ci(pr): cut the Windows PR preflight leg's fixed costs (slice W) (#7378)

PR preflight process (windows-latest) is a CI Gate input and, after the
Bazel fixed-cost work, the merge-group critical path (ends at +247..+303s).
Step timings from merge group 37710264475 and seven same-repo PR runs:
45-115s setup (setup-go, then module and build cache tar extraction),
33-71s "Exercise the real Bash process boundary", 27-58s ./cmd/bd test
compile+link, 5-9s each for three more go test relinks.

- scripts/pr_preflight_test.go: the ~29 pr-preflight.sh runs (about 1.5s
  each under Git Bash on Windows, 42.8s total in the merge group) now run
  as parallel subtests. Each run already had its own temp dir, call log
  and env; the one test that t.Setenv's stays serial. Linux: 4.8s -> 0.8s.
- pr.yml: the hook-timeout step and the bd build fallback step become one
  go test of ./cmd/bd, saving a package re-hash and test-binary relink
  (5-9s on Windows for a sub-second test). Policy tests and the allowlist
  and topology docs follow the step name.
- warm-non-race-cache.sh: on Windows also compile the three CGO_ENABLED=0
  credential-fixture test packages the Windows-only fixture step runs, so
  main.yml's test-windows seeder covers them.

No step, test or gate input is dropped; the job id, name, matrix and
runner are unchanged.

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Julian Knutsen 2026-10-08 02:42:27 +00:00 • committed by GitHub
commit 233d151b37
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
7 changed files with 57 additions and 18 deletions

View file

@ -499,6 +499,13 @@ jobs:
# cold `cmd/bd` test compile in "Exercise generated Git hook timeout
# process boundary" (250-300s); see S2 below (the restore-non-race-cache
# step) for the fix that actually targets it. Reverted to one leg per OS.
#
# Windows leg budget with the caches warm (merge group 37710264475 and PR
# runs, 2026-10-07/08): 45-115s setup (setup-go, then the two cache
# restores, mostly tar extraction), 33-71s for TestPRPreflight (Git Bash
# process creation, now parallel subtests), 27-58s to compile and link
# ./cmd/bd's test binary, and about 25s for the rest. A second Windows leg
# would pay the setup again to save about 20s, so it stays one leg.
pr-preflight-platforms:
name: PR preflight process (${{ matrix.os }})
runs-on: ${{ matrix.runner }}
@ -565,24 +572,21 @@ jobs:
GOCACHE: ${{ runner.temp }}/go-cache/non-race
run: go test '-tags=gms_pure_go' -count=1 -run '^TestTestScriptPrebuiltBinaryContract$' ./scripts
- name: Exercise generated Git hook timeout process boundary
shell: bash
env:
GOCACHE: ${{ runner.temp }}/go-cache/non-race
run: go test '-tags=gms_pure_go' -count=1 -run '^TestGeneratedHookTimeoutProcessBoundary$' ./cmd/bd
# Plain go test's own bd build fallback (BEADS_TEST_BD_BINARY unset),
# One go test of ./cmd/bd runs the generated hook timeout boundary and
# plain go test's own bd build fallback (BEADS_TEST_BD_BINARY unset),
# which Bazel never takes, on every OS developers run go test on: the
# tests tools/bazel/equivalence_allowlist.txt lists as skipped under
# Bazel. Same build as the step above, so no second compile.
- name: Exercise go test's bd build fallback
# Bazel. They were two steps; each go test of ./cmd/bd re-hashes its
# ~1500 packages and relinks the test binary, which cost 5-9s on
# Windows for a fallback run of under one second.
- name: Exercise generated Git hook timeout process boundary and go test's bd build fallback
shell: bash
env:
GOCACHE: ${{ runner.temp }}/go-cache/non-race
run: go test '-tags=gms_pure_go' -count=1 -run '^(TestGOMODCACHENotUnderTestHome|TestGoBuildBDCommandIsCWDIndependent)$' ./cmd/bd
run: go test '-tags=gms_pure_go' -count=1 -run '^(TestGeneratedHookTimeoutProcessBoundary|TestGOMODCACHENotUnderTestHome|TestGoBuildBDCommandIsCWDIndependent)$' ./cmd/bd
# This step recompiles ./cmd/bd's test binary. It must share the restored
# GOCACHE with the hook-timeout step above (same tags, same platform
# GOCACHE with the hook-timeout/bd-build-fallback step above (same tags, same platform
# default CGO_ENABLED, see the script's header) so it is a relink rather
# than a second cold compile: without GOCACHE here it fell back to the
# runner's empty default cache and cold-compiled all of ./cmd/bd on every

