A production container-hosting architecture (ECR through EC2, auto scaling, load balancing, and DNS) provisioned entirely with Terraform and deployed automatically through Bitbucket Pipelines. Releases cost nothing; instance failures fix themselves.
ECR → EC2 → Auto Scaling Group → Application Load Balancer → Route53
The client ran containerized workloads that needed to survive instance failures, absorb traffic spikes, and take new versions without downtime, with no dedicated ops team to babysit any of it.
The answer was a fully codified stack: Docker images flow from ECR onto EC2 instances managed by an Auto Scaling Group, fronted by an Application Load Balancer, with DNS in Route53. Every resource lives in Terraform; every deploy starts with a git push.
Replacing instances naively cut off in-flight requests during every release.
Fixed: ALB connection draining plus ASG instance refresh gated on health checks, so new instances must pass before old ones terminate. Deploys became genuinely zero-downtime.
Manually-created dev and prod diverged constantly, producing "works in dev" failures.
Fixed: one set of Terraform modules with per-environment variable files over remote state, so identical infrastructure is stamped out per environment, changed only through pull requests.
Inherited scaling policies scaled up and down repeatedly under perfectly steady load.
Fixed: re-tuned CloudWatch alarms with proper cooldowns and target-tracking policies based on real traffic patterns, giving a stable fleet and fewer wasted instance-hours.
Zero
downtime deployments after rollout
100%
of infrastructure in Terraform
Minutes
from push to production
24/7
self-healing via health checks