From Silos to Source of Truth: Migrating a Legacy Enterprise ERP

The Legacy Monolith Burden

Many enterprises find themselves trapped by their own success. A massive logistics client came to us with a 15-year-old on-premise ERP system built entirely on a single PHP monolith. While it had served them well, it had become a "Big Ball of Mud." Any update required weeks of regression testing, and during peak shipping seasons, the entire system would slow to a crawl because independent modules (like invoicing and label printing) couldn't be scaled independently.

Implementing the Strangler Fig Pattern

Rebuilding a system of this magnitude from scratch introduces massive business risk. Instead, we utilized the Strangler Fig Pattern—a systematic process of intercepting requests to the old system and gradually overriding them with a new, microservice-based architecture running alongside it in the cloud.

Here is the step-by-step technical blueprint we followed:

• The Traefik API Gateway: We installed a powerful edge router (Traefik) in front of the legacy ERP to act as the traffic cop. Initially, 100% of traffic simply passed straight through to the old system.
• Event Sourcing to Sync State: We implemented Change Data Capture (CDC) via Debezium on the legacy database. Any write to the old database instantly triggered a Kafka event, keeping the new cloud PostgreSQL databases perfectly synchronized.
• Isolating Domains: We identified the "Shipping Rate Calculator" as the lowest-hanging fruit. We built a highly scalable Go microservice to handle this specific function.
• Traffic Redirection: We configured the API gateway to catch all /api/v1/rates requests and route them to the new Go microservice instead of the legacy PHP app. Completely seamlessly to the front-end clients, the new service took over.

Achieving Zero-Downtime Migration

Over a period of 14 months, we systematically "strangled" out the core functionalities: Inventory, Invoicing, Carrier Routing, and finally User Auth. The old monolith slowly shrank until it was nothing but an empty shell that we finally turned off.

By extracting these domains into Dockerized microservices deployed on Kubernetes, cluster auto-scaling now handles seasonal shipping spikes effortlessly. The migration was completed with absolutely zero system downtime, reducing infrastructure costs by 35%.