Building a high-volume metrics pipeline with OpenTelemetry and vmagent

Short summary

Airbnb migrated its production metrics infrastructure from StatsD to OpenTelemetry and Prometheus, reducing CPU overhead from 10% to under 1%. Using a dual-write strategy with their shared metrics library, they rolled out OTLP across 40% of services with low friction while maintaining StatsD as a fallback. High-volume services required delta temporality tuning to manage memory; this pattern applies to any large-scale observability system upgrade.

• •CPU overhead dropped from 10% to under 1% after migrating to OTLP
• •Dual-write approach enabled broad rollout across 40% of services with minimal friction
• •High-cardinality emitters required delta temporality tuning to avoid memory pressure

Generated with AI, which can make mistakes.