Job management

Bamboost includes some tools to manage your jobs. It can:

Create a submission script. Either bash for local execution or sbatch/slurm for remote execution.
Submit jobs on the cluster (or locally).
Monitor the status of your jobs.
Monitor slurm job status on Euler.

I suggest the following workflow:

Create a job runtime script. This script will be the actual execution of your simulation. So it is all the code you do given an input file (just that you won't have an input file).
Create a job creation script. In this script, you define the job parameters and the commands to execute. Also include and copy the job runtime script to the bamboost simulation.
Run the job creation script. This will create a submission script that that can be executed or submitted on the cluster.

Job runtime script

This is your actual simulation code.

Example script.py

import argparse
from bamboost import SimulationWriter
 
def main(sim: SimulationWriter):
 
    parameters = sim.parameters
 
    sim.add_metadata()
    sim.register_git_attributes("path/to/repo")
    sim.add_mesh(...)
 
    # Model calculations
    for i in range(parameters["n_steps"]):
        sim.add_field(...)
        sim.add_field(...)
        sim.add_global_field(...)
        sim.finish_step()
 
    sim.create_xdmf_file()
    sim.finish_sim()
 
if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--simulation", type=str, required=True)
    args = parser.parse_args()
 
    with SimulationWriter.fromUID(args.simulation) as sim:
        main(sim)

Job creation script

This script will create a submission script that can be executed or submitted. Also it serves as an input file. It should do the following:

define parameters
copy the job runtime script to the simulation
copy itself to the simulation
define the commands to execute
create the submission script
(optional) submit the job immediately

Example job.py

import argparse
import os
os.environ["BAMBOOST_MPI"] = "0"
from bamboost import Manager
 
def main(db_path, name: str = None, *, submit: bool = False):
 
    script_file = "path/to/runtime_script"
    ntasks = 8
    mem_per_cpu = "2G"
    note = "This is a test job"
    
    params = {
        "n_steps": 100,
        "E": 1.0,
        "nu": 0.3,
        "mesh": "<mesh-db>:<mesh-id>",
    }
 
    db = Manager(db_path)
    sim = db.create_simulation(
        name=name,
        parameters=params,
        skip_duplicate_check=True,
    )
    sim.change_note(note)
    sim.links["mesh"] = params["mesh"]
    sim.copy_file(script_file)
    sim.copy_file(__file__)
 
    cmd = [
        "source ~/envs/my-euler-modules.sh",
        f"srun python3 $SIMULATION_DIR/{os.path.basename(script_file)} --simulation {sim.get_full_uid()}",
    ]
    sim.create_run_script(
        cmd,
        ntasks=ntasks,
        mem_per_cpu=mem_per_cpu,
        euler=True,
        time="04:00:00",
    )
 
    if submit:
        sim.submit()
 
 
if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("db_path", type=str, help="Path to the database.")
    parser.add_argument(
        "name", type=str, help="Name of the simulation.", default=None, nargs="?"
    )
    parser.add_argument("--submit", "-s", action="store_true", help="Submit the job.")
    args = parser.parse_args()
 
    main(args.db_path, args.name, submit=args.submit)

Job management

Job runtime script

Job creation script

On this page