Scaling a Legacy Monolith to Reduce Cloud Costs by 40%

The Challenge

A rapidly growing B2B logistics platform hit a wall. Their codebase, a massive 5-year-old monolithic Node.js application, was choking under the weight of concurrent user traffic. Simple API requests were taking up to 4 seconds because heavy background tasks (like PDF invoice generation and route optimization algorithms) were blocking the main thread. To keep the app alive, the client was forced to provision massively oversized AWS EC2 instances, burning over $15,000/month in wasted compute costs.

The Acadify Solution Architecture

Acadify engineers executed a "Strangler Fig" migration—systematically pulling out heavy functions from the monolith without ever taking the live system offline.

• Event-Driven Queues: We implemented BullMQ and Redis to decouple the heavy PDF generation and routing logic into asynchronous background workers.
• Kubernetes (EKS) Spot Instances: We containerized the new worker microservices with Docker and deployed them onto an auto-scaling Kubernetes cluster utilizing AWS Spot Instances (which cost 70% less than On-Demand instances).
• Database Connection Pooling: We deployed PgBouncer in front of their PostgreSQL database to manage connection limits, instantly preventing the database lockups that were causing 502 Bad Gateway errors.

The Impact

By separating the web traffic from the heavy data-processing tasks, API response times dropped from an average of 4 seconds down to 120ms. More importantly, because the background workers were scaled elastically on cheap Spot instances, the client's total AWS monthly bill dropped by 40%. The entire migration was handled via strict CI/CD blue/green deployments, resulting in zero seconds of downtime for their users.