When there's no hope left for your Kubernetes nodes, Skuttle will prune them from the Kubernetes API.
Skuttle watches for nodes entering a NotReady state.
If a node has been NotReady for some time, Skuttle will use the node's ProviderID to query the cloud provider and check if it's still available.
Skuttle will only delete a node if the cloud provider reports it as terminated or missing.
Usage of skuttle:
-dry-run
dry run mode to only log instead of scheduling deletion
-kubeconfig string
path to kubeconfig file if not running in-cluster
-log-level string
log level (debug, info, warn, error) (default "info")
-node-selector string
selector used to filter nodes skuttle should manage (default "node.kubernetes.io/node")
-not-ready-duration duration
time duration to tolerate NotReady nodes (default 10m0s)
-providers string
comma-separated list of enabled providers
-refresh-duration duration
refresh duration (default 10s)
Skuttle supports multiple cloud providers at a time, specified with the -providers flag.
The node's ProviderID is expected to be in the format <prefix>://....
<prefix> is used to determine which of the specified cloud providers to query.
The aws provider will handle nodes with a provider ID aws://<region>/<instance ID>.
IAM credentials with permissions to query the existence and state of EC2 instances will need to be available.