kubectl drain - Drain node in preparation for maintenance

Options you will use

--ignore-daemonsets

DaemonSet pods would be recreated on the node anyway: skip them

--delete-emptydir-data

evict pods with emptyDir volumes (their data is lost)

--force

also delete pods with no controller (they are gone for good)

--timeout=D

give up after D (default: wait forever)

--disable-eviction

DELETE instead of evict: bypasses PodDisruptionBudgets. Last resort.

Examples

$ kubectl drain worker-1 --ignore-daemonsets --delete-emptydir-data

the standard maintenance drain

$ kubectl uncordon worker-1

allow scheduling again afterwards

Gotchas

• drain = cordon + evict every pod through the Eviction API. The Eviction API honours PodDisruptionBudgets: "Cannot evict pod as it would violate the pod's disruption budget." and drain retries every 5s - forever, unless --timeout is set.

Taught in

• 17.26 PodDisruptionBudgets and why they matter during a drain
• 18.16 The kubeadm upgrade, step by step
• 18.25 Node lifecycle: cordon, drain, NotReady and taint-based eviction