az account show reports "Error loading command module 'deploymentmanager'"

[bb-froggy]

Related command
az account show

Describe the bug
In a GitHub Action with the windows-latest runner, I install the az Edge built from https://aka.ms/InstallAzureCliWindowsEdge. Running the command az account show says

ERROR: Error loading command module 'deploymentmanager': MGMT_DEPLOYMENTMANAGER
{
   "environmentName": "AzureCloud",
   "homeTenantId": "***",
   "id": "***",
   "isDefault": true,
   "managedByTenants": [],
   "name": "xyz",
   "state": "Enabled",
   "tenantId": "***",
   "user": {
     "name": "***",
     "type": "servicePrincipal"
   }
}

(line breaks were lost in the output and manually inserted back afterwards for readability, but may be at incorrect positions)

This error message does not appear in the normal az installation instead of the Edge built. It does not appear in Linux runners. It wasn't there on 19th February, but was there on 26th February.

To Reproduce
This is the GitHub workflow producing the issue: https://github.com/scepman/scepman-psmodule/blob/main/.github/workflows/test-in-lab.yml. The PowerShell script in the last step calls az account show.

Expected behavior
The deploymentmanager module should be loaded correctly and the error message should not appear.

Environment summary
I have configured an Environment in GitHub and Federated Authentication in Azure that azure/login action uses for authentication.

Additional context
I couldn't reproduce the behavior interactively on a Windows 10 desktop machine. This also makes it harder for me to debug.

[yonzhan]

@jiasli for awareness

[bb-froggy]

Probably caused by this PR: #25523

[jsntcy]

@bb-froggy could you please tell how we can reproduce this issue in our side as we cannot reproduce it on our local machine?

[bb-froggy]

I have found a way to reproduce it now. On my Windows 10 machine with the Edge build of az, running az --help prints out the error message.

For reasons unknown to me, on my Windows 10 machine, az account show does not print it, while it does that on the GitHub runner.

[jsntcy]

@bb-froggy, thanks for the update. I cannot reproduce with az --help on my local machine as well.

• Could you please help run az --help --debug on your machine again to help us get debug log for troubleshooting?
• Could you please help check if "deploymentmanager" is in C:\Users\xxx.azure\commandIndex.json?

[bb-froggy]

I just ran az --help --debug >c:\temp\temp\azoutput-1.txt 2>c:\temp\temp\azoutput-2.txt on the affected Windows 10 machine. This resulted in no console output and produced these two files:
azoutput-1.txt
azoutput-2.txt

The string "deploymentmanager" does not appear in my commandIndex.json

[zhoxing-ms]

@bb-froggy Could you please show us the output of az --version command?
In addition, we guess this problem may be related to your installation environment. Have you installed CLI package in other ways in the local environment before?

[jsntcy]

@bb-froggy,

• Could you please also check if there is "deploymentmanager" folder in "C:\Program Files (x86)\Microsoft SDKs\Azure\CLI2\Lib\site-packages\azure\cli\command_modules"?
• If yes, could you please uninstall old version and install https://aka.ms/InstallAzureCliWindowsEdge again?
• How did you install https://aka.ms/InstallAzureCliWindowsEdge before? uninstall old version first?

[bb-froggy]

• There was a folder deploymentmanager in c:\Program Files (x86)\Microsoft SDKs\Azure\CLI2\Lib\site-packages\azure\cli\command_modules\
• I uninstalled two entries Azure CLI in "Programs and Features" and then installed AzureCliWindowsEdge again.

Now the problem is gone on this machine.

• The last time, I already had the production 2.45.0 of Azure CLI installed and just installed the Edge version once again without uninstalling anything.

I missed running az --version before uninstalling. I remember it gave a json that said that azure-clie and core were version 2.45.0. Now, it is not a json anymore, but still version 2.45.0.

[bb-froggy]

Last time I used az version instead of az --version, that's why it was a Json. I confused the two. Doesn't matter for the issue at hand, though.

[zhoxing-ms]

The last time, I already had the production 2.45.0 of Azure CLI installed and just installed the Edge version once again without uninstalling anything.

@bb-froggy When you install the Edge version on the basis of production version 2.45.0, do you choose Repair as the installation way?

[bb-froggy]

The wizard looked differently for me. It behaved just like there was no existing installation.

Probably the GitHub actions are an environment where this is more easy to reproduce. There, I installed az Edge with this powershell script on windows-latest:

$installUrl = 'https://aka.ms/InstallAzureCliWindowsEdge'
Write-Output "Installing az from  URL $installUrl"

$ProgressPreference = 'SilentlyContinue'
Invoke-WebRequest -Uri $installUrl -OutFile .\AzureCLI.msi
Start-Process msiexec.exe -Wait -ArgumentList '/I AzureCLI.msi /quiet'
rm .\AzureCLI.msi

I can try to create a repository with a minimal example to reproduce this with a GitHub runner, but not today.

[bebound]

@bb-froggy We figure out this.
On windows, if the edge version and official version has same version number, installing the edge version does not remove official version first.
It only creates and overwrites files but does not delete the old file.

You need to uninstall the official before installing the edge version.

[bb-froggy]

Okay, this works for me. In my GitHub Action, I do the following with powershell (not pwsh, which I believe wouldn't work) before installing the edge build:

          # Uninstall any previous version of az
          $azPackage = Get-Package -Name "*Azure CLI"
          if ($null -ne $azPackage) { $azPackage | Uninstall-Package }

This fixed the problem for me.

Since this problem is specific to the Edge build and does not affect the regular release, there is a working solution even for my use case, and you are updating the docs to point out the requirements to prevent this, I consider this issue fixed.