Apache Iceberg V3 in Production: Lessons from a Large-Scale Deployment

Apache Iceberg V3 brings powerful capabilities, but running it reliably in production takes more than a version upgrade. In this session, we share our real-world adoption journey with Iceberg 1.10: table design, write/read path tuning, compaction strategy, and metadata lifecycle management. We’ll unpack the key bottlenecks we faced—small-file growth, snapshot sprawl, merge/update pressure, and query latency variance—and the concrete optimizations that improved both stability and cost efficiency. You’ll leave with practical architecture patterns, anti-patterns, and a battle-tested checklist to move from pilot to production with confidence.

Speakers:

Yuming Wang: eBay, Compute, Lakehouse

Yuming Wang is a member of the Apache Spark Project Management Committee (PMC) and currently leads the build-out and evolution of his company’s compute platform. He focuses on architecting, optimizing, and operating Spark at production scale, while also driving the implementation and operation of a Lakehouse architecture based on Apache Iceberg. He is committed to turning cutting-edge open-source data technologies into reliable and efficient enterprise-grade production systems.

Fei Wang: eBay, Hadoop

Fei Wang is a PMC member of both Apache Kyuubi and Apache Celeborn, currently focusing on the development and evolution of enterprise computing platforms. He specializes in distributed computing, Spark engine optimization, and Lakehouse architecture based on Apache Iceberg, dedicated to translating open-source data technologies into stable and efficient enterprise-grade computing services.