Optimizing Storage Assignment via Pod Scheduling Under Disturbance Factors
Talkat KubeCon + CloudNativeCon North America 2020 Virtual · on Thursday, 2020-11-19· at 14:55· Intrado Virtual Event Platform,·
summary
Uses WaitForFirstConsumer plus Topology Spread Constraints to turn local-disk PV placement into pod scheduling, then retunes kube-scheduler so spread becomes the dominant score. Shows concrete edits: in 1.17 increase even-pods weight (they use weight=500), in 1.18 create a scheduling profile with podTopologySpread weight, and in 1.19 disable no-resource–preferred allocation. Empirical result: 5 Ceph OSD pods go to 4 nodes with tuning vs all to 3 nodes by default, giving rack-level fault tolerance.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 0:29Who am I
- 1:20Agenda
- 4:00Problem Statement
- 5:50Kubernetes
- 6:29Basic idea
- 9:01Port distribution
- 12:42When unsatisfiable
- 16:29Tuning Pod Scheduling
- 18:08Demonstration
- 21:44Key takeaways
related talks
38mDynamic Pod Resource Boundary Adjustment in Web Scale Clusters
18mLet There Be Topology-Awareness in Kube-Scheduler!
38mDynamic Pod Resource Boundary Adjustment in Web Scale Clusters
31mSIG Scheduling Update: Transition From Pod To Workload Scheduling
34mSpreading Apps, Controlling Traffic & Optimising Costs in Kubernetes
30mDeep Dive: Scheduling SIG
39mStaying in Tune: Optimize Kubernetes for Stability and Utilization
24mK8s Pod Scheduling - Deep Dive