View file

@ -927,8 +927,9 @@ Required` requires them to have run remotely and passed.
`tools/bazel/equivalence_allowlist.txt` holds only the two `cmd/bd`
tests of plain `go test`'s own bd build fallback, which Bazel never
takes; `pr-preflight-platforms` runs them on every OS ("Exercise go
test's bd build fallback"). `TestBazelOnlySkipsAreAllowlisted` (under
takes; `pr-preflight-platforms` runs them on macOS and Windows
("Exercise generated Git hook timeout process boundary and go test's bd
build fallback"). `TestBazelOnlySkipsAreAllowlisted` (under
Bazel too) requires every top-level test with a `TEST_SRCDIR`- or
`bazeltest.IsBazel()`-guarded `t.Skip` to have an allowlist `skip`
entry, and no test anywhere to run part of its checks under `go test`

View file

@ -28,3 +28,12 @@ go test -tags gms_pure_go -c -o /dev/null ./cmd/bd
go test -tags gms_pure_go -c -o /dev/null ./scripts
go test '-tags=integration,gms_pure_go' -c -o /dev/null ./scripts
go test '-tags=integration,gms_pure_go' -c -o /dev/null ./scripts/gitattributespolicy
# pr-preflight-platforms' Windows-only "Exercise credential command fixtures on
# Windows" step tests these with CGO_ENABLED=0, which the Windows runner's
# default (CGO_ENABLED=1, gcc on PATH) does not build.
if [[ "$(go env GOOS)" == windows ]]; then
for pkg in ./internal/testutil/credentialcmd ./internal/creds ./internal/storage/dolt; do
CGO_ENABLED=0 go test -tags gms_pure_go -c -o /dev/null "$pkg"
done
fi

View file

@ -616,8 +616,8 @@ func TestPRCIGateRequiresWindowsGlobalPrimeOverride(t *testing.T) {
func TestPRCIGateRequiresGeneratedHookTimeoutProcessBoundary(t *testing.T) {
const (
jobName = "pr-preflight-platforms"
stepName = "Exercise generated Git hook timeout process boundary"
stepCommand = "go test '-tags=gms_pure_go' -count=1 -run '^TestGeneratedHookTimeoutProcessBoundary$' ./cmd/bd"
stepName = "Exercise generated Git hook timeout process boundary and go test's bd build fallback"
stepCommand = "go test '-tags=gms_pure_go' -count=1 -run '^(TestGeneratedHookTimeoutProcessBoundary|TestGOMODCACHENotUnderTestHome|TestGoBuildBDCommandIsCWDIndependent)$' ./cmd/bd"
gateKey = "PR_PREFLIGHT_PLATFORMS"
)

View file

@ -481,7 +481,7 @@ func TestPRRunsGoTestsBazelSkips(t *testing.T) {
}
// The allowlist's remaining skip entries run in pr-preflight-platforms
// (every OS), each named in its go test selector.
fallback := pr.job(t, "pr-preflight-platforms").step(t, "Exercise go test's bd build fallback").Run
fallback := pr.job(t, "pr-preflight-platforms").step(t, "Exercise generated Git hook timeout process boundary and go test's bd build fallback").Run
for _, line := range strings.Split(readPolicyFile(t, root, "tools/bazel/equivalence_allowlist.txt"), "\n") {
body, _, _ := strings.Cut(line, "#")
f := strings.Fields(body)

