The human-centric platform for production ML & AI

Outerbounds

Hi, after setting up AWS with the provided CloudFormation template, all the resources stay active and collect unnecessary cost. How do we make sure the instances terminate? (Especially EC2 and SageMaker). And is it possible to set up a local metadata database for tracking DAG and processes, but recording the artifacts on S3?

You don't want to delete the whole stack completely? Just some individual components?

Destroying the whole stack created by cloudformation is easy. There is a delete command.

<@U076N5S84J3> the cloudformation template is a starter template. you can tweak it to ensure that resources scale down according to your needs (including moving the database to serverless aurora if you so prefer).

you can set up a local metadata database, but your workloads on AWS Batch may not be able to easily communicate to this service. You can always  use the local filesystem as metadata backend by simply providing `python flow.py --metadata=local run` - this allows for artifact tracking to still happen on s3