Migrating a 12-year-old logistics platform off bare metal

Northwind's core dispatch platform had run on the same self-hosted racks for over a decade. Every hardware failure meant a multi-hour outage, and the on-call rotation had grown to dread pager alerts.

A previous attempt at migration had been abandoned after six months when the team discovered undocumented dependencies between the dispatch service and a scheduling job that only one former employee understood.

We started with a full dependency audit, mapping every service-to-service call before writing a single line of infrastructure code. This surfaced the scheduling job dependency that had stalled the earlier attempt.

Migration ran in five phases over ten weeks, moving one service at a time behind a traffic-shifting proxy so any phase could be rolled back independently without affecting the rest of the platform.

Infrastructure was rebuilt as code from day one, replacing years of manual server configuration with a reproducible, version-controlled setup.

The full migration completed with zero unplanned downtime across all five phases.

Incident response time dropped sharply once the on-call team had proper observability and could roll back a single service instead of an entire deployment.

Northwind's infrastructure costs came down as unused legacy capacity was decommissioned.