furiosa-feature-discovery

Overview

The image runs a single binary that periodically detects the FuriosaAI NPUs on the node and writes a feature file for Node Feature Discovery to consume. It depends on NFD being present in the cluster, since NFD is what converts the generated file into node labels. The container starts the binary automatically through its default command.

Configuration

Behavior is controlled by command-line flags:

- --interval — how often, in seconds, the NPU information is refreshed (default 60).
- --output — path of the feature file written for NFD to read (default /etc/kubernetes/node-feature-discovery/features.d/ffd).

The output path must be on a volume shared with NFD so that NFD can read the file the discovery component writes.

Deploying with Helm

The recommended way to deploy on Kubernetes is with the official Helm chart, published at https://github.com/furiosa-ai/helm-charts. The chart installs Furiosa Feature Discovery together with NFD and wires up the shared volume between them. Options can be adjusted in charts/furiosa-feature-discovery/values.yaml before installing.

helm repo add furiosa https://furiosa-ai.github.io/helm-charts
helm repo update
helm install furiosa-feature-discovery furiosa/furiosa-feature-discovery -n <namespace>

Verifying

After deployment, confirm the NPU labels appear on nodes that have FuriosaAI NPUs:

kubectl get nodes -o json | jq '.items[].metadata.labels | with_entries(select(.key | startswith("furiosa.ai/")))'

You can then use these labels in node selectors or affinity rules to schedule NPU workloads.