Question About Shortening Auto-Termination Time in SageMaker Studio

Unfortunately, it is not possible to configure SageMaker Studio to automatically terminate idle sessions sooner than 60 minutes. The minimum idle timeout that can be set for SageMaker Studio applications, including JupyterLab, is 60 minutes (1 hour).

This limitation is built into the SageMaker service. The IdleTimeoutInMinutes setting, which controls how long SageMaker waits after an application becomes idle before shutting it down, has a minimum value of 60 minutes. This applies to both the domain-level and user profile-level configurations.

While it's understandable that you'd want to optimize costs by reducing idle time, especially given the per-second billing with a 1-minute minimum, SageMaker does not currently offer an option to set the auto-termination time below 60 minutes.

If you need more granular control over idle resource management, you might consider exploring custom solutions. For instance, some users implement their own scripts or use external tools to monitor and shut down idle resources more aggressively. However, these approaches would be separate from SageMaker's built-in auto-termination feature and would require additional setup and management.

For cost optimization within the constraints of SageMaker's features, you could:

1. Set the idle timeout to the minimum of 60 minutes.
2. Use CloudWatch Billing Alarms to monitor your usage and costs.
3. Manually shut down resources when you know you won't be using them for a while.
4. Consider using SageMaker notebook instances with lifecycle configurations instead of Studio if you need more control over auto-stopping behavior.

Remember to always manually stop your resources when you're done working to avoid unnecessary charges, especially for periods shorter than the 60-minute auto-termination window.
Sources
IdleSettings - Amazon SageMaker
Update default idle shutdown settings - Amazon SageMaker
MLCOST-16: Stop resources when not in use - Machine Learning Lens

The minimum idle timeout that can be configured for SageMaker Studio applications is 60 minutes. This is a built-in limitation of the service that applies to both domain-level and user profile-level settings.

While this is the good approach for cost management, but given the per-second billing, there is other optimal way to work around this limitation is to use lifecycle configuration scripts for custom automation and configure the IDLE_TIME_IN_SECONDS parameter to your desired number of seconds [1].

Additionally, for shorter work sessions, it's recommended to manually stop resources when finished rather than relying on auto-termination.

Source:

[1] https://github.com/aws-samples/sagemaker-studio-apps-lifecycle-config-examples/tree/main/jupyterlab/auto-stop-idle