Build a data synthetic pipeline using ECS, Eventbridge & Codepipeline for a Data Science Team.
- EC2 GPU On-demand instance
- p2.8x large instance (8 NVIDIA K80 GPU, 32 vCPUs, 488 GiB RAM, $7.2 hr per hour)
- Org VPC with Private Subnets
- AWS ECS (Elastic Container Service) - EC2 launch type with awsvpc ENI ECS task networking; private IP addresses
- ECS Tasks with maximum CPU of 10240 CPU units ( 10 vCPUs) and 100 GB memory per task
- IAM System user for Vendor
- S3 Buckets ( output-vendor-bucket and data-team-bucket)
- Cloudwatch logs access for vendors
- Cloudformation for provisioning of resources
- CodePipeline for CI/CD - gets triggered on a new docker image push to ECR
- Eventbridge(event rules) & Cloudtrail - The event rules, trigger the ECS validation and evaluation script.
- ECR for Docker Image Repository
- Custom ecs-deploy.sh script in Codebuild that updates the task definition and launches the ECS task