Consolidation, Not Refactoring
Six months of consolidation. One deployable. Here's what changed.
Last November, on a routine on-call rotation, we traced a latency spike through seventeen service boundaries only to find connection pool exhaustion in a service untouched for eight months. That was the moment we decided to consolidate.
The migration wasn't the hard part
The rewrite itself took three engineers six weeks. The real difficulty was organizational: convincing platform teams to give up their independently deployable services. Each group had built identity around its microservice. Consolidation felt like ceding territory, even when the data was unambiguous.
We ran the monolith in shadow mode alongside the existing fleet for two weeks, comparing responses on live traffic. P99 latency dropped from 340ms to 82ms. Error budgets expanded by an order of magnitude. Incidents fell 73% in the first month.