[slurm-users] Slurm fair share priority not being applied

Loris Bennett loris.bennett at fu-berlin.de
Fri Dec 8 00:07:39 MST 2017


Lachlan Musicman <datakid at gmail.com> writes:

> On 1 December 2017 at 20:48, Bruno Santos <bacmsantos at gmail.com> wrote:
>
>>  Loris, I think you hit the nail on the head. 
>>
>>  Running sshare -l only shows the root user:
>>  Account User RawShares NormShares RawUsage NormUsage EffectvUsage FairShare LevelFS GrpTRESMins TRESRunMins 
>>  -------------------- ---------- ---------- ----------- ----------- ----------- ------------- ---------- ---------- ------------------------------ ------------------------------ 
>>  root 0.000000 0 1.000000 cpu=0,mem=0,energy=0,node=0 
>>
>>  So I am guessing the user information is indeed not being linked to
>>  the slurmdbd. What do I need to do to set up this?  To be honest the
>>  documentation is very sparse in details of how to set up the
>>  slurmdb.
>
> Did that end up being the problem?
>
> I've had trouble with QoS in the past as well, and now that I use
> sshare -l (actually, I used sshare
> --format=Account,User,FairShare,RawShares,NormShares), I note that
> none of my Accounts have any FairShare. For whatever reason, I
> presumed that NormShare would be sufficient?
>
> Do we need to explicitly list the FS?
>
> Cheers
> L.

Does "none of my Accounts have any FairShare" mean no value of a value
of zero?  If your account don't have any value for FairShare that would
suggest that fairshare isn't set up properly.  If the value is zero,
then all the shares have been used up.

While we never had everyone ending up with zero FairShare, we did
initially have all the active users with zero.  To get around this we
had to increase PriorityWeightFairshare by a couple of orders of
magnitude to be able to distinguish active users from really active
users.

You might want to look at the output of 'sshare -la' and 'sprio -l'.

Cheers,

Loris

-- 
Dr. Loris Bennett (Mr.)
ZEDAT, Freie Universität Berlin         Email loris.bennett at fu-berlin.de



More information about the slurm-users mailing list