Q29: Pod Disruption Budgets (PDB): Protecting high availability during voluntary disruptions and node drains. - DevOps & Kubernetes (50 Qs)

High Availability Under Maintenance

Cluster maintenance operations (node draining, Kubernetes version upgrades, spot instance terminations) are voluntary disruptions. Without a PodDisruptionBudget (PDB), an engineer or automated script running kubectl drain node-1 could terminate all replicas simultaneously, taking down production.

arduino

[ Cluster Admin / Script: kubectl drain node ]
                     |
                     v
+-------------------------------------------------------+
|              API Server Eviction Subresource          |
|                                                       |
|   Evaluates PodDisruptionBudget (PDB):                |
|   "At least 80% of payment-api pods MUST stay alive"  |
+-------------------------------------------------------+
          /                                   \
 [ Replicas Healthy: 5/5 ]           [ Replicas Healthy: 4/5 ]
 Eviction ALLOWED (4/5 = 80%)        Eviction BLOCKED / QUEUED!
 (Node drains safely)                (Waits for new Pod to be ready)

Production PDB Blueprint

yaml

apiVersion: policy/v1
kind: PodDisruptionBudget
metadata:
  name: payment-api-pdb
  namespace: production
spec:
  minAvailable: 80% # Or fixed number: minAvailable: 2
  selector:
    matchLabels:
      app: payment-api

Key PDB Rules

minAvailable vs maxUnavailable: Define either a percentage (80%) or an absolute integer (1).
If a drain request would violate the budget, the eviction request fails with HTTP 429 (Too Many Requests), and the drain controller retries until new replicas become ready on another node.