Flaky test analysis: Xamarin.Android-PR pipeline, April 27 – May 11 2026 (follow-up to #11203)

[simonrozsival]

Flaky test analysis: Xamarin.Android-PR pipeline, April 27 – May 11 2026 (follow-up to #11203)

Follow-up to #11203 — refreshes the flaky-test picture two weeks later, after the fixes that
landed in between (notably #11249 for PerformanceTest and #11257 removing the WearOS lane).

Scope

Period	2026-04-27 → 2026-05-11 (14 days, ending today)
Pipeline	Xamarin.Android-PR (devdiv definition ID 12278)
Method	Identical to #11203 — see that issue for full methodology
PRs analyzed	62 merged PRs
Direct (team-member) PRs	62 (no fork PRs in window)
PRs merged with failed build result	23 (vs. 38 in #11203 ⇒ ~40% drop)
Unique test names observed in (Auto-Retry) runs	80 (vs. 51)
Tests with ≥1 🔴	15 (vs. 33)

Signal categories are the same as #11203: 🔴 = passed on auto-retry in a failed build (highest-confidence flake), 🟢 = passed on auto-retry in a succeeded build, ⚠️ = still failed on auto-retry.

Headline observations

1. PerformanceTest is no longer a red-CI source. The six top Build_* / Install_* /
   DesignTimeBuild_* perf tests from Flaky test analysis: Xamarin.Android-PR pipeline, April 10–24 2026 #11203 each produced zero 🔴 events this window.
   [tests] Skip PerformanceTest on slow CI machines using evaluation time #11249's Assert.Inconclusive slow-machine guard is working.
2. WearOS noise is gone. Four WearOS-only debugger variants from Flaky test analysis: Xamarin.Android-PR pipeline, April 10–24 2026 #11203's top 7
   disappeared because the WearOS lane was removed in [build] Remove WearOS emulator CI lane #11257.
3. Debugger-attach (ApplicationRunsWithDebuggerAndBreaks) remains the largest residual
   flake family.
4. NativeAOT lane is producing fresh flakes — expected as that lane stabilizes.
5. New top offender: TypeAndMemberRemapping(False,MonoVM) — 3 🔴 across 3 different PRs
   ([NativeAOT] Use NativeLinker and invoke lld directly for linking #11256, Split android-reviewer review-rules.md into conditional rule files #11266, LEGO: Pull request from juno/hb_6dddf33b-c6da-43d8-ac04-14d2c339cb00_20260502103235064 to main #11269). Only the (False,MonoVM) parameterization flakes.

Top flaky tests with recommended action

All entries below are in suite MSBuildDeviceIntegration On Device. The "Action" column
applies the same playbook used by #11249 (soft-skip), #11202 ([Ignore]), #11110
(Inconclusive for emulator/env failure), and #10756 (remove).

#	Test	🔴	🟢	⚠️	Recommended action	Notes
1	TypeAndMemberRemapping(False,MonoVM)	3	0	0	Investigate — possible real regression	New high-confidence flake; only (False,MonoVM) parametrization fails; (True,CoreCLR) shows ⚠️1 only. Was 🟢1 in #11203.
2	ApplicationRunsWithDebuggerAndBreaks(False,"guest1","apk",True,MonoVM)	2	1	0	Soft-skip on debugger-attach failure	Recurring from #11203. Pattern matches "device-attach race"; consider Assert.Inconclusive when the debugger-attach step times out specifically.
3	BuildBasicApplicationAndAotProfileIt	2	0	0	Investigate (AOT-profile capture is timing-sensitive)	Recurring from #11203 at same rate (2→2).
4	ApplicationRunsWithDebuggerAndBreaks(False,"guest1","aab",True,MonoVM)	1	3	2	Soft-skip on debugger-attach failure	Same family as #2.
5	ApplicationRunsWithDebuggerAndBreaks(False,null,"aab",True,MonoVM)	1	0	0	Soft-skip on debugger-attach failure	Same family.
6	FastDeployEnvironmentFiles(False,True,MonoVM)	1	0	1	Monitor — only 1 PR	New. Re-check next cycle.
7	AppWithStyleableUsageRuns(True,False,False,MonoVM)	1	0	1	Monitor — only 1 PR	New.
8	CheckResouceIsOverridden(NativeAOT)	1	0	0	Monitor (NativeAOT lane)	New; NativeAOT lane still stabilizing.
9	MonoAndroidExportReferencedAppStarts(False,False,MonoVM)	1	0	0	Monitor	Recurring (2→1).
10	DotNetRunWaitForExit	1	0	0	Monitor — already tracked in #10832	Big improvement (5→1).
11	DotNetRun(True,"managed",NativeAOT)	1	0	0	Monitor (NativeAOT lane)	New.
12	GradleFBProj(True,NativeAOT)	1	0	0	Monitor (NativeAOT lane)	New.
13	MonoAndroidExportReferencedAppStarts(False,True,CoreCLR)	1	0	0	Monitor	New CoreCLR variant.
14	AdbTargetChangesAppBundle	1	0	0	Monitor — only 1 PR	New.
15	EnsureUncaughtExceptionWorks(MonoVM)	1	0	0	Monitor	Big improvement (3→1).

