Ticket 6298

Summary: Sacct AllocTRES returns billing since 17.11.7 upgrade
Product: Slurm Reporter: GSK-ONYX-SLURM <slurm-support>
Component: AccountingAssignee: Albert Gil <albert.gil>
Status: RESOLVED INFOGIVEN QA Contact:
Severity: 4 - Minor Issue    
Priority: ---    
Version: 17.11.7   
Hardware: Linux   
OS: Linux   
Site: GSK Alineos Sites: ---
Atos/Eviden Sites: --- Confidential Site: ---
Coreweave sites: --- Cray Sites: ---
DS9 clusters: --- HPCnow Sites: ---
HPE Sites: --- IBM Sites: ---
NOAA SIte: --- OCF Sites: ---
Recursion Pharma Sites: --- SFW Sites: ---
SNIC sites: --- Linux Distro: ---
Machine Name: CLE Version:
Version Fixed: Target Release: ---
DevPrio: --- Emory-Cloud Sites: ---

Description GSK-ONYX-SLURM 2018-12-27 11:02:54 MST
Hi.
We recently upgraded from 17.02.7 to 17.11.7.  Since the upgrade

    sacct -o alloctres

is returning "billing=n" where n is exactly the same as "cpu=n", eg

uk1sxlx00090 (The Lion): sacct -j 3888251 -o jobid,start,end,elapsed,alloctres%32,allocgres
       JobID               Start                 End    Elapsed                        AllocTRES    AllocGRES
------------ ------------------- ------------------- ---------- -------------------------------- ------------
3888251      2018-11-10T16:25:35 2018-11-13T09:04:10 2-16:38:35  cpu=8,mem=254G,node=1,billing=8
3888251.bat+ 2018-11-10T16:25:35 2018-11-13T09:04:11 2-16:38:36            cpu=8,mem=254G,node=1
uk1sxlx00090 (The Lion): sacct -j 5068765 -o jobid,start,end,elapsed,alloctres%32,allocgres
       JobID               Start                 End    Elapsed                        AllocTRES    AllocGRES
------------ ------------------- ------------------- ---------- -------------------------------- ------------
5068765      2018-11-10T16:08:12 2018-11-12T20:17:58 2-04:09:46   cpu=1,mem=64G,node=1,billing=1        gpu:2
5068765.bat+ 2018-11-10T16:08:12 2018-11-12T20:18:30 2-04:10:18             cpu=1,mem=64G,node=1        gpu:2
uk1sxlx00090 (The Lion):

1. Why the change in "AllocTRES" output?
2. Is this configurable?  Can I get back to the 17.02.7 style output, ie no "billing" in AllocTres ?

Thanks.
Mark.
Comment 5 Albert Gil 2019-01-02 11:44:08 MST
Hi Mark,

The "billing" TRES was added in the 17.11 version:
https://github.com/SchedMD/slurm/blob/slurm-17.11/NEWS#L682

As a TRES, "billing" is accounted and can be used as a limit.
In fact it is part of the default (always there) TRESes, like CPU, Energy, Memory and Node.
So, even if you don't add it in the AccountingStorageTRES, it will be tracked/accounted anyway.
https://slurm.schedmd.com/tres.html
https://slurm.schedmd.com/archive/slurm-17.11.12/tres.html

Why is always there?
Because it's an special TRES that allows the system (and the users) to have a better estimation of the usage/share of the resources than the other TRES because it allows to aggregate with weights all (or any combination) of them in a single value/TRES to represent it.
It is calculated using the TRESBillingWeights (defaults to TRESBillingWeights=CPU=1.0) adding a weight for each desired AccountingStorageTRES.
So, it's not an "output style" but a key feature/system used for example for the fairshare algorithm.
In fact, in newer version you will also see some newer default TRESes like fs/disk, vmen and pages.

Anyway, although "billing" will be always accounted, maybe we could try to add some extra output format options to sacct to allow you to specify what do you want to be shown by sacct. Do you think that it will really help you?


Regards,
Albert
Comment 7 GSK-ONYX-SLURM 2019-01-07 05:49:54 MST
Hi Albert.
Thanks for the explanation.

Whats caught us out is that because theres now an extra comma separated value in the AllocTRES output then feeding that into output thats used as excel CSV caused following data to be misinterpreted.

In the short term I've worked around this, but it might be good if there was a way of controlling the output format of AllocTRES, especially as it could change again in the future.

So would be good if this could be added to a future enhancement wish list.

Please go ahead and close this bug report.

Thanks.
Mark.
Comment 8 Albert Gil 2019-01-07 08:09:54 MST
Thanks Mark.

Closing this bug with INFOGIVEN.
An enhancement is also requested to improve the format options of the AllocTRES for the sacct command.

Albert