The Problem

In a multi-outlet retail business, keeping inventory and transaction data in sync across all locations in real-time is a significant challenge. Traditional batch processing introduces delays that can lead to overselling, inaccurate stock counts, and poor customer experiences.

Architecture Overview

I designed an event-driven architecture using Apache Kafka as the central message broker. Each retail outlet runs its own lightweight microservice that publishes transaction events to Kafka topics. A central consumer service processes these events and updates the master database.

Key Components

Outlet Service — Lightweight Go service running at each outlet, capturing POS transactions and publishing to Kafka
Central Consumer — Processes events from all outlets, handles deduplication and conflict resolution
Stock Sync Engine — Maintains real-time inventory counts across all locations
Dashboard API — Provides real-time analytics and monitoring endpoints

Implementation Details

Each outlet microservice is containerized with Docker and deployed independently. The services use a local SQLite database as a write-ahead log to ensure no transactions are lost even during network outages. When connectivity is restored, events are replayed to Kafka in order.

The central consumer implements exactly-once semantics using Kafka's transactional API combined with idempotent writes to PostgreSQL. This guarantees that even if messages are redelivered, the final state remains consistent.

Results

The system successfully processes thousands of transactions per minute across dozens of outlets with sub-second synchronization latency. Stock accuracy improved from 94% to 99.7%, and the operations team gained real-time visibility into sales and inventory across all locations.

Blog

Automated Golang Deployment to Server with GitHub Actions and Docker