Bangalore Apache Kafka®, Data and Streams Group

Google Maps link

***
Agenda:

• 11:00 - 11:10: Welcome
• 11:10 - 11:50: Nitin Agarwal, Staff Software Engineer I, Confluent
• 11:50 - 12:30: Jeevan D C, Senior Principal, Entain
• 12:30 - 12:40: Break
• 12:40 - 13:20: Samrat Deb, Software Engineer, Uber
• 13:20 - 14:30: Lunch

***
💡 Speaker:
Nitin Agarwal, Staff Software Engineer I, Confluent

Talk:
Queues for Kafka
What if you could use Apache Kafka exactly like a traditional message queue? While Kafka excels at event streaming, workloads with independent work items need per-message acknowledgment, easy retries, and competing consumers rather than strict partition ordering. KIP-932 changes the game by introducing Share Groups in Kafka 4.0. This session explores how Share Groups bring native queuing semantics to Kafka topics, allowing you to scale consumers freely without worrying about partition limits. Come discover how to effortlessly run your queuing workloads on Kafka.
-----
💡 Speaker:
Jeevan D C, Senior Principal, Entain

Talk:
Kafka & Time: The Hardest Dependency
Distributed systems are hard not because they span machines, but because they span time. Clocks lie, "now" doesn't exist, and most events are concurrent — yet we design as if a global timeline is free.

This talk connects real production disasters to the Kafka config and design choices that caused them: partition keys that impose ordering nobody needs, timestamp modes nobody consciously picked, and windowing contracts that silently drop late arrivals. Three acts — time is scary, time is expensive, time is a superpower

-----
💡 Speaker:
Samrat Deb, Software Engineer, Uber

Talk:
Flink & S3: Designing the Native S3 FileSystem
For years, Apache Flink’s S3 integration has forced engineers to make a frustrating compromise: battle Hadoop’s massive 30MB+ dependency tree and classpath conflicts, or use Presto’s lightweight connector and sacrifice exactly-once sink capabilities. To eliminate this technical debt, I designed FLIP-555: Flink Native S3 FileSystem.
In this talk, I will introduce flink-s3-fs-native, a clean-slate implementation built directly on the AWS SDK for Java v2. Completely independent of Hadoop and Presto, it delivers a single, unified connector for both high-speed state checkpointing and exactly-once streaming sinks.
I will walk through the core design decisions and engineering challenges, including:
The Architecture: How we successfully mapped Flink's stream-oriented RecoverableWriter to S3’s multipart upload API.
Modern Cloud-Native I/O: Transitioning from blocking calls to non-blocking asynchronous I/O (via Netty and AWS CRT), alongside native support for IAM Roles (IRSA).
Performance Gains: A deep dive into our benchmark results, which demonstrate a 2x increase in checkpoint throughput (~200 MB/s) and a 50% reduction in checkpoint duration at 15GB+ state sizes.
Join me to learn how this native architecture simplifies deployments, permanently resolves dependency hell, and significantly accelerates Flink checkpointing at scale.