SLURM sbatch 是否自动跨节点复制用户脚本？

Question

Should SLURM (specifically sbatch) automatically copy the user script (not the job configuration script) to the cluster's compute nodes for execution? Upon executing the sbatch file from my login node, the output file is created on one of my compute nodes, but contains the following:

ERROR: could not open file /home/pi/slurm.jl
Stacktrace:
 [1] include at ./boot.jl:328 [inlined]
 [2] include_relative(::Module, ::String) at ./loading.jl:1105
 [3] include(::Module, ::String) at ./Base.jl:31
 [4] exec_options(::Base.JLOptions) at ./client.jl:287
 [5] _start() at ./client.jl:460

I'm running the batch script with sbatch julia.sbatch .

julia.sbatch:

#!/bin/bash
#SBATCH --nodes=4
#SBATCH --ntasks=4
#SBATCH --time=00:15:00
#SBATCH --output=julia.out
#SBATCH --job-name=julia-job

julia slurm.jl

Or should the script (slurm.jl) be located on shared storage accessible to all of the nodes?

Answer 1

Slurm will not copy files other than the submission script to the compute nodes. From the Quick Start User Guide :

Slurm does not automatically migrate executable or data files to the nodes allocated to a job. Either the files must exists on local disk or in some global file system (eg NFS or Lustre).

On most clusters, the /home directory is an NFS filesystem shared on each login and compute node.

Answer 2

You can use sbcast in your script to copy the file to the node.

#!/bin/bash
#SBATCH --nodes=4
#SBATCH --ntasks=4
#SBATCH --time=00:15:00
#SBATCH --output=julia.out
#SBATCH --job-name=julia-job
sbcast slurm.jl slurm.jl
julia slurm.jl