Nubank's experience scaling microservices operations
summary
Nubank describes concrete moves: grows from a few services to 100 microservices across four customer shards, handling 2 billion Kafka messages and ~1,000 instances at peak, and adopts AWS, ECS then Kubernetes, Kafka/Zookeeper, Spark, Databricks and Redshift. It enforces service ownership, dissolves a central DevOps team into platform, data-infra and security squads, and replaces brittle end-to-end tests with decoupled pipelines, showing measurable results like reduced CI time from 15 to 4 hours.
AI-written summary · about
chapters (13)
- 0:00Introduction to Nubank
- 2:20Early startup infrastructure challenges
- 5:51Scaling monitoring and CI issues
- 8:19Building basic monitoring platform
- 11:33IT ops hero pattern problems
- 13:59Enforcing service ownership
- 15:47Spotify squad model adoption
- 18:39Evolution of DevUps guild
- 21:47Infrastructure backlog starvation
- 23:56Creating Platform squad
- 25:46End-to-end test bottleneck
- 27:16Current infrastructure teams
- 29:05Conclusion and takeaways
related talks
30mRunning multiple Nubanks at once
36mModernizing Nubank's Kafka Platform for the Next 10 Years Starting at...
33mData: The Land DevOps Forgot
37mTesting in a Microservices Architecture
05mKeynote: The Vertically Integrated Bank: Scaling from NBFC to Integrated on Kubernet...
27mSlicing 'Big Elephant' systems into micro-services
51mMicroservices and Microfrontends: Scaling, Learning, Growing
39mHow automating cloud operations accelerated time to market at NU