CI: give Windows a persistent Cargo target and enable sccache by jdetter · Pull Request #5946 · clockworklabs/SpacetimeDB

jdetter · 2026-09-16T04:30:49Z

This pull request and its description were written by AI.

Description of Changes

Make Windows self-hosted runners reuse the warmed Cargo target baked into the runner image:

use C:/actions-runner/_work/target, outside the checkout directory;
enable Windows sccache and report its statistics;
keep GitHub-hosted windows-latest, Linux, and macOS behavior unchanged.

The image-warming implementation remains in infra PR 1349 for now and is expected to move into a shared cross-platform Cargo command. This PR only consumes the resulting cache layout, so that implementation change will not require another workflow change.

This is the Windows version of #5848.

Performance

A fresh vm27 image warmed from master commit 190670675f8a26849d15670faac598764cd468eb produced these results in staging run 36007072082:

Upload build artifacts: 5m48s master median → 2m38s (54.6% faster); sccache 84.38% (27 hits, 5 misses), up from vm26's 2.63%.
Build smoketests: 3m33s master median → 1m31s (57.3% faster); sccache 83.78% (31 hits, 6 misses), up from vm26's 0.23%.

This confirms the low vm26 hit rates were primarily cache-age/source-drift rather than a Windows-specific sccache problem. Results are from one staging run, so timings remain subject to runner variance.

API and ABI breaking changes

None.

Rollback safety impact

n/a

Expected complexity level and risk

1 — low risk. CI configuration only; an image without the warm cache falls back to a cold but working build.

Testing

.github/workflows/ci.yml parses successfully
Both cache-consuming Windows build jobs passed against vm27 in Actions run 36007072082
Temporary staging labels removed; final jobs target spacetimedb-windows-runner

Windows CI rebuilds everything from scratch on every run for one reason: the set-cargo-target-dir anchor sends Linux to $HOME/actions-runner/_work/target, outside the workspace, and everything else to $GITHUB_WORKSPACE/target -- which checkout wipes each run. Windows also never had sccache enabled at all. Point Windows at C:/actions-runner/_work/target and enable sccache there, mirroring the Linux setup already on master. The Windows branch is gated on the runner work directory existing, so GitHub-hosted windows-latest jobs keep their current workspace-local behaviour. RUSTC_WRAPPER is set to an explicit C:/sccache/sccache.exe rather than the result of `command -v`: these steps run under Git Bash, whose `command -v` returns an MSYS path (/c/sccache/sccache.exe) that Cargo, a native Windows process, cannot execute. The `command -v` guard is kept only to detect whether the runner image was warmed, so an unwarmed image degrades to a cold but working build instead of failing. The Windows-only actions/cache entries for rusty_v8 and vendored OpenSSL are deliberately left in place for this iteration. Requires a runner image warmed by warm-cache-windows.ps1 in clockworklabs/infra. Without it these paths are simply empty and builds behave as they do today.

Points upload-build-artifacts-windows, smoketest_build_windows and smoketest_partitions_windows at spacetimedb-windows-runner-staging so this PR can be qualified against runner image vm26, which carries the warmed sccache store and dependency-only Cargo target. This commit exists only to measure the effect and MUST be reverted before merge. Leaving it in would send all production Windows CI to the staging pool. The windows-latest matrix in the update-flow job is deliberately untouched -- it runs on GitHub-hosted runners and has no warmed image. Revert with: git revert <this commit>

bfops · 2026-09-22T14:02:18Z

Do you know why the sccache hit rates were so low?

bfops

LGTM!

Route the three self-hosted Windows CI jobs to the staging pool so runner image vm27 can be measured against the prior vm26 qualification.\n\nThis commit is temporary and will be reverted after collecting sccache statistics and job timings.

This reverts commit 886c95f.

jdetter requested review from bfops and cloutiertyler as code owners September 16, 2026 04:30

jdetter marked this pull request as draft September 16, 2026 04:35

jdetter added 5 commits September 16, 2026 00:12

Merge branch 'master' into tars/windows-cache-parity

dc0ecc6

Merge branch 'master' into tars/windows-cache-parity

7dbdfb3

CI: restore production Windows runner labels

64ce9ee

CI: retry after flaky internal smoketest timeout

24de161

jdetter marked this pull request as ready for review September 22, 2026 03:38

jdetter requested a review from rekhoff September 22, 2026 03:40