Sandman was designed to reduce the costs of your AWS SageMaker notebook instances. Using Sandman, you can configure which and when to stop them. You simply need to deploy Sandman's lambda functions and that's it, it will take care of the rest.
You might have in your organization a set of notebook instances that don't need to run 24/7. If you stop these instances in a certain period of time, you can save quite some money. Reducing the uptime from 24/7 to 16/5 can reduce up to 50% of your SageMaker notebook costs.
Before deploying, you can customize where to deploy the lambda functions and at which time you want to start/stop the instances. Open the file variables.tf and modify the default values. You can modify:
- The AWS region you want to deploy Sandman to.
- The start and stop schedules (aka when Sandman will start and stop your instances - UTC time).
- The python runtime (Sandman is compatible with Python >3.6).
Sometimes you have instances that need to run 24/7. Sandman can ignore instances with the tag sandman=ignore. So if you have an instance that you don't want Sandman to touch, just add this tag to your notebook instance.
To deploy Sandman on your account, you need to have awscli configured on your machine and then run ./deploy.sh. This script will create:
- The policy that allows the lambda function to:
sagemaker:StartNotebookInstancesagemaker:StopNotebookInstancesagemaker:ListNotebookInstancessagemaker:ListTagsec2:DescribeSubnetsec2:DescribeVpcsec2:DescribeNetworkInterfaces
- The role that uses the previous policy.
- The lambda functions to start and stop the notebook instances.
- Cloudwatch events that triggers the lambda functions.
$ git clone https://github.com/porthunt/sandman-aws.git
$ cd sandman-aws
$ vim variables.tf # modify the variables
$ ./deploy.shIn case you don't want to have Sandman anymore, just run ./destroy.sh.
Sandman is completely unlicensed. Feel free to just modify it as you want.