Migrations that move the needle: Mongo to Postgres and Lambda to services

The migrations that matter are rarely greenfield. They happen while traffic is live and the team is already overloaded.

Patterns that work

Document store → relational when you have outgrown flexible schemata and need joins, constraints, and predictable query plans. Cut over by bounded domain (one Integrator-style service at a time), dual-write or backfill carefully, and measure p99 before and after.

Sparse Lambda → dedicated service when cold starts, connection churn, and orchestration complexity dominate. A long-lived service with clear SLOs often drops tail latency dramatically — I have seen on the order of −60% p99 on critical endpoints when the boundary was chosen well.

What to avoid

Big-bang cutovers, migrations without observability, and “rewrite everything” programs that outlive the business need. Prefer a vertical slice with a kill switch.

Need a second set of hands for a migration? Book a call.