Operating a Self-Healing Bare-Metal Kubernetes Platform at Global Scale
summary
DigitalOcean describes a self-healing bare-metal Kubernetes platform running storage workloads across 15 regions, 366 nodes, 5,000+ pods with hardware from 40 to 4,500 cores and PVs up to 7 TB. The stack uses Rancher, RKE2, containerd, Calico with BGP, Argo CD/Workflows, Ansible and Terraform; persistent volumes are pre-provisioned and node-affined. Automation enforces OPA policies, reboot/playbook safeguards, automated reconciliation and rapid fleet restore, yielding 99.9% SLA and sub‑second recovery while four engineers operate it.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:03DigitalOcean bare-metal K8s platform
- 7:23Storage workloads need bare metal
- 10:13Bare metal for low-latency storage
- 12:00Bare-metal K8s architecture
- 13:29Networking with BGP peering
- 16:25Cluster-of-clusters topology
- 21:05Multi-region deploy with Argo CD
- 25:23Containing failures with OPA
- 27:33Day-to-day ops and automation
- 30:25Self-healing platform summary
- 33:18Q&A
related talks
20mThe path to self-healing: Re-architecting for massive scale on kubernetes
34mBridging Baremetal and Virtualized Clouds: Operating Telco-Grade Kubernetes at Scale
32mRunning eBay’s High-Performance Workloads with Kubernetes
55mReturn of the Clustering: Kubernetes for Drupal
28mThe Magic of Kubernetes Self-Healing Capabilities
37mRunning Kubernetes in a Manufacturing Line
32mAutomating and Testing Production Ready Kubernetes Clusters in the Public Cloud
19mManaging Millions of Storage Volumes at Scale: Inside DigitalOcean's...