HTTP Response Error when starting AzureML job locally on Windows

Question

Hi everyone,

A client of mine can not trigger an Azure ML job locally with his windows machine and the Python SDK v2.. The problem here is, that it works with my Macbook and with the Macbook of one of my colleagues. The first issue with the windows machine happens already during the stage of user identification via the Azure CLI. The browser is stuck after having succesfully logged into the Azure Portal. The terminal also does not recognize, that he has identified himself and stays kind of in a endless loop. The only way to make this work, is by using the second identification option with a generated code, that he has to insert into the browser. I use zsh and he is using cmd as his terminal. We both use miniconda and the same package dependencies locally.

We use the DefaultAzureCredential(), the remaining setup (compute Cluster, environment, subscription, workspace, ressource group) are also identical. The error happens after this code line here for him:

ml_client.create_or_update(ml_job)

It appears approximately after 10 - 30 seconds, when the loading bar has reached 100%. He gets an HTTP Response Error. The error message is the following one

    raise HttpResponseError(response=response, model=error, error_format=ARMErrorFormat) Info: {     "value": {         "operation": "0193924c719412417d98de0cb1ae32f9",         "request": "0e8a9d2852b83b59"     } }Type: Environment Info: {     "value": "westeurope" }Type: Location Info: {     "value": "westeurope" }Type: Time Info: {     "value": "2024-03-12T17:22:30.6294498+00:00" }Type: InnerError Info: {     "value": {         "code": "BadArgument",         "innerError": {             "code": "CreatRunBadRequest",             "innerError": null         }     } }

Honestly I don´t know, why this error appears, but I assume, that it has to do something with the authentication or credential. Can someone help me with this issue?

Answer 1

@Yannic Vorpahl Could you please check if you are both using the same version of the azure-ai-ml package? If they are different, try to remove and reinstall the latest version.

I would recommend uninstalling the package and installing it afresh since you are reporting issues with setting up your authentication. I would also recommend running a list command to view resources of your workspace with the client before proceeding to create a run.

# Get a list of workspaces in a resource group
for ws in ml_client.workspaces.list():
    print(ws.name, ":", ws.location, ":", ws.description)

I have noticed a similar issue from the python SDK repo but it was a known issue with the version the user had installed. If you are still seeing an issue, could you please open a new issue on the azure-sdk-for-python repo?