Overview

I owned the end-to-end migration of RegisterKaro's production platform from DigitalOcean to AWS (ap-south-1) — a stack serving 50,000+ customers — provisioning the entire environment as code and cutting over with zero downtime.

The problem

The legacy DigitalOcean setup had grown organically: limited isolation, manual provisioning, and a cost/reliability profile that didn't fit a platform this size. We needed a reproducible, multi-AZ environment in-region, with the database, secrets, and compute properly segmented — and the move had to happen without an outage for live customers.

What I built

The whole environment is defined in Terraform and Ansible — no click-ops:

• A multi-AZ EC2 Auto Scaling Group behind an Application Load Balancer, with all compute in private subnets.
• ECR for images, ElastiCache (Valkey) for caching, and Secrets Manager for credentials — TLS and AUTH enforced on the cache.
• Database access locked down: replaced broad 0.0.0.0/0 exposure with VPC peering and an explicit allowlist.
• API health checks decoupled from the message broker so a queue hiccup couldn't fail the load balancer's health signal.

The cutover

I ran a zero-downtime DNS cutover for all 50,000+ customers, kept the old stack on standby through a rollback window, and retired the legacy environment four days early once the new one proved stable. I also validated a ~5-minute recovery objective in a real restore drill before trusting it with production.

Impact

• Reproducible, multi-AZ production environment — fully in code.
• Zero customer-facing downtime during the move.
• A hardened security posture (private subnets, VPC-peered DB, enforced TLS/AUTH) that later underpinned a compliance audit.
• The foundation for the cost and reliability work that followed.