The Right Number of Partitions for a Kafka Topic

Devnexus 2023 – Atlanta 🇺🇸

Apr. 2023

Ricardo Ferreira

Slides

skip-backskip-back  /  skip-forwardskip-forward
download-clouddownload-cloud PDF

Abstract

Every technology has that key concept that people struggle to understand. With databases, is which join clause to use for fetching data from multiple tables. Containers are tricky when you have to pick a storage type given some persistence requirements. With Apache Kafka, the winner is how many partitions to set for a topic. Why this is important? You may ask. Well, sizing Kafka partitions wrongly affects many aspects of the system, such as storage, parallelism, and durability. Worse, it may also affect how much load Kafka can handle. Hence why often the decision about how many partitions to set for a topic is handled by Ops teams, as we see this to be only an infrastructure matter. In reality, this is an architectural design decision that affects even the amount of code you write. This session will peel off the concept of partitions and explain it from the perspective of the Kafka cluster and its clients. It will explain the formula people should use to decide how many partitions to set for a topic, and how to spot a poor decision when they see one.

Previous page The Right Number of Partitions for a Kafka Topic
Next page BOA301: Building your own Apache Kafka connectors

© 2018 - 2026 Ricardo Ferreira

Search is powered by Pagefind. Just hit CTRL+K or CMD+K to start searching.

Powered by Hugo with Dream and Devrel themes.

Open Source

I contribute to LangChain4j, an idiomatic open source Java library for building LLM-powered applications on the JVM. Recent work includes adding native vector search embedding stores so developers can build RAG, recommendation engines, and AI memory systems.

I also ported RedisVL to Go, an open source, AI-native client that brings vector search, semantic caching, LLM memory, semantic routing, rerankers, and an MCP server to the Redis ecosystem for Golang developers.

Public Speaking

I’ve been speaking at conferences since 2008 and doing it full time as part of my work with DevRel since 2018. My talks go deep on the systems I build with: distributed systems and event streaming, AI engineering and vector search, and the data infrastructure that has to hold up when the demo ends and production begins. Some of the events I’ve spoken at include AWS re:Invent, Microsoft Ignite, Google Cloud Next, KubeCon, Oracle OpenWorld, QCon, Strange Loop, Kafka Summit, Pulsar Summit, JavaOne, DevNexus, JFokus, JNation, and All Things Open.

Ricardo Ferreira presenting on the main stage at AI DevWorld
On the main stage at AI DevWorld

You can find my upcoming and past talks on my speaking calendar. Recordings also live on my YouTube channel, and the code I write for talks, demos, and workshops is on my GitHub.

Consulting and Professional Services

If you’d like to hire me as a consultant for your projects, speak at your event, or lead a hands-on workshop for your team, contact me at riferrei@riferrei.com. I can understand the scope of your request and provide a free estimate.

Who am I?

I work at the intersection of AI, data infrastructure, and distributed systems, turning complex technology into things developers can understand and products users love.

Lately, that means hands-on AI engineering: building vector search, semantic caching, agent memory, and RAG into the data layer, and figuring out how to make AI agents secure enough to ship. I contribute to open-source projects like LangChain4j and RedisVL for Golang.

The AI-native work isn’t a pivot. It draws on the same systems-design foundation I’ve built for 20+ years: designing data systems for scale, moving data fast, watching where systems break; now applied to vectors and agents. I have worked on RDBMS and Big Data at Oracle; event streaming with Apache Kafka and Apache Flink at Confluent; observability at Elastic; AI and developer tooling at AWS; and NoSQL and vector stores at Redis. That foundation is exactly what separates AI demos that work on stage from AI systems that survive production.

Social Links