Azure ML mounting datastore while creating deployment

Question

Hi Everyone, Had a question we have the default blob store in the Azure ML studio workspace. I have a scenario where I have stored image in blob and this will be further used for infrencing. I need to know is there any method by which I can mount the blob while creating the deployment or the endpoint

Answer 1

Hi Eashwaran,

A file or a folder from blob storge can be mounted on a compute target inside a job. Below is the code:

from azure.ai.ml import command, Input, MLClient, UserIdentityConfiguration, ManagedIdentityConfiguration
from azure.ai.ml.entities import Data
from azure.ai.ml.constants import AssetTypes, InputOutputModes
from azure.identity import DefaultAzureCredential

# Set your subscription, resource group and workspace name:
subscription_id = "<SUBSCRIPTION_ID>"
resource_group = "<RESOURCE_GROUP>"
workspace = "<AML_WORKSPACE_NAME>"

ml_client = MLClient(
    DefaultAzureCredential(), subscription_id, resource_group, workspace
)

path = "wasbs://******@azuremlexampledata.blob.core.windows.net/titanic.csv"


data_type = AssetTypes.URI_FILE

mode = InputOutputModes.RO_MOUNT


identity = None

inputs = {
    "input_data": Input(type=data_type, path=path, mode=mode)
}

job = command(
    command="head ${{inputs.input_data}}",
    inputs=inputs,
    environment="azureml://registries/azureml/environments/sklearn-1.1/versions/4",
    compute="cpu-cluster",
    identity=identity,
)

ml_client.jobs.create_or_update(job)

However when using a real-time endpoint it is advised to access blobs using managed identity. Here is the supported documentation:
https://learn.microsoft.com/en-us/azure/machine-learning/how-to-access-resources-from-endpoints-managed-identities?view=azureml-api-2&tabs=system-identity-cli

For example:
Set Entra ID Authentication for Blob and ML Studio.
Use Azure Storage Blobs client library to access data from Blob inside deployment. Here is supported documentation:
https://learn.microsoft.com/en-us/python/api/overview/azure/storage-blob-readme?view=azure-python

Feel free to accept this as an answer.

Thankyou for reaching out to the Microsoft QNA Portal.