A Major Manufacturer's Customer-Data Batch: From Source-Code Analysis to Infrastructure Rebuild

What changed

• Starting point
  Existing infrastructure environment lost
• Rebuild method
  From-scratch, driven by source-code analysis
• Execution platform
  Serverless on AWS ECS Fargate

Challenge

A major manufacturer's customer-data processing batch system was still running, but the incumbent vendor had lost the infrastructure environment and almost no configuration information remained. Any overhaul meant working forward while inferring how one piece of code connected to another. On top of that, the batch could not be stopped — so we had to keep production running while working backward from the source code to the infrastructure requirements and rebuilding everything from scratch. It was a demanding set of starting conditions.

Approach

1. 01 Read the existing source code closely and reverse-engineered the required infrastructure (endpoints, triggers, data types)
2. 02 Rebuilt the lost environment from scratch as an event-driven architecture on AWS ECS Fargate + EventBridge
3. 03 Built an S3 → EventBridge → Fargate chain so processing starts within seconds of data arriving
4. 04 Separated processing into per-data-type classes in Python — member data, contract data, usage records, churn determination, and more
5. 05 Externalized connection details and thresholds with AWS SSM Parameter Store
6. 06 Set up CI/CD for the Docker images with AWS CodeBuild for a reproducible build flow

Key success factors

Reverse-engineering from source code

We reconstructed the lost infrastructure requirements from the source code and reproduced the architecture from the ground up.

Responsiveness through event-driven design

The S3 → EventBridge → Fargate chain lets processing start within seconds of data arriving.

Reproducible CI/CD

CodeBuild + Docker made the entire build flow reproducible and removed reliance on individual knowledge.

Externalized configuration

Managing connection details and thresholds in SSM Parameter Store made operational tuning possible without code changes.

Solution

We read the source code closely to reverse-engineer the infrastructure requirements and rebuilt everything from scratch as an event-driven architecture centered on AWS ECS Fargate + EventBridge. We externalized configuration with SSM Parameter Store and set up a reproducible CI/CD pipeline with Docker + CodeBuild. From a "the environment is gone" state, we migrated to a new foundation while keeping production running.