slots were fully utilized
URL: https://github.com/apache/spark/pull/26696#issuecomment-571210997
 
 
   thanks for the update and proposal. I think it would be useful to add in a 
high level description of what exactly the approach is trying to do.  You have 
how its trying to do it and an example but I think what its trying to do is as 
useful.  
   
   Please correct me if I'm wrong, but I think essentially at a high level you 
are proposing to only reset the timer whenever all the currently free resources 
are scheduled on.  So if you start out with 4 executors, you don't reset the 
timer until all of them have been scheduled. At that point you reset the timer. 
Which means that when a running task finishes and you go to start a new one or 
a new executor comes in, it has to wait the timeout.
   And for the fair scheduler case, the way you are passing the flags in 
handles it resetting the timer if it has used its fair share? 
   
   Also just to clarifying a few things:
   
   >  offer 1 resource, no reject - timer is reset
   
   so the assumption here is that we are out of all resources because we either 
had 1 task finish or we added a new executor and if nothing was rejected we 
scheduled that fully and we were fully scheduled before this offer?     This is 
why you have the last line :  offer 1 resource, no reject - no timer reset 
because previous offer was rejected -> since previous offer was rejected there 
are still free resources so we don't reset.
   
   
   

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