Why Eventbrite runs a 700 node Kube cluster just for development

I did a couple interviews with a principal engineer on the DevTools team at Eventbrite that I thought yall might find interesting.

They built a cloud development environment with Docker and Kubernetes. Docker was natural since many of their services were already containerized, and Kubernetes helped with their massive scale. Although their monolith doesn’t run in containers in prod, they containerized it so that it could run in their dev env.

In part 1, Remy shared how Eventbrite decided that building a cloud dev env was worth the investment: [https://kelda.io/blog/eventbrite-interview/](https://kelda.io/blog/eventbrite-interview/)

In part 2, we got into the details of how it works: [https://kelda.io/blog/eventbrite-interview-part-2/](https://kelda.io/blog/eventbrite-interview-part-2/)

Just comment any questions for Remy and I’ll pass them along!

EDIT: There's been some questions about why they run so many nodes. Here's Remy's response

>We tried a few instance types. With big instances, we could not maximize the number of pods so we would underutilize the node resources which would cost more overall. In our case, we use these for development so we don't put too much load on these pods so they can be relatively small and fit well on small/medium nodes.
>There's more information on the EKS limitations [here](https://medium.com/faun/aws-eks-and-pods-sizing-per-node-considerations-964b08dcfad3)

6 thoughts on “Why Eventbrite runs a 700 node Kube cluster just for development”

  1. Heads-up: Link to part 2 in the intro is broken (the one in the end of the article works) and part 3 not linked at all.

  2. Did he consider any alternatives? I’m using docker-compose in dev, k8s in staging and production. Kind of a bummer managing multiple services but there’s no involving a roll-your-own service like yak. Team members more or less know the former.


Leave a Comment