Skip to content

maintenance: keep Python/JAX caches off $HOME on HPC (autolens_inference#13) - #51

Merged
Jammy2211 merged 1 commit into
mainfrom
feature/hpc-cache-off-home
Sep 25, 2026
Merged

Jammy2211 merged 1 commit into
mainfrom
feature/hpc-cache-off-home

Conversation

@Jammy2211

Copy link
Copy Markdown
Contributor

Summary

Keep Python/JAX caches off $HOME (and /tmp) on HPC: the HPC branch of activate.sh now exports XDG/pip/matplotlib/numba/CUDA/Triton/JAX/astropy caches under ${PYAUTO_HPC_CACHE:-$(dirname "$PYAUTO_HPC_BASE")/.cache} (= /mnt/ral/jnightin/.cache on RAL), filling only unset variables. Requested by the RAL admin after node root disks filled with ~/.cache. Part of PyAutoLabs/autolens_inference#13; phase 2 (autolens_profiling, euclid pipeline) is planned/blocked.

Scripts Changed

  • activate.sh

Heart RED override (development only)

  • Authorization (live human, 2026-09-25, this session): answered "Authorise override" to the question naming hpc-cache-off-home — push the 4 branches and open pending-release PRs; merge still needs a separate /prm with green checks; no release.
  • Exact current RED reasons (pyauto-heart readiness):
    • release validation FAILED (stage integrate)
    • workspace validation not passing (4 failed, cloud#35579888156: autolens notebooks/cluster/modeling.ipynb, autolens notebooks/weak/a2744.ipynb, autolens scripts/cluster/modeling.py, +1 more)
    • manifest drift: hub organism blurb (organs present) — 7 mismatch(es) vs PyAutoMind/repos.yaml
  • Passed branch gates: fake-base witness on all four activate.sh (default HPC path, preset/empty JAX cache preserved, SLURM errexit guard quiet, laptop .venv path untouched), bash -n on every edited script; autolens_inference check_submits.py --check 12/12, build_readme.py --check, ruff check/format, pytest scripts/misc/test 57 passed; autogalaxy_assistant --check-provenance 0 errors, --lint-idioms clean, --check-citations 0 missing; in-session diff review. No smoke-covered scripts/ Python changed.
  • This branch does not fix Heart; Heart stays RED for release.

🤖 Generated with Claude Code

…nce#13)

On RAL every node's /home (and /tmp) sits on a small root disk; the RAL admin
found nodes broken by ~/.cache filling it (2026-09-25). A hot-fix lives in the RAL venv's
bin/activate but is lost on a venv rebuild; this makes it permanent in tracked activate.sh.

The HPC branch now exports XDG_CACHE_HOME, PIP_CACHE_DIR, MPLCONFIGDIR, NUMBA_CACHE_DIR,
CUDA_CACHE_PATH, TRITON_CACHE_DIR, JAX_COMPILATION_CACHE_DIR and ASTROPY_CACHE_DIR under
${PYAUTO_HPC_CACHE:-$(dirname "$PYAUTO_HPC_BASE")/.cache}, filling only unset variables
(JAX via ${VAR-...} so an explicit empty value still disables the cache).

Block is inside the PYAUTO_HPC_BASE branch only; laptop .venv users are unaffected.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@Jammy2211 Jammy2211 added the pending-release Awaiting coordinated PyAuto release label Sep 25, 2026
@Jammy2211
Jammy2211 merged commit 7fe7064 into main Sep 25, 2026
1 check passed
@Jammy2211
Jammy2211 deleted the feature/hpc-cache-off-home branch September 25, 2026 15:51
@Jammy2211 Jammy2211 removed the pending-release Awaiting coordinated PyAuto release label Oct 7, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant