Stress fixes: framerate/PerGroup timeouts, overlay tick race, DataGrid scroll poll by codemonkeychris · Pull Request #397 · microsoft/microsoft-ui-reactor

codemonkeychris · 2026-05-24T04:22:52Z

Summary

Three independent stress-flake fixes, bundled per the cadence set by #396.
Each is the implementation of a Net-next-step from INVESTIGATION.md
post-#396.

Cluster T (Path A) — FixtureTimeout => 30s on three render-pump-heavy
fixtures whose mandatory wall-clock floors are tight against the default
15 s budget on loaded CI runners. Not deadlocks; budget violations.
- DataGridParityFixtures.HookPagingFramerateScroll
- AsyncResourceFramerateFixtures.DataGridEditMutation
- NativeDockingSmokeFixture.PerGroupDropTargetVisualDemo
Cluster O — ReconcileHighlightOverlay tick race. DispatcherQueueTimer.Stop() cannot rescind a Tick the dispatcher queue has already dispatched, so a stale tick can tear down a sprite the new refresh still owns. Fix swaps in a fresh timer on refresh; the stale tick's identity check fails and it bails.
Cluster C — DataGrid_ScrollPopulatesData poll. Fixed 800 ms Render
occasionally undershoots realization on CI → cells=0. Now keeps the
800 ms baseline (so the scroll-settle window still fires fetches, and the
sibling ScrollPop_MultipleFetches sanity check still passes) and polls
cells for up to 3 s.

Test plan

dotnet build clean across Reactor + AppTests.Host.
OverlayLifecycle_* self-test family (27 fixtures): all green locally.
DataGrid_ScrollPopulatesData single-shot: all 5 sub-assertions green
(cells=41 / 41, calls=7).
15-iter local stress over both fixtures together: 0/15 failures.
CI Stress (200 iter × 20 shards selftests on this branch) to confirm
Cluster T/O/C all stay at zero hits.

🤖 Generated with Claude Code

…d scroll poll Three independent stress-flake fixes (Clusters T, O, C) per INVESTIGATION.md follow-up to PR #396: Cluster T — bump FixtureTimeout to 30s for three render-pump-heavy fixtures whose budgets are tight on loaded CI runners but not pathological: - DataGridParityFixtures.HookPagingFramerateScroll - AsyncResourceFramerateFixtures.DataGridEditMutation - NativeDockingSmokeFixture.PerGroupDropTargetVisualDemo Cluster O — fix race in ReconcileHighlightOverlay.RefreshOrAdd. DispatcherQueueTimer.Stop() can't dequeue a Tick that the dispatcher queue has already dispatched, so a stale tick can tear down a sprite the refresh still wants alive. Now each refresh swaps in a fresh timer with its own Tick lambda; the stale tick checks ReferenceEquals on ah.Timer and bails when its identity no longer matches. Cluster C — DataGrid_ScrollPopulatesData was relying on a fixed 800ms Render(800) for the scroll-settle → fetch → realization chain (not tracked by _renderPending). Now waits 800ms baseline (preserves the fetch-trigger window so ScrollPop_MultipleFetches still triggers) then polls cells for up to 3s. Validated locally: all 27 OverlayLifecycle_* fixtures pass, all 5 ScrollPop_* sub-checks pass, 15x local stress on both fixtures together is clean. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Copilot

Pull request overview

This PR bundles three stress-flake fixes across Reactor selftests and a dev overlay: (1) increases fixture timeouts for a few render-pump-heavy selftests that can exceed the default watchdog under CI load, (2) fixes a ReconcileHighlightOverlay expiry-timer race by swapping timers on refresh and guarding against stale ticks, and (3) hardens DataGrid_ScrollPopulatesData by polling for realized cells after a baseline settle delay.

Changes:

Increase FixtureTimeout to 30s for three known long-running stress fixtures.
Make ReconcileHighlightOverlay timer expiry resilient to stale dispatcher ticks by creating a fresh timer on refresh and adding an identity guard.
Replace a fixed post-scroll wait with baseline wait + polling for realized DataGrid cells to avoid cells=0 under CI load.

