[IAmJSD]

Closes #987. This follows up #971 (issue #962), which merged at 83459e4 before this commit landed. The commit is cherry-picked unchanged onto main. Test-only.

Changes

1. Control subtest (background_collection_test.go):
   • The store stub now signals started and then blocks on an unbuffered release channel. It's a small blockingRefreshStore helper, and cleanup closes done, so nothing leaks.
   • The non-Lambda control waits for started and asserts collecting is held while the store call is blocked. It then sends on release with a 2s timeout, and asserts collecting is released.
   • If the stub stops blocking, nothing receives on release, so the control fails every time instead of depending on timing.
2. Lambda subtest: the 200ms wait is gone. The guard returns before the synchronous claim of the flag, so the test asserts straight after the call: collecting is false, a non-blocking receive on started gets nothing, and GetGlobalConfig was never called. It now runs in about 0.00s.
3. Guard comments: each test has a one-line comment naming the guard it pins.

Guard	Pinned by
ListRecommendations mayCollect && !s.isLambda (scheduler.go:1183)	LambdaSkipsBackgroundRefresh (also caught by ColdStart_Lambda_NeverMarksOrCollects)
kickColdStartCollection isLambda (:1452)	ColdStart_Lambda_NeverMarksOrCollects only
maybeKickBackgroundRefresh isLambda (:1502)	TestMaybeKickBackgroundRefresh_LambdaGuard/lambda_never_kicks only

Mutation probes

Each mutation was applied, run, and reverted.

Mutation	Result
Guard at 1183 removed	LambdaSkipsBackgroundRefresh and ColdStart_Lambda_NeverMarksOrCollects fail
Guard at 1452 removed	Only ColdStart_Lambda_NeverMarksOrCollects fails
Guard at 1502 removed	Only LambdaGuard/lambda_never_kicks fails; the control still passes
Stub no longer blocks (the wait on release deleted; the reviewer's mutant)	The control fails with "the refresh's store call did not wait for release; the harness no longer blocks it", in 100 of 100 runs with -count=100

Verification

All runs used AWS_CONFIG_FILE=/dev/null AWS_SHARED_CREDENTIALS_FILE=/dev/null. These are unit tests with mocks.

• go test -race -count=1 ./internal/scheduler/: ok, both on the original branch and on this branch, which is cut from main at f44330b.
• go test -race -count=50 -run 'Lambda|LambdaGuard' ./internal/scheduler/: ok.
• go build ./... passes, and go vet and gofmt are clean.
• golangci-lint v2.10.1 reports 0 issues.
• git-secrets (--register-aws plus .gitallowed) is clean on both changed files.
• git merge-tree against current origin/main is clean.

Maintainer: please mirror #962's labels onto #987 and this PR (impact/internal, triaged, priority/p3, severity/low, urgency/this-quarter, effort/xs, type/bug).

🤖 Generated with Claude Code

[coderabbitai]

Warning

Review limit reached

• Run on-demand review

_{This review includes 2 billable files and costs up to $0.50.}

Or wait 3 minutes for your next included review.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

• Configuration used: Repository: LeanerCloud/cloud-commitments-platform/.coderabbit.yaml
• Review profile: CHILL
• Plan: Advanced
• Run ID: 4823c931-b33e-45e5-b311-911860bcc202

📥 Commits

Reviewing files that changed from the base of the PR and between adcd80c and 3f2dafa.

📒 Files selected for processing: 2

• internal/scheduler/background_collection_test.go
• internal/scheduler/scheduler_test.go

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

[cristim]

Independent review of 3f2dafa: no actionable findings

Findings: none.

What I attacked (worktree at the SHA, all probes reverted):

• Guard probe: disabled the guard at scheduler.go:1502 (if false && s.isLambda). TestMaybeKickBackgroundRefresh_LambdaGuard fails 3/3 in the "lambda never kicks" subtest ("collecting flag was claimed on Lambda").
• Harness probe: removed the release/done wait in blockingRefreshStore. The control fails 3/3 after 2s ("store call did not wait for release"). This closes #971 note 1 deterministically: the send on release only succeeds if the store call is blocked.
• Control asserts collecting is held while blocked, then releases and requires the flag to clear.
• Leaks/hangs: started is buffered (1), so a wrongly kicked goroutine cannot block on send; t.Cleanup closes done, which frees the goroutine on every path (including t.Fatal). go test -race -count=10 ./internal/scheduler/ passes.
• The 200ms time.After on the Lambda path is gone (note 2); the guard-to-test mapping comments cover note 3 (lines 1183, 1452, 1502 match the pins).
• scheduler_test.go: only comments added. staleListRecommendations still uses WithBackgroundCollection(context.Background()) (scheduler_test.go:950). Merge with origin/main is clean (git merge-tree).
• Test-only, +76/-35 (additions well under 400).

Gate status:

• Independent review: this comment, at the SHA.
• CI on the SHA: all checks pass, CodeRabbit "rate limited" (optional per gate; exact-SHA local verification above covers it).
• Local verification: macOS, real scenario, fail-before/pass-after via the two mutations above.
• Labels: PR has none; #987 has no triage labels either, so there is nothing to mirror. Gap: #987 needs triage (NON-BLOCKING, maintainer action).
• Author is an external contributor; nothing pushed to their branch.

[cristim]

Correction to the independent review above (3f2dafa): its gate section says all checks pass, but that was wrong when it was posted. Unit Tests, Integration Tests, Build Docker Image and Security Scanning were still pending. The code verdict (no actionable findings) stands. The merge waits for CI to be green on this SHA, and for issue #987 to get triage labels to mirror (it has none).

[cristim]

Gate record for 3f2dafa (origin/main CLAUDE.md "Review gate"):

• Independent review of 3f2dafa: no actionable findings. All three test(scheduler): make the Lambda no-refresh tests bite again #971 follow-up notes (Follow-ups from #971 review: make the LambdaGuard control catch a non-blocking harness #987) are addressed: the 200ms wait is gone, the control is deterministic, and the pin comments match the guards.
• CI on 3f2dafa: CI Success and Run pre-commit hooks pass. The earlier review note that claimed green was corrected on the PR. This is the actual completed state.
• Local verification (macOS):
  • Disabling the Lambda guard (scheduler.go:1502) fails the "lambda never kicks" subtest 3/3.
  • Making the harness non-blocking fails the non-Lambda control 3/3.
  • go test -race -count=10 ./internal/scheduler/ passes.
  • staleListRecommendations still uses WithBackgroundCollection, so the test(scheduler): kick the stale refresh with an unrestricted reader #960 fix is intact. Test-only.
• CR skipped: quota-limited; covered by the independent review plus local verification.
• Labels mirrored from Follow-ups from #971 review: make the LambdaGuard control catch a non-blocking harness #987.

Merging at 3f2dafa with --match-head-commit.