Repository navigation
ci: stop warming unused Windows package archives - #29322
Conversation
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info
📝 Walkthrough
Priority: ⬇️ Low Merge Risk: ⚪ Minimal · up to
|
| Check name | Status | Explanation |
|---|---|---|
| Docstring Coverage |  |
No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff. Analyzed 0 functions across 1… |
| Linked Issues check | |
Check skipped because no linked issues were found for this pull request. |
| Out of Scope Changes check | |
Check skipped because no linked issues were found for this pull request. |
| Title check | |
The title clearly and concisely describes the main change: stopping unnecessary Windows package archive warming. |
| Description check | |
The description follows the required structure and explains the change, motivation, testing, scope, and checklist status. It identifies the linked issue as a maintainer-requested audit, which is allow… |
✨ Finishing Touches
-
📝 Generate docstrings
-
- Commit to this branch
- Create a new PR
-
- Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.
Comment @coderabbitai help to get the list of available commands.
ELI5
Windows cache warmers keep uploading large package archives that ordinary Windows PR checks deliberately avoid downloading. Stop those uploads while preserving the small compiled-module and launcher caches that real checks restore.
What Changed
cache-pnpm-store: false; native preparation, verification records, both launcher caches and probes remain enabled.native/windows-security/**in platform warmers’ main-push triggers. Those sources already invalidate Windows native caches, but previously only selected Linux warming immediately.Why
The 18:51 scheduled run uploaded 698,731,283 bytes (699 MB) of Windows pnpm archives and spent 33.324 seconds saving them (x64 12.255s; ARM 21.069s). Both native and launcher caches were hits. These exact package archive keys were hits six hours earlier, then missing again. Repository cache usage during the audit was 10.687 GB. This shows repeated writes and storage pressure, not which cache evicted another.
Real Windows ARM and x64 consumers use
cache-pnpm-store: false, while restoring their native/launcher caches. Ordinary PR root installs and combined x64 installs also bypass archive restoration in the shared action. The manual x64 terminal-restart workflow still uses the root archive; it retains a cold install/save fallback. No ARM root-archive consumer was found.The reported PR warmer run used 357s ARM + 71s x64. It validates edited Windows workflow commands and cannot seed shared default-branch caches. Keep that infrequent validation: it runs only for the two platform warmer YAML paths, which real Windows consumer workflows do not cover. This change does not claim to eliminate those 428s or establish whole-CI speedup. Removing all warming would also remove useful compiled caches; removing workflow validation would leave changed Windows commands unexercised.
Related audit decisions: retain #24881, #26399, #28081 and #28917, which reduce scheduled or irrelevant allocations. Correct the Windows caller of #24927, rather than revert its measured general download reduction. #29279 cancellation remains compatible.
Other checked CI changes have real benefits: #29136 removes a successful aggregate allocation (observed old job 138s queue + 4s running; new job zero runner/steps), with a hosted negative qualification still failing correctly. #29149 reduced the observed installer build from 831s to 299s, and the whole job from 1170s to 797s; its extra compiler-parity job costs 294s only on qualifying inputs. These compare different hosted runners and commits, not controlled repository-wide averages. Net return from all scheduled warmer jobs remains unproven; cache consumption alone does not prove the producer was necessary.
Linked Issue
Maintainer-requested CI cost audit; no separate issue created.
Visual Proof
N/A — GitHub Actions configuration only; no application UI changes.
Testing
On Linux, the two updated regression assertions failed against the original workflows for the intended Windows archive policy and missing security-trigger reasons. After the fix, 131 tests pass across
ci-dependency-download-cache,ci-pnpm-store-mode,ci-native-cacheandci-pnpm-verification-cache.Changed-code quality, focused Oxlint, Oxfmt,
git diff --check, and actionlint 1.7.12 pass. No TypeScript implementation changes. Both changed Windows jobs passed on exact headab326b454cb69bf7efcdf58b31c82b763db6c19bin run 38831888211. Their logs showcache-pnpm-store: false, the archive lookup/save skipped, and the native and launcher cache hits preserved. Full ready-for-review PR qualification is queued in run 38833157340; merge waits for that workflow and its aggregate gate.Review
Reviewed archive consumers, cold fallback, native/launcher key parity, PR cache scope, event routing and aggregate independence. Required-check names and aggregate failure/conditional handling are unchanged. The package archive fix uses the existing opt-out instead of adding cache-presence or install logic.
Agent skill upstream boundary
Notes
Hosted CI configuration only. No app, SSH execution, folder-workspace, mobile or remote-wire changes. The Windows store opt-out applies equally to x64 and ARM; other platforms retain their existing policy.
Checklist