Show a summary per file

File	Description
tests/Reactor.AppTests.Host/SelfTest/Fixtures/NativeDockingSmokeFixture.cs	Extends timeout for `PerGroupDropTargetVisualDemo` to reduce CI watchdog flakes.
tests/Reactor.AppTests.Host/SelfTest/Fixtures/DataGridScrollFixtures.cs	Adds polling after scroll to wait for realized “Emp-” cells (keeps 800ms baseline).
tests/Reactor.AppTests.Host/SelfTest/Fixtures/DataGridParityFixtures.cs	Extends timeout for `HookPagingFramerateScroll` fixture.
tests/Reactor.AppTests.Host/SelfTest/Fixtures/AsyncResourceFramerateFixtures.cs	Extends timeout for `DataGridEditMutation` fixture.
src/Reactor/Hosting/ReconcileHighlightOverlay.cs	Swaps expiry timer on refresh and guards against stale timer ticks tearing down active sprites.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Files reviewed: 5/5 changed files
Comments generated: 2

            try { existing.Timer.Stop(); } catch { }
-            try { existing.Timer.Start(); } catch { }
+            var refreshedTimer = CreateExpiryTimer(target, existing);
+            existing.Timer = refreshedTimer;
+            refreshedTimer.Start();


            H.Check($"ScrollPop_DataVisible (cells={visibleEmpCells.Count})",
                visibleEmpCells.Count >= 4);


Two independent stress-flake fixes per INVESTIGATION.md session 4 follow-up to PR #397: Cluster F (4/1000 hits in run 26351376710) — FloatingTitleBar_PaneBodyVisible asserted body TextBlock visibility after a single Harness.Render() following DockFloatingWindow.Open. Chrome (TabView, TitleBar absence, TabStripFooter) materializes on that first pump, but the inner pane content lags one or two pumps behind on a loaded CI runner. Same realization-race family as Cluster C; apply the same poll pattern (2s budget, 50ms between pumps) and annotate the check name with the observed body count for future-hit diagnostics. Cluster T-new (2/1000 hits) — EventSubscriptionLeakBaseline timing out at its 30s override. Local timing measurement across 3 fresh-process runs: 14.5 / 15.6 / 15.4 s (avg 15.2s) on a dev box. The fixture is 100 mount/unmount cycles x 2 Harness.Render() = 200 renders + 200 reconcile passes; the work itself is substantial. CI VMs under load have been measured at 2-4x slowdown elsewhere in the doc, easily overshooting the 30s budget. Not a hang — just budget vs CI variance. Bump FixtureTimeout to 60s (~4x local baseline) with a comment block explaining the math. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Eliminates the dominant stress-failure mode (CI VM variance overshooting tight per-fixture budgets). 5 of 8 failures in CI Stress run 26364109146 were 15 s timeout trips across 5 different fixtures - the pattern that per-fixture overrides in PRs #395, #397, #399 chased without converging. Local timing measurements (documented in INVESTIGATION.md) show the Framerate.* family completing in ~7 s on a dev box; CI VMs under contention run 2-4x slower, putting fixture work at ~28 s on the worst tick. 30 s matches the explicit override budget already used by four fixtures and leaves a clean signal gap below the 60 s HangWatchdogLoop dump-on-hang threshold. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

codemonkeychris requested a review from Copilot May 24, 2026 04:26

Copilot started reviewing on behalf of codemonkeychris May 24, 2026 04:26 View session

Copilot AI reviewed May 24, 2026

View reviewed changes

codemonkeychris merged commit aab59c3 into main May 24, 2026
16 checks passed

codemonkeychris deleted the fix/stress-cluster-t-o-c branch May 24, 2026 13:48

codemonkeychris mentioned this pull request May 24, 2026

Stress fixes: Cluster F poll, leak baseline timeout bump #399

Merged

4 tasks

codemonkeychris mentioned this pull request May 25, 2026

test(stress): raise default FixtureTimeout 15s -> 30s #408

Merged

3 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Stress fixes: framerate/PerGroup timeouts, overlay tick race, DataGrid scroll poll#397

Stress fixes: framerate/PerGroup timeouts, overlay tick race, DataGrid scroll poll#397
codemonkeychris merged 1 commit into
mainfrom
fix/stress-cluster-t-o-c

codemonkeychris commented May 24, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		H.Check($"ScrollPop_DataVisible (cells={visibleEmpCells.Count})",
		visibleEmpCells.Count >= 4);

Conversation

codemonkeychris commented May 24, 2026

Summary

Test plan

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Copilot's findings

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants