Core guide
Use this section after completing Start Here. It explains the progression from mental model → robust single job → evidence-based resources → manifest and array → dependency and recovery → accounting and troubleshooting.
Learning path
| Page | Use it when you need to… |
|---|---|
| Start Here | run the tested onboarding checkpoints before adapting the patterns |
| Mental model | distinguish serial code, threaded code, multiple processes, and independent repetitions |
| Robust single job | create an auditable batch script with safe shell behavior, runtime setup, and unique logs |
| Resource right-sizing | select CPU, memory, walltime, and concurrency from evidence |
| Arrays and manifests | parallelize phenotypes, files, chromosomes, simulations, or parameter combinations |
| Dependencies and recovery | fan out, fan in, verify completeness, and rerun only failed elements |
| Monitoring and accounting | understand pending states, inspect live jobs, and compare completed tasks |
| Troubleshooting | follow a repeatable diagnostic sequence rather than changing resources at random |
| Storage and compliance | stage active data safely and keep PHI/sensitive-data boundaries explicit |
| Workflow tools | decide when raw arrays are enough and when a workflow manager removes real complexity |
| AI-assisted SLURM | use ChatGPT/Codex as a drafting and diagnostic assistant without surrendering judgment |
Keep one worker runnable outside SLURM. A clean command-line interface—explicit inputs, one work unit, one output, non-zero exit on failure—is the foundation for arrays, Nextflow, Snakemake, and reliable AI assistance.