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

Re: [HTCondor-users] [HTCondor-Users] condor_userprio calculation related query

On 10/8/19 2:22 AM, Vikrant Aggarwal wrote:
Hello HTCondor Experts,

I have a query regarding condor_userprio output. We are using this output for billing purpose. While running some tests recently we found that condor_userprio is calculating priorities based on allocated resources not used resources. Let's say if one job is starting multiple forked processes then cgroup will opportunistlly allows job to use more CPUs but condor_userprio is calculating the priorityÂonly based on requested_cpus. I confirm this by running test also is this expected behavior? if yes, then what's the best way to charge customer based on actual used resources.

Hello Virkant:

This is the expected behaviour. The accounting in HTCondor is based on the provisioned resources, not the used resource, because provisioning a resource for someone prevents it from being used by anyone else. Otherwise, someone could submit jobs that request lots of cores, but merely sleep, using no cpu resources, and never have their priority decay.

If you want to measure and report the actual used resource, you'll need to look at the job history or event logs on each submit machine.