Flaky test host hang on Windows Debug CI: TaskCancellationIntegrationTests.TaskTool_CancellationToken_GetTaskShowsWorkingBeforeCancel

[PranavSenthilnathan]

Summary

The build (windows-latest, Debug) CI leg intermittently fails because the ModelContextProtocol.Tests test host hangs and is killed after the 7-minute blame inactivity timeout. When the hang was captured, the test executing was:

ModelContextProtocol.Tests.Server.TaskCancellationIntegrationTests.TaskTool_CancellationToken_GetTaskShowsWorkingBeforeCancel

(The blame collector notes "This test may, or may not be the source of the crash.")

This appears to be a flaky / non-deterministic hang — the same assembly passes on the other target frameworks in the same run, and the failure only reproduces intermittently on the Windows Debug leg.

Occurrence

• Run: 29271612002, job build (windows-latest, Debug) (attempt 2)
• Job log: https://github.com/modelcontextprotocol/csharp-sdk/actions/runs/29271612002/job/86940026088
• Observed on PR Enforce PKCE metadata compliance in OAuth client (#730) #1700 (the PR diff only touches OAuth client PKCE metadata validation and its tests — unrelated to task cancellation — so this is very likely a pre-existing flake rather than a regression).

Relevant log excerpt

Passed!  - Failed: 0, Passed: 2273, Skipped: 4, Total: 2277, Duration: 6 m 33 s - ModelContextProtocol.Tests.dll (net10.0)
Passed!  - Failed: 0, Passed: 2273, Skipped: 4, Total: 2277, Duration: 6 m  9 s - ModelContextProtocol.Tests.dll (net8.0)
Passed!  - Failed: 0, Passed: 1974, Skipped: 278, Total: 2252, Duration: 6 m 30 s - ModelContextProtocol.Tests.exe (net472)

The active test run was aborted. Reason: Test host process crashed
Data collector 'Blame' message: The specified inactivity time of 7 minutes has elapsed. Collecting hang dumps from testhost and its child processes.
Data collector 'Blame' message: Dumping this process tree (from bottom).
Data collector 'Blame' message: > 5068 - testhost.
Data collector 'Blame' message:  +- 6860 - ModelContextProtocol.Tests.
Passed!  - Failed: 0, Passed: 2268, Skipped: 4, Total: 2272, Duration: 6 m 27 s - ModelContextProtocol.Tests.dll (net9.0)
Test Run Aborted.
The active Test Run was aborted because the host process exited unexpectedly. Please inspect the call stack above, if available, to get more information about where the exception originated from.
The test running when the crash occurred:
ModelContextProtocol.Tests.Server.TaskCancellationIntegrationTests.TaskTool_CancellationToken_GetTaskShowsWorkingBeforeCancel
This test may, or may not be the source of the crash.

Attachments:
  ...\ModelContextProtocol.Tests_6860_20260713T214733_hangdump.dmp
  ...\testhost_5068_20260713T214741_hangdump.dmp
  ...\Sequence_6dd0728684124ee1a0d9e35a4ba318c2.xml

make: *** [Makefile:19: test] Error 1
##[error]Process completed with exit code 1

Hang dumps (ModelContextProtocol.Tests_6860_*.hangdump.dmp, testhost_5068_*.hangdump.dmp) and the blame Sequence_*.xml are attached to the run as test-result artifacts and should contain the managed call stack of the hung thread.

Notes for debugging

• procdump64.exe could not be found appears earlier in the log, so only the .NET hang dump (not a full crash dump via procdump) was collected — worth downloading the *hangdump.dmp artifacts and inspecting the blocked thread's stack (e.g. via dotnet-dump analyze → clrstack).
• TaskCancellationIntegrationTests.TaskTool_CancellationToken_GetTaskShowsWorkingBeforeCancel coordinates a working→cancel transition; a missed signal / await that never completes is a plausible cause of the inactivity hang. The Sequence XML will confirm whether it was mid-execution or already completed when the host hung.

Expected

The Windows Debug test leg should complete deterministically without hanging.