slurm-script-generator 0.3.0

Generate slurm scripts.

slurm-script-generator

pip install slurm-script-generator

Generate scripts

Generate a slurm script to slurm_script.sh with

generate-slurm-script --nodes 1 --ntasks-per-node 16

#!/bin/bash
########################################################
#            This script was generated using           #
#             slurm-script-generator v0.3.0            #
# https://github.com/max-models/slurm-script-generator #
#      `pip install slurm-script-generator==0.3.0`     #
########################################################

##########################################
#SBATCH --nodes=1                        # number of nodes on which to run
#SBATCH --ntasks-per-node=16             # number of tasks to invoke on each node
##########################################

To save the script to file my_script.sh use --output:

generate-slurm-script --nodes 1 --ntasks-per-node 16 --output my_script.sh

Remove the header comment with --no-header:

generate-slurm-script --nodes 1 --ntasks-per-node 16 --no-header

#!/bin/bash
##########################################
#SBATCH --nodes=1                        # number of nodes on which to run
#SBATCH --ntasks-per-node=16             # number of tasks to invoke on each node
##########################################

You can also generate scripts in Python programmatically:

from slurm_script_generator.slurm_script import SlurmScript

slurm_script = SlurmScript(
    nodes=2,
    ntasks_per_core=16,
    custom_commands=[
        "# Run simulation",
        "srun ./bin > run.out",
    ],
)

print(slurm_script)

#!/bin/bash
########################################################
#            This script was generated using           #
#             slurm-script-generator v0.3.0            #
# https://github.com/max-models/slurm-script-generator #
#      `pip install slurm-script-generator==0.3.0`     #
########################################################

##########################################
#SBATCH --nodes=2                        # number of nodes on which to run
#SBATCH --ntasks-per-core=16             # number of tasks to invoke on each core
##########################################
# Run simulation
srun ./bin > run.out

You can also generate a string representation of the script with generate_script:

script = slurm_script.generate_script()

To export the settings to a json file you can use --export-json:

generate-slurm-script --nodes 2 --export-json setup.json --no-header

#!/bin/bash
##########################################
#SBATCH --nodes=2                        # number of nodes on which to run
##########################################

This json file can used as a basis for creating new scripts

generate-slurm-script --input setup.json --ntasks-per-node 16 --no-header

#!/bin/bash
##########################################
#SBATCH --nodes=2                        # number of nodes on which to run
#SBATCH --ntasks-per-node=16             # number of tasks to invoke on each node
##########################################

Add modules

Add modules with

generate-slurm-script --input setup.json --ntasks-per-node 16 --modules gcc/13 openmpi/5.0 --no-header

#!/bin/bash
##########################################
#SBATCH --nodes=2                        # number of nodes on which to run
#SBATCH --ntasks-per-node=16             # number of tasks to invoke on each node
##########################################
module purge                             # Purge modules
module load gcc/13 openmpi/5.0           # modules
module list                              # List loaded modules

Other

All optional arguments can be shown with

generate-slurm-script -h

usage: generate-slurm-script [-h] [-A NAME] [-b TIME] [--bell] [--bb SPEC]
                             [--bbf FILE_NAME] [-c NCPUS] [--comment NAME]
                             [--container PATH] [--container-id ID]
                             [--cpu-freq MIN[-MAX[:GOV]]] [--delay-boot MINS]
                             [-d TYPE:JOBID[:TIME]] [--deadline TIME]
                             [-D PATH] [--get-user-env] [--gres LIST]
                             [--gres-flags OPTS] [-H] [-I [SECS]] [-J NAME]
                             [-k] [-K [SIGNAL]] [-L NAMES] [-M NAMES]
                             [-m TYPE] [--mail-type TYPE] [--mail-user USER]
                             [--mcs-label MCS] [-n N] [--nice VALUE]
                             [-N NODES] [--ntasks-per-node N]
                             [--oom-kill-step [0|1]] [-O] [--power FLAGS]
                             [--priority VALUE] [--profile VALUE]
                             [-p PARTITION] [-q QOS] [-Q] [--reboot] [-s]
                             [--signal [R:]NUM[@TIME]] [--spread-job]
                             [--stderr STDERR] [--stdout STDOUT]
                             [--switches MAX_SWITCHES[@MAX_TIME]] [-S CORES]
                             [--thread-spec THREADS] [-t MINUTES]
                             [--time-min MINUTES] [--tres-bind ...]
                             [--tres-per-task LIST] [--use-min-nodes]
                             [--wckey WCKEY] [--cluster-constraint LIST]
                             [--contiguous] [-C LIST] [-F FILENAME] [--mem MB]
                             [--mincpus N] [--reservation NAME] [--tmp MB]
                             [-w HOST [HOST ...]] [-x HOST [HOST ...]]
                             [--exclusive-user] [--exclusive-mcs]
                             [--mem-per-cpu MB] [--resv-ports]
                             [--sockets-per-node S] [--cores-per-socket C]
                             [--threads-per-core T] [-B S[:C[:T]]]
                             [--ntasks-per-core N] [--ntasks-per-socket N]
                             [--hint HINT] [--mem-bind BIND]
                             [--cpus-per-gpu N] [-G N] [--gpu-bind ...]
                             [--gpu-freq ...] [--gpus-per-node N]
                             [--gpus-per-socket N] [--gpus-per-task N]
                             [--mem-per-gpu MEM_PER_GPU]
                             [--disable-stdout-job-summary] [--nvmps]
                             [--line-length LINE_LENGTH]
                             [--modules MODULES [MODULES ...]]
                             [--input INPUT_PATH] [--output OUTPUT_PATH]
                             [--export-json JSON_PATH]
                             [--custom-command COMMAND]
                             [--custom-commands COMMAND [COMMAND ...]]
                             [--inline-script COMMAND]
                             [--inline-scripts COMMAND [COMMAND ...]]
                             [--no-header]

