[Infra] SampleAnalyzerIntegrationTest randomly timeouting in CI

Question

[Infra] SampleAnalyzerIntegrationTest randomly timeouting in CI

JanKrivanek opened this issue a month ago · comments

Context

System.TimeoutException : Test failed due to timeout: process 12042 is active for more than 30 sec.

   at Microsoft.Build.UnitTests.Shared.RunnerUtilities.RunProcessAndGetOutput(String process, String parameters, Boolean& successfulExit, Boolean shellExecute, ITestOutputHelper outputHelper) in /_/src/UnitTests.Shared/RunnerUtilities.cs:line 139
   at Microsoft.Build.UnitTests.Shared.RunnerUtilities.ExecBootstrapedMSBuild(String msbuildParameters, Boolean& successfulExit, Boolean shellExecute, ITestOutputHelper outputHelper) in /_/src/UnitTests.Shared/RunnerUtilities.cs:line 67
   at Microsoft.Build.BuildCheck.UnitTests.EndToEndTests.SampleAnalyzerIntegrationTest(Boolean buildInOutOfProcessNode, Boolean analysisRequested) in /home/vsts/work/1/s/src/BuildCheck.UnitTests/EndToEndTests.cs:line 122
   at System.RuntimeMethodHandle.InvokeMethod(Object target, Void** arguments, Signature sig, Boolean isConstructor)
   at System.Reflection.MethodBaseInvoker.InvokeDirectByRefWithFewArgs(Object obj, Span`1 copyOfArgs, BindingFlags invokeAttr)

Sample run: https://dev.azure.com/dnceng-public/public/_build/results?buildId=648080&view=ms.vss-test-web.build-test-results-tab&runId=16013174&resultId=100001&paneView=debug

YuliiaKovalova · Answer 1 · Fri Apr 19 2024 17:10:38 GMT+0800 (China Standard Time)

discussed offline: the execution time can be extended for handling it. Please track it for some time if this approach solves the problem

Gang Wang · Answer 2 · Tue Apr 30 2024 17:10:54 GMT+0800 (China Standard Time)

After observing 9 runs that failed with this test, the failure only happened to the following test cases. From the output logs attached inline, it looks like within 30 seconds the build check analyzer hadn't finished.

SampleAnalyzerIntegrationTest(buildInOutOfProcessNode: True, analysisRequested: True) - Fail buildInOutOfProcessNode_True-analysisRequested_True.txt
SampleAnalyzerIntegrationTest(buildInOutOfProcessNode: False, analysisRequested: True) - Fail buildInOutOfProcessNode_False-analysisRequested_True.txt

Checking with the output log Microsoft.Build.BuildCheck.UnitTests_net8.0_x64.log of fully passed test run, I found there were over 1000 warnings when the analysis was requested. But the warnings except BC0101 couldn't be reproduced locally. This may explain it looks like within 30 seconds the build check analyzer hadn't finished. No idea why CI run has so many extra warnings.

YuliiaKovalova · Answer 3 · Tue Apr 30 2024 17:17:15 GMT+0800 (China Standard Time)

@JanKrivanek , can this fix help with the test performance?
#10084

Jan Krivanek · Answer 4 · Tue Apr 30 2024 17:47:02 GMT+0800 (China Standard Time)

Only slightly - it slashes only about a third of currently detected slowdown of the analyzers.
Though the detected slowdown was in range of about 14% - so it's puzzling that the process runs for over 30 second. It would be interesting to see if it fails with about the same or significantly decreased ratio if we increase the timeout to say - 120 seconds. That might bisect between the slowdown and deadlock theory