CI: spread the slowest tests across the test shards (#3646)

* CI: give the slowest tests a shard each

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* CI: spread the slow tests on pull requests too, and keep both junit files

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* CI: pull requests skip the goldens and the action matrix sample unless the kernel changes

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Docs: which pull requests run the goldens

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* CI: run the tests in one job on a big runner, slowest first

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* CI: allow the 32 vCPU runner label in actionlint, read junit.xml in the golden hint

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
bherbruckandClaude Opus 5.5 authored and GitHub committed 2026-10-20 20:41:58 -04:00
1 parent 4f7ff8c4d8
commit 0379ee76c6
5 files changed
+39 -29

No files matched your search

+7 -6
View File
@@ -1,6 +1,5 @@
# cargo-nextest settings for CI (`cargo xtask test` does not use nextest).
# CI builds the tests once and runs them in shards: 2 for a pull request, 4 for main and the nightly
# (.github/workflows/ci.yml).
# CI builds the tests once and runs them in one job on a 32 vCPU runner (.github/workflows/ci.yml).
[profile.ci]
# Heavy tests (one to twelve minutes each on a CI runner) that pull requests skip. They run on
@@ -34,12 +33,14 @@ slow-timeout = { period = "60s" }
[profile.ci.junit]
path = "junit.xml"
# The action matrix sample runs its designs in parallel worker processes, one per CPU. Run with
# other tests beside it (the heavy ones on main), a pi-case hole that takes 8 s alone (#952 makes
# it a real hole) went past the matrix's 90 s action limit. It runs on its own.
# The action matrix sample runs its designs in parallel workers (ACTION_MATRIX_JOBS, set to match in
# ci.yml). Its workers reserve that many of nextest's threads, so the cores are not oversubscribed
# (a pi-case hole that takes 8 s alone went past the matrix's 90 s action limit when they were),
# and it starts first, beside the other slow tests.
[[profile.ci.overrides]]
filter = "binary(action_matrix) & test(=action_matrix_sample)"
threads-required = "num-cpus"
threads-required = 12
priority = 100
# The slowest tests start first, so a shard does not end with one of them running alone on a
+1
View File
@@ -7,4 +7,5 @@ self-hosted-runner:
- blacksmith-4vcpu-ubuntu-2404
- blacksmith-8vcpu-ubuntu-2404
- blacksmith-16vcpu-ubuntu-2404
- blacksmith-32vcpu-ubuntu-2404
- blacksmith-8vcpu-windows-2025
+28 -20
View File
@@ -8,7 +8,7 @@
# whole workspace, as do pushes to main that change code, the nightly run and Run workflow.
# Documentation, packaging, the other workflows and the GitHub bots' scripts select nothing:
# no native job runs for them, on pull requests or on main.
# Tests are built once and run in shards. A few heavy tests (see .config/nextest.toml) are
# Tests are built once and run in one job on a big runner, slowest first. A few heavy tests (see .config/nextest.toml) are
# skipped on pull requests and run on pushes to main and nightly. The Windows jobs (D3D12 shaders
# and app startup), the reference tests and the doc tests run on main and nightly only, not on
# pull requests (the fast lane, in Select checks); a break there is fixed forward on main. Only
@@ -29,8 +29,7 @@
# reads `ci` from the newest run of a commit, so a run without it would block a PR that passed.
# Cross-compilation is opt-in through Run workflow; release builds remain in release.yml.
# Every Linux job runs on Blacksmith's Ubuntu 24.04 runners (blacksmith-<N>vcpu-ubuntu-2404), sized
# per job: 16 vCPU for Build tests, Clippy and the merge queue's test shards, 8 for compiles, the
# other test shards, doc and reference tests, 4 for the light checks, 2 for Select checks and ci.
# per job: 32 vCPU for the tests, 16 for Build tests and Clippy, 8 for compiles, doc and reference tests, 4 for the light checks, 2 for Select checks and ci.
# The Windows shader job runs on Blacksmith's Windows Server 2025 runners (blacksmith-8vcpu-windows-2025);
# the Windows app startup smoke stays on windows-latest, which has a desktop session that presents frames.
# Pull requests, forks included, the merge queue, pushes to main and the nightly run all use them:
@@ -178,9 +177,12 @@ jobs:
