How Reddit Migrated Petabyte-Scale Kafka from EC2 to Kubernetes

2026-03-17 · Source: ByteByteGo Newsletter · Field: Technology & Digital — Cloud Computing & IT Infrastructure, Software Development & Engineering · Depth: Advanced, medium

Summary

Reddit's Engineering Team successfully migrated its entire Apache Kafka fleet, comprising over 500 brokers and more than a petabyte of live data, from Amazon EC2 virtual machines to Kubernetes with zero downtime and no client application changes. This complex infrastructure migration was guided by strict constraints, including maintaining Kafka availability, preserving metadata, handling tightly coupled client connectivity, and ensuring every step was reversible. The multi-phase process involved introducing a DNS facade, reconfiguring broker IDs, running a mixed EC2/Kubernetes cluster using a forked Strimzi operator, gradually shifting data with Cruise Control, and finally migrating the control plane from ZooKeeper to KRaft. This achievement demonstrates that large-scale infrastructure changes can be executed with minimal risk by employing small, reversible steps and strategic abstraction layers. Key lessons include the value of abstraction, protecting logical state over physical infrastructure, and designing for reversibility to enable confident progress.

Key takeaway

Reddit successfully migrated its petabyte-scale Apache Kafka fleet (500+ brokers) from EC2 to Kubernetes with zero downtime and no client application changes. This was achieved via a DNS facade, a forked Strimzi operator for mixed-cluster operation, and Cruise Control for incremental data shifting, culminating in a ZooKeeper to KRaft control plane migration. This blueprint offers MLOps teams a robust methodology for modernizing critical, stateful data streaming infrastructure without service interruption.

Topics

• Apache Kafka
• Kubernetes
• Infrastructure Migration
• Strimzi
• KRaft

Best for: MLOps Engineer, CTO, VP of Engineering/Data, DevOps Engineer, Software Engineer, Data Engineer

Related on AIssential

• Apache Kafka for Data Engineers — Beginner Guide — Apache Kafka provides a scalable, fault-tolerant, and decoupled architecture for real-time data streaming.
• Article: Two Misconfigurations That Caused Spark OOM Failures on Kubernetes — Cloud migrations can introduce silent infrastructure misconfigurations that cause Spark OOM failures under production load.
• QCon London 2026: Introducing Tansu.io — Rethinking Kafka for Lean Operations — At QCon London 2026, Peter Morgan introduced Tansu.io, an open-source, Apache Kafka-compatible messaging broker designed with stateless brokers that delegate durability entirely to external storage, c…
• How Collaboration Drove Graceland’s Oracle Cloud Migration — Talia Brown, Chief Information and Data Officer at Graceland University, navigated an abrupt digital transformation after the pandemic hit in early 2020, exposing critical needs for remote work and 24/7 student support.
• Nextdoor’s Database Evolution: A Scaling Ladder — Nextdoor's database architecture evolved from a single PostgreSQL instance to a sophisticated distributed system to support millions of users, addressing scaling challenges through a disciplined progr…
• Building the Open Source Alternative to AWS — An open cloud offers transparency, portability, and cost efficiency by open-sourcing its control plane and avoiding data lock-in.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by ByteByteGo Newsletter.