Cluster Autoscaler
Kubernetes component that adds/removes nodes from the cluster based on pod scheduling pressure. If pending pods can't be scheduled due to resource shortage → start a new node. If a node is <50% utilized → drain + terminate.
Integrates with cloud-provider auto-scaling groups: AWS ASG, GCP MIG, Azure VMSS. Pod priority + PodDisruptionBudget are respected on scale-down. Classic gotcha: DaemonSets keep the node alive (Cluster Autoscaler sees them as "can't be moved") → nodes don't scale down. cluster-autoscaler.kubernetes.io/safe-to-evict: "true" annotation needed for logging DaemonSets. Modern alternative: Karpenter (AWS original, now multi-cloud) thinks more aggressively — provisions exact-size nodes per pending workload instead of picking from fixed node groups.