node-life-support

Emergency life-support for Kubernetes nodes.

For when you have kubelets that cannot communicate with their control-plane, but you know the nodes are healthy and want to keep them and their service endpoints online.

Ordinarily, if a kubelet cannot reach or authenticate to kube-apiserver, it will fail to renew its node lease and fail to update its node resource. The node will then become NotReady and service endpoints will be removed.

This small controller periodically renews the node lease and patches node status on behalf of the nodes, so that they remain Ready.

This project may be of particular interest to those who run clusters with remote control-planes, such as AWS EKS clusters extended into AWS Outposts.

Deployment

Using raw manifests:

kubectl apply -f manifests/

Using the Helm chart:

helm install node-life-support chart/node-life-support --namespace node-life-support --create-namespace

Configuration

Environment variables used by the controller:

NODE_LABEL_ALLOWLIST - comma-separated list of node label keys. Only nodes with at least one of these labels will be put on life support. If this is not set, all nodes in the cluster will be put on life support.

Building

1. Build the binary (requires Go >=1.22):

go build -o bin/node-life-support ./

2. Run unit tests:

go test ./...

3. Build container image:

docker buildx build --platform linux/amd64,linux/arm64 .

Contributing

Please read CONTRIBUTING.md and CODE_OF_CONDUCT.md before opening issues or PRs.

License

This project is licensed under the MIT License — see the LICENSE file.