Skip to content

[CI] release-whl-kernel: clean root-owned build artifacts before checkout#23747

Merged
Kangyan-Zhou merged 1 commit intomainfrom
fix-kernel-build-cleanup
Apr 26, 2026
Merged

[CI] release-whl-kernel: clean root-owned build artifacts before checkout#23747
Kangyan-Zhou merged 1 commit intomainfrom
fix-kernel-build-cleanup

Conversation

@Kangyan-Zhou
Copy link
Copy Markdown
Collaborator

Summary

  • The four sgl-kernel build jobs in release-whl-kernel.yml (build-cu129-matrix, build-cu130-matrix, build-rocm-matrix, build-musa43) run on self-hosted nodes that retain the workspace across jobs. Their in-container compile runs as root and leaves root-owned files under sgl-kernel/build/, which causes the next run's actions/checkout@v4 to fail with EACCES: permission denied, rmdir 'sgl-kernel/build/.cmake'.
  • Add a pre-checkout cleanup step that wipes the workspace via a throwaway alpine:3 root container — same pattern already in use in nightly-72-gpu-gb200.yml:132-140 for the same reason.
  • pr-test.yml's kernel-build jobs already have a working sudo rm -rf $GITHUB_WORKSPACE/* cleanup, so they are intentionally left unchanged. release-whl-kernel.yml was the only kernel-building workflow with no cleanup at all.

Test plan

  • Trigger release-whl-kernel.yml (workflow_dispatch) on a PR build and confirm the new "Clean workspace" step runs and the subsequent actions/checkout@v4 succeeds with no EACCES errors.
  • Confirm the four build jobs still produce wheel artifacts (wheel-python3.10-cuda12.9*, wheel-python3.10-cuda13.0*, wheel-python3.10-rocm{700,720}, wheel-python3.10-musa43) as before.

🤖 Generated with Claude Code

…kout

Self-hosted kernel-build nodes retain the workspace across jobs. Prior
builds leave root-owned files under sgl-kernel/build/ (the in-container
compile runs as root and writes to the bind-mounted workspace), which
makes the next actions/checkout@v4 fail with EACCES on rmdir
sgl-kernel/build/.cmake.

Add a pre-checkout step to all four kernel-build jobs (build-cu129-matrix,
build-cu130-matrix, build-rocm-matrix, build-musa43) that wipes the
workspace via a throwaway alpine root container — same pattern already in
nightly-72-gpu-gb200.yml.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@gemini-code-assist
Copy link
Copy Markdown
Contributor

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

@Kangyan-Zhou Kangyan-Zhou merged commit 282b47f into main Apr 26, 2026
43 checks passed
@Kangyan-Zhou Kangyan-Zhou deleted the fix-kernel-build-cleanup branch April 26, 2026 04:39
vguduruTT pushed a commit to vguduruTT/sglang that referenced this pull request May 2, 2026
…kout (sgl-project#23747)

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
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