Crossgen hangs and crashes on MacOs

[Tratcher]

Description

dotnet/AspNetCore's main branch has been experiencing hangs in crossgen on mac since at least 12/26/22. While we haven't been able to capture a dump from those, we did recently get a dump from a segfault in the same area.

Hangs:
https://dev.azure.com/dnceng-public/public/_build/results?buildId=156458&view=logs&j=9895f7d8-b062-540b-461c-07f8695bfcc6&t=a6c33838-4cf4-594d-88cb-801ae7f8db2f&l=533
https://dev.azure.com/dnceng-public/public/_build/results?buildId=145996&view=logs&j=f43b187f-6c15-531c-a59b-e4fba4e9ed20&t=98a09275-a4dd-593e-c21a-1f8bea3be5b1&l=2471
https://dev.azure.com/dnceng-public/public/_build/results?buildId=141806&view=logs&j=f43b187f-6c15-531c-a59b-e4fba4e9ed20&t=98a09275-a4dd-593e-c21a-1f8bea3be5b1&l=2371
https://dev.azure.com/dnceng-public/public/_build/results?buildId=136865&view=logs&j=8e9e2812-09db-5a5d-c249-d96adf9f47fa&t=51e8c4b1-e760-5f88-6965-6fecd74b90ed&l=535
https://dev.azure.com/dnceng-public/public/_build/results?buildId=135884&view=logs&j=f43b187f-6c15-531c-a59b-e4fba4e9ed20&t=98a09275-a4dd-593e-c21a-1f8bea3be5b1&l=2483
https://dev.azure.com/dnceng-public/public/_build/results?buildId=130679&view=logs&j=8e9e2812-09db-5a5d-c249-d96adf9f47fa&t=51e8c4b1-e760-5f88-6965-6fecd74b90ed&l=543
https://dev.azure.com/dnceng-public/public/_build/results?buildId=129390&view=logs&j=8e9e2812-09db-5a5d-c249-d96adf9f47fa&t=51e8c4b1-e760-5f88-6965-6fecd74b90ed&l=533
https://dev.azure.com/dnceng-public/public/_build/results?buildId=127806&view=logs&j=9895f7d8-b062-540b-461c-07f8695bfcc6&t=a6c33838-4cf4-594d-88cb-801ae7f8db2f&l=531
https://dev.azure.com/dnceng-public/public/_build/results?buildId=125508&view=logs&j=9895f7d8-b062-540b-461c-07f8695bfcc6&t=a6c33838-4cf4-594d-88cb-801ae7f8db2f&l=565

Crash:
https://dev.azure.com/dnceng-public/public/_build/results?buildId=157320&view=results
https://dev.azure.com/dnceng-public/_apis/resources/Containers/35147423/MacOS_Test_Dumps?itemPath=MacOS_Test_Dumps%2Fdotnet-50140.1675351152.core

We'll collect additional details in this issue as we find them.

cc: @BrennanConroy @dougbu

Reproduction Steps

dotnet/AspNetCore CI for the main branch on mac.
https://dev.azure.com/dnceng-public/public/_build?definitionId=83&_a=summary

Expected behavior

No hangs or crashes

Actual behavior

Hangs and/or crashes

Regression?

Yes

Known Workarounds

Re-run the job

Configuration

.NET 8

Other information

No response

I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.

[mangod9]

@trylek @janvorli FYI.

[janvorli]

The error message of the System.InvalidOperationException in the dump case says:
Cannot read keys when either application does not have a console or when console input has been redirected. Try Console.Read.

The managed call stack is

00007FFEE40C0F80 0000000115B58CD4 System.ConsolePal.ReadKey(Boolean)
00007FFEE40C0FC0 00000001159C0AC2 IStartupInjectionAssemblyName.Program.Main(System.String[])

I'm not sure why the exception message / stack trace was not written to the stderr, maybe it is the same underlying issue that caused the actual exception.

[BrennanConroy]

Why is crossgen even touching that project? It's a test project located at https://github.com/dotnet/aspnetcore/blob/main/src/Hosting/test/testassets/IStartupInjectionAssemblyName/Program.cs and the crossgen args don't look like they have anything to do with that project:

"/Users/runner/work/1/s/.packages/microsoft.netcore.app.crossgen2.osx-x64/8.0.0-preview.2.23101.9/tools/crossgen2" --targetarch:x64 --targetos:osx -O @"/Users/runner/work/1/s/artifacts/obj/Microsoft.AspNetCore.App.Runtime/Release/net8.0/osx-x64/crossgen/PlatformAssembliesPathsCrossgen2.rsp" "-m:/Users/runner/work/1/s/.packages/microsoft.netcore.app.runtime.osx-x64/8.0.0-preview.2.23101.9/tools/StandardOptimizationData.mibc" -o:"/Users/runner/work/1/s/artifacts/bin/Microsoft.AspNetCore.App.Runtime/Release/net8.0/osx-x64/System.Security.Cryptography.Xml.dll" "/Users/runner/work/1/s/.packages/system.security.cryptography.xml/8.0.0-preview.2.23101.9/lib/net8.0/System.Security.Cryptography.Xml.dll"

[janvorli]

The dump file is not from crossgen2 as you can see from the stack trace. The application start method is IStartupInjectionAssemblyName.Program.Main

[trylek]

For the role of Crossgen2 @ivdiazsa is probably the best to comment as he was one of the two people implementing Crossgen2 in ASP.NET, the other person being @davidwrighton who's currently on parental leave.

[mangod9]

@Tratcher does this still need investigation? From what Jan found it appears this might not be related to cg2?

[Tratcher]

The crash apparently wasn't, but we're still suffering from the regular hangs and could use some help to investigate those. Unfortunately we haven't been able to capture any dumps from the hangs. Are there any other options?

[mangod9]

does it only repro on CI machines or can we setup a local repro for it?

[Tratcher]

We've only managed to repro it on the CI.

[captainsafia]

@mangod9 @amcasey and I are taking a look at this. Not sure if it is a helpful detail, but the issue appears to only repro on x64 macOS. The arm64 macOS build does not hang on the crossgen build.

[mangod9]

Ok interesting. Thanks for additional info -- so is it reproing locally on x64, or still in CI?

[BrennanConroy]

The arm64 macOS build does not hang on the crossgen build.

https://dev.azure.com/dnceng-public/public/_build/results?buildId=147425&view=logs&j=8e9e2812-09db-5a5d-c249-d96adf9f47fa&t=51e8c4b1-e760-5f88-6965-6fecd74b90ed