[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[HTCondor-users] sanity check on preemption

So I wanted to make sure I understood how preemption works with partitionable slots.


Basically, because HTCondor tries to re-use slots, the first step to preemption includes checking in a queued job matches a slot. With partitionable slots, a queued job triggered the creation of a slot from a partionable slot. As such a queued job that wishes to preempt a job running on a partitioned slot needs to first match on that slot before the job running there can be considered for evicted. A job would not preempt by evicting a job, destroying partitioned slot, and creating a new partitioned slot of a different size to match the new job.


Am I understanding that correctly? Or am I way off base?


For some context, I enabled preemption on a cluster with partitionable slots with a PREEMPTION_REQUIREMENTS that looks if a job was running longer than 48 hours and priority is 20% higher. However, there are jobs that keep running after the 48 hour mark even when there are queued jobs waiting to run.


Thanks all!


Gianni Pezzarossi

Computational System Analyst

User Services:Research

Engineering IT Shared Services

University of Illinois @ Urbana-Champaign