PodOverhead: Accounting for Greater Cluster Stability
Talkat KubeCon + CloudNativeCon North America 2019 · on Tuesday, 2019-11-19· at 14:25· Room 1AB - San Diego Convention Center, Upper Level·
summary
Kubernetes 1.16 adds the PodOverhead feature (alpha, feature gate required) to account for resources outside containers so nodes stop overcommitting, especially with sandbox runtimes like Kata Containers and gVisor. It adds RuntimeClass overhead fields to the PodSpec, uses an admission-controller mutation to inject static overhead, and changes kube-scheduler and kubelet to include overhead in requests. Measured examples: a tiny VM-based pod can cost ~150 MB memory extra; an example runtime adds ~230 Mi and raises 100 mCPU request to ~350 mCPU.
AI-written summary · about
chapters (5)
related talks
33mThe Big Kubernetes Mistake Paying More Than Necessary
47mOptimizing Kubernetes Resource Requests/Limits for Cost-Efficiency and Latency
19mBuilding Cost Visibility For Your Kubernetes Clusters
36mTaming the Beast: Advanced Resource Management With Kubernetes
34mKubernetes: Reducing Infrastructure cost and complexity
44mFinOps at scale: Lessons from scaling Kubernetes workloads
40mThe efficient way of Autoscaling Your Workload in Kubernetes
15mDemo: Resource(Cost) optimization on planet scale Kubernetes clusters.