StreamNative Logo
VideoOct 8, 202629 mins

Declarative Data Movement at Scale: Unifying Streaming and Offline-to-Online Pipelines

Unlock Instant Access

Complete the form to start watching.

Session Overview

Discover how Airbnb built a unified, declarative data movement and transformation platform to bridge online services, streaming engines, and offline data warehouses.

In this talk, Justin Cunningham and Sai Govada explain how Airbnb transitioned away from a fragmented, multi-step pipeline plumbing to an intent-driven platform (DMT). Learn how developers declare pipelines in simple YAML configurations, how durable saga orchestration automates provisioning across Apache Flink and Apache Spark, and how automated schema evolution and column-level governance eliminate data drift and manual backfilling at scale.

What You'll Learn

  • The offline vs. online divide: Overcoming business logic drift and high engineering overhead
  • Shifting from imperative scripts to declarative YAML and agent-friendly pipeline definitions
  • Architectural overview: Control plane, schema registry, pluggable compute (Flink & Spark), and storage layers
  • The 4 core pillars: Orchestration, reliability (Skipper durable workflow engine), extensibility, and automated schema governance
  • Coordinating zero-downtime schema evolution and personal data privacy annotations across the pipeline DAG
  • Batch vs. streaming runtimes: Managing real-time Iceberg ingestion and offline-to-online key-value swaps
  • Hard-won production takeaways: Composability, centralized schemas, runtime abstraction, and shadow runs

Chapters

  • 00:00 — Introduction & the online vs. offline data divide
  • 01:26 — The declarative vision: From days of manual setup to 2-minute deploys
  • 04:11 — Evolution & scale of Airbnb's data movement platform
  • 07:37 — Architecture: The 4 layers & control planes
  • 10:36 — Pillar 1: Declarative orchestration & Flink Table API
  • 14:16 — Pillar 2: Reliability & durable sagas with Skipper
  • 15:49 — Pillar 3: Extensibility & modular connectors
  • 17:20 — Pillar 4: Automated schema evolution & privacy governance
  • 22:10 — Offline-to-online batch pipelines with Spark
  • 23:42 — 6 production lessons learned
  • 26:55 — Future roadmap & closing thoughts

About Speaker

Justin Cunningham

Justin Cunningham Justin is a data infra engineer, and has worked leading data platform architecture at various companies including Airbnb, Netflix, and Yelp.

Sai Krishna Chaitanya Govada

Sai Krishna Chaitanya Govada Sai Govada is a Staff Software Engineer on Airbnb's Data Platform, where he leads the design and development of Data Movement platform, that enables teams to define and operate data pipelines as configuration files. His work spans Apache Flink streaming pipelines, Change Data Capture (CDC) architectures, Apache Iceberg-based data warehousing, Kafka-based event infrastructure, and observability systems for large-scale distributed pipelines. Prior to this role, Sai has worked extensively on high-throughput, low-latency data systems processing 100s of billions of events daily. His engineering work at Airbnb has shipped production infrastructure serving hundreds of internal data consumers across trust & safety, machine learning, analytics, and search. Sai is passionate about making data infrastructure accessible.