Ticket 17226 - reservation for a generic resource (GRES)?
Summary: reservation for a generic resource (GRES)?
Status: RESOLVED DUPLICATE of ticket 5771
Alias: None
Product: Slurm
Classification: Unclassified
Component: reservations (show other tickets)
Version: 23.02.3
Hardware: Linux Linux
: --- 4 - Minor Issue
Assignee: Director of Support
QA Contact:
URL:
Depends on:
Blocks:
 
Reported: 2023-07-17 16:52 MDT by Jeff Haferman
Modified: 2023-07-18 10:51 MDT (History)
0 users

See Also:
Site: NPS HPC
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: ---


Attachments

Note You need to log in before you can comment on or make changes to this ticket.
Description Jeff Haferman 2023-07-17 16:52:08 MDT
Our GPU nodes typically have 8 GPUs per node. Is it possible to create a reservation for just a subset of those GPUs? I see from the manual that trackable resources ("TRES") can be specified when creating a reservation, but not generic resources ("GRES").

I'd like to do something like:
sudo scontrol create reservation=gpu1 starttime=2023-07-20T06:00:00 endtime=2023-07-31T06:00:00 users=foobar partition=gpu Nodes=node10 gres=gpu=1


This fails because "gres" is not accepted for reservation creation, so I must set aside all 8 GPUs:

sudo scontrol create reservation=gpu1 starttime=2023-07-20T06:00:00 endtime=2023-07-31T06:00:00 users=foobar partition=gpu Nodes=node10 

I can't be the first person that has asked about this? Is there a work-around to reserve just a subset of the GPUs on a node?
Comment 1 Jason Booth 2023-07-18 10:27:00 MDT
Jeff, please see bug#5771comment#2. Quoted comment below. We are actively looking into adding this feature in 23.11.

> Currently, Slurm can’t explicitly create reservations for GRESs. Supported resources 
> that can be reserved include cores, nodes, licenses, burst buffers, and features.

> However, you could possibly work around this limitation by creating reservations based 
> off a feature. To do that, set a feature for nodes that have GPUs with a string like 
> “k80” and issue a command like the following:

>    scontrol create reservation starttime=now nodes=all duration=15 features=k80 user=root 

> Just be aware that “the reservation creation request can [only] identify... *one* 
> feature that every selected node must contain.” 
> See https://slurm.schedmd.com/reservations.html.

*** This ticket has been marked as a duplicate of ticket 5771 ***
Comment 2 Jeff Haferman 2023-07-18 10:51:54 MDT
Thank you.