Keynote: Pushing the Limits of Kubernetes with Game of Thrones
summary
HBO moves 15,000 pods into production to survive Game of Thrones Season 7 premiere, provisioning clusters on AWS with Terraform modules, autoscaler tweaks and per-node ELBs for production. Modules add reserve minions and backup minions (tainted and affinity-bound) so Prometheus on EBS keeps data during scale downs. Networking uses flannel and surfaces UDP drops and DNS timeouts fixed by kube-dns tuning (cache 10,000, --local answers), plus weekly mega load tests and metric filtering for ~20k containers.
AI-written summary · about
chapters (21)
- 0:00Introduction
- 6:57Our Kubernetes journey
- 7:27High availability
- 8:24Tariffs or modules
- 8:56Problems
- 9:44Improved Module
- 9:58Minion Module
- 10:25Backup Minion Module
- 10:39Reserve Minion Module
- 10:59Summary
- 11:27Networking
- 12:13Services
- 12:45Ingress
- 13:46No balancer
- 14:41Cube DNS
- 15:19Resolving DNS
- 15:55DNS Cache Size
- 16:16Memory Address Flag
- 16:59DNS Server Flag
- 17:16Telemetry
- 19:28Conclusion
related talks
34mKeynote: Kubernetes1.6 and the Open Source Roadmap
20mKeynote: Kubernetes - Don't Stop Believin'
17mKeynote: Nine lessons from Kubernetes in Enterprise
14mKeynote: Kubernetes: The Video Game
17mKeynote: Kubernetes: Lessons from the Trenches
06mKeynote: Open Innovation: Kubernetes on Google Cloud
08mSponsored Keynote: Free Range Kubernetes: Feeding, Watering & Herding - Brian Goodman-Jones
53mManaging Containers at Scale with CoreOS and Kubernetes
