fix(stress): per-fixture hang watchdog + WaitForIdle cap raise

[codemonkeychris]

Summary

Two race-driven flakes surfaced in the recent 500x selftest stress run on main (CI Stress run 26513480330): 4 of 10000 iterations failed (~99.96% pass), with 2 of 4 in NativeDocking_* fixtures. Investigation showed both root causes were in test infrastructure, not the new docking window manager work (#418/#420/#421/#415).

Root causes

1. ReactorHost.WaitForIdleAsync(maxYields: 10) silently returned after 10 low-priority dispatcher yields whether or not the render loop had actually settled. Under CI VM contention, callers like Harness.Render() moved on against a half-rendered tree — producing flakes like NativeDocking_RoleAware_SplitDocArea_CloseNonLast failing to FindText("body:d1") on iteration 3/500.

2. SelfTestRunner.HangWatchdogLoop used a global 60 s constant equal to EventSubscriptionLeakBaseline's own FixtureTimeout. On worst-case CI ticks (~60 s for 200 renders + 200 reconciles, as the fixture's own comment acknowledged), the two raced — the watchdog won and FailFasted the shard instead of letting the fixture's graceful timeout fire.

Changes

File	Change
src/Reactor/Hosting/ReactorHost.cs	WaitForIdleAsync default cap 10 → 50; Debug.WriteLine when the cap actually hits so the next flake leaves a greppable trail
tests/.../SelfTest/SelfTestRunner.cs	FixtureProgress carries a per-fixture HangThreshold = max(60 s, FixtureTimeout + 30 s); watchdog uses that, not the global constant
tests/.../SelfTest/SelfTestFixtureBase.cs	doc comment updated to reflect new watchdog rule
tests/.../Fixtures/NativeDockingReliabilityFixture.cs	EventSubscriptionLeakBaseline FixtureTimeout 60 → 120 s; heartbeat H.Check after warmup and every 25 cycles

The heartbeat H.Check calls don't reset the watchdog (it's still elapsed-since-fixture-start), but they make ok lines visible in the log so a future hang reveals which 25-cycle window it lived in, not just "no progress".

Test plan

• src/Reactor and tests/Reactor.AppTests.Host both compile clean (0 errors, only pre-existing XML-comment / nullability warnings)
• Kick off a 1000x CI Stress run targeting selftests on this branch
• Confirm no flakes in the NativeDocking_* fixtures or DataGrid_KeyboardAndPrivateRenderPaths
• Confirm EventSubscriptionLeakBaseline per-iteration time stays well under the 120 s budget (target: still ~15 s local, ~30-60 s under CI contention)
• Spot-check stress logs for [Reactor.WaitForIdle] yield cap hit lines — if any appear, the 50-yield cap is being approached and we'll want to investigate the underlying render loop

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR hardens the selftest stress infrastructure to reduce race-driven flakes under CI contention by improving “wait until UI settles” behavior and making the hang watchdog threshold fixture-aware (so long-budget fixtures don’t get fail-fasted prematurely).

Changes:

• Increase ReactorHost.WaitForIdleAsync yield cap (10 → 50) and add a diagnostic trace when the cap is hit.
• Make the hang watchdog threshold per-fixture (max(60s, FixtureTimeout + 30s)) instead of a single global ceiling.
• Extend EventSubscriptionLeakBaseline fixture timeout (60s → 120s) and add periodic TAP “heartbeat” checks for better log locality during stress.

Show a summary per file

File	Description
src/Reactor/Hosting/ReactorHost.cs	Raises WaitForIdleAsync yield cap and logs when the cap is hit.
tests/Reactor.AppTests.Host/SelfTest/SelfTestRunner.cs	Introduces per-fixture hang thresholds for the watchdog and adjusts progress publication.
tests/Reactor.AppTests.Host/SelfTest/SelfTestFixtureBase.cs	Updates documentation to reflect the new watchdog threshold rule.
tests/Reactor.AppTests.Host/SelfTest/Fixtures/NativeDockingReliabilityFixture.cs	Increases leak baseline timeout and adds TAP heartbeat checks.

Copilot's findings

Tip

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

• Files reviewed: 4/4 changed files
• Comments generated: 4

[codemonkeychris]

Stress run summary for https://github.com/microsoft/microsoft-ui-reactor/actions/runs/26526776797:

• Overall selftest reliability: 997/1000 iterations passed = 99.7%.
• Scope: 1000 selftest iterations across 40 shards; unit/integration shards were skipped.
• Docking-window-specific signal: 1 docking-related failure in 1000 iterations, or roughly 99.9% docking iteration reliability for this run.

Failed tests observed during the run:

Shard / Iteration	Test	Failure
Shard 5 / iteration 1	DataGrid_ScrollPopulatesData	ScrollPop_InitialRender - assertion failed
Shard 12 / iteration 17	AsyncResource.Framerate.DataGridScroll	exceeded 30s
Shard 40 / iteration 25	NativeDocking_A11y_HostLandmarkAndPaneAutomationIds	A11y_PaneAutomationId_ActiveTabFound - assertion failed