Deep Dive Into the Latest Kubernetes Scheduler Features
Talkat KubeCon + CloudNativeCon North America 2019 · on Thursday, 2019-11-21· at 16:25· Room 6C - San Diego Convention Center, Upper Level·
summary
Kubernetes scheduler core is refactored into a scheduling framework alpha with extension points (preFilter, filter, score, normalizeScore, reserve, permit, bind, postFilter) and legacy predicates/priorities wrapped as plugins, a migration that required 100+ PRs. Performance work cuts scheduling latency by about 2x for very large (~5k) clusters and reduces preferred pod‑affinity complexity from O(n·P) to O(P). New features include a permit-based gang scheduling primitive, pod topology/even‑spreading GA work, daemonset scheduling moved into the scheduler, and scheduler‑aware pod overhead accounting (scheduler side implemented, kubelet side pending).
AI-written summary · about
chapters (6)
- 0:00Kubernetes scheduler intro
- 4:59Scheduling framework deep dive
- 14:42K8s 1.17 performance
- 18:21Graduated K8s 1.17 features
- 20:27Pod overhead updates
- 23:16Q&A
related talks
30mDeep Dive: Scheduling SIG
34mKubernetes Scheduling Features or How Can I Make the System Do What I Want?
35mSIG Scheduling Deep Dive
36mDeep Dive: SIG Scheduling
29mSIG Scheduling Deep Dive
20mHow Scheduling in Kubernetes Works
30mSIG Scheduling Intro & Updates
30mEnhancing the Kubernetes Scheduler for Diverse Workloads in Large Clusters