Before patching or rebooting a node, move its work elsewhere. kubectl cordon stops new Pods; kubectl drain cordons and then evicts every Pod through the eviction API, so the controllers recreate them on other nodes.
Drain stops rather than lose something by accident. DaemonSet Pods would come straight back, so it needs --ignore-daemonsets. Pods with no controller would vanish, so it needs --force. Pods with emptyDir data would lose it, so it needs --delete-emptydir-data. Read each error; each names a decision.
Evictions respect PodDisruptionBudgets: if evicting a Pod would take a set below its minAvailable, drain waits. When a budget allows no disruptions at all, add capacity first rather than deleting the budget. After maintenance, kubectl uncordon returns the node; Pods don't move back by themselves, but new ones can land there.
Empty a node, check budgets, return it.