Taming Unbounded Resources with Node Feature Discovery
summary
Node Feature Discovery (NFD) labels real hardware and drivers so Kubernetes can schedule by actual peripherals, demonstrated on a touring pi seven-node cluster (1 master + 6 workers) that exposes usb i2c gpio labels after editing the NFD configmap and using a custom ARM image. Hot‑plug relabeling works but has observable lag; loaded kernel modules often false‑posit if the peripheral is absent. For scale, use GitOps — fleet tested with a million clusters — and watch out for local script execution and device‑presence verification.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:11What is Node Feature Discovery
- 1:21Why do I need a labeler
- 1:42Automating resource management
- 2:07Unbounded resources
- 2:54Go fish
- 3:40Demo architecture
- 6:34Installing Node Feature Discovery
- 8:14Configuring Node Feature Discovery
- 15:59Node Feature Discovery Worker
- 16:23Master
- 16:43Installation
- 17:23Configuration Map
- 19:58Git Ops
- 21:50Git Ops Pipeline
- 25:29Summary
- 29:21Links
related talks
33mNFD and My Adventure in the Cloud Native Project Jungle
34mStreamline Your Cluster with Automated Node Labels,Taints, and Annotations
33mNFD and My Adventure in the Cloud Native Project Jungle
25mK8s Labels Everywhere! Decluttering With Node Profile Discovery.
26mMake Kubernetes Greater for NFV@Arm
35mResource Management WG Deep Dive
26mResource Orchestration of HPC on Kubernetes: Where We Are Now and the Journey Ahead!
18mLet There Be Topology-Awareness in Kube-Scheduler!