slurm schema needs to be developed

[shahzebsiddiqui]

Since we don't have a way to inject sbatch keys at the moment if user specify in run section as follows

job_test:
  description: Hello World example with OpenMP using gnu compiler
  type: script
  env:
    srcdir: $WORK/buildtest-stampede2/tests/apps/openmp/src
    exe: hello_sbatch
  run: |
    #SBATCH -N 1
    #SBATCH -n 1
    #SBATCH -p normal
    #SBATCH -t 00:10
    gcc -fopenmp -o $exe $srcdir/omp_hello.c
    OMP_NUM_THREADS=2
    ./$exe

The generated script is the following

(buildtest-framework) login4.stampede2(1113)$ cat .buildtest/hello/job_test.sh
#!/bin/bash
srcdir=$WORK/buildtest-stampede2/tests/apps/openmp/src
exe=hello_sbatch
#SBATCH -N 1
#SBATCH -n 1
#SBATCH -p normal
#SBATCH -t 00:10
gcc -fopenmp -o $exe $srcdir/omp_hello.c
OMP_NUM_THREADS=2
./$exe

The #SBATCH keys need to be specified via executor and buildtest needs to add them before added at the top even before the environment section. If not we will get an error when running the script. Reordering the sbatch directive on top we will get the job submitted properly.

[vsoch]

This isn't a bug for the script type, it means that a custom type is needed for slurm, one that can take in the SBATCH variables and then write them to file before the rest of the script.

[shahzebsiddiqui]

closing this since it is now implemented in #329