[slurm-users] User can bypass the account limit and I don't understand why

Michał Kadlof michal.kadlof at pw.edu.pl
Tue Sep 5 08:35:02 UTC 2023


Hello,

I need your help. I have a user who is able to bypass his account Limit, 
and I can't figure out why.

We have two partitions with lower tier (all hosts) named "long", and 
higher tier (only one host) named "sfglab". Here is it's config:

PartitionName=long Nodes=dgx-[1-4],sr-[1-3] MaxTime=10-0 State=UP 
PriorityTier=10000 QOS=long OverSubscribe=NO DefaultTime=2-00:00:00 
PreemptMode=suspend TRESBillingWeights=CPU=1,Mem=0.062G,GRES/gpu=72.458

PartitionName=sfglab Nodes=dgx-1 MaxTime=10-0 State=UP 
PriorityTier=20000 PreemptMode=off OverSubscribe=NO AllowAccounts=sfglab 
AllowGroups=sfglab TRESBillingWeights=CPU=1,Mem=0.062G,GRES/gpu=72.458 
Hidden=NO

Partition long also has assigned QOS (named "long") with limit: 
cpu=450,gres/gpu=28,mem=6T

And also we have account (also named "sfglab") with limits:

sacctmgr modify account where name=sfglab set 
GrpTRES=cpu=300,gres/gpu=20,mem=5T


As a result user is able to allocate more 32 GPUs...
I don't understand why :(

We use slurm 23.02.2.

-- 
best regards | pozdrawiam serdecznie
*Michał Kadlof*
Head of the high performance computing center 	Kierownik ośrodka 
obliczeniowego HPC
Eden^N cluster administrator 	Administrator klastra obliczeniowego Eden^N
Structural and Functional Genomics Laboratory 	Laboratorium Genomiki 
Strukturalnej i Funkcjonalnej
Faculty of Mathematics and Computer Science 	Wydział Matematyki i Nauk 
Informacyjnych
Warsaw University of Technology 	Politechnika Warszawska
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20230905/ec852075/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4788 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20230905/ec852075/attachment.bin>


More information about the slurm-users mailing list