I have found an issue though. I have a user that has the largest priority on Niagara but his job doesn't run. It is stuck on the queue waiting for resources.
I simplified the script and tested on our TDS and I could reproduce the problem.
For example the following batch script:

>#!/bin/bash -l
>#SBATCH --ntasks=1
>#SBATCH --time=00-11:30
>#SBATCH --ntasks=1
>#SBATCH --nodes=1
>#SBATCH --cpus-per-task=1
>#SBATCH --mem-per-cpu=14000
>#SBATCH -J memory_per_cpu_test
>#SBATCH -A scinet
>#SBATCH --mail-type=ALL
>#SBATCH --output=slurm-%j.out
> 
>echo "#########################################"
>echo " SLURM submission batch script stdout "
>echo "#########################################"
> 
>source ~/scripts/sbatch_job_envs.sh 
> 
>env > env_${SLURM_JOB_ID}
> 
>srun -l hostname
>sleep 600

is very similar to the user's one. However it gets stuck on the queue:

>squeue
>   JOBID PARTITION       NAME                 USER              ACCOUNT ST TIME_LIMIT       TIME  TIME_LEFT           START_TIME   CPUS   PRIORITY  NODES NODELIST(REASON)
>    5008   compute memory_per              bmundim               scinet PD   11:30:00       0:00   11:30:00                  N/A      1      22739      1 (Resources)

Even thought TDS cluster is free of jobs. If I comment out the following
line:

>#SBATCH --mem-per-cpu=14000

the job runs normally. Do you know why? I will attach the slurm.conf and cgroup.conf in a minute.

Thanks,
Bruno.