Airbnb's observability team just did something most large-scale engineering organizations only dream about: they rebuilt their entire metrics pipeline from the ground up, and they did it with open-source components. This isn't a story about a proprietary silver bullet or a flashy new tool. It's a story about simplification at a scale that most of us will never touch, yet the lessons apply to every team wrestling with data complexity.
The team moved away from StatsD and their custom Veneur aggregation layer, replacing it with a stack built on OpenTelemetry Protocol (OTLP), the OpenTelemetry Collector, and VictoriaMetrics' vmagent. The result is a system that ingests over 100 million samples per second in production. That number is staggering, but what matters more is the reasoning behind the migration. They didn't switch for the sake of novelty. They switched because the old pipeline had become a bottleneck, a patchwork of proprietary pieces that required constant maintenance and offered little flexibility. By consolidating on open standards and tools, they reduced operational overhead while gaining a clearer path forward.
For you, this isn't just a case study from a company with more servers than most countries. It's a signal that the era of homegrown observability backends is ending. If you've been holding onto a custom metrics pipeline because it works or because migrating feels risky, this is the permission slip you've been waiting for. The OpenTelemetry ecosystem has matured to the point where it can handle the most demanding production environments. You don't need a team of specialists to build and maintain your own aggregation layer anymore. The tooling exists, it's proven, and it's free to adopt.
What's particularly refreshing here is the emphasis on simplicity. Too often, engineering leaders equate sophistication with complexity. Airbnb's new pipeline flips that assumption. By standardizing on OTLP and using vmagent for collection, they reduced the number of moving parts. That's not a step backward; it's a step toward resilience. Fewer components mean fewer failure modes, and open standards mean you're not locked into a vendor's roadmap. This is the kind of pragmatic thinking that should guide your own infrastructure decisions, whether you're running ten services or ten thousand.
The takeaway is straightforward: your metrics pipeline should be boring, reliable, and built on open standards. If you're still maintaining custom aggregation logic or proprietary agents, start planning your exit. Not because today's system is broken, but because tomorrow's scale will be. Airbnb's move proves that a modern, open-source stack can handle the load. The question isn't whether you can make the switch. It's whether you can afford to wait until you have no choice.