SageMaker endpoint stuck in "Updating" status

Hi, Here are a few suggestions that may help:

1.- Retry Deleting the Endpoint: Even though the endpoint is in the "Updating" status, you can still try to delete it. The DeleteEndpoint API operation should be able to delete the endpoint, even if it's in an unusual state. Try deleting the endpoint again and see if it's successful this time.

2.- Delete the Endpoint Configuration: If you're still unable to delete the endpoint, try deleting the associated endpoint configuration first. The DeleteEndpointConfig API operation allows you to delete the endpoint configuration, which may then allow you to delete the stuck endpoint. https://docs.aws.amazon.com/sagemaker/latest/dg/realtime-endpoints-delete-resources.html

3.- Use the AWS CLI or SDK: You can try deleting the endpoint and configuration using the AWS CLI or an AWS SDK, such as the AWS SDK for Python (Boto3). This may provide more detailed error messages that could help identify the root cause.

Delete the endpoint using the AWS CLI

aws sagemaker delete-endpoint --endpoint-name <endpoint-name>

Delete the endpoint configuration using the AWS CLI

aws sagemaker describe-endpoint-config --endpoint-config-name <endpoint-config-name>

4.- Check for Dependent Resources: Ensure that there are no other resources, such as models or other configurations, that are still dependent on the endpoint you're trying to delete. These dependencies may be preventing the deletion from completing successfully. Example command to delete model via AWS CLI aws sagemaker describe-endpoint-config --endpoint-config-name <endpoint-config-name> https://docs.aws.amazon.com/sagemaker/latest/dg/realtime-endpoints-delete-resources.html

I will recommend to try again, even point 1 and 2, if any of this steps works please contact AWS Support. They can investigate the issue further.

Hi,

I have a related question: why can't we delete an endpoint in the "creating" status? Why do I need to wait until the timeout?

For best practices, I launch the endpoint creation and monitor the logs in CloudWatch using live tail logs. If at any point I realize that there is something that needs to be adjusted (such as forgotten files, conflicting dependencies, etc.), I would prefer to delete the endpoint in the "creating" status, make the necessary updates, and then start another endpoint creation process—without losing time and energy waiting for the timeout to reach.

I hope this issue can be addressed.