Using new Windows runners crashes when conda is called from python script

[atalman]

Using following workflow:
https://github.com/pytorch/test-infra/blob/main/.github/workflows/windows_job.yml

Creates an issue with windows worker when similar code is executed:

def get_anaconda_output_for_package(pkg_name_str):
    import subprocess as sp

    # If we are installing using conda just list package name
    if installation_str.find("conda install") != -1:
        cmd = "conda list --explicit"
        output = sp.getoutput(cmd)
        for item in output.split("\n"):
            if pkg_name_str in item:
                return item
        return f"{pkg_name_str} can't be found"
    else:
        cmd = "conda list -f " + pkg_name_str
        output = sp.getoutput(cmd)
        # Get the last line only
        return output.strip().split('\n')[-1]

following error is observed:

output: 'conda' is not recognized as an internal or external command,
    main()
  File "C:\actions-runner\_work\builder\builder\pytorch\builder\test\smoke_test\smoke_test.py", line 198, in main
operable program or batch file.
    smoke_test_cuda(options.package)
  File "C:\actions-runner\_work\builder\builder\pytorch\builder\test\smoke_test\smoke_test.py", line 99, in smoke_test_cuda
    raise RuntimeError(
RuntimeError: CUDA version issue. Loaded: operable program or batch file. Expected: 11.7

[seemethere]

Oh this is probably because the subprocess itself doesn't have the same environment variables set up, I wonder if there's a way to pass the environment variables to the subprocess

[atalman]

Thanks! I will look into this issue.