Keynote: Kubernetes at GitHub
summary
GitHub runs 20% of production services on on‑prem Kubernetes, including github.com, using multiple clusters each with roughly 1,500 CPUs and 5 TB RAM per cluster and seven node types (preferred nodes: 40 CPU threads, 128 GB RAM, 2 local SSDs). Operational tooling includes node problem healer, cube‑tesla, cube‑health‑proxy, a kubectl chat wrapper and a gob load balancer, and a canary deploy workflow that routes a small fraction of traffic for safety; persistent local volumes are next.
AI-written summary · about
chapters (12)
- 0:00The road to kubernetes
- 2:58Embracing kubernetes
- 4:36Infrastructure and scaling
- 5:57Data center automation
- 7:13Production workload design
- 8:49Networking and load balancing
- 9:16Operational tooling
- 10:49Deployment flow
- 12:48Review lab environments
- 14:12Canary deployments
- 16:16Standardization benefits
- 17:27Future goals
related talks
36mUsing K8s to Power Microservices at Scale: The GitHub Way
20mKeynote: Kubernetes - Don't Stop Believin'
17mKeynote: Backstage with Kubernetes
17mKeynote: Extending Kubernetes: Application Orchestration to 1000’s of Clusters in...
22mThe Kubernetes Community
06mKeynote: Open Innovation: Kubernetes on Google Cloud
06mKeynote: The Cloud Native Ecosystem in China
07mKeynote: High Performance Computing on Google Kubernetes Engine