Tests with elevated ⚠️ (retry-fail) counts

DotNetRunWithDeviceParameter (⚠️2) — already tracked in #10832; no 🔴 hits this window.
Other ⚠️1 tests are not strong-enough signals to act on yet.

Tests appearing in #11203 with no 🔴 hits this window

• The whole PerformanceTest family (Build_*, Install_*, DesignTimeBuild_*) — resolved by [tests] Skip PerformanceTest on slow CI machines using evaluation time #11249.
• All WearOS-only ApplicationRunsWithDebuggerAndBreaks variants — resolved by [build] Remove WearOS emulator CI lane #11257.
• SupportDesugaringStaticInterfaceMethods(MonoVM) — no signal this window.

Proposed concrete follow-ups

1. Investigate TypeAndMemberRemapping(False,MonoVM) — file a focused issue. Three
   independent 🔴 hits in one window across unrelated PRs ([NativeAOT] Use NativeLinker and invoke lld directly for linking #11256, Split android-reviewer review-rules.md into conditional rule files #11266, LEGO: Pull request from juno/hb_6dddf33b-c6da-43d8-ac04-14d2c339cb00_20260502103235064 to main #11269), one
   parametrization only.
2. Generalize the soft-skip pattern to ApplicationRunsWithDebuggerAndBreaks — when the
   debugger-attach step itself fails (not the assertion under test), call
   Assert.Inconclusive the same way [tests] Skip PerformanceTest on slow CI machines using evaluation time #11249 does for slow-machine perf tests and Use Assert.Inconclusive for emulator acquisition failures #11110 does
   for emulator acquisition.
3. Investigate BuildBasicApplicationAndAotProfileIt — recurring at a steady rate.
4. NativeAOT lane: leave at "monitor" for one more cycle; if any of CheckResouceIsOverridden(NativeAOT), DotNetRun(...,NativeAOT), GradleFBProj(True,NativeAOT) recur, treat as real bugs and escalate.
5. Repeat this analysis in ~2 weeks to confirm trend and catch regressions.

Existing related issues

• Flaky test analysis: Xamarin.Android-PR pipeline, April 10–24 2026 #11203 — prior analysis (this issue's predecessor)
• [Testing] Consider increasing timeouts for PreformanceTest #11031 — PerformanceTest timeouts (largely addressed by [tests] Skip PerformanceTest on slow CI machines using evaluation time #11249)
• [tests] test failures on dnceng-public #10832 — broad "test failures on dnceng-public" tracker; covers DotNetRunWaitForExit and DotNetRunWithDeviceParameter
• Figure out WearOS test failures #7506 — Figure out WearOS test failures — closed-in-effect by [build] Remove WearOS emulator CI lane #11257 removing the lane
• [tests] Skip InflateCustomView_ShouldNotLeakGlobalRefs until global ref leak is fully resolved #11202 — example of [Ignore] pattern for a known-leaking test

Caveats

Same as #11203: only the last build per PR is sampled; only (Auto-Retry) runs carry test
detail; no error-message content collected; ⚠️1-count tests are weak signals and may be noise.

Methodology

Same script and data path as #11203. See that issue for the full methodology section.