PyJobRunner 2023.12.1

Job runner for scientific computing workloads

Jobrunner is a command line tool to manage and deploy computational experiments, organize workflows, and enforce a directory-based hierarchy to enable reuse of files and bash scripts along a directory tree. Organization details of the tree are encoded in Jobfiles, which provide an index of files and scripts on each node and indicate their relationship with Jobrunner commands. It is a flexible tool that allows users to design their own directory structure and maintain consistency with the increase in complexity of their workflows and experiments.

Read our paper here: https://arxiv.org/abs/2308.15637

Installation

Stable releases of Jobrunner are hosted on Python Package Index website (https://pypi.org/project/PyJobRunner/) and can be installed by executing,

pip install PyJobrunner

Note that pip should point to python3+ installation package pip3.

Upgrading and uninstallation is easily managed through this interface using,

pip install --upgrade PyJobrunner
pip uninstall PyJobRunner

The following installation option can be used to allow for using customization specific to instruments.

pip install PyJobruner --user --install-option="--with-instruments"

This allow for the use of the instrument field in the Jobfile

There maybe situations where users may want to install Jobrunner in development mode $\textemdash$ to design new features, debug, or customize options/commands to their needs. This can be easily accomplished using the setup script located in the project root directory and executing,

./setup develop --with-instruments

Development mode enables testing of features/updates directly from the source code and is an effective method for debugging. Note that the setup script relies on click, which can be installed using,

pip install click

The jobrunner script is installed in $HOME/.local/bin directory and therfore the environment variable, PATH, should be updated to include this location for command line use.

Dependencies

python3.8+ click toml pyyaml alive-progress

Statement of Need

Use of software for data acquisition, analysis, and discovery in scientific studies has allowed integration of sustainable software development practices into the research process, enabling physics-based simulation instruments like Flash-X (https://flash-x.org) to model problems ranging from pool boiling to stellar explosions. However, the design and management of software-based scientific studies is often left to individual researchers who design their computational experiments based on personal preferences and the nature of the study.

Although applications are available to create reproducible capsules for data generation (https://codeocean.com), they do not provide tools to manage research in a structured way which can enhance knowledge related to decisions made by researchers to configure their software instruments. A well-organized lab notebook and execution environment enables systematic curation of the research process and provides implicit documentation for software configuration and options used to perform specific studies. This in turn enhances reproducibility by providing a roadmap towards data generation and contributing towards knowledge and understanding of an experiment.

Jobrunner addresses this need by enabling the management of software environments for computational experiments that rely on a Unix-style interface for development and execution. The design and operation of the tool allow researchers to efficiently organize their workflows without compromising their design preferences and requirements. We have applied this tool to manage performance and computational fluid dynamics studies using Flash-X.

Example

Application of Jobrunner can be understood better with an example design of a computational experiment. Consider an experiment named Project representative of a publicly available dataset (https://github.com/Lab-Notebooks/Outflow-Forcing-BubbleML) for the work presented in (https://arxiv.org/pdf/2306.10174.pdf). The directory tree has the following structure,

$ tree Project

├── Jobfile
├── environment.sh
├── sites/
├── software/
├── simulation/

Each node in the tree is organized to capture information related to different aspects of the experiments. The node sites/ for example stores platform specific information related to compilers and modules required to build the software stack described in the node software/. Information provided in these nodes capture the execution environment of the computational experiment.

Following is the design of the sites/ node for the example above,

$ tree Project/sites
├── sites/
    ├── sedona/
        ├── modules.sh

The site-specific subnode sites/sedona/ consists of commands to load platform specific compilers and libraries required to build Flash-X which is the instrument used to perform the experiments.

