Inside Kubernetes Resource Management (QoS) – Mechanics and Lessons from the Field
summary
Explains how Kubernetes turns pod requests and limits into cgroup settings on the node, with requests map to cgroups shares, limits to CFS quota/period and the scheduler placing pods based on requests and allocatable (not current usage). Describes runtime behavior: memory triggers evictions while CPU is throttled, changed eviction ranking since 1.9 (usage then priority), and edge cases like daemonsets and hypervisor alignment. Recommends enforcing resource quotas and LimitRanges, starting with Guaranteed pods, tuning kubelet eviction thresholds, and using CPU manager or VPA for tighter control.
AI-written summary · about
chapters (42)
- 0:00Introduction
- 0:26Audience
- 2:25Agenda
- 2:39Whats missing
- 4:10What who cares
- 5:03Operating system basics
- 5:30Linux kernel
- 7:23Grouping
- 8:27Linux Kernel Scheduler
- 9:13Running Containers
- 9:57Container Isolation
- 12:25Groups
- 12:41Quality of Service
- 12:52equivalence class
- 13:50quality of service lifecycle
- 14:23stable resources
- 14:36beta resources
- 15:12requests and limits
- 17:02API Server
- 18:28Scheduling
- 19:00Allocatable
- 19:29Scheduling Algorithm
- 20:29Node Groups
- 20:43CPU and Memory Requests
- 22:07CPU and Memory Requests Example
- 23:34Resource Pacification
- 23:55Hierarchy
- 24:36Overcommit
- 25:53Eviction Threshold
- 26:16Eviction Order
- 27:58Resource Management Parameters
- 29:01Guaranteed Resource Classes
- 29:27Enforce Defaults
- 29:42Protect Critical System Parts
- 30:10Stress Testing
- 31:10Cloud Monitoring
- 31:36Java Runtime
- 32:34Download API
- 32:46eviction thresholds
- 33:26CPU Manager
- 33:58Kernel Shared Resource
- 34:18Questions
related talks
36mTaming the Beast: Advanced Resource Management With Kubernetes
32mTaking Care of Your Control Plane With API Priority and Fairness and Resource Quotas
18mLet There Be Topology-Awareness in Kube-Scheduler!
34mUnlocking Performance: Topology-Aware CPU Scheduling W...
10mKubernetes: Changing the Way We Think and Talk About Computing
10mNever Go Limitless – On the Importance of ResourceQuotas at the Edge
24mK8s Pod Scheduling - Deep Dive
34mPanel: Improving and Managing Kubernetes at Scale