tgravescs commented on issue #26614: [SPARK-29976][CORE] New conf for single 
task stage speculation
URL: https://github.com/apache/spark/pull/26614#issuecomment-558773705
 
 
   Yes you still need the config for the timeout, you just don't need a second 
one when to apply that config. ie when you have <= second config task number 
then use the config, otherwise use the normal speculation logic.
   
   I was originally thinking the total number of tasks <= number of slots on 1 
executor, then apply the timeout config, seemed the most straightforward and 
obvious to the user.  I'm fine with either way though as long as it can be 
explained to user.  I think using delta does complicate things again as it uses 
the new algorithm sometimes and then the original algorithm at other times.  My 
initial thought is to keep it simple in initial implementation, they can always 
turn the spark.speculation.quantile down when you have a larger number of 
tasks, but lots of corner cases again.  The thing with 1 task is that the 
current settings will never work for it because you need at least 1 to compare 
against.
   
   Note you will ask for a new executor if you speculate and the executors are 
all used.  It might not be on a different node though.

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