Large-Scale K8s Cluster Operation and Management
summary
JDOS is a Kubernetes-based platform running millions of containers with a single cluster over 9000 nodes and 20,000 services, using a heavily tailored Kubernetes 1.6. Engineers cut API server load by 99% by disabling unnecessary dynamic ConfigMap mounts, implement in‑place rolling upgrades to avoid rescheduling, replace the default IP store with a custom service (MySQL backend) to reach ~20,000 allocations per minute, and speed scheduling by random sampling a subset of ~500 nodes, reducing placement time from minutes to seconds.
AI-written summary · about
chapters (5)
- 0:00Start
- 9:04Configure Map Optimization
- 15:02Automated Recovery
- 16:49The Scheduler Optimization
- 25:38Process Inspection
related talks
35mAuto Operation of Hundreds of K8s Clusters in Yahoo! Japan
38mCost-Effective Scheduling of a Massive Number of Containers in Kubernetes
50mScale Kubernetes to Manage 60K Nodes Through Architectural Extension
38mCost-Effective Scheduling of a Massive Number of Containers in Kubernetes
34mCreating Multistage Development Environments with a K8s Cluster of 8000 Pods and Growing
36mOptimizing Resource Usage in Kubernetes
41mKubernetes the Very Hard Way
40mOperating a Self-Healing Bare-Metal Kubernetes Platform at Global Scale