Pod disruption budgets: make node drains measurable

Operational decision

A document API runs four replicas and needs three ready during one-node maintenance. Set minAvailable to three, inspect currentHealthy and disruptionsAllowed, and verify replacement Pods can schedule before draining. The manifest is a starting policy for a single namespace. Check that the Deployment template carries the same app label and that the replicas span nodes. If the budget blocks a drain, diagnose unavailable replicas, capacity, and readiness first. Using a forceful deletion to finish a maintenance window bypasses the availability contract. A PDB also does not guarantee zero user errors; validate the client path while a replica leaves service.

apiVersion: policy/v1
kind: PodDisruptionBudget
metadata: {name: document-api-budget, namespace: documents}
spec:
  minAvailable: 3
  selector:
    matchLabels: {app: document-api}

Cost and verification

A strict budget can delay node replacement indefinitely when a workload is already unhealthy or spare capacity is missing. Four replicas cost more than one, but the extra instance permits a controlled eviction while maintaining the three-instance floor. Track drain duration and rejected evictions alongside availability; a successful policy is one operators can exercise without bypassing it. Budget health depends on readiness probes, so a false positive readiness check can make the protection misleading.

Common Mistakes

• Do not assume a PDB protects against node failure.
• Do not set a three-Pod minimum on three replicas and expect an immediate drain.
• Do not bypass the Eviction API to hide an unhealthy workload.

Connected lessons

• DevOps: delivery, infrastructure, and reliable operations
• Kubernetes Deployment: rolling update capacity
• Kubernetes probes: startup, readiness, and liveness
• Capacity and load tests: identify the next bottleneck

Advanced follow-up

• Cluster upgrade drill: preserve a path through each version step

Advanced follow-up

• Pod priority: reserve recovery capacity without evicting the wrong work

Advanced follow-up

• Node consolidation: calculate the capacity needed to evict safely