View file

@ -95,6 +95,7 @@ type preflightRun struct {
func TestPRPreflightSupportsGitHubCLIWithoutClosingIssuesReferences(t *testing.T) {
t.Run("numeric lookup uses exact canonical GraphQL coordinates", func(t *testing.T) {
t.Parallel()
run := runPRPreflightWithFakeGH(t, preflightFixture{})
if run.err != nil {
t.Fatalf("pr-preflight failed with old gh field set: %v\n%s", run.err, run.output)
@ -137,6 +138,7 @@ func TestPRPreflightSupportsGitHubCLIWithoutClosingIssuesReferences(t *testing.T
})
t.Run("canonical PR URL and zero count preserve warning behavior", func(t *testing.T) {
t.Parallel()
run := runPRPreflightWithFakeGH(t, preflightFixture{
prArg: "https://github.com/octo/repo/pull/7",
graphQLResponse: closingIssuesZero,
@ -226,6 +228,7 @@ func TestPRPreflightSupportsGitHubCLIWithoutClosingIssuesReferences(t *testing.T
}
for _, test := range tests {
t.Run(test.name, func(t *testing.T) {
t.Parallel()
run := runPRPreflightWithFakeGH(t, test.fixture)
if exitCode(run.err) != 2 {
t.Fatalf("exit = %d, want 2; error=%v\n%s", exitCode(run.err), run.err, run.output)
@ -318,6 +321,7 @@ func TestPRPreflightChecksUseLatestRunPerName(t *testing.T) {
`]`
t.Run("stale failures and cancellations are superseded by newer green runs", func(t *testing.T) {
t.Parallel()
run := runPRPreflightWithFakeGH(t, preflightFixture{rollup: staleGreenRollup})
if !strings.Contains(run.output, "No failed or pending status checks reported.") {
t.Fatalf("expected stale rollup entries to be superseded:\n%s", run.output)
@ -328,6 +332,7 @@ func TestPRPreflightChecksUseLatestRunPerName(t *testing.T) {
})
t.Run("a genuinely failing latest run still blocks", func(t *testing.T) {
t.Parallel()
freshRedRollup := `[` +
`{"name":"CI Gate / Required","status":"COMPLETED","conclusion":"SUCCESS","completedAt":"2026-07-27T10:00:00Z"},` +
`{"name":"CI Gate / Required","status":"COMPLETED","conclusion":"FAILURE","completedAt":"2026-07-28T10:00:00Z"}` +
@ -339,6 +344,7 @@ func TestPRPreflightChecksUseLatestRunPerName(t *testing.T) {
})
t.Run("a commit status cannot supersede a same-named check run", func(t *testing.T) {
t.Parallel()
// GitHub treats a check run and a commit status sharing a name as
// independent required gates; both must pass.
mixedRollup := `[` +
@ -352,6 +358,7 @@ func TestPRPreflightChecksUseLatestRunPerName(t *testing.T) {
})
t.Run("identically named jobs in different workflows stay independent", func(t *testing.T) {
t.Parallel()
crossWorkflowRollup := `[` +
`{"__typename":"CheckRun","workflowName":"Main","name":"Test","status":"COMPLETED","conclusion":"FAILURE","completedAt":"2026-07-28T10:00:00Z"},` +
`{"__typename":"CheckRun","workflowName":"Nightly","name":"Test","status":"COMPLETED","conclusion":"SUCCESS","completedAt":"2026-07-28T11:00:00Z"}` +
@ -363,6 +370,7 @@ func TestPRPreflightChecksUseLatestRunPerName(t *testing.T) {
})
t.Run("an in-progress rerun with a zero completion time supersedes an old failure as pending", func(t *testing.T) {
t.Parallel()
// gh exports absent CheckRun times as the Go zero time, not null.
rerunRollup := `[` +
`{"__typename":"CheckRun","workflowName":"Main","name":"CI","status":"COMPLETED","conclusion":"FAILURE","startedAt":"2026-07-27T09:00:00Z","completedAt":"2026-07-27T10:00:00Z"},` +
@ -396,6 +404,7 @@ func TestPRPreflightPRGateHealthSample(t *testing.T) {
`]`
t.Run("all failed across 3+ heads with 5+ decisive runs warns and names the workflow", func(t *testing.T) {
t.Parallel()
run := runPRPreflightWithFakeGH(t, preflightFixture{baseRunList: greenBase, prGateRunList: sixRunsThreeHeadsAllFailed})
if run.err != nil {
t.Fatalf("expected warn-only exit, got error: %v\n%s", run.err, run.output)
@ -407,6 +416,7 @@ func TestPRPreflightPRGateHealthSample(t *testing.T) {
})
t.Run("stays warn-only even under PR_PREFLIGHT_BLOCK_RED_BASE=1", func(t *testing.T) {
t.Parallel()
// The warn-only property is the load-bearing one: mybd's pr-babysit
// classifies any unrecognized [block] line as a genuine merge blocker
// and parks merge lanes on it, and it withholds the base-fix
@ -426,6 +436,7 @@ func TestPRPreflightPRGateHealthSample(t *testing.T) {
})
t.Run("one success in the workflow's sample stays silent", func(t *testing.T) {
t.Parallel()
mostlyFailedOneSuccess := strings.Replace(sixRunsThreeHeadsAllFailed, `"conclusion":"failure","headBranch":"branch-c","workflowName":"PR CI","workflowDatabaseId":101,"createdAt":"2026-07-28T05:00:00Z"`, `"conclusion":"success","headBranch":"branch-c","workflowName":"PR CI","workflowDatabaseId":101,"createdAt":"2026-07-28T05:00:00Z"`, 1)
