Unable to Train a model using AutoML

Question

I am using the Microsoft Non-Production MCAPS-Hybrid-xxxxx subscription and trying to do the MS Learn lab for training a model using Auto ML. The first issue I had in this environment was that I need to add the storage account permissions to my account and use Entra ID authentication to allow the uploading of the MLTable and diabetes.csv file to the workspace. In the next section, for using Auto ML, I get this error after I submit the training job:

Encountered user error while fetching data from Dataset. Error: UserErrorException:

Message: Failed to download mltable yaml with error: 

Error Code: UserError

Error Message: Getting permission error when trying to access MLTable,please make sure proper access is configured on storage:

Error Code: ScriptExecution.StreamAccess.Authentication

Native Error: error in streaming from input data sources

StreamError(PermissionDenied(Some(NoIdentityOnCompute)))

=> permission denied when access stream. Reason: Some(NoIdentityOnCompute)

PermissionDenied(Some(NoIdentityOnCompute)) | URI: azureml://subscriptions/212ea8e7-7b04-4383-9eec-6ff78fb90709/resourcegroups/rg-dp100-labs/workspaces/mlw-dp100-labs2/datastores/workspaceblobstore/paths/UI/2026-01-12_164002_UTC/Train a model using AutoML/MLTable

Error Message: Authentication failed when trying to access the stream. Make sure you have correct permissions set up. Ok(NoIdentityOnCompute) For troubleshooting help, see: https://learn.microsoft.com/en-us/azure/machine-learning/how-to-workspace-diagnostic-api?view=azureml-api-2| session_id=29dbe198-9678-4ec0-aa62-29320bca4867| session_id=29dbe198-9678-4ec0-aa62-29320bca4867

InnerException None

ErrorResponse 

{

"error": {

    "code": "UserError",

    "message": "Failed to download mltable yaml with error: \nError Code: UserError\nError Message: Getting permission error when trying to access MLTable,please make sure proper access is configured on storage: \nError Code: ScriptExecution.StreamAccess.Authentication\nNative Error: error in streaming from input data sources\n\tStreamError(PermissionDenied(Some(NoIdentityOnCompute)))\n=> permission denied when access stream. Reason: Some(NoIdentityOnCompute)\n\tPermissionDenied(Some(NoIdentityOnCompute)) | URI: azureml://subscriptions/212ea8e7-7b04-4383-9eec-6ff78fb90709/resourcegroups/rg-dp100-labs/workspaces/mlw-dp100-labs2/datastores/workspaceblobstore/paths/UI/2026-01-12_164002_UTC/Train a model using AutoML/MLTable\nError Message: Authentication failed when trying to access the stream. Make sure you have correct permissions set up. Ok(NoIdentityOnCompute) For troubleshooting help, see: https://learn.microsoft.com/en-us/azure/machine-learning/how-to-workspace-diagnostic-api?view=azureml-api-2| session_id=29dbe198-9678-4ec0-aa62-29320bca4867| session_id=29dbe198-9678-4ec0-aa62-29320bca4867"

}

}

I added owner, contributor RBAC access on the storage account for my user account and the managed identity for the ML workspace. The lab specifies serverless compute, so I don't know if this is an issue in this environment. I know that Key Access and SAS tokens are disabled in this tenant. Any suggestions would be apricated.

Answer 1

Hi Dan Gilmore,

Thank you for sharing the details! The error you’re seeing usually indicates a permission or authentication issue when accessing the MLTable or datastore during AutoML job submission.

Here are some steps to resolve it:

1. Verify Storage Permissions

Ensure your account has Storage Blob Data Contributor or equivalent role on the storage account linked to your Azure ML workspace. AutoML needs access to read/write data assets.

• https://learn.microsoft.com/en-us/azure/storage/common/storage-auth?view=azureml-api-2

2. Check Authentication on Compute

The error NoIdentityOnCompute suggests the compute instance or cluster lacks identity to access the datastore.

• Assign a Managed Identity to your compute and grant it access to the storage account.
• https://learn.microsoft.com/en-us/azure/machine-learning/how-to-identity-based-service-authentication?view=azureml-api-2

3. Validate MLTable and Data Asset

Confirm the MLTable YAML and data asset paths are correct and accessible. If using a lab environment, re-upload the dataset and re-register the MLTable.

• https://learn.microsoft.com/en-us/azure/machine-learning/how-to-create-data-assets?view=azureml-api-2

4. Troubleshoot AutoML Job

Check the child run logs in the Azure ML Studio under Outputs + Logs for detailed error traces.

References:

• Set up AutoML training
• Troubleshoot AutoML experiments

If the issue persists after these checks, please share:

• Your workspace setup details in private chat (subscription, resource group)
• Compute type and identity configuration
• Any additional error logs from std_log.txt

This will help us provide further guidance.

Thankyou!

Answer 2

You might want to consider setting up a user-assigned managed identity as described in https://learn.microsoft.com/en-us/azure/machine-learning/how-to-use-serverless-compute?view=azureml-api-2&tabs=python

---

If the above response helps answer your question, remember to "Accept Answer" so that others in the community facing similar issues can easily find the solution. Your contribution is highly appreciated.

hth

Marcin