git diff --no-renames --name-only -z "$BASE_SHA...$HEAD_SHA" > "$RUNNER_TEMP/ci-paths"
while IFS= read -r -d '' path; do
case "$path" in crates/kernel/*|vendor/truck-*) kernel=true ;; esac
# What can change a robustness golden: the kernel and the crates a design is built with.
# What a pull request re-checks the goldens for: the kernel, truck and the goldens
# themselves (a re-record is checked before it is queued). The goldens are frozen, so
# on any other change they would only re-check unchanged results; the merge queue,
# main and the nightly run every seed before anything lands.
case "$path" in
crates/kernel/*|crates/engine/*|crates/doc/*|crates/sketch/*|crates/geom/*|vendor/truck-*|tests/robustness/*) goldens=true ;;
crates/kernel/*|vendor/truck-*|tests/robustness/*) goldens=true ;;
esac
# What the Windows D3D12 jobs exercise: the GPU crate and its shaders, the app and
# its startup, the viewport's wgpu glue, the vendored egui-wgpu, their scripts.
@@ -387,22 +389,22 @@ jobs:
run: sccache --show-stats
tests:
name: Native Linux tests (${{ matrix.shard }}/4)
name: Native Linux tests
needs: [changes, checks, build-tests]
if: ${{ !cancelled() && needs.changes.outputs.native == 'true' && (github.event_name != 'pull_request' || needs.checks.result == 'success') && needs.build-tests.result == 'success' }}
runs-on: ${{ github.event_name == 'merge_group' && 'blacksmith-16vcpu-ubuntu-2404' || 'blacksmith-8vcpu-ubuntu-2404' }}
# One job on a big runner: nextest hands tests to its threads as they free up, the slowest
# first (priority in .config/nextest.toml), so they start at once and no core idles at the
# end. Four shards cut by `slice` each paid the setup (about 6 minutes) and a shard that drew
# several slow tests ran for 20 minutes while the others waited.
runs-on: blacksmith-32vcpu-ubuntu-2404
timeout-minutes: 60
strategy:
fail-fast: false
matrix:
# Pull requests test their affected crates in 2 shards; main and the nightly use 4.
shard: ${{ fromJSON(github.event_name == 'pull_request' && '[1, 2]' || '[1, 2, 3, 4]') }}
env:
# Pictures (snapshots, screenshots, ui.render, the viewport's offscreen tests) are drawn on
# the GPU. The runner has none: Mesa's lavapipe (software Vulkan) draws them, and a test that
# would skip for want of an adapter fails instead.
SOLIDCRAFT_GPU: software
SOLIDCRAFT_REQUIRE_GPU: "1"
ACTION_MATRIX_JOBS: "12"
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
- name: Install a software Vulkan driver (lavapipe)
@@ -420,30 +422,36 @@ jobs:
env:
EVENT_NAME: ${{ github.event_name }}
GOLDENS: ${{ needs.changes.outputs.goldens }}
SHARD: ${{ matrix.shard }}
SHARDS: ${{ github.event_name == 'pull_request' && 2 || 4 }}
# The robustness golden seeds a pull request skips: each takes one to twenty minutes under
# CI load, together about four fifths of the goldens' time. Pushes to main, the merge
# queue and the nightly run every seed.
SLOW_SEEDS: 81|45|20|24|95|02|57|40|18|48|32|46|85|61|77|34|14
run: |
# Pull requests skip the heavy tests (default-filter in .config/nextest.toml).
filter=()
# Pull requests skip the heavy tests (default-filter in .config/nextest.toml); `default()`
# below is that filter, so a pull request still runs only what it ran before.
base='default()'
if [ "$EVENT_NAME" = pull_request ]; then
# The action matrix sample (about 6 minutes on all cores) runs in the merge queue, on
# main and nightly, not on every pull request push.
base='default() & not (binary(action_matrix) & test(=action_matrix_sample))'