run := runPRPreflightWithFakeGH(t, preflightFixture{baseRunList: greenBase, prGateRunList: mostlyFailedOneSuccess})
if run.err != nil {
@ -437,6 +448,7 @@ func TestPRPreflightPRGateHealthSample(t *testing.T) {
})
t.Run("all failed but only 2 distinct heads stays silent", func(t *testing.T) {
t.Parallel()
twoHeadsAllFailed := `[` +
`{"conclusion":"failure","headBranch":"branch-a","workflowName":"PR CI","workflowDatabaseId":101,"createdAt":"2026-07-28T10:00:00Z","url":"https://github.com/o/r/actions/runs/1"},` +
`{"conclusion":"failure","headBranch":"branch-a","workflowName":"PR CI","workflowDatabaseId":101,"createdAt":"2026-07-28T09:00:00Z","url":"https://github.com/o/r/actions/runs/2"},` +
@ -455,6 +467,7 @@ func TestPRPreflightPRGateHealthSample(t *testing.T) {
})
t.Run("empty sample stays silent", func(t *testing.T) {
t.Parallel()
run := runPRPreflightWithFakeGH(t, preflightFixture{baseRunList: greenBase, prGateRunList: "[]"})
if run.err != nil {
t.Fatalf("expected exit 0, got error: %v\n%s", run.err, run.output)
@ -465,6 +478,7 @@ func TestPRPreflightPRGateHealthSample(t *testing.T) {
})
t.Run("mixed workflows: only the all-red workflow fires, and it is named", func(t *testing.T) {
t.Parallel()
mixedWorkflows := `[` +
`{"conclusion":"success","headBranch":"branch-a","workflowName":"PR CI","workflowDatabaseId":101,"createdAt":"2026-07-28T12:00:00Z","url":"https://github.com/o/r/actions/runs/201"},` +
`{"conclusion":"success","headBranch":"branch-b","workflowName":"PR CI","workflowDatabaseId":101,"createdAt":"2026-07-28T11:30:00Z","url":"https://github.com/o/r/actions/runs/202"},` +
@ -488,6 +502,7 @@ func TestPRPreflightPRGateHealthSample(t *testing.T) {
})
t.Run("undetermined base health skips the sample", func(t *testing.T) {
t.Parallel()
// With no decisive base runs at all, preflight has just warned that it
// could not determine base health; diagnosing "red for every PR while
// the base branch shows green" would contradict that line.
@ -501,6 +516,7 @@ func TestPRPreflightPRGateHealthSample(t *testing.T) {
})
t.Run("red base skips the sample even when a workflow would otherwise fire", func(t *testing.T) {
t.Parallel()
redBaseRunList := `[{"conclusion":"failure","workflowName":"CI","createdAt":"2026-07-28T09:00:00Z","url":"https://github.com/o/r/actions/base-1"}]`
run := runPRPreflightWithFakeGH(t, preflightFixture{
baseRunList: redBaseRunList,
@ -518,6 +534,7 @@ func TestPRPreflightPRGateHealthSample(t *testing.T) {
})
t.Run("unreadable/non-JSON sample stays silent", func(t *testing.T) {
t.Parallel()
run := runPRPreflightWithFakeGH(t, preflightFixture{baseRunList: greenBase, prGateRunList: "gh: rate limit exceeded"})
if run.err != nil {
t.Fatalf("expected exit 0, got error: %v\n%s", run.err, run.output)
@ -528,6 +545,7 @@ func TestPRPreflightPRGateHealthSample(t *testing.T) {
})
t.Run("BASE_RUN_LIST positively drives the base-branch dispatch, not the PR-gate one", func(t *testing.T) {
t.Parallel()
greenBaseRunList := `[{"conclusion":"success","workflowName":"CI","createdAt":"2026-07-28T12:00:00Z","url":"https://github.com/o/r/actions/base-2"}]`
run := runPRPreflightWithFakeGH(t, preflightFixture{baseRunList: greenBaseRunList})
if run.err != nil {
@ -542,6 +560,13 @@ func TestPRPreflightPRGateHealthSample(t *testing.T) {
})
}
// runPRPreflightWithFakeGH runs scripts/pr-preflight.sh once in its own temp
// dir, call log and environment, so callers that do not t.Setenv may run it
// from parallel subtests. They do: on Windows, Git Bash process creation makes
// each run cost about 1.5s, and the ~29 runs in this file used to take 42s of
// pr-preflight-platforms' Windows leg one after another (gastownhall/beads
// merge group 37710264475). TestPRPreflightFixtureIgnoresHostShellAndGitHubState
// stays serial because it rewrites the process environment.
func runPRPreflightWithFakeGH(t *testing.T, fixture preflightFixture) preflightRun {
t.Helper()

View file

@ -14,7 +14,7 @@
# Plain go test's own bd build fallback (BEADS_TEST_BD_BINARY unset), which
# Bazel never takes (bd comes in through BEADS_TEST_BD_BINARY). pr.yml's
# pr-preflight-platforms runs both on Linux, macOS and Windows ("Exercise go
# test's bd build fallback").
# pr-preflight-platforms runs both on macOS and Windows ("Exercise generated
# Git hook timeout process boundary and go test's bd build fallback").
cmd/bd TestGOMODCACHENotUnderTestHome skip # no in-test go build under Bazel; pr-preflight-platforms runs it
cmd/bd TestGoBuildBDCommandIsCWDIndependent skip # no in-test go build under Bazel; pr-preflight-platforms runs it