ci: one cross-build runner per target, one target per OS on PRs - #951
Open
u9g wants to merge 5 commits into
Open
ci: one cross-build runner per target, one target per OS on PRs#951u9g wants to merge 5 commits into
u9g wants to merge 5 commits into
Conversation
The single cross-build job compiled all four zig targets in one goreleaser invocation on one 4-vCPU runner: ~560 C/C++ translation units (webrtc APM + portaudio) x 4 targets, ~17 minutes of the job's ~19. Split into a matrix so each target gets its own runner, and drop the arm64 targets on pull requests -- main still builds all four before release.
setup-go's default key is setup-go-<os>-<arch>-<image>-go-<ver>-<hash(go.sum)>, with no workflow or job component, so cross-build and the Go Test workflow compete for one key. Test wins -- cross-build's post step spends 61s tarring 700MB and then reports: Failed to save: Unable to reserve cache with key setup-go-Linux-x64-ubuntu24- go-1.26.3-d55f71f0... So the zig cross objects are never stored, and what cross-build restores each run is Test's native-gcc `go test -race` cache, which no zig target can use: a 706MB restore that changed the build from 19m45s to 19m19s. Scope the cache per target with a restore-keys prefix so a go.sum bump keeps the webrtc/portaudio C++ objects.
Both jobs hit the same setup-go key collision as cross-build: the key has no workflow or job component, so test.yaml claims it first and these jobs restore objects they can't use -- native-gcc for the zig Windows cross-build, `-race` for the darwin build. windows.yaml's cross-build recompiles the same ~560 webrtc/portaudio TUs plus a test binary per package at 6m12s a run, and runs twice per push.
rektdeckard
approved these changes
Aug 28, 2026
rektdeckard
left a comment
rektdeckard
left a comment
Member
There was a problem hiding this comment.
Looks good, I think we can live with ARM build tests only on main
| ${{ case( | ||
| github.event_name == 'pull_request', | ||
| fromJSON('["lk-linux-amd64", "lk-windows-amd64"]'), | ||
| fromJSON('["lk-linux-amd64", "lk-linux-arm64", "lk-windows-amd64", "lk-windows-arm64"]') |
Member
There was a problem hiding this comment.
ARM builds only happen on merge to main then? I suppose that's OK since these take the lion's share of the time on PRs, but we do lose some confidence.
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
2 participants
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.
The
Cross-build linux+windows (zig)job takes ~19 minutes, ~17 of which are a singleGoReleaser snapshot buildstep. It compiles all four zig cross targets in one goreleaser invocation on oneubuntu-latest(4 vCPU) runner:.cc/.cunderpkg/apm+ 114 underpkg/portaudio≈ 560 translation units-O2throughzig ccNothing is reused between runs either:
setup-go's cache key is an exactgo.sumhash with norestore-keys, and even when it hits (699 MB restored in run 33115504417) the step still took 19m19s vs 19m45s cold.This PR does the cheap structural half:
lk-linux-amd64,lk-windows-amd64); pushes tomainstill build all four, so nothing reaches a release untested.The
.crosscache key gains the target id, since each runner now populates only its own.cross/<target>/.Expected: ~19 min → ~5 min on PRs.
Not addressed here: why a warm 699 MB
GOCACHEbuys nothing. Worth a follow-up.