Kubernetes documentation - Taints and Tolerations

Fetched 2026-10-04. Page meta description is the canonical one-line definition: 'Node affinity is a property of Pods that attracts them to a set of nodes (either as a preference or a hard requirement). Taints are the opposite -- they allow a node to repel a set of pods. Tolerations are applied to pods. Tolerations allow pods with matching taints. Tolerations allow scheduling but don't guarantee scheduling: the scheduler also evaluates other parameters as part of its function.' Three taint effects NoSchedule / PreferNoSchedule / NoExecute; NoExecute is the one that evicts pods that do not tolerate it, which is the mechanism that turns a GPU node fault into a serving outage for a pod that cannot tolerate or re-place. The 'allow but don't guarantee' clause is the core scheduling fact for LLM serving: toleration is necessary but not sufficient, so a replica can stay Pending even though its taint is tolerated.