How GitHub’s Open‑Source GLB Load Balancer Achieves High‑Performance Scaling

GitHub’s HTTP, Git and SSH services handle billions of connections daily; to meet this demand the company built the GitHub Load Balancer (GLB), whose open‑source version has just been released.

GLB is a load balancer developed internally at GitHub that runs directly on bare metal to achieve superior performance. Initially, service operators relied on vertical scaling and HAProxy for load balancing, which required specialized hardware and dedicated gigabit networks for failover. As the workload grew, a more portable solution that could run in a typical data‑center was needed, leading to the creation of GLB.

The core of GLB’s algorithm is Rendezvous (or Highest Random Weight, HRW) hashing. This technique lets operators smoothly add or remove controller and proxy nodes without interrupting user services, because each request is consistently mapped to the same backend unless that backend becomes unavailable.

GLB addresses many challenges faced by engineers at GitHub: horizontal scalability, high availability, support for high‑pressure connections, testability at any software level, and resistance to typical DDoS attacks. The system can assign a primary and a backup backend for each request, ensuring that even if a node fails, traffic is immediately redirected to the secondary server.

The article also explains the general concept of a load balancer: distributing incoming requests across multiple servers to avoid overloading any single machine. Simple routers cannot perform application‑level balancing, which is why virtual load balancers such as LVS (Linux Virtual Server) are used. GLB builds on these ideas, adding health checks and the primary/backup assignment strategy to improve reliability during disaster recovery scenarios.

GLB Director, the implementation of this load‑balancing approach, is now available for download from GitHub.

GitHub - github/glb-director: GitHub Load Balancer Director and supporting tooling.