Understanding Scalability and Performance in the Kubernetes Master
summary
The talk examines Kubernetes master scalability limits around and beyond 5,000 nodes by identifying bottlenecks in components such as etcd, the API server, and the scheduler. It explains how Alibaba ran benchmarks and profiled the master, then applied specific master-side tweaks and tunings that yielded over 100× performance improvements, enabling smooth operation of 10,000-node clusters.
AI-written summary · about
chapters (7)
- 0:00Start
- 1:07Background
- 2:30Experience on scalability
- 13:49Efficient node heartbeats
- 17:31API Server load balancing
- 22:18List-Watch & Cacher
- 32:14Controller failover
related talks
36mUnderstanding Scalability and Performance in the Kubernetes Master
42m2000 Nodes and Beyond: How We Scaled Kubernetes to 60,000-Container Clusters
34mA story of managing kubernetes cluster with 15k nodes & various workloads
48mScale and Performance Testing of Kubernetes: Answers for Specific Applications
34mPanel: Improving and Managing Kubernetes at Scale
34mA Huge Cluster or Multi-Clusters? Identifying the Bottleneck
30mScaling Kubernetes to Thousands of Nodes Across Multiple Clusters, Calmly
31mHow We Scale up to 2k Nodes for Batch Jobs Using Cluster Autoscaler