Skip to main content

Atlas now has a second staging workload root on the shared staging foundation

ยท One min read
Atlas Infra

Atlas now has a dedicated terraform/staging2/ root for a second staging workload plane.

This rollout keeps the state isolated but deliberately avoids cloning the entire staging foundation.

terraform/staging2 now:

  • keeps its own backend state key at staging2/terraform.tfstate
  • reuses the VPC, public/private subnets, approved security groups, and MSK cluster already owned by terraform/staging
  • reuses the shared staging application ECR repositories instead of creating *2 repositories
  • creates a second ALB, ECS cluster, runtime secrets, dashboard/scoring/Camunda/Kafka UI services, RDS instances, and Valkey replication group with resource names ending in 2

Operationally, the new root also formalizes an important workflow decision: the ECS task definitions for events ingestion, dashboard backend, and scoring are treated as bootstrap artifacts only. Terraform still defines the base contract, but later manual AWS-side task-definition revisions are allowed to drift without Terraform forcing the live container definitions back to the committed image and sidecar shape.

That tradeoff keeps staging2 useful as a fast test plane while preserving the existing operator workflow for manual runtime adjustments.