Staying in Tune: Optimize Kubernetes for Stability and Utilization
Talkat KubeCon + CloudNativeCon North America 2019 · on Thursday, 2019-11-21· at 17:20· Room 6F - San Diego Convention Center, Upper Level·
summary
Explains how setting pod requests vs limits drives scheduling, overcommit and eviction behavior, and shows that memory is incompressible so hitting the eviction threshold 100Mi or allocatable edge can trigger OOM-driven Docker kills and cluster-wide cascades. Recommends tuning kube-reserved and system-reserved to account for daemon usage, lowering eviction pressure transition (default 5m), using monitoring (Prometheus) and Vertical Pod Autoscaler to iterate requests/limits, and enabling Docker live-restore to reduce blast radius.
AI-written summary · about
chapters (11)
- 0:00Start
- 3:21Challenges
- 5:05What can we do about it?
- 5:35Unbounded Resource Consumption
- 8:41Limits and Requests
- 9:32Scheduling
- 10:27Overcommit
- 18:36Allocatable Memory & Ephemeral Storage
- 25:16Defaults
- 28:49Summary
- 29:18Staying in Tune Optimize Kubernetes for Stability and
related talks
47mOptimizing Kubernetes Resource Requests/Limits for Cost-Efficiency and Latency
29mRight-Size Your Kubernetes Deployments
36mTaming the Beast: Advanced Resource Management With Kubernetes
36mDon't Be Greedy: Rightsize Your Kubernetes Cluster With Prometheus
34mLoad Testing Kubernetes: How to Optimize Your Cluster Resource Allocation in Production
36mKubernetes Confessions: Real Mistakes We’ve All Made (and What They Teach Us)
11mHow Jenga & Tetris Can Help Run More Containers with Lesser Costs...
32mOptimizing Application Performance on Kubernetes