karpenter

Karpenter is a Kubernetes node autoscaler designed for flexibility, performance, and simplicity. It improves the efficiency and cost of running workloads on Kubernetes clusters through a four-step loop: watching for pods the Kubernetes scheduler has marked unschedulable, evaluating the scheduling constraints those pods request (resource requests, node selectors, affinities, tolerations, and topology spread constraints), provisioning nodes that meet the pods' requirements, and removing nodes once they are no longer needed. Karpenter is a multi-cloud project. The core repository is complemented by provider implementations maintained by various organizations, including AWS, Azure, AlibabaCloud, Bizfly Cloud, Clever Cloud, Cluster API, Exoscale, GCP, Hetzner, Huawei Cloud, IBM Cloud, Proxmox, Oracle Cloud Infrastructure (with both an Oracle-maintained and a Zoom-maintained provider), Akamai/Linode (Alpha), and UpCloud (community and kubekanvas-maintained variants). The project is developed in the open under the Kubernetes SIGs organization. Community engagement happens through the Kubernetes Slack, with dedicated channels for users (#karpenter) and contributors (#karpenter-dev). The project holds bi-weekly working group meetings and weekly issue triage meetings, with published Zoom links, a Google Calendar, and meeting notes. Contribution is welcomed via the issue tracker, which highlights good-first-issue and help-wanted labels, and a contributor guide is provided. Participation is governed by the Kubernetes Code of Conduct. A collection of recorded talks spanning 2021 through 2025 covers topics such as workload consolidation, autoscaling comparisons, and automating cluster updates.