Authentication issues when starting notebook compute

Hello @bleeksiel_lhc,

I understand that you're experiencing issues with authentication when trying to access your compute instance in Azure Machine Learning Studio.

Here are a few steps you can take to troubleshoot the problem:

Please ensure that you have the required permissions to access both the workspace and the compute instance. Specifically, you need the "Can attach to" permission on the compute resource to connect successfully.

As you're no longer seeing the authentication notification, please try manually re-authenticating to your Azure Machine Learning workspace. You can do this by signing out and then signing back in, which may help re-establish the required permissions.

If the issue persists after re-authenticating, consider restarting the compute instance. This can help resolve any residual permission or authentication-related problems.

Make sure that WebSocket communication is allowed on your network, as it's required for connecting to notebooks. Verify that your network permits WebSocket connections to *.instances.azureml.net and *.instances.azureml.ms.

If you're encountering a kernel error, reviewing the Jupyter logs can help identify the cause. You can access the logs by running the command sudo journalctl -u jupyter in the terminal. If kernel crashes occur frequently, it may be beneficial to switch to a compute instance with higher memory capacity.

If your compute instance is deployed in a workspace with a private endpoint, ensure you're accessing it from within the associated virtual network and that any required custom DNS settings are correctly configured.

Please refer this Troubleshooting.

I hope this helps. Do let me know if you have further queries.

Thank you!

The user has the "ML Data Scientist" role, so permissions are not an issue.

The workaround for seems to be to start the compute instance from the "Compute" section of AML studio, wait until it has started and only then open a notebook. After that everything works fine.

As I said, the disappearance of the "click here to authenticate" prompt is the problem, not any of the issues in the troubleshooting (which I did before posting the original question).

[Update 2025-05-05]

The workaround doesn't work. Not only does the user get kicked out, but other users get kicked out as well. Since the workspace is temporarily inaccessible the suggested solution of restarting the compute is not possible. Also tried

• Clearing Browser cache
• Signing out and signing in
• Rebooting client PC

Possibly related https://stackoverflow.com/questions/78569327/user-xxx-does-not-have-access-to-compute-instance-yyy-azure-machine-learning, though it isn't clear what is meant by "upload the DNS to get the new route from the compute instances IP and it's name"

Hi @bleeksiel_lhc,

If your source is secured using a Virtual Network, it's important to ensure that DNS resolution is correctly directing traffic to the private IP address defined in your DNS configuration. You can check this by running an nslookup on the fully qualified domain name (FQDN).

If the FQDN does not resolve to the expected private IP, you may need to create or verify the private endpoint, fix any broken virtual network links, or add the appropriate CNAME or A records in your DNS server, especially if you're using a custom DNS setup.

If DNS resolution is working as expected and the FQDN resolves correctly, the next step is to capture a network trace using browser developer tools or check your firewall and proxy logs. This will help you identify whether any required host URLs are being blocked. Ensure that these URLs are whitelisted in your firewall or proxy to allow proper connectivity.

Kindly refer this Troubleshoot private endpoint connection.

Host URL whitelisting - Configure inbound and outbound network traffic.

Thank you!