Cannot evict pod: disruption budget
The Eviction API refused because evicting this pod would leave fewer pods available than its PodDisruptionBudget allows. kubectl drain retries forever (or until --timeout).
Diagnose in order
Section titled “Diagnose in order”- Find the budget and how many disruptions it allows right now:
kubectl get pdb -n shop. Look atALLOWED DISRUPTIONS. Zero means nothing can be evicted. - Find out why it’s zero. Compare
minAvailable(ormaxUnavailable) with the replica count, and check whether the other replicas are actually Ready.
Usual causes
Section titled “Usual causes”| Situation | Why disruptions allowed is zero |
|---|---|
One replica, minAvailable: 1 |
Evicting the only pod would always violate it |
minAvailable equal to the replica count |
Same, at any scale |
| Other replicas not Ready | The budget is already used up by unhealthy pods |
| All replicas on the node being drained | Evicting one leaves the rest; a second drain step blocks |
Run at least two replicas spread across nodes, and express the budget as maxUnavailable: 1 rather than a minAvailable that equals the replica count. If the other replicas are unhealthy, fix them first: the PDB is doing its job.
Bypassing it (kubectl drain --disable-eviction or deleting the pod directly) causes exactly the outage the budget exists to prevent, so only do it with the service owner’s agreement.
Read next: Upgrades.