Back to archive

A PodDisruptionBudget Cannot Save You From a Dead Node

Planned disruption and unplanned failure are different beasts. Checkout needs both PDBs and topology spread.

Production idea

PDBs, topology spread constraints, node failure, zones, and replica placement.

kubectl get pdbkubectl describe pdb checkoutkubectl get pods -o wide
  • Use PDBs for voluntary disruption.
  • Spread replicas across zones.
  • Keep enough replicas to survive one zone loss.
  • Test node drains before production maintenance.