Improve DevX at scale with Serverless
summary
Philips migrates CI/CD to GitHub with an open-source, AWS-based system that provisions self-hosted action runners on demand via a Serverless control plane. The system auto-scales VMs (including Windows and ARM), uses AWS Spot Instances and Direct Connect for corporate network access, and supports 8000+ developers across 8000+ repositories. It launches about 20,000 jobs a day, caches VM images to cut cold-starts, and reports cost and reliability trade-offs from production use.
AI-written summary · about
chapters (22)
- 0:00Introduction and context
- 3:01Shift to inner source model
- 7:19Challenges with hosted CI/CD runners
- 9:48Building serverless runners with AWS
- 12:19How serverless runners work
- 15:52Deploying infrastructure with Terraform
- 17:22Open-sourcing the solution
- 21:27Cost efficiency and sustainability
- 24:56Improving developer experience
- 30:17Optimizing Windows runner speed
- 33:32Handling massive workflow spikes
- 36:30Monitoring and stability issues
- 38:22Open source contributions and value
- 41:49Using Backstage for software catalog
- 43:17Automating compliance checks
- 45:18Alternative runner controller setup
- 46:45Global team workflows and pools
- 48:11Access control in inner source
- 49:37Cost comparison with alternatives
- 51:31Central team vs team autonomy
- 52:55Ensuring spot instance availability
- 53:17Summary and Q&A
related talks
01h 02mFrom Monolith to Serverless
41mCI/CD in the Serverless Era
26mScaling Enterprise Development with Cloud IDEs
14mState of Serverless DevEx & Observability
42mEnterprise Serverless Adoption. An Experience Report
17mGoing Full Silo - Using Serverless to Build for Dev Simplicity
31mOptimizing Developer Productivity with Serverless Technology
29mGoing Serverless