Rather than rebuild from scratch, we took an incremental approach to breaking the monolith apart. First, we mapped out logical boundaries in the existing application — GPS tracking, driver notifications, route history, and reporting could each stand on their own as separate services. We containerized each one using Docker so they could be deployed, updated, and scaled independently.

Migration happened one service at a time, starting with reporting since it was the lowest-risk piece, and working up to GPS tracking last, once we’d confirmed the approach held up in production without issues.

The database ended up being the harder part of the project. The original system had a single shared database that every part of the application read and wrote to directly, so separating services meant untangling which service actually owned which data — without breaking historical reports that dispatch managers pulled from daily. A few edge cases around older trip records took extra work to sort out, since some of that data didn’t cleanly map to a single service.

We also set up per-service monitoring using Prometheus, with alerts routed to the team’s Slack, so if the notification service started failing, it would get flagged within minutes instead of someone noticing only after drivers started calling in.