App Arguments

Every python-s3 input and what it does

The following covers every input available in the portal form and the Tapis job request. Most jobs need only the two required inputs; everything else defaults to off. For a decision-oriented guide, see Choosing Inputs.

1. File input¶

Input Directory (required)¶

A single directory staged by Tapis into the job working directory.

What should be inside:

• your main script (Python, Tcl, or an input file for your own binary)

• any supporting files your script needs (models, data, configs)

• optional helper files

  • requirements.txt (for PIP_REQUIREMENTS)

  • pre/post scripts (for PRE_SCRIPT / POST_SCRIPT)

  • ZIP bundles referenced by UNZIP_INPUTS

The wrapper cds into this directory before running anything. Relative paths in your script should assume this directory is the working directory.

2. App arguments¶

Input Script (required)¶

The filename of the input script passed to the executable.

Rules:

• filename only (no path)

• must exist inside the Input Directory (or inside a bundle expanded by UNZIP_INPUTS)

Examples: run_analysis.py, model.tcl, input.dat

Arguments (optional)¶

Free-form arguments appended after the Input Script.

Example:

--NodalMass 4.19 --outDir outCase1

Final command structure:

[MPI_LAUNCHER] <BINARY> <Input Script> <Arguments...>

3. Scheduler inputs¶

TACC Scheduler Profile (fixed)¶

The app uses the python_default profile, which loads python/3.12.11 and nothing else. All other module state is controlled explicitly through EXTRA_MODULES, so runs are reproducible rather than dependent on login-node defaults.

Allocation¶

The DesignSafe portal injects your TACC allocation automatically. From dapi, pass allocation="..." to ds.jobs.generate.

4. Environment variables¶

All optional. Empty values are skipped entirely. “True-like” means True or 1 (case-insensitive).

4.1 Execution¶

BINARY (default python3)¶

The executable for the main run. Three forms:

• Name (no /): resolved on PATH after EXTRA_MODULES load — e.g. OpenSeesMP

• ./name: a program shipped inside the Input Directory; the execute bit is restored automatically if staging dropped it

• Absolute path: e.g. $WORK/apps/mysolver for your own compiled code

Resolution happens after the pre-script (so a pre-script may build or stage the binary). If the binary cannot be found, the job fails with a clear hint before the main run starts. python is normalized to python3. The resolved absolute path is recorded in job-summary.json.

USE_MPI (default False)¶

Use True for OpenSeesMP/SP and mpi4py. Use False for serial code and single-node parallelism via threading or concurrent.futures.

MPI_LAUNCHER (default ibrun)¶

The launch command used when USE_MPI is true-like. ibrun is correct on TACC systems; the option exists so the same wrapper ports to systems using srun or mpirun.

4.2 Modules¶

EXTRA_MODULES¶

Comma-separated TACC modules to load before anything runs, e.g.

opensees,hdf5/1.14.4

A module that fails to load aborts the job. Modules the main run needs must go here, not in a pre-script — module load inside a pre/post script only affects that script’s own process.

4.3 Python packages and environments¶

PIP_REQUIREMENTS¶

A requirements-style file in the Input Directory, e.g. requirements.txt. The job fails immediately if the file is missing — a silently skipped install is a debugging trap.

PIP_PACKAGES¶

Comma-separated packages, e.g. scipy,h5py. Installed in one pip install call.

PYTHON_ENV¶

Where installs go depends on this setting:

• Unset (default): when either pip input is set, a temporary per-job virtual environment (.job-venv) is created with access to the module-provided numpy/scipy/mpi4py stack, activated for the whole job, and deleted on exit — nothing leaks into $HOME, no packages carry between jobs, the archive stays small. No environment is created when no installs are requested.

• Set (e.g. $WORK/envs/myproject): that environment is used instead — created if missing, activated, kept after the job — so repeat jobs skip reinstall time. Keep persistent environments in $WORK/$SCRATCH, not inside the Input Directory (they would be archived with your results).

Either way the environment is activated for pre/post scripts and the main run, and its path is recorded as python_env in job-summary.json. If you override BINARY to a different Python interpreter by absolute path, it bypasses the environment and will not see the installed packages.

4.4 Input preparation¶

UNZIP_INPUTS¶

Comma-separated ZIP files in the Input Directory to expand before anything else runs (.zip suffix optional):

UNZIP_INPUTS = mydata,meshes

Use this when your input directory would otherwise contain many small files: Tapis stages each file of a directory input as its own transfer task (~40 s per file under tenant load), so one bundled ZIP stages in seconds where a hundred loose files can take an hour. Because expansion runs first, the pre-script, requirements file, and main script can all live inside the bundle. A listed ZIP that is missing fails the job immediately.

4.5 Pre/post scripts¶

PRE_SCRIPT¶

Script run after environment setup but before the main run — use it for staging, generating inputs, or copying data from $WORK/$SCRATCH.

• if relative, resolved inside the Input Directory

• .py runs via python3; executables run directly; anything else runs via bash

• a missing or failing pre-script aborts the job before the main run — a broken setup never burns SUs on the compute phase

POST_SCRIPT¶

Script run after a successful main run (same resolution rules) — use it for post-processing, packaging, or moving outputs. Skipped if the main run fails.

POST_SCRIPT_REQUIRED (default False)¶

By default a failing post-script logs a warning and the job still succeeds — a plotting glitch should not mark a ten-hour simulation FAILED. Set True when the post-processing output is the deliverable (e.g., aggregating a parameter sweep).

5. Outputs and the run record¶

Every job writes two log artifacts next to tapisjob.out:

• job-summary.json — machine-readable run record: resolved binary, Python environment, exact command, per-stage exit codes and timings. Written on success and failure. See the example on the app page.

• The Tapis archive (in My Data by default) contains the Input Directory, tapisjob.out, and job-summary.json.

Large outputs are best packaged or relocated by a POST_SCRIPT — see Choosing Inputs for recipes.

6. Typical patterns¶

Minimal starter configurations for General Python, OpenSeesMP, OpenSeesPy, and mpi4py are collected in Choosing Inputs; the end-to-end verified example is on the Python App page.