Slurm job submission options

options:
  -h, --help            show this help message and exit
  -A, --account NAME    charge job to specified account (default: None)
  -b, --begin TIME      defer job until HH:MM MM/DD/YY (default: None)
  --bell                ring the terminal bell when the job is allocated
                        (default: None)
  --bb SPEC             burst buffer specifications (default: None)
  --bbf FILE_NAME       burst buffer specification file (default: None)
  -c, --cpus-per-task NCPUS
                        number of cpus required per task (default: None)
  --comment NAME        arbitrary comment (default: None)
  --container PATH      Path to OCI container bundle (default: None)
  --container-id ID     OCI container ID (default: None)
  --cpu-freq MIN[-MAX[:GOV]]
                        requested cpu frequency (and governor) (default: None)
  --delay-boot MINS     delay boot for desired node features (default: None)
  -d, --dependency TYPE:JOBID[:TIME]
                        defer job until condition on jobid is satisfied
                        (default: None)
  --deadline TIME       remove the job if no ending possible before this
                        deadline (default: None)
  -D, --chdir PATH      change working directory (default: None)
  --get-user-env        used by Moab. See srun man page (default: None)
  --gres LIST           required generic resources (default: None)
  --gres-flags OPTS     flags related to GRES management (default: None)
  -H, --hold            submit job in held state (default: None)
  -I, --immediate [SECS]
                        exit if resources not available in "secs" (default:
                        None)
  -J, --job-name NAME   name of job (default: None)
  -k, --no-kill         do not kill job on node failure (default: None)
  -K, --kill-command [SIGNAL]
                        signal to send terminating job (default: None)
  -L, --licenses NAMES  required license, comma separated (default: None)
  -M, --clusters NAMES  Comma separated list of clusters to issue commands to
                        (default: None)
  -m, --distribution TYPE
                        distribution method for processes to nodes (default:
                        None)
  --mail-type TYPE      notify on state change (default: None)
  --mail-user USER      who to send email notification for job state changes
                        (default: None)
  --mcs-label MCS       mcs label if mcs plugin mcs/group is used (default:
                        None)
  -n, --ntasks N        number of processors required (default: None)
  --nice VALUE          decrease scheduling priority by value (default: None)
  -N, --nodes NODES     number of nodes on which to run (default: None)
  --ntasks-per-node N   number of tasks to invoke on each node (default: None)
  --oom-kill-step [0|1]
                        set the OOMKillStep behaviour (default: None)
  -O, --overcommit      overcommit resources (default: None)
  --power FLAGS         power management options (default: None)
  --priority VALUE      set the priority of the job (default: None)
  --profile VALUE       enable acct_gather_profile for detailed data (default:
                        None)
  -p, --partition PARTITION
                        partition requested (default: None)
  -q, --qos QOS         quality of service (default: None)
  -Q, --quiet           quiet mode (suppress informational messages) (default:
                        None)
  --reboot              reboot compute nodes before starting job (default:
                        None)
  -s, --oversubscribe   oversubscribe resources with other jobs (default:
                        None)
  --signal [R:]NUM[@TIME]
                        send signal when time limit within time seconds
                        (default: None)
  --spread-job          spread job across as many nodes as possible (default:
                        None)
  --stderr, -e STDERR   File to redirect stderr (%x=jobname, %j=jobid)
                        (default: None)
  --stdout, -o STDOUT   File to redirect stdout (%x=jobname, %j=jobid)
                        (default: None)
  --switches MAX_SWITCHES[@MAX_TIME]
                        optimum switches and max time to wait for optimum
                        (default: None)
  -S, --core-spec CORES
                        count of reserved cores (default: None)
  --thread-spec THREADS
                        count of reserved threads (default: None)
  -t, --time MINUTES    time limit (default: None)
  --time-min MINUTES    minimum time limit (if distinct) (default: None)
  --tres-bind ...       task to tres binding options (default: None)
  --tres-per-task LIST  list of tres required per task (default: None)
