Session creation failure with compute instance in Azure ML Studio – Custom application deployment failed

Question

Hello,

I am working in Azure Machine Learning Studio where I have a compute instance running, and I have successfully created a connection with Azure OpenAI service.

As part of my testing, I was working with Prompt Flow and attempted to create a compute session using the running compute instance. However, I encountered the following error:

Session status messages: Custom application deployment failed after retry with new compute sessionXXXXXXXXXXXX, abort session creation

The diagnostics log said it try to install session with config like below:
2025-06-11 12:01:07 +0 Information Try to Install session with config: {"SessionId":<SESSION ID>,

"VmSize":null,"MaxIdleTimeSeconds":null,"Identity":null,"ComputeName":null,"EnableMultiContainer":false, "ChatGroupExecutionEnvironments":null,"Language":0,"BaseImage":null,"Packages": "promptflow==1.18.0","azure-ai-ml==1.27.1"], "RequirementsFileName":"requirements.txt","RuntimeImage":null}

Could you please help on fixing this? thanks.

Answer 1

Hi Vishal Shinde,

Error typically indicates a failure during the environment setup phase for a Prompt Flow session in Azure Machine Learning. Based on the diagnostic logs, the session configuration appears to be missing key parameters like ComputeName, VmSize, and BaseImage, which may lead to deployment failures.To help resolve the issue, please consider the following steps:

• Compute Instance Setup: Ensure that a healthy and running compute instance is explicitly selected when starting the Prompt Flow session. Restarting the compute instance and creating a new session can also help resolve transient issues. Troublshooting,, Create/Manage a compute session
• Environment and Dependencies: Use a simplified requirements.txt file with only the necessary packages: promptflow==1.18.0 azure-ai-ml==1.27.1 Alternatively, create a custom environment in Azure ML Studio using a Conda YAML or Docker configuration. If you're using a custom image, ensure you're building on a supported base image and all required dependencies are included. You may refer to How to customize base image
• Base Image Configuration: Specify a compatible base image such as: mcr.microsoft.com/azureml/openmpi4.1.0-ubuntu20.04
• Docker Image Accessibility: If you're in a private link-enabled workspace or behind restricted networks, ensure access to external image sources like pkg-containers.githubusercontent.com and ghcr.io is allowed. Lack of access can cause environment build failures.
• Session Configuration Parameters: The diagnostic output shows several fields like VmSize, Identity, and MaxIdleTimeSeconds are set to null. Setting these explicitly when creating the session can help avoid default misconfigurations.
• Permissions and Role Assignments: Ensure your user account and the workspace-managed identity have appropriate roles, such as Reader and Storage Blob Data Reader, especially for accessing backing storage. If role assignments were recently added, please allow 5–10 minutes for propagation.
• Region Considerations: In rare cases, changing to a different region may help if certain resources are experiencing regional constraints.

For your reference: Azure ML Workspace cannot create prompt flow compute session.
I hope this information helps. Thank you!

Answer 2

Hi @Vishal Shinde Please check which region you have created your resources in. I would suggest try creating your resources in EastUS2 or Sweden Central, as these are shown as Supported locations, look at the snip below.

Thanks

Answer 3

Hi Vishal Shinde (Persistent Systems Inc),
Have you tried this on your end? Could you please check and confirm, and let us know if you encounter any further issues?