Hi, William Dear <william.d...@i3-corps.com> writes:
> It looks like Brian's suggestion of using SACCT will be the fast answer in > the short term so I'll just have to write my own script to aggregate the > output. I was hoping for a canned solution such as XDMoD but haven't found > one that quite > fits our needs. If there's a list of recommended supporting applications for > SLURM I would appreciate that. > > One example of how the canned reporting doesn't meet our needs is that my > users self limit their arrays such as "--array=1-12000%100". Technically, > the initial job isn't waiting on anything but itself since it only runs 100 > at a time but > all the pending array jobs still show up as waiting. If the partition > resources are too low and the job is running less than 100 then it actually > is waiting on another job. The challenge will be determining when a job is > self limiting vs > waiting on a different job. What is the use-case for having users need to self-limit? We just rely on the cap for the maximum number of jobs in an array and on fairshare to do the rest. Cheers, Loris > Thanks, > > William Dear > > ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ > From: slurm-users <slurm-users-boun...@lists.schedmd.com> on behalf of Loris > Bennett <loris.benn...@fu-berlin.de> > Sent: Monday, May 16, 2022 9:04 AM > To: Slurm User Community List <slurm-users@lists.schedmd.com> > Subject: Re: [slurm-users] Performance tracking of array tasks > > CAUTION: This email originated from outside of the organization. Do not click > links or open attachments unless you recognize the sender and know the > content is safe. > > Hi William, > > William Dear <william.d...@i3-corps.com> writes: > >> Could anyone please recommend methods of tracking the performance of >> individual tasks in a task array job? I have installed XDMoD but it is >> focused solely on the Job level with no information about >> tasks. >> >> My users almost exclusively use task arrays to run embarrassingly parallel >> jobs. After the job is complete I would like to see run time and peak RAM >> usage per task so that we can correctly size the >> reservations for future jobs. It would also be very helpful to break this >> down by node so that I can identify poorly performing nodes. >> >> William Dear > > I'm not sure what you mean by a 'task array job'. A job can have > multiple tasks within it - I don't think you will be able to get data on > such individual tasks very easily. However, a job array is just a sort > of convenient wrapper around a bunch of jobs. Each element of a job > array still has its own job ID, so you can extract job data the same way > you do for a non-array job. > > Cheers, > > Loris > > -- > Dr. Loris Bennett (Herr/Mr) > ZEDAT, Freie Universität Berlin Email loris.benn...@fu-berlin.de > > _____________________________________ > Confidentiality Notice - The information contained in this e-mail and any > attachments to it may be legally privileged and include confidential > information. If you are not the intended recipient, be aware that any > disclosure, > distribution or copying of this e-mail or its attachments is prohibited. If > you have received this e-mail in error, please notify the sender immediately > of that fact by return e-mail and permanently delete the e-mail and any > attachments > to it.