This repository is not maintained anymore. For a similar solution, please check https://github.com/PlayFab/thundernetes

Scaling multiplayer Dedicated Game Servers on Azure Kubernetes Service

Scaling Dedicated Game Servers (DGS) is a hard problem. They're stateful (having the bulk of player data stored in server memory), can't be explicitly shut down (since players might be still enjoying their game) and, as a rule of thumb, their connection with the players must be of minimal latency, especially for real-time multiplayer games.

This repository aims to provide a solution/guidance/building blocks for managing containerized dedicated game servers using the Kubernetes orchestrator on Azure using the managed Azure Kubernetes Service (AKS). However, you could probably use parts of the project to scale memory-stateful workloads.

~ This is currently a work in progress. Not recommended for production use ~

Documentation

• Installation
• Architecture
• Controllers
• Development and e2e testing
• Autoscaling
• FAQ
• Other Kubernetes resources

Other

Demos

In order to demonstrate this project, we've built a simple "echo" UDP server in Node.js and we've also adapted the OpenArena open source game so it can work with our solution. Both Dockerfiles are stored in the demos folder whereas the corresponding Kubernetes deployment YAML files are located in the artifacts/examples folder.

Docker Hub Images

Images used for this project are hosted on Docker Hub:

• OpenArena game sample
• A Node.js UDP echo server
• API Server
• Controllers

Architecture Diagram

Thanks

To Brian Peek and Andreas Pohl for the countless discussions we had (and still have!) about scaling DGSs. Moreover, I'd like to express my gratitude to the awesome people on #sig-api-machinery channel on Kubernetes Slack for answering a lot of my questions during the development of this project.

Presentation

Checkout a presentation I gave about this project on Open-conf conference in Greece, on March 2019.

This is not an official Microsoft product.