Flaky test: RunDotnetProcessAsync_Cancellation_KillsProcessTree intermittently fails in CI

[azchohfi]

DotNetServiceTests.RunDotnetProcessAsync_Cancellation_KillsProcessTree (added in #686) fails intermittently in the Build and Package workflow on pull requests that cannot affect it.

Source: src/winapp-CLI/WinApp.Cli.Tests/DotNetServiceTests.cs#L1695

Observed

PR	Scope of the change	Result
#726	sample test harness only (no product code)	failed
#728	two YAML config files (dependabot.yml, codeql-config.yml)	failed
main @ 51666c0d	—	passed

Neither PR touches anything that could influence process-tree teardown, and the same commit passed on main, so this looks intermittent rather than broken.

Failing runs:

• Invoke build tools directly in sample tests instead of Invoke-Expression #726 - https://github.com/microsoft/winappCli/actions/runs/31525372392
• Cover NuGet and npm in Dependabot, and use the CodeQL security-extended suite #728 - https://github.com/microsoft/winappCli/actions/runs/31525574112

Signal

The failure takes ~15s before reporting, which suggests a timeout elapsing while waiting on child-process teardown rather than an assertion mismatch. A loaded CI runner plausibly exceeds whatever fixed wait the test uses.

Likely fix: wait deterministically on child-process exit (poll for the PIDs to disappear, or await the process handles) instead of a fixed delay/timeout.

Possibly related

CrashDumpServiceWorkflowTests.AnalyzeDumpAsync_RealManagedDump_RunsClrMdManagedEnumeration (added in #644) failed in the same #728 run. Different subsystem and much older, so it may be an unrelated second flake — worth confirming before folding it in here.

[azchohfi]

Another failure in the same family, so this looks systemic rather than specific to #686.

DotNetServiceTests.RunDotnetInheritedAsync_CancelledMidBuild_KillsChildProcessTree (added in #612) failed on the re-run of the same #728 job: https://github.com/microsoft/winappCli/actions/runs/31525574112

Running tally of intermittent failures seen on PRs that cannot affect them:

Test	Added in	Failure time
RunDotnetProcessAsync_Cancellation_KillsProcessTree	#686	~15s (looks like a timeout)
RunDotnetInheritedAsync_CancelledMidBuild_KillsChildProcessTree	#612	834ms (fast, so likely an assertion)
AnalyzeDumpAsync_RealManagedDump_RunsClrMdManagedEnumeration	#644	~5.7s

The first two are both process-tree-kill cancellation tests from different PRs, which suggests the shared teardown/kill helper is the unreliable part rather than either individual test. The two distinct failure shapes (timeout vs fast assertion) hint at a race with more than one losing outcome.

Worth looking at whichever helper both cancellation tests rely on for child-process teardown, rather than fixing them one at a time.