Ticket 2218

Summary: possible to setup per-association limits for specific qos?
Product: Slurm Reporter: Doug Jacobsen <dmjacobsen>
Component: slurmdbdAssignee: Tim Wickberg <tim>
Status: RESOLVED INFOGIVEN QA Contact:
Severity: 4 - Minor Issue    
Priority: --- CC: da
Version: 15.08.4   
Hardware: Cray XC   
OS: Linux   
Site: NERSC 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 Doug Jacobsen 2015-12-03 05:15:14 MST
Hello,

We have a very special purpose partition "realtime" that I need to gate access to in a rather detailed way.  Only certain users have access to it, each user will potentially have different limits to it (e.g., GrpTRES, MaxJobs, etc).

Is there a way I can add a new association for these users specifically for the a particular qos?

What I'm after to to allow these users to operate with their usual limits in other qos (e.g., normal), but get different limits when using the realtime qos.

nid00837:~ # sacctmgr add user dmj cluster=cori qos=premium account=mpccc
 Nothing new added.
nid00837:~ # sacctmgr show assoc where user=dmj cluster=cori
   Cluster    Account       User  Partition     Share GrpJobs       GrpTRES GrpSubmit     GrpWall   GrpTRESMins MaxJobs       MaxTRES MaxTRESPerNode MaxSubmit     MaxWall   MaxTRESMins                  QOS   Def QOS GrpTRESRunMin
---------- ---------- ---------- ---------- --------- ------- ------------- --------- ----------- ------------- ------- ------------- -------------- --------- ----------- ------------- -------------------- --------- -------------
      cori    fungalp        dmj                    1                                                                       bb/cray=0                                                                  normal
      cori       m342        dmj                    1                                                                       bb/cray=0                                                                  normal
      cori   fnglanot        dmj                    1                                                                       bb/cray=0                                                                  normal
      cori      plant        dmj                    1                                                                       bb/cray=0                                                                  normal
      cori      mpccc        dmj                    1                                                                                                                                                  normal
nid00837:~ #
nid00837:~ # sacctmgr modify user set MaxJobs=1 where user=dmj qos=premium
 Nothing modified
nid00837:~ # sacctmgr show assoc where user=dmj cluster=cori
   Cluster    Account       User  Partition     Share GrpJobs       GrpTRES GrpSubmit     GrpWall   GrpTRESMins MaxJobs       MaxTRES MaxTRESPerNode MaxSubmit     MaxWall   MaxTRESMins                  QOS   Def QOS GrpTRESRunMin
---------- ---------- ---------- ---------- --------- ------- ------------- --------- ----------- ------------- ------- ------------- -------------- --------- ----------- ------------- -------------------- --------- -------------
      cori    fungalp        dmj                    1                                                                       bb/cray=0                                                                  normal
      cori       m342        dmj                    1                                                                       bb/cray=0                                                                  normal
      cori   fnglanot        dmj                    1                                                                       bb/cray=0                                                                  normal
      cori      plant        dmj                    1                                                                       bb/cray=0                                                                  normal
      cori      mpccc        dmj                    1                                                                                                                                                  normal
nid00837:~ #



Thanks,
Doug
Comment 1 Tim Wickberg 2015-12-03 08:50:56 MST
So you'd want to set limits against (User, QOS) pairs so that the specific limits vary by (user, qos), but isn't just defined in the QOS?

Slurm can't do that currently. I think you'd have to just define additional QOS's with the respective limits set in them, and allow users access to those particular QOS's.

Obviously there's a bit more overhead in establishing those additional QOS's, but is there any reason that wouldn't work for you?
Comment 2 Doug Jacobsen 2015-12-03 09:04:36 MST
that would work, however I still need a way to authorize a user to access a particular qos -- is that possible?
Comment 3 Danny Auble 2015-12-03 09:52:35 MST
Doug, could you use partition associations instead for this?  It should be rather straight forward.  You can also limit the QOS an association has access to by setting it in the association with sacctmgr, just QOS+=realtime, or whatever your qos is.
Comment 4 Tim Wickberg 2015-12-10 05:55:29 MST
Doug - Haven't heard back on this one in a week... do partition associations work for what you're after?
Comment 5 Doug Jacobsen 2015-12-10 06:43:59 MST
Sorry, I've been out of town.  I ended up setting enforcement to include
qos and then automated managing which qos each user is allowed to access.
So that worked out.

----
Doug Jacobsen, Ph.D.
NERSC Computer Systems Engineer
National Energy Research Scientific Computing Center <http://www.nersc.gov>
dmjacobsen@lbl.gov

------------- __o
---------- _ '\<,_
----------(_)/  (_)__________________________


On Thu, Dec 10, 2015 at 11:55 AM, <bugs@schedmd.com> wrote:

> *Comment # 4 <http://bugs.schedmd.com/show_bug.cgi?id=2218#c4> on bug 2218
> <http://bugs.schedmd.com/show_bug.cgi?id=2218> from Tim Wickberg
> <tim@schedmd.com> *
>
> Doug - Haven't heard back on this one in a week... do partition associations
> work for what you're after?
>
> ------------------------------
> You are receiving this mail because:
>
>    - You reported the bug.
>
>
Comment 6 Tim Wickberg 2015-12-10 07:56:35 MST
No problem, just wanted to make sure you weren't stalled on our response. Closing now.

- Tim