Akri 0.1: Edge Device Discovery for Kubernetes

Introduction

On November 19, 2020, Microsoft announced Akri 0.1, an open-source project from the CNCF Edge Working Group. Akri (Greek for “edge”) abstracts hardware devices—cameras, sensors, industrial controllers—so they can be scheduled like native Kubernetes resources.

Device Discovery Handlers

• Pluggable discovery handlers detect devices via ONVIF, udev, OPC UA, and more.
• Each discovered device becomes a Custom Resource (Instance) bound to a Kubernetes node.
• Operators can mix handlers in a cluster, covering multiple device protocols concurrently.

Broker Pods & Workloads

1. Broker Pods run on the node hosting the device, exposing standardized interfaces to workloads.
2. Pods request devices via Kubernetes annotations, keeping scheduling declarative.
3. Brokers manage access control, ensuring exclusive device usage per workload.

Edge & Hybrid Use Cases

• Industrial IoT: onboard PLCs and sensors into Kubernetes for data collection and monitoring.
• Smart cities: connect IP cameras and environmental sensors to ML inference services.
• Retail/Healthcare: manage USB peripherals and barcode scanners securely at the edge.

Getting Started

kubectl apply -f https://raw.githubusercontent.com/project-akri/akri/main/deploy/manifests/install.yaml

Deploy an ONVIF configuration:

apiVersion: akri.sh/v0
kind: Configuration
metadata:
  name: onvif
spec:
  discoveryHandler:
    discoveryDetails:
      onvif:
        discoveryDurationSeconds: 10

Summary

Akri 0.1 gives edge operators a unified way to discover, allocate, and manage hardware devices within Kubernetes-native workflows.