Skip to content

docs: local-disk tiering design note (spike-verified) - #8

Merged
ligon merged 2 commits into
mainfrom
docs/local-disk-tiering
Sep 9, 2026
Merged

docs: local-disk tiering design note (spike-verified)#8
ligon merged 2 commits into
mainfrom
docs/local-disk-tiering

Conversation

@ligon

@ligon ligon commented Sep 9, 2026

Copy link
Copy Markdown
Owner

Design note for a disposable working tree on /local/job$ID with the shared mirror as the durable repo and git as the sync (post-commit hook + periodic WIP snapshot on refs/sucoder/wip/<name>). Spike-verified on n0036.savio4; results table in the note. Also links persistent-presence.org from the docs index.

Companion finding, to be fixed separately: no deadline watchdog runs for confined jobs (_start_slurm_timer is only reached via _ensure_slurm_node, cli.py:1130).

🤖 Generated with Claude Code

https://claude.ai/code/session_01StcQgXQDE4F6eJer1sEXRb

Sue the Coder and others added 2 commits September 9, 2026 10:43
Design note for keeping the agent's working tree and caches on the
compute node's /local/job$ID while the shared mirror stays the durable
repository: a post-commit hook publishes each commit to the mirror, and
a periodic snapshot of the dirty tree lands on refs/sucoder/wip/<name>.
Replaces the orphan-prone all-on-/local --local-disk layout.

Includes measured filesystem facts (~/mirrors is a Lustre symlink;
flock fails only on NFS $HOME), a manual spike on n0036.savio4 that
verified every claim (clone 3.5s, commit-to-published 1.15s, snapshot
1.27s, restore byte-identical), and the finding that no deadline
watchdog runs for confined jobs because _start_slurm_timer is only
reached on the salloc path.

Also links persistent-presence.org from the docs index, which was
published but unlisted.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01StcQgXQDE4F6eJer1sEXRb
…pdate

The docs job failed twice in a row on PR #8 with
  E: Failed to fetch https://dl.google.com/linux/chrome-stable/deb/dists/stable/main/binary-amd64/Packages.gz  Hash Sum mismatch
before reaching org-publish.  Only emacs-nox from Ubuntu's repos is needed.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01StcQgXQDE4F6eJer1sEXRb
@ligon
ligon merged commit 790e97c into main Sep 9, 2026
4 checks passed
@ligon
ligon deleted the docs/local-disk-tiering branch September 12, 2026 20:59
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