Airbnb engineering blog

The post describes how Airbnb reworked its Observability as Code (OaC) workflow to provide pre-deployment visibility and fast feedback for alert behavior. By adding local-first development, Change Reports, and bulk backtesting integrated with Prometheus rule evaluation, engineers reduced alert validation cycles from weeks to minutes, migrated 300,000 alerts to Prometheus, and dramatically cut alert noise. The article emphasizes compatibility with existing standards, operational guardrails for large-scale simulation, and the importance of owning the end-to-end developer experience.

This post summarizes Airbnb’s 2025 research program across major conferences, highlighting work in applied machine learning for search and personalization, production LLM systems, causal and adaptive experimentation, and data systems. Key technical takeaways include interleaving and counterfactual evaluation for faster ranking assessment, map-specific ranking metrics and bimodal listing embeddings for improved search, agent-in-the-loop and incremental summarization for LLM-based customer support, Bayesian methods for adaptive experiments, and SQL:Trek for automated index design.

Cosmo W. Q examines strategies for safely applying dynamic configuration changes in large-scale systems. The post highlights operational patterns—validation, gradual rollouts, monitoring, automated rollback, and governance—and discusses tooling and processes to reduce risk when changing configuration across distributed services.

Anna Sulkina recounts her path to joining Airbnb, discussing her background and experiences that led her to the company. The piece is a personal profile rather than a technical deep dive.

New techniques we’re using at Airbnb to improve and maintain performance of SwiftUI features at scale. The post describes performance-improvement strategies and practices for SwiftUI-based features in Airbnb’s apps.

How to achieve high availability with distributed database on Kubernetes. The article discusses approaches Airbnb uses to run distributed databases reliably on Kubernetes for high availability.

How Airbnb upgrades tens of thousands of pods on dozens of Kubernetes clusters to new Istio versions. The post explains the strategies and tooling used to perform large-scale, low-risk Istio upgrades across many clusters.

A more powerful engine and a simpler API for our data-oriented mesh. The article covers modernizing Viaduct to provide improved capabilities and a cleaner developer-facing interface for data services.

How Airbnb combines GraphQL infra, product context, and LLMs to generate and maintain convincing, type-safe mock data at scale. The post describes integrating product context and language models to produce and maintain mocks for GraphQL schemas.

The post describes how Airbnb launched 20+ locally preferred payment methods in just over a year by modernizing its payments platform with a domain-driven, connector/plugin-based architecture and a Multi-Step Transactions (MST) framework. It details three standardized payment archetypes (redirect, async, direct), a YAML-driven Payment Method Config with code generation to accelerate integrations, a PSP emulator for reliable testing, and a centralized observability framework to maintain reliability at scale.

Shravan Gaonkar explains Airbnb's shift from static rate limiting to an adaptive traffic-management approach for its key-value store, outlining the motivations and high-level architecture for the adaptive system. The post highlights how adaptive controls improve stability and throughput under variable load and discusses operational trade-offs and observed outcomes.

The post explains Airbnb’s rearchitecture of Mussel from a legacy v1 storage backend to Mussel v2, built on a NewSQL backend with a Kubernetes-native control plane to reduce operational complexity and improve scalability. It details key design decisions — a stateless Dispatcher, Kafka-backed durable writes, dynamic range sharding and presplitting, optimized bulk-load and TTL processes, and a blue/green dual-write migration pipeline — and reports production results including a zero-downtime petabyte migration, sustained 100k+ streaming writes/sec, and p99 reads under 25 ms.

The article describes how Airbnb redesigned search ranking for its map interface by explicitly modeling how user attention flows across map pins. It presents three technical approaches—uniform attention with bounded booking-probability selection, a tiered pin design (regular pins and mini‑pins) to prioritize attention, and a recentering algorithm that discounts peripheral attention—and reports A/B test gains in bookings, discovery metrics, and reduced map moves, with further details in a KDD ’24 paper.