How to prevent that a new blob storage container is created for each job

I have created a resource group, workspace and a compute instance. Every time I run a job a new blob storage container is created to store the Python files linked to that job, instead of using the default container azureml-blobstore-

romungi-MSFT 43,696 Reputation points Microsoft Employee

2023-10-03T10:11:01.0766667+00:00

@Jeroen Hamers (imec-OnePlanet) Are you sure the runs are creating a new container rather than a folder or directory? If you are creating a new run from the ML portal or the SDK the runs should not create a new container, the runs or jobs will use the default blobstore or azureml container to record the runs. Also, have you used an already existing storage account as your workspace storage account or was it created along with your workspace? Thanks!!
Jeroen Hamers (imec-OnePlanet) 20 Reputation points

2023-10-03T11:26:19.3233333+00:00

Hi, Thanks for the quick response! To answer your question:

Are you sure the runs are creating a new container rather than a folder or directory?

Yes, I checked it on the blob storage. I would expect that it would use the default container which I did not change.

Also, have you used an already existing storage account as your workspace storage account or was it created along with your workspace?

No, a new storage account was created when I have setup the workspace.

Best regards,

Jeroen

Jeroen Hamers (imec-OnePlanet) 20

The commands I used to setup everything are:

az login

az group create --name <resource_group_name> --location <location>

az configure --defaults group=<resource_group_name>

az ml workspace create --name <workspace_name>

az configure --defaults workspace=<workspace_name>

az ml compute create --name <compute_name> --size STANDARD_DS11_V2 --type ComputeInstance

The following yaml file was used to setup the job:

$schema: https://azuremlschemas.azureedge.net/latest/commandJob.schema.json
experiment_name: <experiment_name>
description: <description>
environment: azureml:<environment_name>:<version>
compute: azureml:<compute_name>
type: command
inputs:
  input_folder: 
    type: uri_folder
    path: azureml://datastores/<blob_storage_name>/paths/<folder_name>
    mode: ro_mount
outputs:
  output_folder:
    type: uri_folder
    path: azureml://datastores/<blob_storage_name>/paths/<folder_name>
code: src
command: >-
  python <python_file>
  --input_folder ${{inputs.input_folder}} 
  --output_folder ${{outputs.output_folder}}

And finally to run the job I used:

az ml job create --file <job_file_name>.yml --web

Tume ‪ 10 Reputation points

2023-11-08T06:30:34.4133333+00:00

I'm having the same issue. A new blob container is being created for every job
Yu, Tao 0 Reputation points

2023-12-15T18:04:41.1466667+00:00

This is a huge problem for our storage account and each job's snapshot is in individually created container.
Faeh, Kevin 0 Reputation points

2024-01-10T15:21:22.24+00:00

I am having the same issue. The created container has partially the name of the default blob container and the name of the job. Did anyone find a solution?
Faeh, Kevin 0 Reputation points

2024-01-10T15:22:21.07+00:00

I am having the same issue. The created container has partially the name of the default blob container and the name of the job.
Aziz Nadirov 0 Reputation points

2024-01-24T11:16:50.81+00:00

I also have the same issue

1 answer

Diondra Peck 10 Reputation points Microsoft Employee

2024-02-15T04:26:33+00:00

This is expected behavior as of azure-ai-ml version 1.7.0. Code snapshot creation was transitioned from directory-based within a container to container-based within a datastore in order to accommodate users with fine RBAC (Resource Based Access Control) policies that don't permit container creation for all users of a workspace. The change wasn't announced in the 1.7.0 CHANGELOG; however, given strong user feedback and confusion about the change, we will be adding it to the CHANGELOG retroactively to create clarity: https://github.com/Azure/azure-sdk-for-python/pull/34332.

The AzureML SDK team really appreciates the feedback and are looking to see how we can move forward in a way that maintains the RBAC benefits of the change while addressing UX concerns.
Please sign in to rate this answer.

2 people found this answer helpful.

0 comments No comments
Sign in to comment

Share via

How to prevent that a new blob storage container is created for each job

1 answer