[slurm-users] Conflicting --nodes and --nodelist

Diego Zuccato diego.zuccato at unibo.it
Thu Jun 3 06:28:44 UTC 2021


Il 01/06/2021 15:37, Brian Andrus ha scritto:

Tks Brian and Marcus.
Seems I misinterpreted the docs.
We need to target one of two identical nodes for performance comparisons 
between runs (students have to learn how to scale their jobs).
I'll have to use a feature, since it seems requesting a partition is 
overridden by environment var $SBATCH_PARTITION (that I'm using to 
specify a default partition *set*, with each partition containing only 
omogeneus nodes, so that by default jos only get omogeneus nodes from a 
single partition).

BYtE,
  Diego

> That is expected behavior as Marcus pointed out.
> 
> I suspect you may be doing something like targeting one of two systems 
> that each have a node-locked license for some software, or have some 
> different specs.
> 
> In this case, you may want to use the FEATURES option when defining 
> those nodes and then request that feature when submitting your job.
> 
> Brian Andrus
> 
> 
> On 6/1/2021 4:15 AM, Diego Zuccato wrote:
>> Hello all.
>>
>> I just found that if an user tries to specify a nodelist (say 
>> including 2 nodes) and --nodes=1, the job gets rejected with
>> sbatch: error: invalid number of nodes (-N 2-1)
>> The expected behaviour is that slurm schedules the job on the first 
>> node available from the list.
>> I've found conflicting info about the issue. Is it version-dependant?
>> If so, we're currently using 18.08.5-2 (from Debian stable). Should we 
>> expect changes when Debian will ship a newer version? Is it possible 
>> to have the expected behaviour?
>>
>> Tks.
>>
> 


-- 
Diego Zuccato
DIFA - Dip. di Fisica e Astronomia
Servizi Informatici
Alma Mater Studiorum - Università di Bologna
V.le Berti-Pichat 6/2 - 40127 Bologna - Italy
tel.: +39 051 20 95786



More information about the slurm-users mailing list