Is it possible to generate credentials to interact with Azure from inside a job running on local compute? If I start a job like this: And the script itself, ./src/main.py , looks like: The job fails. because there are no credentials, however it would succeed on a compute instance. error_log.txt

Credentials in local compute on Azure ML

Chinedum Echeta 0 Microsoft Employee

Is it possible to generate credentials to interact with Azure from inside a job running on local compute?

If I start a job like this:

User's image

And the script itself, ./src/main.py, looks like:

User's image

The job fails. because there are no credentials, however it would succeed on a compute instance. error_log.txt

YutongTie-MSFT 48,581

Hello Chinedum,

Thanks for reaching out to us, could you please share the document you are referring to?

There is a quick sample below, does it help you achieve your goal?

from azure.ai.ml import MLClient
from azure.identity import DefaultAzureCredential
# authenticate
credential = DefaultAzureCredential()
SUBSCRIPTION="<SUBSCRIPTION_ID>"
RESOURCE_GROUP="<RESOURCE_GROUP>"
WS_NAME="<AML_WORKSPACE_NAME>"
# Get a handle to the workspace
ml_client = MLClient(
    credential=credential,
    subscription_id=SUBSCRIPTION,
    resource_group_name=RESOURCE_GROUP,
    workspace_name=WS_NAME,
)

Regards,

Yutong

Chinedum Echeta 0 Reputation points Microsoft Employee

2024-06-18T14:22:13.1+00:00

HI Yutong,

The problem is not with creating a client, but I want to be able to generate Azure credentials within a job running on local compute.

I am not referring to a specific document, the code I shared screenshots of is a contrived example to demonstrate the problem I am trying to solve. I cannot write it as inline code as the post gets automatically
deleted by moderation.

Chinedum Echeta 0 Microsoft Employee

Example

.
├── config.json
├── main.py
└── src
    └── script.py

config.json

{
    "subscription_id": "<SUBSCRIPTION_ID>",
    "resource_group": "<RESOURCE_GROUP>",
    "workspace_name": "<AML_WORKSPACE_NAME>"
}

main.py

from azure.ai.ml import MLClient, command
from azure.identity import DefaultAzureCredential

ml_client = MLClient.from_config(DefaultAzureCredential())

job = command(
    compute="local",
    code="./src/",
    command="python main.py",
    environment="AzureML-AI-Studio-Development@latest",
)

ml_client.jobs.create_or_update(job)

script.py

from azure.identity import DefaultAzureCredential

print(DefaultAzureCredential().get_token("https://management.azure.com/.default"))

To reproduce, log in to Azure with AZ CLI, then run python main.py. Ensure Docker is running. The script completes successfully, but viewing the result in Azure ML shows that the job fails, because it is not authenticated against Azure and cannot retrieve a token.

Share via

Credentials in local compute on Azure ML