ci(rust): fix cache thrashing that forced cold compiles every run#1201
Open
gaspb wants to merge 2 commits into
Open
ci(rust): fix cache thrashing that forced cold compiles every run#1201gaspb wants to merge 2 commits into
gaspb wants to merge 2 commits into
Conversation
The Rust CI critical path (Test job) was ~17 min, ~80% of it recompiling dependencies from scratch. Root cause: sccache (GHA backend) and Swatinem/rust-cache shared the single 10 GB per-repo Actions cache and evicted each other. Even main-push runs reported "No cache found", sccache was failing >50% of its writes (445/816 in Test), and it gave ~0% hit rate on Clippy, which it cannot cache at all. - Drop sccache everywhere; rely on rust-cache alone so the whole 10 GB budget serves the dependency cache. - Only save the cache from main (save-if) so PR/dependabot branches restore main's warm cache instead of evicting it. - Trim CI debug info to line-tables-only (dev+test, CI env only): faster codegen/linking and a much smaller target to cache. Local dev is unchanged. Free disk space step kept intentionally (past disk-pressure issues). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0176TaTFYYnLPUbQ6zthn7hD
Image builds swung 24-41 min because the cargo-chef `cook` layer (all deps, ~12 min in release) kept getting evicted from the shared 10 GB GHA cache: 8 parallel images each doing cache-to=gha,mode=max blow the budget and evict each other. Proof from one main run: meteroid-api's cook layer MISSED (11m49s rebuild, 26 min total) while the identical meteroid-scheduler cook layer was CACHED (9m53s total). - Move layer cache to the registry (GHCR), one ref per image: no 10 GB cap so the cook layer stays warm. Only main writes it, so branches can't evict it. - Wire mold as the linker: it was installed in every image but unused because the .cargo/config.toml mold config is .dockerignored. Set RUSTFLAGS=-Clink-arg=-fuse-ld=mold (verified locally: readelf shows the mold stamp in the linked binary) so release links stop using the slow default ld. Expected: meteroid-api (critical path) ~26 min cold -> ~10 min warm, roughly halving the workflow. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0176TaTFYYnLPUbQ6zthn7hD
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Attempt reducing cache miss / eviction & improving build time