# file: Project/sites/sedona/modules.sh
#
# Load Message Passing Interface (MPI) and
# Hierarchical Data Format (HDF5) libraries
module load openmpi
module load hdf5

There are situations where requirements for Flash-X are not available as modules and may have to be built from their respective source. This is usually the case when a specific version of the library or compiler is desired. The software/ node provides configuration details for these,

$ tree Project/software

├── software/
    ├── Jobfile
    ├── setupFlashX.sh
    ├── setupAMReX.sh

Here the script setupAMReX.sh provides commands to get the source code for AMReX(https://github.com/AMReX-Codes/amrex) and build it for desired version and configuration. The script setupFlashX.sh sets the version for Flash-X to perform the experiments. The Jobfile indicates the use of these files by assigning them to specific Jobrunner commands,

# file: Project/software/Jobfile
#
# Run these scripts during jobrunner setup command
job:
  setup:
    - setupAMReX.sh
    - setupFlashX.sh

The environment.sh file at the root of the Project directory sources the site-specific modules.sh and sets environment variables for compilation and execution.

# file: Project/environment.sh
#
# Set project home using realpath of current directory
export PROJECT_HOME=$(realpath .)

# Enter site information and source the modules
SiteName="sedona"
SiteHome="$PROJECT_HOME/sites/$SiteName"
source $SiteHome/modules.sh

# Set environment variables required for Makefile.h
export MPI_HOME=$(which mpicc | sed s/'\/bin\/mpicc'//)
export HDF5_HOME=$(which h5pfc | sed s/'\/bin\/h5pfc'//)

# Assign path for local AMReX installation
export AMREX2D_HOME="$PROJECT_HOME/software/AMReX/install-$SiteName/2D"
export AMREX3D_HOME="$PROJECT_HOME/software/AMReX/install-$SiteName/3D"

# Path to Flash-X
export FLASHX_HOME="$PROJECT_HOME/software/Flash-X"

The Jobfile at this node assigns the use of environment.sh,

# file: Project/Jobfile

instrument: Flash-X

# Scripts to include during jobrunner setup and submit commands
job:
  setup:
    - environment.sh
  submit:
    - environment.sh

During the invocation of jobrunner setup software/ command, environment.sh, setupAMReX.sh and setupFlashX.sh are combined using the information in Jobfiles and executed in sequence to build the software stack.

The computational experiments are described in the node simulation/ and organized under different studies, FlowBoiling, EvaporatingBubble and PoolBoiling as shown below,

$ tree Project/simulation

├── simulation/
    ├── FlowBoiling/
    ├── EvaporatingBubble/
    ├── PoolBoiling/
        ├── Jobfile
        ├── flashSetup.sh
        ├── flashRun.sh
        ├── pool_boiling.par
        ├── earth-gravity/
            ├── Jobfile
            ├── earth_gravity.par
        ├── low-gravity/
            ├── Jobfile
            ├── low_gravity.par

The Jobfile under subnode simulation/PoolBoiling provides details for the files and scripts at this level

# file: Project/simulation/PoolBoiling/Jobfile
#
job:
  # list of scripts that need to execute during setup
  setup:
    - flashSetup.sh

  # target executable created during setup
  target: flashx

  # input for the target
  input:
    - pool_boiling.par

  # list of scripts that need to execute during submit
  submit:
    - flashRun.sh

During the invocation of jobrunner setup simulation/PoolBoiling command, environment.sh and flashSetup.sh are combined using the information in Jobfiles and executed in sequence to build the target executable flashx. The software stack built in the previous step is available through the environment variables in environment.sh.

The subnode simulation/PoolBoiling contains two different configurations earth_gravity and low_gravity to represent a parametric study of the boiling phenomenon under different gravity conditions. Each configuration contains its respective Jobfile,

# file: Project/simulation/PoolBoiling/earth_gravity/Jobfile
#
job:
  # input for the target
  input:
    - earth_gravity.par

Scientific instruments like Flash-X require input during execution which is supplied in the form of parfiles with a .par extension. The parfiles along a directory tree are combined to create a single input file when submitting the job. For example, invocation of jobrunner submit simulation/PoolBoiling/earth_gravity combines pool_boiling.par and earth_gravity.par that is used to run the target executable flashx using the combination of environment.sh and flashRun.sh.