  --use-min-nodes       if a range of node counts is given, prefer the smaller
                        count (default: None)
  --wckey WCKEY         wckey to run job under (default: None)
  --cluster-constraint LIST
                        specify a list of cluster constraints (default: None)
  --contiguous          demand a contiguous range of nodes (default: None)
  -C, --constraint LIST
                        specify a list of constraints (default: None)
  -F, --nodefile FILENAME
                        request a specific list of hosts (default: None)
  --mem MB              minimum amount of real memory (default: None)
  --mincpus N           minimum number of logical processors per node
                        (default: None)
  --reservation NAME    allocate resources from named reservation (default:
                        None)
  --tmp MB              minimum amount of temporary disk (default: None)
  -w, --nodelist HOST [HOST ...]
                        request a specific list of hosts (default: None)
  -x, --exclude HOST [HOST ...]
                        exclude a specific list of hosts (default: None)
  --exclusive-user      allocate nodes in exclusive mode for cpu consumable
                        resource (default: None)
  --exclusive-mcs       allocate nodes in exclusive mode when mcs plugin is
                        enabled (default: None)
  --mem-per-cpu MB      maximum amount of real memory per allocated cpu
                        (default: None)
  --resv-ports          reserve communication ports (default: None)
  --sockets-per-node S  number of sockets per node to allocate (default: None)
  --cores-per-socket C  number of cores per socket to allocate (default: None)
  --threads-per-core T  number of threads per core to allocate (default: None)
  -B, --extra-node-info S[:C[:T]]
                        combine request of sockets, cores and threads
                        (default: None)
  --ntasks-per-core N   number of tasks to invoke on each core (default: None)
  --ntasks-per-socket N
                        number of tasks to invoke on each socket (default:
                        None)
  --hint HINT           Bind tasks according to application hints (default:
                        None)
  --mem-bind BIND       Bind memory to locality domains (default: None)
  --cpus-per-gpu N      number of CPUs required per allocated GPU (default:
                        None)
  -G, --gpus N          count of GPUs required for the job (default: None)
  --gpu-bind ...        task to gpu binding options (default: None)
  --gpu-freq ...        frequency and voltage of GPUs (default: None)
  --gpus-per-node N     number of GPUs required per allocated node (default:
                        None)
  --gpus-per-socket N   number of GPUs required per allocated socket (default:
                        None)
  --gpus-per-task N     number of GPUs required per spawned task (default:
                        None)
  --mem-per-gpu MEM_PER_GPU
                        real memory required per allocated GPU (default: None)
  --disable-stdout-job-summary
                        disable job summary in stdout file for the job
                        (default: None)
  --nvmps               launching NVIDIA MPS for job (default: None)
  --line-length LINE_LENGTH
                        line length before start of comment (default: None)
  --modules MODULES [MODULES ...]
                        Modules to load (e.g., --modules mod1 mod2 mod3)
                        (default: [])
  --input INPUT_PATH    path to input json file (default: None)
  --output OUTPUT_PATH  json path to save slurm batch script to (default:
                        None)
  --export-json JSON_PATH
                        path to export yaml for generating the slurm script to
                        (default: None)
  --custom-command COMMAND
                        Add a custom command at the end of the script (e.g.
                        --custom-command 'mpirun -n 8 ./bin > run.out')
                        (default: None)
  --custom-commands COMMAND [COMMAND ...]
                        Add custom commands at the end of the script (e.g.
                        --custom-commands '# Run simulation' 'mpirun -n 8
                        ./bin > run.out') (default: [])
  --inline-script COMMAND
                        Inline script to add at the end of the script (e.g.
                        --inline-script script.sh) (default: None)
  --inline-scripts COMMAND [COMMAND ...]
                        Add inline scripts at the end of the script (e.g.
                        --inline-scripts script1.sh script2.sh) (default: [])
  --no-header           Do not include the header comment in the generated
                        script (default: False)