Deploying the world's largest GitLab instance 12 times daily

What this post added

This post details the technical implementation and operational challenges of deploying the world's largest GitLab instance (GitLab.com) up to 12 times daily without downtime. It elaborates on the automated deployment pipeline built using GitLab CI/CD, including the build process for Omnibus packages and Cloud Native GitLab (CNG) images, the creation of auto-deploy branches, and the deployment sequence. It specifically highlights the canary deployment strategy with environment-based validation across Staging Canary, Production Canary, Staging Main, and Production Main stages, including QA checkpoints and traffic monitoring. The post also delves into technical architecture considerations for hybrid infrastructure (VMs and Kubernetes), the challenges of multi-version compatibility during deployments (database schema ahead of code), and the atomic upgrade mechanisms for Gitaly using tableflip.