Kubernetes at the Far Edge: Harnessing IoT with Lightweight Clusters and Akri
Watch talk on YouTubeThe far edge
- Resource constraint computing
- Limited connectivity
- More and smaller clusters
Why kubernetes
- Automation, Scalability and resilience
- Workload Portability through containers
- Orchestration
- Declarative state
Enter k0s
- Minimal footprint as static binary
- Simplified edge cluster management
Managing disconnected edge nodes
- Needs: Remote managability
- Idea: Centralized, remote Control Plane (that only does control plane)
- Challenge: Network disconnections (kubernetes usually moves workload)
Akri
- Discovery of iot devices
- Exposes IoT devices as k8s resources
- Handels workload scheduling for leaf devices
Demo
Can be found in the video
Q&A
- What about image distribution: Depends on networking conditions, k0s supports interna. images delivered as tar.gz
- What can the broker do: Anything that a pod can interact with
- What about reboots: Well akri had some problems in the demo, kubelet seems to start the containers again
Random Notes
- Akri Kinda reminded me of the gpu-operator with extra resouce capacity for attached devices