Skip to content

Distribute image archives through direct fabric - #135

Closed
FujitsuPolycom wants to merge 2 commits into
codex/cuda-restore-terminologyfrom
codex/direct-fabric-image-fanout
Closed

Distribute image archives through direct fabric#135
FujitsuPolycom wants to merge 2 commits into
codex/cuda-restore-terminologyfrom
codex/direct-fabric-image-fanout

Conversation

@FujitsuPolycom

Copy link
Copy Markdown
Owner

Result

Status: implemented, with GPU-free command and state-machine coverage. Direct-fabric throughput and live image import remain unqualified.

Add scripts/fanout_image_archive.py to download one checksum-bound image archive on a configured seed rank and forward it through the four-rank direct-link chain derived from the validated site topology. Management SSH starts and inspects work; rsync binds to each source fabric address and connects to the adjacent peer address.

The utility provides:

  • plan-only default behavior;
  • read-only --verify mode;
  • explicit confirmation for execution;
  • --create-only archive distribution without Docker import;
  • bounded final and partial paths beneath an explicit target directory;
  • exact-file reuse and conflict rejection without overwriting final files;
  • resumable curl --continue-at - seed downloads;
  • resumable rsync --partial --append-verify fabric hops;
  • SHA-256 verification after every hop; and
  • rank, hop, archive, topology, and optional image-import evidence.

No command from this branch was run against live hosts.

Validation

  • python -m pytest scripts/test_fanout_image_archive.py scripts/test_sparkring_site.py -q: 141 passed
  • ruff check --select E,F,W --ignore E501 scripts/fanout_image_archive.py scripts/test_fanout_image_archive.py: passed
  • Offline planning against the sanitized GLM-5.3 TP4 site produced the expected 0→1→2→3 fabric chain and no remote operation.
  • The broader python -m pytest scripts -q -rs run produced 508 passed and 3 skipped, plus one unrelated README assertion failure that reproduces unchanged on the stacked base.

This draft is stacked on #131 so its site/profile terminology matches that branch.

Add a plan-first four-rank archive fan-out utility that seeds one checksum-bound file and forwards it through the direct-link chain derived from the validated site topology. Preserve resumable partial downloads and rsync hops, reject conflicting final files, verify SHA-256 after every hop, support read-only verification and create-only execution, and emit rank/hop evidence before optional image import. Validation: 141 fan-out and site-topology tests passed; Ruff E/F/W checks passed. The broader scripts suite has one pre-existing README assertion failure also reproduced on the stacked base.
@FujitsuPolycom

Copy link
Copy Markdown
Owner Author

The resulting runtime, operator contract, and evidence are consolidated in retained draft stack #146#147#150. Independent page-base research remains in #149. Closing this superseded draft and deleting only its remote head branch.

@FujitsuPolycom
FujitsuPolycom deleted the codex/direct-fabric-image-fanout branch August 31, 2026 01:55
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