One Cluster to Rule Them All - Multi-Tenancy for 1k Users
summary
A single Kubernetes cluster ran for about 1,000 users and research projects using kops, a GitLab-to-namespace integrator and a custom auth bridge called kuglocken, reaching 2,500 namespaces and ~17 nodes with ~150 concurrent users. The main bottlenecks were namespace and control-plane metadata (etcd, Prometheus scaling, per-namespace logging in Graylog), plus GPU sharing limits and storage latency solved with local M.2 SSDs; lessons include the need for quotas, better monitoring, logging design, and stricter admission controls before scaling.
AI-written summary · about
chapters (6)
- 0:00Introduction and team background
- 7:54Multi-tenancy challenges and solutions
- 16:41Defining tenants and Kubernetes roles
- 21:59Logging and monitoring limitations
- 26:14Real-world production issues
- 32:20Key lessons and future steps
related talks
32mScaling up to your first 10 million users - The Kubernetes Edition
32mScaling Kubernetes Application to 1M Users
30mScaling Kubernetes to Thousands of Nodes Across Multiple Clusters, Calmly
38mScaling Kube-Apiserver to Thousands of Tenants – Lessons Learned From kcp
03mFrom 5 to 1,300+ Clusters: Declarative Scaling for Private Kubernetes
46mThe Journey of Migrating 2500 Apps to Multi-Cluster Kubernetes and Zero Trust Istio
34mA Huge Cluster or Multi-Clusters? Identifying the Bottleneck
50mScale Kubernetes to Manage 60K Nodes Through Architectural Extension