revert(ci): drop turbopackFileSystemCacheForBuild and restore actions/cache#4500
Conversation
…/cache Build times grew monotonically from ~200s to 1305s (~6x) after #4478 landed. The Turbopack FS cache flag combined with sticky-disk persistence caused unbounded cache growth. Reverting both — keeping the poweredByHeader and optimizePackageImports trim from the original PR.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
PR SummaryMedium Risk Overview Updates the CI workflow to restore Reviewed by Cursor Bugbot for commit 37e272b. Configure here. |
Greptile SummaryThis PR reverts a CI regression introduced in #4478 by removing
Confidence Score: 5/5This PR is safe to merge — it reverts two tightly scoped CI settings that caused unbounded cache growth and restores a well-understood caching pattern. Both changes are narrow and low-risk: removing an experimental Next.js flag that had no effect on application behavior, and swapping an eviction-free disk mount for the standard GitHub Actions cache with a lockfile-keyed strategy. The Bun, node_modules, and Turbo sticky-disk caches are untouched. There is no production code change. No files require special attention. Important Files Changed
Reviews (1): Last reviewed commit: "revert(ci): drop turbopackFileSystemCach..." | Re-trigger Greptile |
Summary
experimental.turbopackFileSystemCacheForBuild: truefromapps/sim/next.config.tsactions/cache@v5for.next/cache(replaces sticky-disk mount)poweredByHeader: false,optimizePackageImportstrim)Why
After #4478 merged, CI build times grew monotonically:
The Turbopack FS build cache flag is the necessary condition — without it, Turbopack doesn't write FS cache during
next build, so persistence is moot. Combined with sticky-disk's lack of key-based eviction (7-day idle only), the cache grew unbounded across runs and dominated mount/restore time. Reverting both restores the previous behavior.Type of Change
Testing
Will verify build duration on this PR returns to the ~200s baseline.
Checklist