
12/11/2016 · Sid Sijbrandij
What this post added
This post details GitLab's decision to move from cloud hosting to purchasing their own bare-metal servers for GitLab.com to improve performance and scalability. It outlines the proposed hardware specifications for chassis, CPUs, disks (including the use of SSDs for caching with Bcache), memory, and network configuration. The plan involves a move towards a fault-tolerant CephFS cluster and the migration of application servers and PostgreSQL to bare metal, while keeping CI Runners in the cloud for easier autoscaling. The post also discusses the strategy of using a common node design for interchangeability and managing blended workloads on Kubernetes for optimal resource utilization.