Issues with windows runner - windows-latest

[sophia-guo]

windows-2016 is not supported any more. We'd like to update to windows-latests.

The issue with windows-latests:

• Originally openjdk, system tests will fail at the beginning part of the build with message C:\cygwin64\bin\make.exe' failed with exit code 2 with all jdk_version , which suggests it might be an issue related with cygwin on windows-latest.

make[1]: *** [compile.mk:45: compile] Error 1
make[1]: Leaving directory '/cygdrive/d/a/run-aqa/run-aqa/aqa-tests/TKG'
make: *** [makefile:67: compile] Error 2
Error: The process 'C:\cygwin64\bin\make.exe' failed with exit code 2

https://github.com/adoptium/run-aqa/actions/runs/3016813951/jobs/4849941896

• Checked recent nightly run: system tests are good with window-latest. openjdk tests with jdk17 failed with specific tests. openjdk test with jdk8 and jdk11 failed at the beginning part of build with different error message. https://github.com/adoptium/run-aqa/actions/runs/3185357794/jobs/5194843504

C:\cygwin64\bin\git.exe clone --depth 1 -q --reference-if-able D:\a\run-aqa\run-aqa/openjdk_cache https://github.com/adoptium/jdk11.git openjdk-jdk
info: Could not add alternate for 'D:\a\run-aqa\run-aqa/openjdk_cache': reference repository 'D:\a\run-aqa\run-aqa/openjdk_cache' is not a local repository.
Username for 'https://github.com/': 
Error: The operation was canceled.

Feels like it's not a cygwin with windows-latest issue any more.

[sophia-guo]

The current issue only happened to jdk8 and jdk11, which don't have the repo https://github.com/adoptium/jdk8 or https://github.com/adoptium/jdk11. jdk17 does have the repo https://github.com/adoptium/jdk17.

Need to see the step of C:\cygwin64\bin\git.exe clone --depth 1 -q --reference-if-able D:\a\run-aqa\run-aqa/openjdk_cache https://github.com/adoptium/jdk8.git openjdk-jdk

[sophia-guo]

Current failure was caused by the workaround to git clone system and openjdk material, which is git clone jdk with wrong repo https://github.com/adoptium/jdk${version}.git.

run-aqa/src/runaqa.ts

Lines 290 to 294 in fa4002f

	if (buildList === 'openjdk' && version != '') {
	process.chdir('openjdk')
	// Shallow clone the adoptium JDK version - quietly - if there is a reference repo obtain objects from there - destination is openjdk-jdk
	await exec.exec(`git clone --depth 1 -q --reference-if-able ${process.env.GITHUB_WORKSPACE}/openjdk_cache https://github.com/adoptium/jdk${version}.git openjdk-jdk`)
	process.chdir('../')

Update https://github.com/adoptium/jdk${version}.git to https://github.com/adoptium/jdk${version}u.git can fix this issue. However the workaround means run-aqa will only work with master branch, openjdk, aqa-systemtest, STF cann't be customized or specified.

[sophia-guo]

Remove the workaround the failure would be the same as https://github.com/adoptium/run-aqa/actions/runs/3016813951/jobs/4849941896. job failed as no bash available. Github runner windows-latest has [Bash 5.1.16(1)-release] installed (https://github.com/actions/runner-images/blob/main/images/win/Windows2022-Readme.md). With Cygwin running bash it looks good C:\cygwin64\bin\bash.exe ./get.sh --sdk_resource github-hosted --sdkdir D:\a\run-aqa\run-aqa

The problem happened when:

• running bash by java program, which actually might be same as the second one as java program is also called by ant in our system.
  https://github.com/adoptium/TKG/blob/master/src/org/openj9/envInfo/MachineInfo.java#L48-L56
• running bash by ant https://github.com/adoptium/TKG/blob/master/scripts/build_test.xml#L128

The error message suggest:

	Windows Subsystem for Linux has no installed distributions.
	
	Distributions can be installed by visiting the Microsoft Store:
	
	https://aka.ms/wslstore

This only happened to github runner, which installs cygwin to runner and edit the PATH environment variable to add the Cygwin.

Jenkins windows agent in adoptium are also installed cygwin and run the tests with cygwin, but didn't have the same issue.

[ultramancoder]

Hi! According to my (limited) understanding, ProcessBuilder passes the command as bash. Internally that'll invoke CreateProcess apis of windows which resolve a command without absolute path using logic here: https://learn.microsoft.com/en-us/windows/win32/api/processthreadsapi/nf-processthreadsapi-createprocessa#parameters . As we can see cygwin dirs or the PATH is not searched to resolve the command to run.

However, when running directly in actions or usings actions/exec library, it calls https://github.com/actions/toolkit/blob/b36e70495fbee083eb20f600eafa9091d832577d/packages/io/src/io.ts#L231 internally. This function searches the path manually for trying to find tool in path so it works when invoking without java/ant.

If this understanding is correct, then either the process builder code in TKG will need to be updated to do the same kind of resolving the command from the path or alternatively the full path to bash executable will need to be defined.

I am not sure why it works in jenkins but if you could point me to where that is setup, I would like to look into it and figure out why.

[smlambert]

Our Jenkins nodes are setup using Ansible playbooks (found in the infrastructure repository https://github.com/adoptium/infrastructure/blob/master/ansible/playbooks/AdoptOpenJDK_Windows_Playbook/main.yml).

[ultramancoder]

Thanks for the link, @smlambert.

Another thing I just noticed is that the functional tests do pass on windows. For instance see, https://github.com/adoptium/run-aqa/actions/runs/3380432551/jobs/5613604498.

Note that these logs have the same WSL error message mentioned earlier in this issue and the one we can also see in system tests that fail on windows-latest. Therefore, I think the WSL error may not be the cause of failures. I'll try to reproduce the failure locally, maybe that helps in getting more insight into the issue.

[smlambert]

thanks for digging into it @ultramancoder !

[ultramancoder]

Hi again! I looked at https://github.com/adoptium/infrastructure/blob/master/ansible/inventory.yml to find what windows server adoptium's temurin builds were using.

Github's windows-latest points to Windows 2022 server. In adoptium's windows build farm, there are mostly Windows 2012 servers, a few Windows 2016 Servers and 1 Windows 2019 server. Since there are no Windows 2022 servers there, this issue probably never showed up there.

[ultramancoder]

I also looked into reproducing the errors locally and it does fail there as well. I modified the makefiles to change ant to ant -v to get more verbose logs. Where it failed for me was https://github.com/adoptium/aqa-tests/blob/66436e6253ac9dd65a9fbc6cd5ad0e258df34ed4/external/build_image.sh#L69. docker was unable to recognise /cygdrive paths, passing the file paths through cygpath -w made the script work and progress the build but it later failed due to other local issues on PC. As of now, I am not sure if this docker issue is also the one happening in CI. Will need more investigation.

Might I suggest to stick with Windows 2016 Runners till the adoptium build farm's have a 2022 server?

[ultramancoder]

Hi! One more interesting development, some checks pass on windows-latest when impl is switched to hotspot instead of openj9 in the workflow. https://github.com/ultramancoder/run-aqa/actions/runs/3434222084/jobs/5725336952 . The ones that fail, fail on all platforms with an error about test groups.