Is it possible to use different cpus for different job steps with slurm?

Question

I deployed a slurm (version 19.05.5) control and compute node locally. I want to be able to run job steps with srun inside an sbatch script using one cpu per job step and in parallel.

So I wrote the following script date.sh:

#!/bin/sh

sleep 10
date +%s

And the following sbatch script:

#!/bin/sh
#SBATCH --job-name=TestJob
#SBATCH --time=00:05:00
#SBATCH --cpus-per-task=1
#SBATCH --ntasks=3

srun -l --exclusive --ntasks=1 date.sh &
srun -l --exclusive --ntasks=1 date.sh &
srun -l --exclusive --ntasks=1 date.sh &
srun -l --exclusive --ntasks=1 date.sh &
srun -l --exclusive --ntasks=1 date.sh &
wait

The behaviour I expected was that the first three job steps run in parallel, in three different tasks, using three different cpus. But the result is the following:

0: 1652629977
0: 1652629977
0: 1652629977
srun: Job 42 step creation temporarily disabled, retrying
srun: Job 42 step creation temporarily disabled, retrying
srun: Step created for job 42
srun: Step created for job 42
0: 1652629987
0: 1652629987

So all the runs were done as a same task, which means one cpu was used. How do I force the runs to be done in different tasks? Other than spreading the srun commands over multiple scripts and calling sbatch on each one with the sbatch option ntasks=1.

More info:

• Removing --exclusive results in all the 5 steps being run simultaneously in 1 task.
• Removing --exclusive and --ntasks=1 results in running each step 3 times in 3 different tasks.
• Removing --ntasks=1 results in running each step 3 times in 3 different tasks. But it prints out warnings about not being able to create job steps, then notifications are printed when the job steps are created.