Gitlab Runners at Scale as a Service
summary
Built a Fleet-style GitLab runners-as-IaaS that uses the GitLab Instance Executor with a Terraform module + Packer-built AMIs to autoscale EC2 executors, support instance and Docker executors, mount S3/EFS/FSx, and export reproducible images; it runs ~350K jobs/month, sustains up to 25k parallel capacity (peak 3k), spans ~40 AWS accounts and reports cost-by-job via DataDog. Known issues: customers struggle with Terraform/Packer, inter-layer versioning, and chronic overprovisioning; plan is a UI to abstract Terraform and image management.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:29Managing GitLab runners
- 5:46Scalable runner requirements
- 7:51Fleeting runners plugin
- 10:43Fleeting runners architecture
- 13:08Fleeting runners features
- 21:07Observability and cost
- 25:51Current challenges
- 29:28Future vision
- 32:41Closing remarks
- 33:26Q&A
related talks
29mDeploying CI/CD Infrastructure with Terraform and Packer
29mScaling Across the Enterprise
29mBest Practices for Kubernetes Runners
26mRunning GitLab and GitLab CI in Kubernetes for the Enterprise
35mGitLab CI and Kubernetes, Build Test and Deploy Your Projects
02h 47mA GitOps Architecture from Scratch: Gitlab, Ansible, Terraform, Kub and AWS
36mBuilding a CI/CD driven infrastructure for managing Kubernetes clusters
29mEphemeral Cloud Infrastructure for Acceptance Tests with Terraform Enterprise & GitLab
