The Problem #
A GPS fleet tracking system handling real-time vehicle tracking was built on Laravel (PHP) with MySQL. As the number of devices and location data volume grew, the system experienced bottlenecks:
- Response times getting progressively slower on reporting and history endpoints
- Database connections frequently exhausted by complex queries on the massive device data table
- Expensive maintenance: every change required understanding an increasingly complex PHP codebase
The Solution #
Full backend conversion from Laravel → Go using Clean Architecture pattern, with clear database role separation: MySQL for domain data, PostgreSQL + TimescaleDB for location data (time-series).
Architecture #
Request → Controller → UseCase → Repository → MySQL (domain) / PostgreSQL+TimescaleDB (location)
↕
Gateway → External API / RabbitMQ
Each layer has clear responsibilities:
- Entity: 33 domain models (device, geofence, vehicle, driver, invoice, etc.)
- UseCase: 40 business logic layers
- Repository: 82 data access files (read/write split, master-slave aware)
- Route: 370+ API endpoints
Tech Stack #
| Component | Technology |
|---|---|
| Language | Go 1.26 |
| HTTP Framework | GoFiber v2 |
| Database | MySQL (domain data) + PostgreSQL/TimescaleDB (location/time-series data) |
| ORM/Driver | pgx/v5 (raw SQL performance) |
| Messaging | RabbitMQ (amqp091-go) |
| Cache | Ristretto (in-memory) |
| Migration | golang-migrate |
| Validation | go-playground/validator |
| Auth | JWT + unified auth |
Key Features #
- Real-time tracking: device location data processed and stored in TimescaleDB hypertable
- Geofencing: entry/exit alerts, scheduled geofences
- Dashboard & reporting: device summary, route history, fuel monitoring
- Fleet management: vehicle, driver, device assignment
- Billing: invoice, partner, payment integration (BCA VA)
- Multi-tenancy: enterprise, user group, shared link
- IoT Hub: device commands, incident tracking
Results #
- ~194K lines of Go code replacing the PHP codebase
- 370+ API endpoints with full parity against the Laravel version
- Automatic read/write split: heavy queries routed to slave DB
- TimescaleDB hypertable for the device data table receiving millions of records per day
- Efficient build time and binary, single binary deployment via Docker
Lessons Learned #
- Clean Architecture enables gradual migration: each endpoint could be converted one by one without breaking changes, allowing v2 (Laravel) and v3 (Go) to run in parallel during the transition
- pgx/v5 > ORM for high-throughput: full control over SQL queries delivers more predictable performance compared to ORM abstractions
- TimescaleDB is a game-changer: hypertables + compression keep millions of location records per day stored efficiently and fast to read.