Streaming from Apache Iceberg - Building Low-Latency and Cost-Effective Data Pipelines

Apache Flink is a very popular stream processing engine featuring sophisticated state management, even-time semantics, exactly-once state consistency. For low latency processing, Flink jobs typically consume data from streaming sources like Apache Kafka. Apache Iceberg is a widely adopted data lake technology supporting numerous features like snapshot isolation, transactional commit, fast scan planning. While Iceberg was originally designed for batch, it can also be used as a streaming source in Flink. This not only lowers the processing delays from hours or days to just minutes, but also significantly reduces the infrastructure cost and operational burden.

In this talk, we will explain the design of the Flink Iceberg source that we contributed to Apache Iceberg open source project. We will compare the Kafka and Iceberg sources for streaming read and present performance evaluation results of the Iceberg streaming read. We will discuss how the Iceberg streaming source can power many common stream processing use cases (like ETL, feature engineering). It enables users to build low-latency streaming pipelines chained by Iceberg that are cost effective and easy to operate.


Steven Wu

Software Engineer @Apple and Apache Iceberg PMC

Steven Wu is a software engineer at Apple. He is working at the AIML data platform team focusing on stream processing and data lake technologies. Previously, he worked at Netflix where he helped build the real-time data infrastructure. He is passionate about building scalable distributed systems and empowering people with data.

Read more
Find Steven Wu at:


Tuesday Jun 13 / 11:50AM EDT ( 50 minutes )


Salon E


Stream Processing Apache Flink Apache Iceberg Data Pipelines Architecture


From the same track

Session Streaming

Laying the Foundations for a Kappa Architecture - The Yellow Brick Road

Tuesday Jun 13 / 10:35AM EDT

In the ever changing landscape of big data, focus is slowly moving away from batch and towards realtime analytics. Data Science workflows are evolving to adapt to this changing landscape.

Speaker image - Sherin Thomas
Sherin Thomas

Staff Software Engineer @Chime

Session Serverless

The Rise of the Serverless Data Architectures

Tuesday Jun 13 / 01:40PM EDT

For a while, it looked like Serverless was just a convenient way to run stateless functions in the cloud. But in the last year we’ve seen the rapid rise in serverless data stores.

Speaker image - Gwen Shapira
Gwen Shapira

Founder @Nile, PMC Member @Kafka

Session Data Architecture

Building a Large Scale Real-Time Ad Events Processing System

Tuesday Jun 13 / 02:55PM EDT

Two years ago, we embarked on building DoorDash's ad platform from the ground up. Today, our platform handles over 2 trillion events every day and our advertising business has experienced significant growth in recent years, becoming a key area of focus for the company.

Speaker image - Chao Chu
Chao Chu

Software Engineer @DoorDash

Session Architecture

Enabling Remote Query Execution Through DuckDB Extensions

Tuesday Jun 13 / 04:10PM EDT

DuckDB is a high-performance, embeddable analytical database system that has gained massive popularity in the last few years.

Speaker image - Stephanie Wang
Stephanie Wang

Founding Engineer @MotherDuck


Unconference: Modern Data Architecture & Engineering

Tuesday Jun 13 / 05:25PM EDT

What is an unconference? An unconference is a participant-driven meeting. Attendees come together, bringing their challenges and relying on the experience and know-how of their peers for solutions.

Speaker image - Ben Linders
Ben Linders

Independent Consultant in Agile, Lean, Quality and Continuous Improvement