Skip to content

package-ci and gitleaks: CI_POOL=on pools them over every free runner, the Oracle arm64 one included - #16

Merged
Soul-Craft merged 2 commits into
mainfrom
ci/oracle-pool
Oct 16, 2026
Merged

Soul-Craft merged 2 commits into
mainfrom
ci/oracle-pool

Conversation

@Soul-Craft

Copy link
Copy Markdown
Contributor

What

  • package-ci.yml (verify) and gitleaks.yml: while CI_RUNNER is hf, the org variable CI_POOL=on routes them to the label soulcraft-free instead of soulcraft-hf. Unset (today), nothing changes.
  • gitleaks.yml downloads the tarball of the runner's architecture (RUNNER_ARCH X64 or ARM64), each checked against its own published SHA-256 (8.30.1: linux_x64 551f6fc8..., linux_arm64 e4a487ee...).
  • package-ci.yml: runners named soulcraft-oracle-* skip the GitHub uv cache: they keep theirs on disk (UV_CACHE_DIR), and the GitHub cache would only download it again and upload it whole.

Why

soulcraft-free is carried by the never-sleeping Oracle Cloud Always Free A1 runners (linux-arm64, workspace ci/oracle-runner/, SoulCraft-Org/workspace PR to follow) and, once they get the label, by the Hugging Face runners: the package gate then pools over every free runner instead of queueing on two.

Only jobs proven on arm64 move. Proof, for $0 on GitHub's arm64 runners in this public repository (throwaway branch probe/arm64-oracle, deleted), public code only (ComfyUI Base at its newest tag, whose testbed provisions the same six packs as every package's):

ubuntu-24.04-arm ubuntu-24.04
testbed --server (ComfyUI v0.41.0, frontend 1.55.17, torch 2.14.1) 62 s, 1,322 node types 65 s, 1,322 node types
Chromium for playwright 1.62.0, rendering the server's frontend 27 s, yes 25 s, yes
base verify.sh (repo, zip) 672 passed / 12 skipped; 444 / 240 identical
gitleaks 8.30.1, checksum-checked linux_arm64 clean linux_x64 clean

Runs 38936500441 and 38937829563. Every testbed requirement also resolves wheels-only on aarch64 at the same 148 versions as x64 (uv pip compile --only-binary :all:).

actionlint 1.7.12: no new findings. This PR's own gitleaks check runs the x64 path.

After merge

Nothing until someone sets CI_POOL. The order is in workspace ci/oracle-runner/README.md (Launch, steps 6 and 7). Back: gh variable delete CI_POOL --org SoulCraft-Org.

🤖 Generated with Claude Code

Soul-Craft and others added 2 commits October 16, 2026 11:26
… (label soulcraft-free)

The package gate and gitleaks are proven on linux-arm64 (probe runs 38936500441, 38937829563: the
base testbed, Chromium 1.62.0, verify.sh and gitleaks match x64 exactly), so with the org variable
CI_POOL=on they may run on any runner labelled soulcraft-free, the never-sleeping Oracle Cloud A1
runner (workspace ci/oracle-runner/) included. Unset, nothing changes: label soulcraft-hf.
gitleaks downloads the tarball of the runner's architecture, each checked against its own SHA-256.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… disk)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@Soul-Craft
Soul-Craft requested a review from a team as a code owner October 16, 2026 17:31
@Soul-Craft
Soul-Craft merged commit f702002 into main Oct 16, 2026
3 checks passed
@Soul-Craft
Soul-Craft deleted the ci/oracle-pool branch October 16, 2026 17:33
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