Reply
Explorer
Posts: 12
Registered: ‎06-27-2017

Spark Dynamic Resource Allocation Executor Timeout

spark.dynamicAllocation.executorIdleTimeout = 60

 

Can this be lowered? Is there any benefit/harm?

 

My question is regarding preemption. Are executors of a preempted container "idle". or is this idle step skipped during preemption?

 

Highlighted
Posts: 57
Topics: 0
Kudos: 8
Solutions: 5
Registered: ‎05-15-2018

Re: Spark Dynamic Resource Allocation Executor Timeout

Hello @regeamor

 

Thank you for posting the query with us.

 

Basically when you enable Dynamic allocation it gracefully remove the Idle containers which were idle for (60s as default) But when you lower the value will remove the executors frequently depending upon the executor's usage (just like the tasks getting allocated to those executors and the amount of tasks required)

 

https://spark.apache.org/docs/latest/job-scheduling.html#graceful-decommission-of-executors

 

Also, the above said behaviour is upto Spark (and yarn container pre-emption wont be aware of it)

Thanks,
Satz
Announcements