Computational jobs are typically submitted using schedulars like slurm to efficiently manage and allocate computational resources on large supercomputing systems. The details of the schedular with desired options is supplied by extending the Jobfile at root of the Project directory,

# file: Project/Jobfile
#
# Scripts to include during jobrunner setup and submit commands
job:
  setup:
    - environment.sh
  submit:
    - environment.sh

# schedular command and options to dispatch jobs
schedular:
  command: slurm
  options:
    - "#SBATCH -t 0-30:00"
    - "#SBATCH --job-name=myjob"
    - "#SBATCH --ntasks=5"

Jobrunner also provides features to keep the directory structure clean. Results and artifacts from computational runs can be designated for archiving or cleaning by extending the Jobfile for each study,

# file: Project/simulation/PoolBoiling/earth_gravity/Jobfile
#
job:
  # input for the target
  input:
    - earth_gravity.par

  # clean slurm output and error files
  clean:
    - "*.out"
    - "*.err"

  # archive flashx log and output files
  archive:
    - "*_hdf5_*"
    - "*.log"

Jobrunner commands

Setup

jobrunner setup <JobWorkDir> creates a job.setup file using job.setup scripts defined in Jobfiles along the directory tree. Jobrunner executes each script serially by changing the working directory to the location of the script. A special environment variable JobWorkDir provides the value of <JobWorkDir> supplied during invocation of the command.

Working directory: /Project/simulation/PoolBoiling
Parsing Jobfiles in directory tree

job.setup: [
        /Project/environment.sh
        /Project/simulation/PoolBoiling/flashSetup.sh
        ]

Submit

jobrunner submit <JobWorkDir> creates a job.submit file using job.submit scripts and schedular.options values defined in Jobfiles along the directory tree. schedular.command is used to dispatch the resulting script.

Working directory: /Project/simulation/PoolBoiling/earth_gravity
Parsing Jobfiles in directory tree

schedular.command:
        slurm
schedular.options: [
        #SBATCH -t 0-30:00
        #SBATCH --job-name=myjob
        #SBATCH --ntasks=5
        ]
job.input: [
        /Project/simulation/PoolBoiling/pool_boiling.par
        /Project/simulation/PoolBoiling/earth_gravity/earth_gravity.par
        ]
job.target:
        /Project/simulation/PoolBoiling/flashx
job.submit: [
        /Project/environment.sh
        /Project/simulation/PoolBoiling/flashRun.sh
        ]

Along with the job.submit script, job.input and job.target files are also created in <JobWorkDir> using values defined in Jobfiles.

Archive

jobrunner archive --tag=<tagID> <JobWorkDir> creates archives along the directory tree using the list of values defined job.archive. The archives are created under the sub-directory jobnode.archive/<tagID> and represent the state of the directory tree during the invocation.

Export

jobrunner export --tag=<pathToArchive> <JobWorkDir> exports directory tree and archives objects to an external directory <pathToArchive> to preserve state and curate execution environment.

Clean

jobrunner clean <JobWorkDir> removes Jobrunner artifacts from the working directory

Examples

Functionality of Jobrunner is best understood through example projects which can be found in following repositories:

• akashdhruv/Multiphase-Simulations: A lab notebook to manage development of Flash-X

• Lab-Notebooks/Outflow-Forcing-BubbleML: Reproducibility capsule for research papers (https://arxiv.org/pdf/2306.10174.pdf) (https://arxiv.org/pdf/2307.14623.pdf)

• Lab-Notebooks/Flow-Boiling-3DL: Execution environment for running three-dimensional flow boiling simulations on high performance computing systems.

Citation

@article{dhruv_dubey_2023,
  author  = {Dhruv, Akash and Dubey, Anshu},
  journal = {Computing in Science & Engineering},
  title   = {Managing Software Provenance to Enhance Reproducibility in Computational Research},
  year    = {2023},
  volume  = {25},
  number  = {3},
  pages   = {60-65},
  doi     = {10.1109/MCSE.2023.3314288}
}

Acknowledgements

This material is based upon work supported by Laboratory Directed Research and Development (LDRD) funding from Argonne National Laboratory, provided by the Director, Office of Science, of the U.S. Department of Energy under Contract No. DE-AC02-06CH11357.