kubectl drain hangs: a PodDisruptionBudget blocks eviction
kubectl drain respects PodDisruptionBudgets, so a strict budget can stall a node upgrade. Find the budget, then fix replicas or the budget, not the drain.
What you see
kubectl drainnever finishes and keeps retrying evictions- A node upgrade or scale-down is stuck waiting on one pod
What it means
Drain evicts pods safely: it lets their containers terminate gracefully and respects the PodDisruptionBudgets you have set. If evicting a pod would drop its workload below the budget, the eviction is refused and drain waits.
A budget of minAvailable: 100%, or one equal to the replica count, can never be satisfied during a drain. That is a configuration problem, not a drain bug.
How to confirm the cause
- 1List budgets and how many disruptions they allow
kubectl get pdb -A
ALLOWED DISRUPTIONSof0is the blocker. - 2See the budget's detail
kubectl describe pdb PDB_NAME -n NAMESPACE
Check the selector, the minimum or maximum, and the current healthy count.
- 3Check the workload has enough healthy replicas
kubectl get deployment NAME -n NAMESPACE
A single-replica workload with a budget cannot be drained.
Causes and fixes
The budget is too strict for the replica count
- HOW TO TELL
ALLOWED DISRUPTIONSis 0 even though all pods are healthy.- ROOT CAUSE
- The budget requires every pod to stay available, so no eviction is ever allowed.
- FIX
- Run more replicas than the budget's minimum, or relax the budget, so at least one pod can be evicted at a time.
Pods are already unhealthy
- HOW TO TELL
- Some pods are not Ready, so the healthy count is at the minimum.
- ROOT CAUSE
- Evicting a healthy pod would drop below the budget while another is already down.
- FIX
- Fix the unhealthy pods. The docs recommend setting the
AlwaysAllowunhealthy pod eviction policy on budgets so misbehaving pods can be evicted during a drain.
DaemonSet pods block the drain
- HOW TO TELL
- Drain stops with a message about DaemonSet-managed pods.
- ROOT CAUSE
- DaemonSet pods are ignored by design and cannot be evicted normally.
- FIX
- Pass
--ignore-daemonsets, as the docs show.kubectl drain --ignore-daemonsets NODE_NAME
How to stop it coming back
- Set budgets as
maxUnavailable: 1for most workloads, not 100% available. - Run at least two replicas of anything with a budget.
- Test a drain in staging before a maintenance window.
Keep going
- Runbook: Node NotReady: the kubelet stopped reporting
- Runbook: Pod stuck in Terminating
- Cheat sheet: kubectl
- Lesson: the kubernetes topic in Learn
Sources
Kubernetes documentation (kubernetes.io), read 18 Nov 2026. The docs version selector listed v1.33 to v1.37.