exoscale-sks-lifecycler

Lifecycle automation of Exoscale SKS cluster nodes.

Current capabilities:

• evict nodes, which are not on the desired version
• evict nodes, which are defined via a labelSelector

Quality-of-life features:

• before a node is evicted, pods are re-scheduled (in order to maintain high availability for applications):
  • a pod managed by a Deployment has its rollout restarted, in order to cause no downtime for single-replica deployments
  • a pod managed by a DaemonSet is not evicted
  • a pod without an ownerReference is directly evicted

Getting Started

Configuration

Configuration is done via environment variables.

Required configuration

export EXOSCALE_API_KEY=EXO...
export EXOSCALE_API_SECRET=cIzK1...
export EXOSCALE_API_ZONE=at-vie-1
export KUBECONFIG=/path/to/kubeconfig.yaml

export EXOSCALE_SKS_LIFECYCLER_DESIRED_K8S_VERSION=v1.28.7
export EXOSCALE_SKS_LIFECYCLER_SKS_CLUSTER_ID=905ff...

Optional configuration

EXOSCALE_SKS_LIFECYCLER_EVICT_NODES_LABELSELECTOR lets you define nodes, which you want to evict from the cluster, via a labelSelector.

export EXOSCALE_SKS_LIFECYCLER_EVICT_NODES_LABELSELECTOR="node.kubernetes.io/instance-type=cpu.extra-large,key2=val2"

Run

The program loops over all nodes in the cluster, and then exits!

Run directly via go from the project's root directory:

go run main.go nodepool cycle

Or use the container image ghcr.io/whizus/exoscale-sks-lifecycler.

Development

The application is written in Go and uses spf13/cobra and spf13/viper libraries to provide CLI functionality and simple configuration.

To get started with development, see CONTRIBUTING.md.