API Priority and Fairness: Kube-APIServer Flow-control Protection
Talkat KubeCon + CloudNativeCon North America 2020 Virtual · on Friday, 2020-11-20· at 17:55· Intrado Virtual Event Platform,·
summary
Kubernetes adds an APIPriorityAndFairness feature gate (alpha in 1.18) that classifies requests into priority levels and flow schemas, implements shuffle sharding of fixed-size queues and a fair queuing variant scheduler, and exposes concurrency shares in priorityLevelConfiguration. Demo on a local kmd cluster shows changing a priority’s shares from 1 to 15 and the server recalculating that priority’s capacity limit to 120. Goal: protect loopbacks, controllers and aggregated servers from DOS and avoid deadlocks; debug endpoints land in 1.19 and e2e tests target 1.20.
AI-written summary · about
chapters (25)
- 0:00Intro
- 0:30Topic
- 1:35About Me
- 2:18Team
- 2:48Summary
- 3:59Higher Goal
- 4:44Self-Protection
- 5:11Kube-APIServer Loopback
- 6:30Priority Inversion
- 7:30Deployment of Doom
- 8:18Kubelet Amuck
- 9:52Multi-Tenancy
- 10:26Non Goal
- 13:08Linux Qdisc
- 15:14Classful Flow-Control System
- 16:02Classifier
- 19:37Queue Scheduler
- 20:04FQ: Fair Queuing
- 20:25FQ for Server Requests
- 21:07APF System
- 21:23Documentation
- 21:38FlowSchema (Exempt)
- 21:59FlowSchema (Catch-All)
- 22:17Priority LevelConfiguration
- 27:46Non-Blocking Items
related talks
33mKubernetes API Server Performance Clinic: Auditing, and Priority...
35mSIG Scheduling Deep Dive
34mKubernetes Scheduling Features or How Can I Make the System Do What I Want?
38mDeep Dive: API Machinery SIG
34mRunning prioritised workload in Kubernetes
34mLatest Kubernetes Scalability Improvements
42mDeep Dive Into the Latest Kubernetes Scheduler Features
30mDeep Dive: Scheduling SIG