04/03/2023, 7:44 PM
Dear Pulumi people 🙂 I have manually created an EC2 instance and have created all other resources (S3, Lambda, IAM, ...) with pulumi. Next, I would like to set up the EC2 instance with pulumi and add an autoscaling behavior. I guess EC2 is the right choice because we need GPU power. Which workflows or articles do you recommend to set it up such, that I get 0 or more EC2 instances, depending on the load? How should the load be detected? By AWS or via some other measurement (e.g. a metric derived from prefect 2 workflow orchestration). Thanks in advance 🙏


04/03/2023, 8:33 PM
this is a really open ended question that depends a lot on the context of your situation. One way is to use AWS ASGs and autoscale based on CloudWatch metrics for GPU usage. Another possibility is with a container orchestrator like K8s/EKS, which can be set up to autoscale as well, assuming your workloads are already containerized.


04/05/2023, 7:35 AM
Yes, I intentionally wanted to ask it open first 🙂 OK, I guess I will go for the AWS Autoscaling Group first 👍 Are there already workflows available on how to create a such via pulumi?