Moving a large-scale metrics pipeline from StatsD to OpenTelemetry / Prometheus

Airbnb shares their production-tested approach to migrating a large-scale metrics pipeline from StatsD to OpenTelemetry and Prometheus. The migration prioritizes collecting metrics at full scale first to expose bottlenecks before validating the new system. They reduce CPU overhead from 10% to under 1%, address high-cardinality performance issues with delta temporality, and implement streaming aggregation via vmagent to achieve an order of magnitude cost reduction.