This is a Kubernetes controller that we developed to tackle broken nodes in our Cluster, which we discovered to be a problem after our migration from Spot Ocean to Karpenter (version 1.0.0). We offered this implementation as a contribution to Karpenter in this PR which was later on closed in favour of their own, more cloud provider centric, implementation.
It's a very simple controller that watches for the taint node.kubernetes.io/unreachable and if the taint has been there for more than 10 minutes it
deletes the Nodeclaim, which will lead to the termination of the node.