Return to Project Vault

Lead Full-Stack Engineer & Architect | AHCC | 2019

//Context & Scale

AHCC is an enterprise organization managing mission-critical business systems on single-point-of-failure on-premise physical infrastructure. Co-located relational databases and local disk storage dependencies caused acute operational bottlenecks: automated backups caused severe disk I/O degradation, system updates required risky manual intervention, and traffic surges were hard-bounded by bare-metal capacity. Retained as Lead Full-Stack Engineer and Architect to modernize the platform, author declarative Infrastructure as Code (IaC), containerize services, and execute a zero-downtime cloud migration of an active 550 GB PostgreSQL cluster to AWS.

//Architectural Decisions

  • ->Asynchronous Multi-Phase Database Sync: Configured asynchronous logical replication between on-premise PostgreSQL and AWS Aurora PostgreSQL Multi-AZ, verifying consistency via row-level checksum verification until replication lag stabilized below 15ms prior to final cutover.
  • ->Split-Brain Prevention Protocol: Reduced DNS TTL on Route 53 to 60 seconds 72 hours before cutover, enforcing temporary read-only database locks on the on-premise source at T-0 to reject stale writes during the final delta drain.
  • ->Connection Spikes with PgBouncer: Deployed PgBouncer in transaction pooling mode in front of the Aurora PostgreSQL cluster, enabling the infrastructure to absorb bursts of up to 4,500 concurrent connections without backend process exhaustion.
  • ->Stateless Container Decoupling: Decoupled local disk dependencies by migrating user asset pipelines to Amazon S3 with pre-signed upload URLs and CloudFront CDN distribution, packaging applications into stateless Docker containers on AWS ECS.

//Engineering Execution

  • ->Declarative Terraform Provisioning: Authored modular Terraform manifests to provision VPC networking, multi-AZ subnets, security policies, ECS task definitions, and Aurora failover rules in version control.
  • ->Multi-Service Containerization: Packaged multi-language services (PHP, Node.js, and Go) into lightweight multi-stage Docker images deployed to AWS ECS clusters.
  • ->Cutover Rehearsal & Validation: Executed simulated staging cutovers to validate replication consistency, database foreign key constraints, and read/write failover procedures before live execution.

//Measurable Impact

  • ->Zero Customer Downtime: Migrated the 550 GB enterprise PostgreSQL database and active services to AWS within a scheduled sub-minute DNS cutover window with zero data corruption.
  • ->4,500 Concurrent Connection Throughput: PgBouncer transaction pooling prevented database pool exhaustion during high-concurrency traffic bursts.
  • ->Automated Provisioning: Reduced infrastructure spin-up times from several days of manual server setup to automated, reproducible Terraform execution.