test(infra): cover scheduler/hosted-service loop orchestration + idempotency (#243)

[thomasluizon]

Closes the remaining /prod-readiness tests-audit gaps for four background services with behaviour tests that drive the tick loop and orchestration (not just the pure helpers), each able to fail if the behaviour breaks. Timers are driven via the hosted lifecycle + injected abstractions — no wall-clock sleeps. Test-only; no source change was required.

What's added

OpenAiBatchPollerService

• PollPendingBatches_CompletedAndInFlightBatches_FinalizesCompletedAndLeavesInFlight — a completed batch is downloaded/persisted/finalized and its files deleted, while an in_progress batch stays Submitted and is never downloaded or cleaned up.
• ExecuteAsync_HostedLifecycle_RunsOnePollTickThenStopsGracefully — StartAsync runs one poll tick (signalled off the unit-of-work save), StopAsync cancels the loop cleanly (ExecuteTask completed, not faulted).

StreakFreezeAutoActivationService

• ActivateMissedDayFreezes_ExistingFreezeForMissedDate_DoesNotDoubleActivate — DB-backed idempotency: a freeze already present for the missed date blocks a second freeze/consume/push.
• ActivateMissedDayFreezes_GuardAlreadyRecordedForMissedDate_DoesNotReactivateOrRepush — a recorded SentStreakFreezeAlert guard blocks re-activation.
• ExecuteAsync_HostedLifecycle_RunsOneFreezePassThenStopsGracefully — the hosted loop activates one freeze then cancels cleanly.

PlayNotificationCleanupService

• ExecuteAsync_HostedLifecycle_RunsOnePurgePassThenStopsGracefully — the hosted loop runs the purge on tick and respects the 30-day Play / 90-day Stripe retention windows (records inside the window survive), then stops cleanly. The purge signal keys off the last DB log op to avoid racing the shared context.

XpAwardLogBackfillHostedService

• RunBackfillAsync_RunTwiceOnRestart_SecondRunSkipsAndDoesNotDoubleBackfill — idempotent across a restart: the second run is a no-op (flag/rows written once, "already complete" logged).
• StartAsync_CancelledDuringStartupDelay_StopsCleanlyWithoutRunningBackfill — StopAsync during the startup delay unwinds without running the backfill and without faulting.

Verification

• dotnet build — succeeds.
• dotnet test (Orbit.Infrastructure.Tests) — 1933 passed, 0 failed.

Refs thomasluizon/orbit-ui-mobile#243

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[claude]

Review Complete

Scope: PR #377 — test(infra): cover scheduler/hosted-service loop orchestration + idempotency (#243)
Recommendation: APPROVE

Summary

This is a test-only PR (four files under tests/Orbit.Infrastructure.Tests/Services/, no src/ changes) adding hosted-lifecycle and idempotency-behavior tests for OpenAiBatchPollerService, StreakFreezeAutoActivationService, PlayNotificationCleanupService, and XpAwardLogBackfillHostedService. Each new test was checked against its corresponding source service (src/Orbit.Infrastructure/Services/*.cs and ScheduledServiceBase.cs) to verify the assertions actually match production behavior and that the new StartAsync/await <signal>/StopAsync pattern is race-free.

Key correctness checks performed:

• ScheduledServiceBase.ExecuteAsync (src/Orbit.Infrastructure/Services/Hosting/ScheduledServiceBase.cs:24-48) runs the first tick immediately (no startup delay), so waiting on a signal fired mid-tick and then calling StopAsync is deterministic — BackgroundService.StopAsync awaits ExecuteTask to full completion, so there's no read/write race on shared DbContext/state between the background loop and the post-StopAsync assertions in any of the four new lifecycle tests.
• PlayNotificationCleanupServiceTests's new SignalOnStripePurgeLogger correctly keys off EventId == 5 (LogStripeEventsPurged), which only fires when purgedEvents > 0 (PlayNotificationCleanupService.cs:78) — the test seeds an expired Stripe row so the signal is guaranteed to fire.
• StreakFreezeAutoActivationService's two new idempotency tests correctly exercise the two guard checks in StageFreeze (StreakFreezeAutoActivationService.cs:130-134) and assert no freeze/push occurs.
• OpenAiBatchPollerServiceTests's completed-vs-in-flight test correctly matches ProcessBatchAsync's branching (completed → finalize/delete files; in_progress → left untouched, no delete).
• XpAwardLogBackfillHostedServiceTests's two new tests correctly cover restart idempotency (flag write blocks a second backfill) and cancellation during the 5s startup delay (no backfill runs, ExecuteTask completes non-faulted).

No narration comments were added. No TODO/workarounds/dead code/type-safety issues found.

Findings

Critical

None.

High

None.

Medium

None.

Low / Info

None posted (per the rubric's signal gate — no concretely actionable Medium+ issues surfaced; two very minor observations were considered and dropped as non-actionable).

Subagents

Agent	Verdict
security-reviewer	N/A — gate is "any src/ code changed"; this diff touches only tests/, no src/ files
contract-aligner	N/A — no DTO, Controller route, or shared-type change

Validation

Check	Result
Build (dotnet)	N/A — skipped per caller instruction; separate CI check (Build) covers it
Tests (dotnet)	N/A — skipped per caller instruction; separate CI check (Unit Tests) covers it

What's good

• Genuinely exercises the hosted BackgroundService lifecycle (StartAsync/StopAsync/ExecuteTask) rather than only the pure tick-body helper methods.
• Uses TaskCompletionSource signals tied to a real side effect (last DB op / mock invocation) instead of wall-clock Task.Delay sleeps — avoids flakiness while staying deterministic.
• New tests are tightly scoped to the specific idempotency/orchestration behavior claimed in the PR body and each one maps cleanly to a specific guard/branch in the corresponding service.

Recommendation

No changes requested — ready to merge (subject to the separate Build/Unit Tests/SonarCloud CI checks passing).