fi
if [ "$EVENT_NAME" != pull_request ]; then
# The boolean property tests draw new random inputs every run; they run only in the
# nightly property-tests job, so a known kernel bug cannot turn main red at random.
filter=(--ignore-default-filter -E 'not (package(solidcraft-kernel) & binary(boolean_props))')
base='not (package(solidcraft-kernel) & binary(boolean_props))'
elif [ "$GOLDENS" = true ]; then
# A pull request that can change a robustness golden (tests/robustness/goldens.json)
# also checks every seed but the slow ones, in the same run so they share its cores.
filter=(--ignore-default-filter -E "default() | (package(solidcraft-engine) & test(/^tests::robustness_goldens::/) & not test(/^tests::robustness_goldens::seed_($SLOW_SEEDS)\$/))")
base="(default() & not (binary(action_matrix) & test(=action_matrix_sample))) | (package(solidcraft-engine) & test(/^tests::robustness_goldens::/) & not test(/^tests::robustness_goldens::seed_($SLOW_SEEDS)\$/))"
fi
junit=${CARGO_TARGET_DIR:-target}/nextest/ci/junit.xml
rm -f "$junit"
# Extract into ./target: tests that run a built binary (env!("CARGO_BIN_EXE_*")) have
# the build job's path, which is the same checkout path here.
run=(cargo nextest run --archive-file tests.tar.zst --workspace-remap . --extract-to . --extract-overwrite)
if ! "${run[@]}" --profile ci --partition "slice:$SHARD/$SHARDS" --no-tests=pass "${filter[@]}"; then
status=0
"${run[@]}" --profile ci --no-tests=pass --ignore-default-filter -E "$base" || status=1
if [ "$status" != 0 ]; then
# A failed golden gets the hint to re-record it (junit.xml lists the failures).
if [ -f "$junit" ] && tr '\n' ' ' < "$junit" | grep -q '<testcase name="tests::robustness_goldens::[^"]*"[^>]*>[[:space:]]*<failure'; then
echo "::error::A robustness golden changed. If the new result is intended, re-record it in this pull request and say why:"
@@ -455,7 +463,7 @@ jobs:
if: always()
uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7.0.2
with:
name: junit-${{ matrix.shard }}
name: junit
path: ${{ env.CARGO_TARGET_DIR || 'target' }}/nextest/ci/junit.xml
retention-days: 7
+1 -1
View File
@@ -199,7 +199,7 @@ Check that validation after merging. They are listed in
`.config/nextest.toml` and the [testing page](docs/book/src/dev/testing.md):
- `solidcraft-engine`: `tests::hostile_params_never_panic`, everything in `tests::robustness` and
the seeds of `tests::robustness_goldens` (`seed_00` … `seed_95`; pull requests matching the workflow's geometry paths run all but the slowest, `SLOW_SEEDS` in `ci.yml`),
the seeds of `tests::robustness_goldens` (`seed_00` … `seed_95`; pull requests that change the kernel, truck or the goldens run all but the slowest, `SLOW_SEEDS` in `ci.yml`),
`pattern_suppress_tests::enclosure_arc_extinction_keeps_a_real_sharp_inner_lip_after_resize_and_reopen`,
`pattern_suppress_tests::enclosure_datums_follow_wall_height_and_board_depth_with_material_proofs`,
`cmd::plastic::tests::enclosure_snap_fit_follows_resized_front_wall`, the `atomic_commands` contract tests
+2 -2
View File
@@ -77,8 +77,8 @@ Pull requests skip a few heavy tests (each takes one to twelve minutes); they ru
`main`, nightly, and in `cargo xtask test`:
- `solidcraft-engine`: `tests::hostile_params_never_panic`, everything in `tests::robustness` and
the seeds of `tests::robustness_goldens` (`seed_00` … `seed_95`; pull requests that can change
them run all but the slowest, `SLOW_SEEDS` in `ci.yml`),
the seeds of `tests::robustness_goldens` (`seed_00` … `seed_95`; pull requests that change the kernel, truck or
the goldens run all but the slowest, `SLOW_SEEDS` in `ci.yml`),
`pattern_suppress_tests::enclosure_arc_extinction_keeps_a_real_sharp_inner_lip_after_resize_and_reopen`,
`pattern_suppress_tests::enclosure_datums_follow_wall_height_and_board_depth_with_material_proofs`,
`cmd::plastic::tests::enclosure_snap_fit_follows_resized_front_wall`, the `atomic_commands` contract tests