Hi Beauty

# Problem

The operation ran on 3 monolithic EC2 instances with PM2, local Mongo and hand-configured Nginx — no IaC, secrets hardcoded in .env files and manual deploys. On top of that, the cloud provider was about to shut down service, forcing a full migration to a new AWS account under time pressure and with no room for business downtime.

# Solution

I designed the new infrastructure on AWS EKS 1.33 from scratch using Terraform as the single source of truth: VPC, node groups (on-demand + spot), NLB with ACM TLS, ECR, ElastiCache Redis. I installed the PSMDB operator to run 3-replica MongoDB clusters per country with HA (quorum), centralized 42 secrets in AWS Secrets Manager synced to the cluster via ExternalSecrets Operator + Reloader. Dockerized and wrote K8s manifests for 21 apps (Express, NestJS and Next.js), reusable GitHub Actions workflows with OIDC (no long-lived keys) and ArgoCD/manual deploys by dispatch. Cutover of 22 DNS records via Cloudflare API, full backup of the 3 legacy mongos and incremental S3 sync from legacy bucket to the new one via CronJob. Post-migration I found a \$regex query saturating the Mongo primary to 100% CPU (getGuide over 718k docs) and fixed it with equality lowercase + compound index: 27s → 0ms.